A tilt sensor drift elimination method based on bias tracking algorithm

Abstract

The present application belongs to the technical field of vehicle-mounted communication antenna servo control, and particularly relates to a tilt angle sensor drift elimination method based on a bias tracking algorithm, comprising the following steps: synchronously collecting original tilt angle readings and chip temperature readings of the tilt angle sensor at a fixed collection period; calculating a short-time standard deviation of the original tilt angle readings in a sliding window, determining as stationary when the short-time standard deviation is lower than a stationary threshold, determining as disturbance when the short-time standard deviation is not lower than the stationary threshold, obtaining a stationary state flag, and taking the square of the short-time standard deviation as an observation noise variance; calculating a temperature rise rate according to the chip temperature readings, and outputting a high value as a bias process variance when the stationary state flag is stationary and the temperature rise rate is higher than a thermal equilibrium threshold. The present application solves the problem of transient bias drift of the tilt angle sensor caused by chip temperature rise, so that the output error of the corrected tilt angle value continuously meets the leveling accuracy requirement in the task period.

Description

Technical Field

[0001] This invention belongs to the field of vehicle-mounted communication antenna servo control technology, specifically relating to a tilt sensor drift elimination method based on offset tracking algorithm. Background Technology

[0002] Vehicle-mounted communication antennas are widely used in scenarios such as field mobile operations, emergency rescue, and mobile command. The antenna servo control unit relies on tilt sensors to obtain the vehicle's attitude angle in real time, driving the leveling mechanism to maintain the antenna platform at a horizontal reference position, thereby ensuring that the antenna beam pointing accuracy meets the link establishment requirements. During continuous mission periods of 4 to 8 hours, the antenna servo control unit needs the tilt sensor to continuously output attitude angle data with errors not exceeding the leveling accuracy requirements to support real-time leveling control of the antenna under conditions such as vehicle movement, parking, and terrain undulations.

[0003] To suppress the output error of tilt sensors, existing technologies typically employ a factory-grade temperature compensation calibration method: before the sensor leaves the factory, the zero-point offset of the sensor is measured at multiple temperature nodes in a temperature-controlled environment chamber, and a calibration lookup table of zero-point offset as temperature changes is established. When the device is in use, the chip temperature reading is read, and the original tilt angle reading is statically compensated after interpolation from the table to eliminate the zero-point offset error at steady-state temperature.

[0004] However, the aforementioned factory-calibrated temperature compensation method only calibrates the steady-state bias of the sensor after it reaches thermal equilibrium at each temperature node. It cannot describe the transient characteristics of the bias drift over time as the chip temperature dynamically rises from ambient temperature to thermal equilibrium temperature after the sensor is powered on. In actual use of automotive antennas, the temperature rise process after the sensor is powered on lasts from several minutes to tens of minutes. During this period, the bias drift can reach tens of millimeter levels, far exceeding the leveling accuracy requirements. The factory calibration lookup table has no ability to compensate for the transient bias during this stage. Furthermore, the factory calibration results are fixed in the equipment parameters and cannot be adaptively corrected according to equipment aging, seasonal changes in ambient temperature, and individual differences, leading to a continuous decline in compensation accuracy after long-term use. Summary of the Invention

[0005] This invention provides a tilt sensor drift elimination method based on a bias tracking algorithm to solve the technical problems of inaccurate temperature drift compensation, easy interference from disturbances, and slow convergence upon power-up in existing tilt sensors.

[0006] This invention provides a tilt sensor drift elimination method based on a bias tracking algorithm, comprising the following steps: The original tilt angle readings and chip temperature readings of the tilt sensor are collected synchronously at a fixed acquisition cycle; The short-time standard deviation of the original tilt angle reading is calculated within a sliding window. If the short-time standard deviation is lower than the static threshold, it is judged as static; if it is not lower than the static threshold, it is judged as disturbance. The static state indicator is obtained, and the square of the short-time standard deviation is used as the observation noise variance. The temperature rise rate is calculated based on the chip temperature reading. When the static state is marked as static and the temperature rise rate is higher than the thermal equilibrium threshold, a high value is output as the bias process variance. When the static state is marked as static and the temperature rise rate is not higher than the thermal equilibrium threshold, a low value is output as the bias process variance. When the static state is marked as disturbance, the bias process variance remains unchanged. Using the original tilt angle reading as the observation, and the bias process variance and observation noise variance as noise parameters, the prediction step and update step of the augmented Kalman filter are run to output the true tilt angle estimate and bias estimate. The corrected tilt angle value is obtained by subtracting the bias estimate from the true tilt angle estimate. When the static state is marked as static, the historical drift rate is updated by dividing the bias estimate by the power-on duration. At the next power-on, the product of the historical drift rate and the warm-up duration is injected into the augmented Kalman filter as the initial bias value.

[0007] Its effects are as follows: This method integrates the determination of vehicle motion state and sensor thermal state into the noise parameter adaptive adjustment framework of augmented Kalman filter, so that the observation noise variance and bias process variance match the current actual working conditions in each acquisition cycle. When the vehicle body is disturbed, the observation weight is automatically reduced and the bias process variance is frozen. When the temperature rises rapidly, the bias is kept to track quickly. When the temperature is at thermal equilibrium, the bias estimation jitter is compressed. This achieves continuous and reliable estimation of bias drift under all working conditions. After correction, the output error of the tilt angle value is kept within the leveling accuracy requirement range during the 4 to 8 hour mission period.

[0008] Furthermore, the short-time standard deviation is calculated as follows: In each acquisition cycle, the current original tilt angle reading is stored in a sliding window of a preset length. The length of the sliding window is not less than the vehicle vibration cycle. After the sliding window is full, the average value of all original tilt angle readings in the window is calculated to obtain the window mean. The square of the difference between each original tilt angle reading and the window mean is calculated. The average of all squared values ​​is taken and then the square root is taken to obtain the short-time standard deviation.

[0009] Furthermore, the temperature rise rate is calculated as follows: in each acquisition cycle, the current chip temperature reading is stored in a temperature window of a preset length. After the temperature window is full, the average value of the readings in the second half of the temperature window is subtracted from the average value of the readings in the first half, and then divided by the time interval between the center of the two segments to obtain the temperature rise rate.

[0010] Its effect is as follows: the temperature rise rate is calculated by dividing the difference in chip temperature readings within the time period corresponding to the temperature window length by the corresponding time interval. Compared with the difference between two adjacent steps, the effective calculation time interval is extended from 50 milliseconds to several seconds, reducing the impact of single-step measurement noise on the temperature rise rate estimation to a negligible level. This ensures that the thermal equilibrium threshold judgment result has a stable ability to distinguish between the rapid temperature rise stage and the thermal equilibrium stage, and avoids frequent erroneous switching of the bias process variance between high and low values ​​due to the jitter of the temperature rise rate estimation.

[0011] Furthermore, the high-end value is greater than the low-end value, and the ratio of the high-end value to the low-end value is not less than the preset value.

[0012] Its effects are as follows: the ratio of high-level value to low-level value is maintained at no less than 100 times, the ratio of the offset estimation update step size of the rapid temperature rise stage to the thermal equilibrium stage is no less than 10 times, the offset drift is effectively tracked in the rapid temperature rise stage, and the jitter amplitude of the offset estimation is compressed to less than one-tenth of that in the rapid temperature rise stage in the thermal equilibrium stage, so that the output stability of the corrected tilt angle value in the thermal equilibrium stage is significantly better than that in the rapid temperature rise stage.

[0013] Furthermore, the state vector of the augmented Kalman filter is a two-dimensional vector, with the first component being the true tilt component and the second component being the bias component; the state transition matrix is ​​a second-order identity matrix, indicating that the true tilt component and the bias component remain approximately unchanged in adjacent acquisition cycles; the observation matrix is ​​a row vector with both elements being 1, indicating that the original tilt reading is equal to the sum of the true tilt component and the bias component plus the observation noise.

[0014] Furthermore, the prediction update includes a prediction step and an update step. In the prediction step, the prior state vector is obtained by multiplying the state transition matrix by the posterior state vector of the previous period. The prior covariance matrix is ​​calculated using the state transition matrix, the posterior covariance matrix of the previous period, and the process noise covariance matrix. The process noise covariance matrix is ​​a diagonal matrix, with its first diagonal element being the fixed true tilt process variance and its second diagonal element being the bias process variance. In the update step, the innovation is obtained by subtracting the product of the observation matrix and the prior state vector from the original tilt reading. The Kalman gain vector is calculated using the prior covariance matrix, the observation matrix, and the observation noise variance. The posterior state vector is obtained by adding the product of the Kalman gain vector and the innovation to the prior state vector. The second component of the posterior state vector is the posterior bias estimate.

[0015] Furthermore, the augmented Kalman filter is initialized as follows: upon power-up, the initial value of the true tilt component of the state vector is set to the original tilt reading acquired for the first time, and the initial value of the bias component is set to 0; the initial covariance matrix is ​​a diagonal matrix, the initial variance corresponding to the true tilt component is determined based on the system's dynamic measurement accuracy index, the initial variance corresponding to the bias component is determined based on the sensor's factory zero-point deviation range, and the initial variance corresponding to the bias component is greater than the initial variance corresponding to the true tilt component.

[0016] Its effect is that by setting the initial variance corresponding to the bias component to be greater than the initial variance corresponding to the true tilt component, the augmented Kalman filter updates the bias component by a larger step size than the true tilt component in the initial stage of power-on. This causes the bias estimate to converge quickly from zero to the true bias level in the first few acquisition cycles, reducing the output error of the tilt value after correction in the initial stage of power-on and shortening the time required for the filter to reach stable output accuracy.

[0017] Furthermore, it also includes a historical drift rate learning step: when the static state is marked as static and the temperature rise rate is not higher than the thermal equilibrium threshold, the difference between the posterior bias estimates of two adjacent steps is divided by a fixed acquisition period to obtain the current step bias drift rate. The average of the bias drift rates for a consecutive preset number of steps is taken to obtain the current drift rate average. The current drift rate average is appended to the historical drift rate list. The historical drift rate list retains the most recent preset number of power-on records, and deletes the earliest record when the limit is exceeded. The average of all records in the historical drift rate list is taken as the basis for setting the low value for the next power-on.

[0018] Its effect is as follows: by using the moving average method to persistently store the historical drift rate across power-on cycles, the prediction basis of the initial bias value comes from the weighted average result of multiple power-on processes, rather than the measured value of a single power-on process. This effectively suppresses the impact of abnormal historical drift rate values ​​caused by the decline in bias estimation quality due to long-term disturbance of the vehicle body during a single power-on process on subsequent power-on processes. As a result, the prediction accuracy of the initial bias value gradually converges with the increase of the cumulative number of power-on cycles of the device, thereby continuously shortening the convergence time of the filter after each power-on.

[0019] Furthermore, before executing the historical drift rate learning step, a dual-axis consistency verification step is also included: the current offset ratio is obtained by taking the ratio of the current pitch axis posterior offset estimate to the roll axis posterior offset estimate; if the historical offset ratio has not been initialized, the average of the current offset ratios over a consecutive preset number of steps is used as the historical offset ratio and written into the memory, and the historical drift rate learning step is executed; if the historical offset ratio has been initialized, the absolute value of the difference between the current offset ratio and the historical offset ratio is calculated. If the absolute value is not greater than the ratio tolerance threshold, the historical drift rate learning step is executed; if the absolute value is greater than the ratio tolerance threshold, the historical drift rate learning step is skipped.

[0020] Furthermore, the historical bias ratio is updated as follows: only within the acquisition period where the absolute value is not greater than the ratio tolerance threshold and the historical drift rate learning step has been completed, the historical bias ratio is replaced by the average of the current bias ratios that have passed the dual-axis consistency verification step for the most recent consecutive preset number of steps.

[0021] The beneficial effects are as follows: This invention addresses the problem of zero-point bias drift caused by chip temperature rise in tilt sensors in vehicle-mounted antenna servo systems, leading to a decrease in pointing accuracy after prolonged antenna operation. It uses a sliding window to sense the vehicle's motion state in real time via short-time standard deviation and the temperature rise rate to sense the sensor's thermal state in real time. These two state information are mapped to the observation noise variance and bias process variance of an augmented Kalman filter, respectively, driving the filter to adaptively adjust the bias estimation strategy under different operating conditions, achieving real-time estimation and elimination of bias drift. The historical drift rate learning mechanism across power-on cycles gradually improves the prediction accuracy of the initial bias value as the equipment is used, shortening the filter's convergence time after each power-on. The entire method runs at the servo controller software layer without adding any hardware, continuously outputting corrected tilt angle values ​​that meet the leveling accuracy requirements throughout the entire task period. Attached Figure Description

[0022] Figure 1 This is a flowchart of a tilt sensor drift elimination method based on an offset tracking algorithm.

[0023] Figure 2 This is a schematic diagram of the overall system architecture for a tilt sensor drift elimination method based on the bias tracking algorithm.

[0024] Figure 3 A comparison curve showing the change of tilt angle output error with power-on time for different methods.

[0025] Figure 4 The graph shows the degradation curves of the bias estimation error under different degrees of vehicle body disturbance. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] After a vehicle-mounted antenna servo system is deployed in a field location, the antenna needs to be continuously aligned with the target for an extended period, typically 4 to 8 hours. As the core sensing element for antenna leveling, the tilt sensor's output tilt angle readings experience zero-point offset drift due to internal chip temperature rise during prolonged operation, causing the antenna to gradually deviate from the target. Taking a 2.4-meter vehicle-mounted C-band scattering antenna as an example, after the vehicle completes deployment and comes to a stop, the servo controller powers on, and the antenna enters continuous alignment and tracking mode. The zero-point offset drift caused by internal sensor temperature rise typically exceeds the leveling accuracy requirements 2 to 3 hours after power-on. This method runs in the servo controller's software layer without adding any hardware. It executes continuously with a fixed 50-millisecond acquisition cycle, estimating and eliminating offset drift in real time, and outputting the corrected tilt angle value to the servo controller, ensuring the antenna's pointing accuracy throughout the entire mission. The processing flow for the roll and pitch axes is identical; the following explanation uses the pitch axis as an example.

[0028] An embodiment of the tilt sensor drift elimination method based on the bias tracking algorithm provided by this invention: like Figure 1 As shown, the tilt sensor drift elimination method based on the offset tracking algorithm includes the following steps: S101 synchronously acquires the raw tilt angle readings and chip temperature readings of the tilt sensor at a fixed acquisition cycle.

[0029] In one embodiment, after the servo controller is powered on, it synchronously reads two streams of raw data via the sensor serial port at a fixed acquisition period of 50 milliseconds: the raw tilt angle reading of the tilt sensor's pitch axis and the chip temperature reading of the sensor's built-in thermometer. The two data streams are acquired synchronously within the same acquisition period, ensuring a strict temporal correspondence between the chip temperature reading and the raw tilt angle reading.

[0030] The raw tilt angle reading is a measurement directly output by the sensor, containing two superimposed components: the true attitude angle of the antenna mounting surface relative to the horizontal plane, and the zero-point offset drift caused by temperature rise. Since these two components cannot be directly separated in the raw reading, they need to be estimated and separated through subsequent filtering steps. The chip temperature reading reflects the internal thermal state of the sensor; the rate of temperature rise directly determines the rate of offset drift, therefore, the chip temperature reading is the primary basis for subsequent temperature rise rate calculations.

[0031] The fixed acquisition period is 50 milliseconds, corresponding to 20 samples per second. The determination of this sampling frequency is based on the following: First, the typical frequency of engine idling vibration is 2 to 10 Hz, and the 50-millisecond sampling period corresponds to a 20 Hz sampling frequency, which satisfies the Nyquist sampling theorem and can completely capture the vehicle vibration signal; Second, the offset drift is a slow change process on the order of minutes, and the 50-millisecond acquisition interval is much smaller than the drift time scale, so it will not cause the omission of drift information.

[0032] To ensure that the augmented Kalman filter can immediately enter an effective working state after power-on, an initialization is performed upon power-on, which includes the following four parts.

[0033] First, the initial values ​​of the state vector are set. The state vector is a two-dimensional vector, with the first component being the true tilt angle component and the second component being the bias component. Upon power-up, the initial value of the true tilt angle component is set to the initial tilt angle reading acquired for the first time. This means that at the moment of power-up, the bias drift has not yet accumulated, and the initial tilt angle reading is approximately equal to the true attitude angle, making it reasonable to use this as the starting point estimate for the true tilt angle component. The initial value of the bias component is set to 0. This means that at the moment of power-up, the bias drift has not yet been generated, and 0 is used as the initial bias estimate. The subsequent filter will gradually correct this estimate as the temperature rises.

[0034] Second, the initial covariance matrix is ​​set as a diagonal matrix, reflecting the uncertainty of the estimated values ​​of the two components at the initial moment. The initial variance corresponding to the true tilt component is 0.01 square degrees, corresponding to an initial uncertainty of 0.1 degrees, determined based on the system's dynamic measurement accuracy index, consistent with the sensor's nominal measurement accuracy order of magnitude. The initial variance corresponding to the bias component is 0.04 square degrees, corresponding to an initial bias uncertainty of 0.2 degrees, determined based on the sensor's factory zero-point deviation range, and can be corrected using data-driven corrections by referring to the zero-point deviation index in the sensor datasheet. The initial variance corresponding to the bias component is greater than that corresponding to the true tilt component, reflecting that the uncertainty of the bias at the initial moment of power-on is higher than the uncertainty of the true attitude angle, prompting the filter to maintain a higher update sensitivity for the bias estimate in the initial stage, thus accelerating the bias convergence speed.

[0035] Third, the process noise covariance matrix is ​​set. The process noise covariance matrix is ​​a diagonal matrix, with its first diagonal element being the true tilt angle process variance, which is set to a fixed value of 1×10⁻⁶. -8 Square degree reflects the fact that the true attitude angle of the vehicle when stationary remains approximately constant within adjacent acquisition periods, with minimal variation; its second diagonal element is the variance of the bias process, which is initially set to a high value of 1×10 at power-on. -4 The square degree corresponds to the operating condition where the bias drift rate is relatively fast during the cold start stage. Subsequently, S103 will adaptively adjust the bias drift rate based on the temperature rise rate and the static state flag.

[0036] Fourth, the initial value of the observation noise variance is set. The initial value of the observation noise variance is taken from the nominal measurement noise variance specified in the sensor datasheet, with a typical value of 1×10⁻⁶. -4 Squared degrees. This initial value is only used in the transition phase before the sliding window is filled. After the sliding window is filled, the observation noise variance will be calculated and replaced in real time by S102 based on the short-time standard deviation.

[0037] After initialization, the augmented Kalman filter has all the initial conditions to perform prediction update iterations. In each subsequent acquisition cycle, the posterior state vector and posterior covariance matrix of the previous cycle are used as the input for the prediction step of the current cycle, forming a continuous iterative data stream.

[0038] The above initialization method corresponds to the following subordinate scheme: The initialization method of the augmented Kalman filter is as follows: When powered on, the initial value of the true tilt component of the state vector is set to the original tilt reading of the first acquisition, and the initial value of the bias component is set to 0; the initial covariance matrix is ​​taken as a diagonal matrix, the initial variance corresponding to the true tilt component is determined according to the dynamic measurement accuracy index of the system, the initial variance corresponding to the bias component is determined according to the zero-point deviation range of the sensor at the factory, and the initial variance corresponding to the bias component is greater than the initial variance corresponding to the true tilt component.

[0039] Specifically, taking a 2.4-meter vehicle-mounted C-band scattering antenna as an example, the sensor datasheet indicates a measurement accuracy of 0.1 degrees, corresponding to an initial variance of 0.01 square degrees for the true tilt component; the factory zero-point deviation range is ±0.2 degrees, corresponding to an initial variance of 0.04 square degrees for the bias component. The ratio of these two is 4, meaning the initial variance of the bias component is significantly larger than that of the true tilt component. Therefore, the filter updates the bias component with a larger step size during the initial power-on phase, quickly bringing the bias estimate from 0 to the true bias value within the first few acquisition cycles, thus avoiding large deviations in the tilt value after initial correction. If the sensor model is changed, the two initial variances can be directly re-determined based on the measurement accuracy and zero-point deviation indicators in the new sensor datasheet, without modifying other parameters.

[0040] S101 establishes a synchronous acquisition mechanism for the original tilt angle reading and chip temperature reading each time it is powered on, as well as sets all the initial parameters of the augmented Kalman filter. This provides the data source and filter start state for the subsequent continuous iterative processing from S102 to S105, and establishes the data flow starting point of the entire compensation method.

[0041] S102, calculate the short-time standard deviation of the original tilt angle reading within the sliding window. If the short-time standard deviation is lower than the static threshold, it is judged as static; if it is not lower than the static threshold, it is judged as disturbance, thus obtaining the static state indicator. The square of the short-time standard deviation is used as the observation noise variance.

[0042] In one embodiment, within each acquisition cycle, the current raw tilt angle reading is stored in a sliding window of a preset length. The sliding window serves to retain a sequence of raw tilt angle readings over a recent period to extract statistical features reflecting the current motion state of the vehicle. The sliding window is maintained using a first-in, first-out (FIFO) approach: a new raw tilt angle reading is written in each acquisition cycle, while the oldest reading is discarded, and the window always retains the readings within the most recent preset length of steps.

[0043] The calculation of short-time standard deviation is the core means of determining whether the vehicle body is stationary. In order to ensure that the calculation results can reliably distinguish between stationary and disturbed states, the length of the sliding window must meet the following constraints: the window duration is not less than the typical period of vehicle body vibration, so as to ensure that the window can completely contain the data of at least one vibration period, so that the standard deviation under the disturbed state can fully reflect the vibration amplitude, rather than just capturing a local segment of the vibration waveform.

[0044] The following subordinate scheme further defines the specific calculation method of short-time standard deviation: In each acquisition cycle, the current raw tilt angle reading is stored in a sliding window of a preset length. The length of the sliding window is not less than the vehicle vibration cycle. After the sliding window is full, the average value of all raw tilt angle readings in the window is calculated to obtain the window mean. The square of the difference between each raw tilt angle reading and the window mean is calculated. The average value of all squares is taken and then the square root is taken to obtain the short-time standard deviation.

[0045] The above calculation process uses the standard population standard deviation formula, which is an existing technical formula in statistics, and its specific form is as follows:

[0046] In the formula, σ is the short-time standard deviation in degrees; N is the sliding window length, which is the total number of steps for the original tilt angle readings within the window. This is the initial tilt angle reading at step i within the window, in degrees; This is the average of all original tilt angle readings within the window, i.e., the window mean, in degrees.

[0047] The physical meaning of this formula is: to calculate the dispersion of all raw tilt angle readings within a window relative to the window mean. When the vehicle is stationary, the true attitude angle is uniquely determined by the ground slope and remains constant. The random fluctuations in the sensor output only come from measurement noise, and the readings are closely distributed around the mean, with a low standard deviation. When the vehicle is disturbed, the true attitude angle changes with the vehicle's movement, the dispersion of the readings increases significantly, and the standard deviation rises accordingly. The difference in the magnitude of the standard deviation between the two states is significant, demonstrating reliable distinguishing ability.

[0048] For example, a fixed acquisition period of 50 milliseconds and a sliding window length of 20 steps, corresponding to a duration of 1 second, are used. The typical period of engine idling vibration is 0.1 to 0.5 seconds. The 1-second window duration is not less than the upper limit of the above vibration period, which meets the requirement that the window length is not less than the vehicle vibration period. This ensures that the window can completely cover the entire idling vibration period and that the standard deviation calculation results under disturbance conditions stably reflect the vibration amplitude. If the engine vibration period of the actual deployed vehicle exceeds 0.5 seconds, the sliding window length can be increased accordingly, with twice the measured vibration period corresponding to the number of steps as the data-driven setting value.

[0049] After the sliding window has moved 20 steps, the short-time standard deviation is calculated according to the above formula. It is then compared with the stationary threshold, and a stationary state flag is output: when the short-time standard deviation is lower than the stationary threshold, the stationary state flag is output as stationary, indicating that the vehicle is currently stationary; when the short-time standard deviation is not lower than the stationary threshold, the stationary state flag is output as disturbance, indicating that there is an external disturbance to the vehicle.

[0050] The method for setting the stationary threshold is as follows: A data-driven approach is preferred. During the initial deployment of the device, 100 consecutive steps of raw tilt angle readings are collected under a known stationary state. The measured standard deviation of these 100 steps is calculated, and twice this standard deviation is used as the stationary threshold and written into the parameter configuration. Taking a 2.4-meter vehicle-mounted C-band scattering antenna and its associated sensor as an example, the sensor resolution is 0.01 degrees, and the measured standard deviation under stationary conditions is approximately 0.0025 degrees. Twice this deviation yields a stationary threshold of 0.005 degrees. This value is half the sensor resolution, ensuring that normal measurement noise does not trigger disturbance detection, while maintaining sufficient sensitivity to minor disturbances to the vehicle body. If initial deployment testing is not possible, half the sensor resolution can be directly used as a conservative default value for the stationary threshold.

[0051] Understandably, the size of the stationary threshold directly determines the sensitivity of the stationary state flag, thus affecting the timing of bias process variance adjustment in S103 and the triggering condition for historical drift rate learning in S105. If the stationary threshold is too large, slight vehicle disturbances may still be misjudged as stationary, causing the bias estimate to continuously update during disturbances and introducing false drift. If the stationary threshold is too small, normal sensor measurement noise may be misjudged as disturbance, causing the bias process variance to freeze frequently and reducing the continuity of bias tracking. A stationary threshold at half the sensor resolution strikes a balance between these two misjudgment risks, ensuring the stationary state flag has a stable accuracy rate under actual operating conditions.

[0052] While outputting the stationary state flag, the square of the current short-time standard deviation is used as the observation noise variance at the current moment and directly passed to the augmented Kalman filter update step in S104. The square of the short-time standard deviation is mathematically equivalent to the sample variance of the original tilt angle readings within the window. Its dimension is square degrees, which is consistent with the dimension of the observation noise variance in the Kalman filter framework, and can be directly substituted into the Kalman gain calculation of the update step.

[0053] Understandably, using the square of the short-time standard deviation as the observation noise variance enables real-time adaptive adjustment of the observation noise variance according to the vehicle's motion state. When the vehicle is stationary, the short-time standard deviation is at the level of measurement noise, the observation noise variance is small, the Kalman gain is large, the filter assigns a high confidence weight to the current raw tilt angle reading, and the bias estimation update step size is large, enabling rapid tracking of bias drift. When the vehicle is disturbed, the short-time standard deviation increases with the vibration amplitude, the observation noise variance increases synchronously, the Kalman gain automatically decreases, the filter reduces the confidence weight of the current raw tilt angle reading, and reduces the contamination of the bias estimation during the disturbance period. This mechanism ensures that the value of the observation noise variance always matches the current actual observation quality without manual intervention.

[0054] S102 simultaneously outputs two intermediate quantities in each acquisition cycle: a stationary state flag and an observation noise variance. The former provides S103 with the basis for determining the timing of bias process variance adjustment, while the latter provides S104 with an observation noise variance that matches the current vehicle motion state in real time. This achieves one standard deviation calculation and reuse in two places. Without increasing the amount of additional computation, the subsequent filtering steps can adaptively adjust the estimation strategy according to the actual state of the vehicle, effectively suppressing the interference of disturbance periods on the bias estimation.

[0055] S103 calculates the temperature rise rate based on the chip temperature reading. When the static state is marked as static and the temperature rise rate is higher than the thermal equilibrium threshold, it outputs a high value as the bias process variance. When the static state is marked as static and the temperature rise rate is not higher than the thermal equilibrium threshold, it outputs a low value as the bias process variance. When the static state is marked as disturbed, it keeps the bias process variance unchanged.

[0056] In one embodiment, within each acquisition cycle, the current temperature rise rate is calculated based on the chip temperature reading acquired by S101, and combined with the static state flag output by S102, the current bias process variance is output according to three conditional branches.

[0057] The physical meaning of bias process variance is: in the state equation of an augmented Kalman filter, it describes the uncertainty of the change in the bias component between two adjacent acquisition periods. A larger bias process variance means the filter assumes a greater change in the bias component within the current acquisition period, resulting in a larger update step size for the bias estimate and faster tracking of bias drift. Conversely, a smaller bias process variance means the filter assumes a smaller change in the bias component within the current acquisition period, resulting in a smaller update step size for the bias estimate and a smoother, more stable bias estimate.

[0058] The temperature rise rate is calculated based on the following subordinate scheme: the difference between the chip temperature reading in the current acquisition cycle and the chip temperature reading before the previous temperature window length step is divided by the product of the temperature window length and the fixed acquisition cycle to obtain the temperature rise rate.

[0059] The temperature rise rate is calculated using a first-order difference estimation method based on the piecewise mean difference. The method is as follows: the difference between the current chip temperature reading and the chip temperature reading a certain number of steps before the temperature window is divided by the corresponding time interval to obtain the average temperature rise rate within that time interval. The specific form is as follows:

[0060] In the formula, The rate of temperature rise at the current moment, expressed in degrees Celsius per second; This is the chip temperature reading for the current acquisition cycle, in degrees Celsius. This is the chip temperature reading at the M-th step before the current acquisition cycle, in degrees Celsius; M is the temperature window length, in steps. To ensure a fixed acquisition period, a value of 0.05 seconds is used. This represents the time interval corresponding to the temperature window, in seconds.

[0061] Understandably, the use of a piecewise mean difference first-order difference estimation method to calculate the temperature rise rate, rather than simply taking the difference between two adjacent steps, is to suppress the interference of single-step measurement noise in the chip temperature readings on the temperature rise rate estimation. The time interval between two adjacent differences is only 50 milliseconds, and the single-step measurement noise of the chip temperature readings is on the order of approximately 0.01 degrees Celsius, making it difficult to accurately determine the thermal equilibrium state. Extending the time interval to the time period corresponding to the length of the temperature window is equivalent to performing a time averaging of the temperature rise rate, making the estimation result smooth and stable, and reliably distinguishing between the rapid temperature rise stage and the thermal equilibrium stage.

[0062] The temperature window length M is set as follows: The temperature window length is taken as 5% to 10% of the temperature rise timescale, corresponding to the number of steps. Taking a 2.4-meter vehicle-mounted C-band scattering antenna as an example, the typical temperature rise time from power-on to thermal equilibrium is approximately 20 minutes, corresponding to 24,000 steps. Taking 5% of this, approximately 1,200 steps, corresponds to a duration of 60 seconds. Considering computational resource constraints, the actual temperature window length is set to 40 steps, corresponding to a duration of 2 seconds, achieving a balance between smoothing effect and response speed. If the temperature rise timescale is shorter, the temperature window length can be shortened accordingly to 20 steps.

[0063] After the temperature rise rate is calculated, it is compared with the thermal equilibrium threshold. Combined with the static state flag, the bias process variance is output according to the following three condition branches.

[0064] The first conditional branch: The static state is marked as static, and the temperature rise rate is higher than the thermal equilibrium threshold, outputting a high value as the bias process variance. This condition corresponds to the sensor being in a rapid temperature rise phase, with a fast bias drift rate, requiring a large bias process variance to ensure the filter maintains its ability to quickly track the bias drift.

[0065] The second conditional branch: The static state is indicated by being stationary, and the temperature rise rate is not higher than the thermal equilibrium threshold. A low output value is used as the bias process variance. This condition corresponds to the sensor having entered the thermal equilibrium stage, where the bias drift rate is extremely slow. A small bias process variance is needed to keep the bias estimation smooth and stable, and to suppress the interference of measurement noise on the bias estimation.

[0066] The third conditional branch: The stationary state is marked as a disturbance, and the bias process variance remains unchanged, that is, the bias process variance output from the previous acquisition cycle is maintained. This condition corresponds to the presence of external disturbances to the vehicle body, which reduces the reliability of the original tilt angle readings. In this case, the bias process variance is frozen to avoid erroneous updates to the bias estimate caused by spurious observations during the disturbance period.

[0067] The following subordinate scheme further limits the ratio of high-end value to low-end value: the ratio of high-end value to low-end value shall not be less than the preset value.

[0068] Understandably, in this embodiment, the preset value is 100, and the ratio between the high-end value and the low-end value is maintained at no less than 100 times to ensure that the tracking speed of the filter for the bias component has a sufficiently significant difference between the two thermal states. During the rapid temperature rise phase, the change in bias drift within each acquisition cycle is approximately the square root of the high-end value; during the thermal equilibrium phase, the change in bias drift within each acquisition cycle is approximately the square root of the low-end value. The ratio of these two is equal to the square root of the ratio of the high-end value to the low-end value, i.e., no less than 10 times. This means that the bias tracking step size during the rapid temperature rise phase is at least 10 times that during the thermal equilibrium phase, enabling effective tracking of the bias drift during the rapid temperature rise phase, while simultaneously compressing the jitter amplitude of the bias estimation to less than one-tenth of that during the rapid temperature rise phase during the thermal equilibrium phase.

[0069] The specific methods for setting the high and low values ​​are as follows: During the initial deployment of the device, record the total amount and duration of the bias drift from power-on to thermal equilibrium. Divide the total bias drift by the total number of steps in the rapid temperature rise phase to obtain a typical value for the bias drift increment per step in the rapid temperature rise phase. Take the square of this typical value as the high value. For example, with a 2.4-meter vehicle-mounted C-band scattering antenna, the total bias drift from power-on to thermal equilibrium is approximately 0.3 degrees Celsius. The rapid temperature rise phase lasts approximately 600 steps, with an offset drift increment of approximately 0.0005 degrees Celsius per step. Taking the square of this increment yields a high value of approximately 2.5 × 10⁻⁶. -7 Square degree, rounded to 1×10 -6 Square degree. During the thermal equilibrium phase, the bias drift is extremely slow, with each step changing by approximately 0.00001 degrees. Squared, this yields a low-end value of approximately 1 × 10⁻⁶ degrees. -10 Square degree, rounded to 1×10 -10 The square degree, with a ratio of 10000, meets the requirement of not less than 100. If initial deployment testing is not feasible, a higher value of 1×10 can be directly used. -4 Squareness, low-grade value is 1×10 -6 As a conservative default value, the ratio of square degree to the two is 100, which meets the minimum ratio requirement.

[0070] The method for setting the thermal equilibrium threshold is as follows: During the initial deployment of the device, record the measured temperature rise rate when the sensor enters thermal equilibrium. Take 1.5 times this measured value as the thermal equilibrium threshold and write it into the parameter configuration. Taking a 2.4-meter vehicle-mounted C-band scattering antenna as an example, the measured temperature rise rate when the sensor enters thermal equilibrium is approximately 0.007 degrees Celsius per second. Taking 1.5 times this rate yields a thermal equilibrium threshold of approximately 0.01 degrees Celsius per second. If it is not possible to conduct an initial deployment measurement, 0.02 degrees Celsius per second can be directly used as a conservative default value for the thermal equilibrium threshold.

[0071] S103 adaptively switches the bias process variance between three states—high value, low value, and constant value—based on the temperature rise rate and static state flag during each acquisition cycle. This enables the bias tracking capability of the augmented Kalman filter to match the current thermal state of the sensor and the vehicle's motion state in real time. It maintains effective tracking of bias drift during the rapid temperature rise phase, suppresses bias estimation jitter during the thermal equilibrium phase, and freezes the bias process variance during the disturbance phase to prevent false updates. This provides S104 with process noise parameters adapted to the current operating conditions.

[0072] S104 uses the original tilt angle reading as the observation and the bias process variance and observation noise variance as noise parameters to run the prediction step and update step of the augmented Kalman filter, and outputs the true tilt angle estimate and bias estimate.

[0073] In one embodiment, within each acquisition cycle, the bias process variance output by S103 and the observation noise variance output by S102 are used as the noise parameters at the current moment, and the current raw tilt angle reading is used as the observation. The prediction step and update step of the augmented Kalman filter are executed sequentially, and the true tilt angle estimate and bias estimate at the current moment are output.

[0074] The current raw tilt angle reading synchronously acquired in step S101 is used as the observation input for the augmented Kalman filter. The augmented Kalman filter internally maintains a two-dimensional state vector, where the first component is defined as the true tilt angle component; the second component is defined as the bias component, representing the zero-point bias drift caused by chip temperature rise. Based on the observation relationship that the raw tilt angle reading equals the sum of the true tilt angle component and the bias drift component, an observation equation is constructed, which is characterized as:

[0075] In the formula, For the first The raw tilt angle readings for each acquisition cycle. For the true tilt component, For bias components, To observe noise.

[0076] The augmented Kalman filter performs recursive calculations based on a preset state transition relation. This relation is characterized by the fact that, within a short time interval of a fixed acquisition period, the changes in both the total tilt component and the bias component are extremely small and can be approximated as constant. The observation equation is modeled as follows: the original tilt reading equals the sum of the total tilt component and the observation noise.

[0077] Within each acquisition cycle, the augmented Kalman filter performs a prediction step and an update step sequentially.

[0078] In the prediction step, the filter uses the posterior state vector and its posterior covariance matrix obtained from the previous cycle to extrapolate the prior state vector for the current cycle based on the state transition relationship, and calculates the prior covariance matrix for the current cycle. When calculating the prior covariance matrix, the filter superimposes a process noise covariance matrix to characterize the uncertainty introduced during the prediction process. The process noise covariance matrix is ​​a diagonal matrix; its diagonal elements corresponding to the total tilt angle component take preset fixed constants, while the diagonal elements corresponding to the bias component are filled with the bias process variance output in real time by step S103 based on the temperature rise rate and the static state flag.

[0079] In the update step, the filter compares the total tilt angle component in the prior state vector obtained from the prediction step with the currently acquired raw tilt angle reading, calculating the deviation between the two as information. Subsequently, the filter uses the prior covariance matrix, the observation matrix, and the observation noise variance output from step S102 to calculate the optimal Kalman gain vector at the current time. The Kalman gain vector is used to determine the weighted fusion weight between the prior estimate and the current observation: when the observation noise variance output from step S102 is small, indicating that the vehicle is stationary and the observation quality is high, the Kalman gain increases, and the filter assigns a higher confidence weight to the current raw tilt angle reading; when the observation noise variance output from step S102 is large, indicating that the vehicle is disturbed and the observation quality is reduced, the Kalman gain decreases, and the filter reduces its dependence on the current raw tilt angle reading, instead placing more trust in the extrapolation results from the prediction step.

[0080] The filter uses the Kalman gain vector to weight the innovation, and then adds the weighted result as a correction to the prior state vector obtained in the prediction step to obtain the posterior state vector for the current cycle. Simultaneously, the filter updates the posterior covariance matrix for the current cycle, for use in the prediction step of the next acquisition cycle.

[0081] After the iterative calculations of the prediction and update steps described above, the augmented Kalman filter outputs a posterior state vector at the end of each acquisition cycle. The first component of this posterior state vector is the estimated true tilt angle at the current moment, and the second component is the estimated bias at the current moment. These two estimates are then passed to step S105 to calculate the corrected tilt angle value.

[0082] S104 completes a full iteration of the prediction step and update step within each acquisition cycle, outputting the current true tilt angle estimate and bias estimate. Both are components of the posterior state vector. After optimal weighted fusion using Kalman gain, they are statistically the minimum mean square error estimates of the current true tilt angle component and bias component, providing a reliable state estimation basis for the correction output of S105 and the learning of historical drift rates.

[0083] S105 obtains the corrected tilt angle value by subtracting the bias estimate from the true tilt angle estimate. When the stationary state is marked as stationary, the historical drift rate is updated by dividing the bias estimate by the power-on duration. At the next power-on, the product of the historical drift rate and the warm-up duration is injected into the augmented Kalman filter as the initial bias value.

[0084] In one embodiment, within each acquisition cycle, the corrected back tilt angle value at the current moment is obtained by subtracting the offset estimate value output by S104 from the true tilt angle estimate value output by S104, and the corrected back tilt angle value is output to the antenna servo control unit in real time as the angle reference for antenna pointing adjustment.

[0085] The calculation method for the corrected tilt angle is as follows:

[0086] In the formula, This is the corrected backslope angle value at the current moment, in degrees; This is the posterior estimate of the true tilt angle at the current moment, output by S104, in degrees. This is the current time-time bias posterior estimate output by S104, in degrees.

[0087] Understandably, the state equation of the augmented Kalman filter models the original tilt angle reading as the sum of the true tilt angle component and the bias component. Therefore, the first component of the posterior state vector, the true tilt angle estimate, theoretically no longer contains the bias component. Subtracting the bias estimate from the true tilt angle estimate here is an explicit secondary stripping of the filter output. This aims to further eliminate residual bias components caused by an overly large initial covariance matrix setting during the initial convergence phase. After the filter fully converges, the bias estimate tends to stabilize, and the difference between the true tilt angle estimate and the bias estimate approaches zero. The secondary stripping does not introduce additional error. During filter convergence, the secondary stripping accelerates the elimination of bias residues and shortens the time required for the corrected tilt angle value to reach stable accuracy.

[0088] For example, taking the 600th step after powering on a 2.4-meter vehicle-mounted C-band scattering antenna as an example, at this point the filter is still in the convergence process. The estimated true tilt angle is approximately 2.153 degrees, the estimated offset is approximately 0.021 degrees, and the corrected tilt angle is 2.153 degrees minus 0.021 degrees, i.e., 2.132 degrees. If the secondary stripping is not performed and the estimated true tilt angle of 2.153 degrees is directly used as the output, the antenna pointing error is approximately 0.021 degrees, corresponding to a beamwidth of approximately 1.2 degrees for the 2.4-meter C-band antenna. The 0.021-degree error accounts for 1.75% of the beamwidth, which has a perceptible impact on the antenna pointing accuracy in the early stages of convergence. After performing the secondary stripping, the corrected tilt angle of 2.132 degrees is closer to the true attitude angle, and the antenna pointing error is significantly reduced.

[0089] Historical drift rate update: In each acquisition cycle, determine whether the stationary status flag output by S102 is stationary. If the stationary status flag is stationary, then update the historical drift rate by dividing the current offset estimate by the current power-on duration.

[0090] The historical drift rate is calculated as follows:

[0091] In the formula, Historical drift rate, in degrees per second; This is the current bias estimate, in degrees. The power-on duration at the current moment is the cumulative duration from the current power-on moment to the corresponding moment of the current data collection cycle, expressed in seconds.

[0092] Understandably, the physical root of bias drift is the zero-point shift caused by changes in internal stress and electrical characteristics of the sensor chip as its temperature rises from the ambient temperature upon power-up to its thermal equilibrium temperature. During multiple power-ups of the same sensor model, the shape of the chip's temperature rise curve exhibits high consistency, and the accumulation pattern of bias drift with power-up duration also shows high repeatability. Therefore, dividing the measured bias estimate during the current power-up process by the power-up duration yields the historical drift rate, which reflects the typical drift characteristics of the sensor under the current ambient temperature conditions and can serve as a basis for predicting the initial bias value during the next power-up. The historical drift rate is updated only when the stationary state is marked as stationary, ensuring the reliability of the bias estimate and preventing false bias estimates during vehicle disturbances from contaminating the learning results of the historical drift rate.

[0093] The following subordinate scheme further defines the storage and update method of historical drift rate: the historical drift rate is persistently stored across power-on cycles using a moving average method. At the end of each power-on process, the historical drift rate that is finally updated in this power-on process is weighted and averaged with the previously stored historical drift rate to update the persistently stored historical drift rate.

[0094] Specifically, the historical drift rate moving average across power-on cycles is updated as follows:

[0095] In the formula, The average historical drift rate persisted after the nth power-on process, in degrees per second; The historical drift rate is the final update of the nth power-on process, in degrees per second. The average historical drift rate is the persistent storage value after the (n-1)th power-on process, in degrees per second. The learning weight for the current power-on process is a dimensionless real number that is greater than 0 and less than or equal to 1.

[0096] The learning weight is set as follows: the reciprocal of the cumulative number of power-ups is used as the learning weight. Specifically, for the nth power-up, the learning weight is 1 divided by n, making the average historical drift rate equivalent to the arithmetic mean of the historical drift rates over the previous n power-ups. This convergence occurs gradually with increasing power-up count. When the cumulative number of power-ups exceeds 20, the learning weight is fixed at 0.05, ensuring the average historical drift rate maintains dynamic tracking capability of the drift characteristics over the most recent approximately 20 power-ups, while suppressing the excessive influence of a single abnormal power-up on the average historical drift rate. Taking a 2.4-meter vehicle-mounted C-band scattering antenna as an example, the device is powered on approximately 2 to 4 times per day. 20 power-ups correspond to a usage cycle of approximately 5 to 10 days, which can cover the impact of seasonal environmental temperature changes on drift characteristics, ensuring the average historical drift rate remains adaptable to current environmental conditions.

[0097] Understandably, using a moving average method to persistently store historical drift rates across power-on cycles, rather than retaining only the historical drift rates from a single power-on process, can effectively suppress the interference of outliers in historical drift rates caused by the degradation of bias estimation quality due to prolonged vehicle disturbances during a single power-on process on the prediction of initial bias values ​​in subsequent power-on processes. The weighted averaging of multiple power-on processes makes the average historical drift rate tend towards the true drift characteristics of the sensor under current environmental conditions, and the prediction accuracy gradually improves with the increase in the number of power-on processes.

[0098] Bias initial value injection: Upon the next power-on, the product of the historical drift rate average stored in persistent memory and the warm-up time is used as the bias initial value and injected into the state vector initialization process of the augmented Kalman filter.

[0099] The initial bias value is calculated as follows:

[0100] In the formula, This is the initial bias value, in degrees; This represents the average historical drift rate for persistent storage, expressed in degrees per second. The warm-up time is the time interval from the moment of power-on to the moment when the augmented Kalman filter begins to output the corrected backtilt angle value, in seconds.

[0101] The state vector initialization method for the augmented Kalman filter is as follows: the first component of the posterior state vector, the true tilt angle estimate, is initialized to the original tilt angle reading at power-on; the second component, the bias estimate, is initialized to the initial bias value; the posterior covariance matrix is ​​initialized as a diagonal matrix, with the first diagonal element being 1×102. -4Square degree, where the second diagonal element is 10% of the square of the initial bias value, reflects the initial uncertainty of the prediction result for the initial bias value.

[0102] Understandably, injecting the product of the historical drift rate average and the warm-up time as the initial bias value into the filter ensures that the augmented Kalman filter already contains a reasonable prediction of the current bias drift in its initial state vector after power-on, rather than re-converging from zero bias. This mechanism allows the filter to have an initial bias estimate close to the actual bias magnitude at the beginning of power-on, significantly shortening the convergence time required for the filter to output a reliable corrected tilt angle value from initialization. Taking a 2.4-meter vehicle-mounted C-band scattering antenna as an example, without injecting the initial bias value, the filter typically requires about 600 steps to converge from zero bias to a bias estimation error below 0.01 degrees; after injecting the initial bias value, the initial bias estimation error is already below 0.05 degrees, and the number of steps required for the filter to converge to a bias estimation error below 0.01 degrees is reduced to about 120 steps, increasing the convergence speed by about 5 times.

[0103] The warm-up time is set as follows: Prioritize a data-driven approach. During the initial deployment of the device, record the time required for the sensor to reach a stable output after power-on correction of the tilt angle value. Take the average of these times as the warm-up time and write it into the parameter configuration. For example, with a 2.4-meter vehicle-mounted C-band scattering antenna, the measured warm-up time is approximately 30 seconds, corresponding to 600 acquisition cycles. During this period, the filter completes initial convergence, and the output error of the corrected tilt angle value drops to within 0.01 degrees. If an initial deployment test is not possible, 60 seconds can be used as a conservative default value for the warm-up time.

[0104] The S105 outputs the corrected tilt angle value in each acquisition cycle and continuously updates the historical drift rate in a static state. Through a persistent learning mechanism across power-on cycles, the prediction accuracy of the initial bias value gradually improves with the accumulation of device usage time. Upon the next power-on, the product of the average historical drift rate and the warm-up time is injected into the filter's initial state, shortening the filter's convergence time. This ensures that the corrected tilt angle value can achieve reliable output accuracy in the early stages of power-on, meeting the requirements of vehicle-mounted antennas for fast and accurate pointing under frequent power-on and power-off conditions.

[0105] like Figure 2 As shown, the diagram illustrates the functional modules, including the tilt sensor, short-time standard deviation calculation module, temperature rise rate calculation module, augmented Kalman filter, and servo controller. Arrows indicate the input of the raw tilt angle reading and chip temperature reading, as well as the data flow of signals such as the stationary status flag and observation noise variance to the Kalman filter, ultimately showing the entire link relationship from the output corrected tilt angle value to the servo controller.

[0106] like Figure 3As shown, a comparison curve of the tilt angle output error versus the factory temperature compensation calibration method during sensor operation is presented. The existing method shows significant drift and large fluctuations in error during temperature rise. However, after adopting the temperature drift compensation method provided by this invention, the tilt angle output error is effectively suppressed, and the overall error mean square value with power-on time is reduced by about 15% to 20% compared with the prior art, indicating that this invention can effectively eliminate transient bias drift.

[0107] like Figure 4 As shown in the figure, a comparison of the degradation curves of the bias estimation error of the present invention and the prior art as a function of short-time standard deviation is presented under harsh operating conditions such as vehicle engine idling vibration. It can be seen from the figure that in the normal static state range, the errors of the two are similar; when the short-time standard deviation increases and enters the harsh perturbation state range shown by the shaded area, the estimation error of the prior art degrades significantly, while the degradation of the present invention is significantly slowed down due to the implementation of observation noise variance adjustment and bias process variance freezing mechanism. The bias estimation error is reduced by about 20% compared with the prior art, which demonstrates the excellent robustness of the present invention under vehicle body disturbance conditions.

[0108] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for eliminating tilt sensor drift based on offset tracking algorithm, characterized in that, Includes the following steps: The original tilt angle readings and chip temperature readings of the tilt sensor are collected synchronously at a fixed acquisition cycle; The short-time standard deviation of the original tilt angle reading is calculated within a sliding window. If the short-time standard deviation is lower than the static threshold, it is judged as static; if it is not lower than the static threshold, it is judged as disturbance. The static state indicator is obtained, and the square of the short-time standard deviation is used as the observation noise variance. The temperature rise rate is calculated based on the chip temperature reading. When the static state is marked as static and the temperature rise rate is higher than the thermal equilibrium threshold, a high value is output as the bias process variance. When the static state is marked as static and the temperature rise rate is not higher than the thermal equilibrium threshold, a low value is output as the bias process variance. When the static state is marked as disturbance, the bias process variance remains unchanged. Using the original tilt angle reading as the observation, and the bias process variance and observation noise variance as noise parameters, the prediction step and update step of the augmented Kalman filter are run to output the true tilt angle estimate and bias estimate. The corrected tilt angle value is obtained by subtracting the bias estimate from the true tilt angle estimate. When the static state is marked as static, the historical drift rate is updated by dividing the bias estimate by the power-on duration. At the next power-on, the product of the historical drift rate and the warm-up duration is injected into the augmented Kalman filter as the initial bias value.

2. The tilt sensor drift elimination method based on offset tracking algorithm according to claim 1, characterized in that, The short-time standard deviation is calculated as follows: In each acquisition cycle, the current original tilt angle reading is stored in a sliding window of a preset length. The length of the sliding window is not less than the vehicle vibration cycle. After the sliding window is full, the average value of all original tilt angle readings in the window is calculated to obtain the window mean. The square of the difference between each original tilt angle reading and the window mean is calculated. The average value of all squares is then taken as the square root to obtain the short-time standard deviation.

3. The tilt sensor drift elimination method based on offset tracking algorithm according to claim 1, characterized in that, The temperature rise rate is calculated as follows: In each acquisition cycle, the current chip temperature reading is stored in a temperature window of a preset length. After the temperature window is full, the average value of the readings in the second half of the temperature window is subtracted from the average value of the readings in the first half, and then divided by the time interval between the center of the two segments to obtain the temperature rise rate.

4. The tilt sensor drift elimination method based on offset tracking algorithm according to claim 1, characterized in that, The high-end value is greater than the low-end value, and the ratio of the high-end value to the low-end value is not less than the preset value.

5. The tilt sensor drift elimination method based on offset tracking algorithm according to claim 1, characterized in that, The state vector of the augmented Kalman filter is a two-dimensional vector, with the first component being the true tilt component and the second component being the bias component; the state transition matrix is ​​a second-order identity matrix, indicating that the true tilt component and the bias component remain approximately unchanged in adjacent acquisition cycles; The observation matrix is ​​a row vector with both elements being 1, indicating that the original dip reading is equal to the sum of the true dip component and the offset component, plus the observation noise.

6. The tilt sensor drift elimination method based on offset tracking algorithm according to claim 5, characterized in that, The prediction update includes a prediction step and an update step. In the prediction step, the prior state vector is obtained by multiplying the state transition matrix by the posterior state vector of the previous period. The prior covariance matrix is ​​calculated using the state transition matrix, the posterior covariance matrix of the previous period, and the process noise covariance matrix. The process noise covariance matrix is ​​a diagonal matrix, with its first diagonal element being the fixed true tilt process variance and its second diagonal element being the bias process variance. In the update step, the innovation is obtained by subtracting the product of the observation matrix and the prior state vector from the original tilt angle reading. The Kalman gain vector is calculated using the prior covariance matrix, the observation matrix, and the observation noise variance. The posterior state vector is obtained by adding the product of the Kalman gain vector and the innovation to the prior state vector. The second component of the posterior state vector is the posterior bias estimate.

7. The tilt sensor drift elimination method based on offset tracking algorithm according to claim 5, characterized in that, The augmented Kalman filter is initialized as follows: upon power-up, the initial value of the true tilt component of the state vector is set to the original tilt reading acquired for the first time, and the initial value of the bias component is set to 0; the initial covariance matrix is ​​a diagonal matrix, the initial variance corresponding to the true tilt component is determined based on the system's dynamic measurement accuracy index, the initial variance corresponding to the bias component is determined based on the sensor's factory zero-point deviation range, and the initial variance corresponding to the bias component is greater than the initial variance corresponding to the true tilt component.

8. The tilt sensor drift elimination method based on offset tracking algorithm according to claim 1, characterized in that, It also includes a historical drift rate learning step: when the static state is marked as static and the temperature rise rate is not higher than the thermal equilibrium threshold, the difference between the posterior bias estimates of two adjacent steps is divided by a fixed acquisition period to obtain the current step bias drift rate. The average of the bias drift rates for a consecutive preset number of steps is taken to obtain the average drift rate of this step. The average drift rate of this step is added to the historical drift rate list. The historical drift rate list retains the most recent preset number of power-on records and deletes the earliest record when the limit is exceeded. The average of all records in the historical drift rate list is taken as the basis for setting the low value for the next power-on.

9. The tilt sensor drift elimination method based on offset tracking algorithm according to claim 8, characterized in that, Before performing the historical drift rate learning step, a two-axis consistency verification step is also included: the current bias ratio is obtained by taking the ratio of the current pitch axis posterior bias estimate to the roll axis posterior bias estimate; If the historical bias ratio has not been initialized, the average of the current bias ratio over a consecutive preset number of steps is used as the historical bias ratio and written into the memory, and the historical drift rate learning step is executed; if the historical bias ratio has been initialized, the absolute value of the difference between the current bias ratio and the historical bias ratio is calculated. If the absolute value is not greater than the ratio tolerance threshold, the historical drift rate learning step is executed; if the absolute value is greater than the ratio tolerance threshold, the historical drift rate learning step is skipped.

10. The tilt sensor drift elimination method based on offset tracking algorithm according to claim 9, characterized in that, The historical bias ratio is updated only when the absolute value is not greater than the ratio tolerance threshold and the historical drift rate learning step has been completed. The historical bias ratio is replaced by the average of the current bias ratios that have passed the dual-axis consistency verification step for the most recent consecutive preset number of steps.

Images