An adaptive zero drift compensation method and system for a pressure sensor

Abstract

This invention relates to the field of pressure sensor measurement technology, specifically to an adaptive zero-point drift compensation method and system for pressure sensors. The method includes: using the difference between the real-time output under no-pressure conditions and the initial zero-point signal as the zero-point drift signal input to an RC low-pass filter; calculating the information entropy of the zero-drift signal within a sliding time window, and determining the target time constant and dynamic fusion weights through nonlinear mapping; adjusting the digital potentiometer array to adjust the time constant to the target value; generating an instantaneous compensation curve and presetting the capacitor voltage when power-on or the temperature change rate exceeds a threshold; calculating the input-output voltage difference and its first derivative to obtain a lead correction term, and weighting and fusing it with the filtered output according to the dynamic fusion weights to obtain the zero-point compensation voltage; and cutting off the drift input and introducing a reverse polarity traction voltage when compensation saturates. The solution of this invention can improve the response speed, accuracy, and operational stability of zero-point drift compensation.

Description

Technical Field

[0001] This invention relates to the field of pressure sensor measurement technology, and specifically to an adaptive zero-point drift compensation method and system for pressure sensors. Background Technology

[0002] Pressure sensors are indispensable core sensing devices in industrial control, automotive electronics, aerospace, and medical equipment. Their measurement accuracy and stability directly impact the operational safety and efficiency of the entire system. In practical applications, pressure sensors inevitably experience zero-point drift due to the material properties of their internal sensing elements, the discreteness of their manufacturing processes, and the complexity of the operating environment. Zero-point drift refers to the phenomenon where, in the absence of pressure excitation, the sensor's output signal deviates from its initial zero-point reference and changes slowly or irregularly over time. This drift is often superimposed on the actual pressure measurement signal, causing significant measurement errors. Because of the strong randomness and time-varying nature of zero-point drift signals, traditional parameter compensation methods cannot achieve a balance between noise filtering and rapid drift tracking, making it difficult to guarantee high-precision compensation results.

[0003] In actual service conditions, pressure sensors frequently encounter situations such as initial system power-on or drastic changes in external ambient temperature. Under these abrupt changes, the internal thermal stress of the sensor fluctuates dramatically, causing the zero-point drift to exhibit strong transient nonlinear characteristics. Conventional low-pass filtering compensation mechanisms often suffer from severe response lag and phase delay due to inherent time constant limitations, failing to track and cancel the rapidly drifting signal in a timely manner. When the system is in a long-term extreme environment or encounters severe interference, the continuously accumulating zero-point compensation signal can easily reach or even exceed the hardware range saturation boundary of the back-end processing circuit or digital-to-analog converter. Once hardware saturation occurs, traditional compensation systems often fall into a deadlock or failure state, unable to recover to the normal linear adjustment range, resulting in the paralysis of the entire pressure sensor's measurement function. There is an urgent need in this field for a pressure sensor zero-point drift compensation method that can adjust parameters according to the characteristics of the drift signal, has rapid response capability under abrupt changes, and can prevent and escape from hardware saturation, in order to improve the measurement accuracy and long-term reliability of the sensor in extreme environments.

[0004] Chinese patent application CN111721466A discloses a method and system for correcting zero-point drift of a pressure sensor. The method acquires multiple voltage values ​​measured by the pressure sensor at a certain temperature, inputs the temperature and voltage values ​​into a preset temperature compensation model for temperature compensation to obtain corresponding compensation pressure values, calculates the average of the compensation pressure values, and corrects each compensation pressure value based on a preset algorithm to obtain a corrected pressure value, thereby reducing the impact of ambient temperature changes and residual stress inside the resistor on measurement accuracy. However, the correction process of this method mainly relies on a pre-calibrated temperature compensation model and offline post-processing of the measurement results. Its control logic focuses on temperature-induced drift correction under relatively steady-state conditions, and it lacks real-time characterization of the disorder of noise during operation. Under dynamic disturbance conditions such as the coexistence of high-frequency noise and low-frequency gradual changes or drastic temperature changes, it is difficult to simultaneously achieve noise suppression and drift tracking, and its adaptive compensation capability is still lacking.

[0005] Chinese patent application CN121632443A discloses a method and system for zero-point drift compensation of pressure sensors in a high-voltage DC converter valve cooling system. It constructs a multi-factor coupled drift model with temperature, vibration frequency, and electromagnetic interference intensity as inputs, collects environmental parameters and pressure signals in real time, and periodically collects a reference pressure signal. Short-term dynamic compensation is achieved based on model prediction and reference deviation correction. The model is updated through incremental learning according to the drift trend, while simultaneously performing anomaly monitoring and early warning. Although this method improves the compensation adaptability in complex environments through multi-factor modeling and reference calibration, its compensation effect is highly dependent on the accuracy of model prediction and periodic reference calibration. The overall modeling and computation process is relatively complex. In scenarios where the compensation amount reaches the hardware capability limit or encounters strong disturbances or sudden changes, it lacks the processing capability to quickly exit the abnormal state and restore stable output at the execution level. There is still room for improvement in balancing operational stability and response speed. Summary of the Invention

[0006] To address the problem of hardware saturation and the inability of traditional compensation systems to automatically recover to the normal linear adjustment range, leading to the paralysis of the entire pressure sensor's measurement function, this invention proposes an adaptive zero-point drift compensation method for pressure sensors in its first aspect. The method includes: acquiring the initial zero-point signal of the sensor under a known zero-pressure reference; using the difference between the real-time output of the sensor in a pressure-free excitation state and the initial zero-point signal as the zero-point drift signal and inputting it to an RC low-pass filter; calculating the information entropy of the zero-point drift signal within a sliding time window; determining the target time constant and dynamic fusion weight based on the information entropy using a nonlinear mapping function; adjusting the digital potentiometer array associated with the resistor of the RC low-pass filter to adjust the filter's time constant to the target time constant; and when the system is powered on or the sensor housing temperature change rate exceeds a predetermined threshold... When the temperature is at its maximum, an instantaneous compensation curve is generated based on the current temperature value and the rate of temperature change. An initial compensation preset voltage is calculated on the instantaneous compensation curve, and the capacitor voltage of the RC low-pass filter is preset as the initial compensation preset voltage. The difference between the input voltage and the output voltage of the RC low-pass filter and its first derivative are calculated, and the linear combination of the two is used as a lead correction term. The lead correction term and the output voltage of the RC low-pass filter are weighted and fused according to the dynamic fusion weight to generate a zero-point compensation voltage. When the zero-point compensation voltage reaches or exceeds the hardware saturation boundary, the input of the zero-point drift signal is cut off, and an internal traction voltage with opposite polarity to the saturation boundary and a preset amplitude shrinkage is used as the new input. The zero-point drift signal input is restored after the amplitude of the zero-point drift signal falls back to within the saturation margin boundary.

[0007] By using the difference between the real-time output under no-pressure excitation state and the initial zero-point signal as the zero-point drift signal and processing it through an RC low-pass filter, the target time constant and dynamic fusion weights are determined by nonlinear mapping using the information entropy within the sliding time window. This allows the filter time constant to adaptively adjust according to the disorder of the zero-drift signal, avoiding the problem of a fixed time constant being difficult to balance between high-frequency noise and low-frequency slow changes. When the system powers on or the temperature change rate exceeds a threshold, an instantaneous compensation curve is generated based on the current temperature and the temperature change rate, and the capacitor voltage is preset, shortening the transition process required for the filter to establish a stable output. The lead correction term and the filter output are weighted and fused according to the dynamic fusion weights to generate the zero-point compensation voltage, achieving a balance between noise suppression and drift tracking. When the compensation voltage reaches the hardware saturation boundary, the drift signal input is cut off and a reverse polarity traction voltage is introduced. The input is restored only after the signal falls back to within the saturation margin boundary, avoiding long-term output lock-up due to integral entanglement, thereby improving the response speed and operational stability of zero-point drift compensation.

[0008] Further, calculating the information entropy of the zero-point drift signal within the sliding time window includes: storing the zero-point drift signal acquisition values ​​for consecutive preset sampling periods into a one-dimensional buffer array to form a time series within the sliding time window; counting the frequency of the zero-point drift signal acquisition values ​​in the one-dimensional buffer array across multiple equidistant numerical intervals; dividing the frequency of each numerical interval by the total number of preset sampling periods to obtain the probability of each interval; calculating the natural logarithm of each probability, multiplying the natural logarithm by the corresponding probability itself to obtain a product, summing the products of all intervals and taking the opposite number to obtain the information entropy within the sliding time window.

[0009] By storing the zero-point drift signal acquisition values ​​of consecutive preset sampling periods into a one-dimensional buffer array to form a time series, and then converting the frequency of multiple equidistant numerical intervals into probabilities, the information entropy is obtained by multiplying the natural logarithm of each probability with itself and then summing and inverting the result. This transforms the disorder and high-frequency disturbance state of the zero-drift signal in the current window into feature values ​​that can be directly used in the calculation, avoiding the error caused by judging the noise level based solely on the instantaneous amplitude. This provides a stable and reliable criterion for determining the subsequent time constant and fusion weight.

[0010] Furthermore, the preset sampling period is 256, and each sampling period is... to The number of the plurality of equally spaced numerical intervals is 16 to 32, and the range of the zero-point drift signal acquisition values ​​is [missing information]. to It is then evenly divided into each equidistant numerical interval; the interval with a frequency of zero is assigned a value of 1×10. -6 The offset constant is then used in the calculation of the information entropy.

[0011] Further, the target time constant and dynamic fusion weights are determined by a nonlinear mapping function based on the information entropy, including: multiplying the information entropy by a first constant gain with positive polarity and adding a first fixed bias, substituting it into the hyperbolic tangent function for calculation and taking the absolute value to obtain an adjustment coefficient, multiplying the adjustment coefficient by the system reference time constant to obtain the target time constant; squaring the information entropy, multiplying it by a second constant gain with positive polarity and adding a second fixed bias, substituting it as an input parameter into the Sigmoid function for calculation, and outputting the dynamic fusion weights whose value range is limited to 0 and 1.

[0012] By multiplying the information entropy by a positive first constant gain and superimposing a first fixed bias, and then substituting the absolute value of the hyperbolic tangent function, the adjustment coefficient is obtained. This coefficient is then multiplied by the system reference time constant to obtain the target time constant, making the time constant change smoothly and boundedly with the degree of noise disorder. At the same time, the information entropy is squared, multiplied by a second constant gain, and superimposed with a second fixed bias, and then substituting the result into the sigmoid function. This limits the dynamic fusion weights to a normalized range between 0 and 1, preventing weights from going out of bounds and causing weighted fusion distortion. As a result, reasonable adjustment parameters can be output in both noise enhancement and drift stabilization states.

[0013] Furthermore, the first constant gain ranges from 0.5 to 2.0, the first fixed bias is a positive value from 0.0 to 1.0, and the system reference time constant is... The value range of the second constant gain is 0.1 to 0.5, and the value range of the second fixed bias is -2.0 to 0.

[0014] Further, adjusting the digital potentiometer array associated with the RC low-pass filter resistor to adjust the time constant of the RC low-pass filter to the target time constant includes: dividing the target time constant by the fixed capacitance value in the RC low-pass filter to calculate the target series resistance value; subtracting the internal contact bottom resistance value of the digital potentiometer array in a zero-resistance state from the target series resistance value; if the subtraction result is less than zero, clamping the net resistance value to zero; otherwise, using the subtraction result as the net resistance value; dividing the net resistance value by the single-stage tap step resistance value and rounding down to obtain the instruction control step value; and writing the instruction control step value to the digital potentiometer array through the serial peripheral interface bus to control the on / off state of the corresponding multiplexer network to connect and match the resistance units.

[0015] The target series resistance is calculated by dividing the target time constant by the fixed capacitor value. Then, the internal contact bottom resistance of the digital potentiometer array in zero-resistance state is subtracted, and the result is clamped to zero when it is less than zero. This ensures that the set resistance value fully takes into account the inherent bottom resistance of the actual hardware and avoids equivalent resistance deviation caused by ignoring the bottom resistance. The net resistance is divided by the single-stage tap step resistance value and rounded down to obtain the instruction control step value. This value is then written through the serial peripheral interface to control the on / off state of the multiplexer network. This allows the adjustable resistor network to approach the target value with a matched resistance unit, improving the consistency between the actual time constant adjustment result and the target setting.

[0016] Further, generating an instantaneous compensation curve based on the current temperature value and the temperature change rate includes: discretely segmenting and quantizing the current temperature value and the temperature change rate of the sensor housing, then concatenating them using bitmasks according to high and low address segments to obtain synthetic addressing instruction parameters; performing a read operation on the storage area of ​​the storage array offset from the starting address of the synthetic addressing instruction parameters to extract five consecutive control point offset parameters; adding the five consecutive control point offset parameters bit by bit to the coordinates of five pre-cached reference control points to obtain an actual Bessel compensation control point group; and inputting the actual Bessel compensation control point group into a polynomial interpolation equation operator to generate a corresponding coordinate sampling mapping value set to obtain the instantaneous compensation curve.

[0017] By discretely segmenting the current temperature value and the sensor housing temperature change rate, and then concatenating the high and low address segment bit masks, a synthetic addressing instruction parameter is generated. Based on this, the offset parameters of five consecutive control points are read from the offset address of the storage array, and then added bit by bit with the pre-cached reference control point coordinates to obtain the actual Bessel compensation control point group. Finally, the polynomial interpolation equation operator is input to generate a set of coordinate sampling mapping values, so that the instantaneous compensation curve can be quickly customized according to the current thermal disturbance state, avoiding the slow convergence of the filtering process when the temperature changes drastically, and improving the adaptability of the compensation curve to temperature drift.

[0018] Further, the input voltage and output voltage of the RC low-pass filter are read separately under the same sampling clock. The input voltage and output voltage under the current sampling clock are subtracted to obtain a first difference. The first difference is subtracted from the corresponding difference under the previous sampling clock and divided by the sampling period to obtain the first derivative of the difference. The first difference is multiplied by a first static coefficient to obtain a first product. The first derivative of the difference is multiplied by a second static coefficient with time dimension to obtain a second product. The first product and the second product are added to obtain the lead compensation term. The margin ratio coefficient is obtained by subtracting the dynamic fusion weight using a constant 1. The lead compensation term is multiplied by the margin ratio coefficient to obtain a first weighted value. The output voltage of the RC low-pass filter is multiplied by the dynamic fusion weight to obtain a second weighted value. The first weighted value and the second weighted value are added to generate the zero-point compensation voltage.

[0019] Furthermore, when the zero-point compensation voltage reaches or exceeds the hardware saturation boundary, the input of the zero-point drift signal is cut off, and an internal traction voltage with opposite polarity and a preset shrinkage value is used as the new input. This includes: when the comparison detects that the zero-point compensation voltage reaches or exceeds the system's preset hardware saturation boundary voltage, controlling a multiplexer to disconnect the input channel of the zero-point drift signal; obtaining the reference values ​​of the polarity and amplitude of the hardware saturation boundary voltage, calculating the value with opposite polarity and amplitude minus the preset shrinkage value, and using this as the internal traction voltage; and inputting the internal traction voltage as the new input signal into the RC low-pass filter through a closed backup channel.

[0020] In a second aspect, the present invention provides an adaptive zero-point drift compensation system for a pressure sensor, comprising a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, an adaptive zero-point drift compensation method for a pressure sensor according to the present invention is implemented.

[0021] The technical effects of this invention are as follows: The difference between the output under no-pressure excitation and the initial zero-point signal is used as the zero-point drift signal. The target time constant and dynamic fusion weights are determined through nonlinear mapping based on the information entropy within the sliding time window. This allows the filtering stage to adaptively adjust according to the degree of noise disorder, avoiding the problem of a fixed time constant failing to simultaneously suppress high-frequency noise and track low-frequency drift. A digital potentiometer array, after incorporating internal low resistance, approximates the target resistance value, improving the consistency between the actual time constant adjustment and the set value. When the system is powered on or the temperature change rate exceeds a threshold, an instantaneous compensation curve is generated and a preset capacitor voltage is applied, shortening the transition process required for output stabilization and enhancing adaptability to temperature drift. The lead correction term and the filtered output are weighted and fused according to dynamic fusion weights, prioritizing smoothing and noise suppression during severe disturbances and trend tracking during stable conditions. When the compensation voltage reaches the saturation boundary, the drift input is cut off and a reverse polarity traction voltage is introduced to neutralize accumulated charge. The input is restored after the signal returns to normal, avoiding output lock-up due to integral entanglement. This comprehensively improves the response speed, accuracy, and operational stability of zero-point drift compensation. Attached Figure Description

[0022] Figure 1 This is a flowchart of an adaptive zero-point drift compensation method for a pressure sensor according to an embodiment of the present invention;

[0023] Figure 2 This is a schematic diagram of the feedforward compensation scheme in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the rapid release of hardware saturation in an embodiment of the present invention; Figure 4The present invention provides a structural block diagram of an adaptive zero-point drift compensation system for a pressure sensor according to an embodiment of the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] Reference Figure 1 This invention proposes an adaptive zero-point drift compensation method for a pressure sensor, comprising the following steps: S1, acquire the initial zero-point signal of the sensor under the known zero-pressure reference, and use the difference between the real-time output of the sensor under the pressureless excitation state and the initial zero-point signal as the zero-point drift signal and input it to the RC low-pass filter.

[0026] During system initialization, the microcontroller's Memory Access Controller (DMA) works in conjunction with a high-precision analog-to-digital converter (ADC) to continuously acquire 100 cycles of sensorless analog signals. The arm_mean_f32 function in the CMSIS-DSP software library is called to calculate the average of the acquired floating-point array, and this average is stored in non-volatile memory as the initial zero-point signal. During operation, the ADC acquires the real-time output voltage at a sampling rate of 1kHz. The microprocessor's arithmetic logic unit calls the arm_sub_f32 function to calculate the difference sequence between the real-time output and the initial zero-point signal, which serves as the digital zero-point drift signal. This digital signal is converted into an analog voltage signal by the ADC, impedance-matched by a voltage follower composed of operational amplifiers with unity voltage gain, and the output is connected to the input of a hardware RC low-pass filter consisting of a series resistor and a parallel capacitor.

[0027] S2, calculate the information entropy of the zero-point drift signal within the sliding time window, and determine the target time constant and dynamic fusion weights based on the information entropy through a nonlinear mapping function.

[0028] A circular buffer with 256 floating-point variables is allocated in the microcontroller's random access memory as a sliding time window. Based on a first-in, first-out (FIFO) algorithm, the latest data is updated and the oldest data is discarded during each sampling. A histogram statistical algorithm is used, employing the C standard math library (math.h) to divide the voltage difference data in the circular buffer into 32 independent sub-intervals according to amplitude ranges. The frequency of data falling into each sub-interval is counted and divided by the total number to calculate the discrete probability distribution. Based on Shannon's information entropy principle, the log2f function from the math.h library is called to perform logarithmic operations on each non-zero probability and calculate the sum of the information entropy of the probability distribution. A system is constructed using the natural constant... The Sigmoid mapping algorithm with base 0.01 uses the obtained information entropy value as the input variable. When the information entropy increases, indicating an increase in high-frequency random noise, the target time constant value calculated by the mapping algorithm increases monotonically and the dynamic fusion weight value increases monotonically. When the information entropy decreases, indicating that the zero-point drift is in a low-frequency steady state, the target time constant value decreases monotonically and the dynamic fusion weight value decreases monotonically.

[0029] In some embodiments, calculating the information entropy of the zero-point drift signal within a sliding time window includes: The zero-point drift signal acquisition values ​​of consecutive preset sampling periods are stored in a one-dimensional buffer array to form a time series within the sliding time window; Statistical analysis of the frequency of zero-point drift signal acquisition values ​​in multiple equally spaced intervals within a one-dimensional buffer array; Divide the frequency of each numerical interval by the total number of preset sampling periods to obtain the probability of each interval; Calculate the natural logarithm of each probability, multiply the natural logarithm by the corresponding probability itself to obtain the product, sum the products of all intervals and take the opposite number to obtain the information entropy within the sliding time window.

[0030] The preset sampling period can be configured as follows: to Example A sliding time window is constructed using a circular buffer, with the window length set to [value missing]. Each sampling period, for example ,correspond The real-time data storage depth forms a time series. During data acquisition, the microcontroller will store data within a range such as... to The zero-point drift signal acquisition values ​​are evenly divided into... In a number of equally spaced numerical intervals, optionally The values ​​range from 16 to 32; this example uses 16 values. The width of each interval is calculated as follows: The system iterates through the 256 cached data points, using a counter array to accumulate the number of data hits in each interval, resulting in a frequency distribution array. The microcontroller then uses its floating-point unit to divide the frequency of each interval by the total number of 256, converting this into a corresponding probability of occurrence. Intervals with a frequency of zero are either masked or assigned a probability. The minimum offset constant is used to calculate the product of each probability and its natural logarithm using standard mathematical library functions. An accumulator is then used to sum the products across all intervals, and the sum is inverted to obtain the information entropy. This information entropy, as a quantization metric, represents the degree of disorder and high-frequency noise disturbance of the zero-drift signal within the current sliding window.

[0031] In some embodiments, the target time constant and dynamic fusion weights are determined based on information entropy through a nonlinear mapping function, including: Multiply the information entropy by the first constant gain with positive polarity and add the first fixed bias, substitute it into the hyperbolic tangent function for calculation and take the absolute value to obtain the adjustment coefficient; multiply the adjustment coefficient by the system reference time constant to obtain the target time constant; After squaring the information entropy, multiply it by a positive second constant gain and add a second fixed bias. Substitute this as an input parameter into the Sigmoid function to calculate the output, which is a dynamic fusion weight whose value range is limited to 0 and 1.

[0032] The selectable range for the first constant gain is 0.5 to 2.0, with an example of 1.2. The first fixed bias is set to a positive value between 0.0 and 1.0, with an example of 0.5. The reference time constant is preset to... The microprocessor substitutes the real-time calculated information entropy into the equation of the hyperbolic tangent function and takes its absolute value. For example, when the information entropy equals 1.25, the calculated adjustment coefficient is approximately 0.964. This adjustment coefficient is multiplied by... The target time constant was then obtained to be approximately To generate the dynamic fusion weights, the information entropy is squared to amplify the difference between high and low perturbation features. The squared information entropy value is multiplied by a second constant gain, selectable from 0.1 to 0.5 (0.2 in this example), and then superimposed with a second fixed bias, selectable from -2.0 to 0 (-1.5 in this example). The resulting parameters are then substituted into the Sigmoid function to perform a nonlinear mapping. This process restricts the dynamic fusion weights to a normalized continuous space of 0 to 1, which can be used as floating-point multiplication weight coefficients in software algorithms when the outputs of subsequent lead correction and smoothing filtering are mixed.

[0033] S3 adjusts the digital potentiometer array associated with the RC low-pass filter resistor to adjust the filter's time constant to the target time constant.

[0034] Based on fundamental circuit principles, the time constant equals the equivalent resistance multiplied by the capacitance. Given a known and fixed capacitance value for the RC low-pass filter film, the microprocessor executes a floating-point division instruction to divide the target time constant by the capacitance value to calculate the target resistance value. The digital potentiometer array uses three MCP41100 high-resolution digital potentiometers cascaded in series to form an adjustable resistor network. The microcontroller uses a built-in binary search lookup algorithm to retrieve the corresponding set of tap configuration digits based on the target resistance value. The HAL_SPI_Transmit function from the STM32 Hardware Abstraction Layer (HAL) library is then called to send this set of digits to the control registers of the three digital potentiometers via the SPI bus. By changing the conduction node of the internal solid-state multiplexer, the actual equivalent resistance value connected to the hardware circuit is altered, causing the actual operating time constant of the hardware RC low-pass filter to approach and stabilize at the target time constant.

[0035] In some embodiments, adjusting a digital potentiometer array associated with an RC low-pass filter resistor to adjust the filter's time constant to a target time constant includes: The target series resistance value is calculated by dividing the target time constant by the value of the capacitor fixed in the RC low-pass filter. Subtract the internal contact bottom resistance of the digital potentiometer array in zero resistance state from the target series resistance value; if the subtraction result is less than zero, clamp the net resistance value to zero, otherwise use the subtraction result directly as the net resistance value; divide the net resistance value by the single-stage tap step resistance value and round down to obtain the instruction control step value. The serial peripheral interface bus writes instructions to the digital potentiometer array to control the step value, and controls the on / off of the corresponding multiplexer network to connect the series-matched resistance units.

[0036] The employed RC low-pass filter features high-precision surface-mount capacitors on the hardware board-level circuitry. The fixed capacitor values ​​can be selected from 1 microfarad to 10 microfarads; this example uses 10 microfarads. The microcontroller reads the currently dynamically generated target time constant in floating-point format, for example... Divide by the value of the solidified capacitor to calculate the current required target series resistance value of the system. Because digital potentiometer arrays, such as the MCP41100, still exhibit internal multiplexer resistance even when set to zero taps, the internal contact resistance of the potentiometer array is pre-measured and calibrated. ,from Subtracting this minimum resistance value from the calculated net resistance value required for internal matching of the digital potentiometer, we obtain the net resistance value. The microprocessor calculates the single-stage tap step resistance value based on the hardware specifications of the digital potentiometer, such as a total nominal resistance of 100kΩ and an 8-bit digital potentiometer with 256 resolution levels. .Will The net resistance value divided by The process involves rounding down to the nearest integer, generating a control step value of 24. This control step value is then cascaded and distributed to the registers of the three digital potentiometers using integer division and modulo algorithms. For a fully loaded potentiometer, the maximum level digital word is written, and the remainder step value is written to the secondary potentiometer. The on / off states of the internal multiplexers at each level are independently configured via the SPI bus, achieving a closed-loop overall resistance distribution for the cascaded network. The main control chip sends 16-bit data frames at a 4MHz clock frequency via the SPI bus to control the on / off state of the MOSFET switching network within the digital potentiometer array, achieving a resistance value of [value missing]. Adding the bottom resistance to form The actual equivalent resistance is approximated by the target setting and the switching of the time constant is achieved.

[0037] S4. When the system is powered on or the temperature change rate of the sensor housing exceeds a predetermined threshold, an instantaneous compensation curve is generated based on the current temperature value and the temperature change rate. The initial compensation preset voltage is calculated on the instantaneous compensation curve, and the capacitor voltage of the RC low-pass filter is preset as the initial compensation preset voltage.

[0038] A surface-mount NTC thermistor voltage divider circuit acquires ambient electrical signals in real time. The Steinhart-Hart equation conversion algorithm is used to obtain the current precise Celsius temperature value. The temperature change gradient between adjacent sampling periods is calculated using a first-order backward difference numerical calculation method, serving as the temperature change rate. A state machine conditional branch statement is set to monitor the system's reset power-on flag or compare the calculated temperature change rate with a stored absolute threshold of 0.5°C per second in flash memory. When any of these trigger conditions are met, the current temperature value and temperature change rate are used as a two-dimensional input vector, substituted into a pre-trained bivariate cubic polynomial regression model. The output yields transient curvature characteristic parameters, linear drift rate, and initial bias. These are used as the quadratic coefficient, linear coefficient, and constant term (intercept) respectively, constructing a parabolic instantaneous compensation curve equation with time as the variable. The time variable in this polynomial equation is set to zero, and the constant term is extracted as the intercept term, yielding the initial compensation preset voltage value. The microcontroller calls the HAL_DAC_SetValue function to enable the independent auxiliary digital-to-analog converter channel to output the absolute voltage, controls the general-purpose input / output pin to output a high-level signal to turn on the external CD4053 analog switch, and directly connects the output of the auxiliary digital-to-analog converter across the capacitor of the RC low-pass filter. Fast charging: Once the preset voltage on the capacitor plates is established, immediately control the pin to output a low level to disconnect the analog switch.

[0039] In some embodiments, generating an instantaneous compensation curve based on the current temperature value and the rate of temperature change includes: After discretely segmenting and quantizing the current temperature value and the rate of change of the sensor housing temperature, the bitmasks of the high and low address segments are concatenated to obtain the synthetic addressing instruction parameters. Perform a read operation on the storage area in the storage array that synthesizes addressing instruction parameters with the starting address offset, and extract the offset parameters of five consecutive control points; The offset parameters of five consecutive control points are added bit by bit to the coordinates of five pre-cached reference control points to obtain the actual Bezier compensation control point group. The actual Bessel compensation control point set is input into the polynomial interpolation equation operator to generate the corresponding coordinate sampling mapping numerical set, and the instantaneous compensation curve is obtained.

[0040] The current temperature value is sampled with a resolution of 1℃ and mapped to a range, for example, -40℃ to 125℃ is represented as an 8-bit unsigned integer from 0 to 165; the calculated temperature change rate is discretized and represented as a 4-bit value from 0 to 15 within the range of -5℃ / s to +5℃ / s. The microprocessor performs bitwise operations, shifting the 8-bit temperature value left by 4 bits and concatenating it with the 4-bit temperature change rate using a bitwise OR logic to form a 12-bit synthetic addressing instruction parameter. This synthetic addressing instruction parameter is multiplied by the total number of bytes occupied by each set of offset parameters, for example, 10 bytes, and the result is added to the starting address base address of the flash memory, for example, 0x08020000, to address the location of the target memory area. Five 16-bit signed format control point offset parameters are extracted using a bus memory read instruction. The microcontroller retrieves the coordinates of five reference Bessel control points pre-cached in the on-chip RAM. It then adds the extracted offset parameters bit-by-bit to the ordinate of each reference control point, completing the spatial correction of the vertices of the initial control polygon to obtain the actual Bessel compensation control point set. This set of points is input into a polynomial interpolation operator to perform fourth-order Bessel polynomial operations. The interpolation parameter is incremented from 0 to 1 in a fixed step size of 0.01, iteratively calculating 101 corresponding mapped numerical nodes. This set of nodes constitutes a customized instantaneous compensation curve under the current instantaneous thermal disturbance state. The ordinate coordinate of the first discrete node in the mapped numerical node set where the interpolation parameter is 0 is extracted and used as the initial compensation preset voltage.

[0041] S5, calculate the difference between the input voltage and the output voltage of the RC low-pass filter and its first derivative, and use the linear combination of the two as the lead correction term.

[0042] The microcontroller's dual-channel synchronous analog-to-digital converter peripheral module is configured to simultaneously and continuously acquire the analog voltages of the signal input and output nodes of the RC low-pass filter. Vector subtraction is performed within the arithmetic logic unit to obtain the real-time voltage difference sequence between the internal and external nodes. To obtain the first derivative of the difference, a first-order backward differential numerical derivative algorithm is implemented in the firmware. The rate of change of the difference is calculated by subtracting the voltage difference of the previous sampling period from the current sampling period's voltage difference and then dividing by the sampling period's time step. Fixed empirical proportional and differential coefficients are set in the microcontroller. The arm_mac_f32 multiply-accumulate joint operation function in the CMSIS-DSP library is called to sum the result of multiplying the difference by the proportional coefficient and the result of multiplying the first derivative of the difference by the differential coefficient. The value stored in the register is the digital quantity of the lead compensation term with phase lead compensation characteristics.

[0043] In some embodiments, the difference between the input voltage and the output voltage of the RC low-pass filter and its first derivative are calculated, and the linear combination of the two is used as the lead compensation term, including: Under the same sampling clock, read the input voltage and output voltage of the RC low-pass filter respectively; Subtract the input voltage from the output voltage at the current sampling clock to obtain the first difference; Subtract the corresponding difference from the previous sampling clock from the first difference, and divide the result of the subtraction by the sampling period to obtain the first derivative of the difference. Multiply the first difference by the first static coefficient to obtain the first product, multiply the first derivative of the difference by the second static coefficient with time dimension to obtain the second product, and add the first product and the second product to obtain the lead correction term.

[0044] Employing a high-precision 16-bit analog-to-digital converter with multi-channel parallel synchronous sampling capability, interrupted by a single hardware timer, for example, with a sampling period set to... This means a trigger frequency of 1kHz drives the corresponding ADC channel to perform data latching, ensuring that the input and output voltages of the RC low-pass filter are read within the same sampling clock cycle. In the microcontroller's arithmetic logic unit, the input voltage at the current clock scale is subtracted from the output voltage to calculate the current first difference. For example, when the input is... The output is At that time, the first difference is Retrieve the first difference value corresponding to the previous sampling clock cycle from the buffer, for example... The current first difference The increment of the difference is obtained by subtracting the first difference from the previous period. Dividing by the sampling period parameter of 0.001 seconds, the first derivative of the difference is calculated as follows: In the subsequent synthesis operation, a first static coefficient with a preset selectable range of 0.1 to 1.5 is used; in this example, 0.5 is chosen, which is the difference from the first value. Multiplying them together gives 4 The first product; set a second static coefficient, which can be selected from 0.01 seconds to 0.05 seconds, for example, 0.02 seconds, and the first derivative. Multiply to get The second product. Add the two products together to obtain... This result is the advance correction term with phase advance prediction characteristics.

[0045] S6 weights and fuses the lead compensation term with the output voltage of the RC low-pass filter according to the dynamic fusion weight to generate the zero-point compensation voltage.

[0046] The lead compensation term calculated in the pre-processing steps is set as source data sequence 1, and the actual output voltage of the RC low-pass filter synchronously acquired by the analog-to-digital converter is set as source data sequence 2. A linear interpolation weighted fusion algorithm model is constructed in the microprocessor. The constant 1 is calculated, and the dynamic fusion weight is subtracted to obtain the complementary weight. Then, source data sequence 1 is multiplied by the complementary weight, and source data sequence 2 is multiplied by the dynamic fusion weight. At the program level, the arm_scale_f32 vector scaling function of the CMSIS-DSP library is called sequentially to complete the above two floating-point multiplication operations. Then, the arm_add_f32 vector addition function is called to add the corresponding positions of the two scaled intermediate results, merging them to output a single-dimensional scalar sequence, which is then sent to the system's main control digital-to-analog converter. The voltage signal generated in this way is the zero-point compensation voltage against signal phase delay.

[0047] In some embodiments, the lead compensation term and the output voltage of the RC low-pass filter are weighted and fused according to dynamic fusion weights to generate a zero-point compensation voltage, including: The residual ratio coefficient is obtained by subtracting the dynamic fusion weight from the constant 1. Multiply the advance correction term by the residual ratio coefficient to obtain the first weighted value; The output voltage of the RC low-pass filter is multiplied by the dynamic fusion weight to obtain the second weighted value; The first weighted value is added to the second weighted value to generate the zero-point compensation voltage.

[0048] The microprocessor utilizes a floating-point dynamic fusion weight calculated using a nonlinear mapping of the Sigmoid function, with a range limited to 0.0 to 1.0. For example, a real-time value of 0.65 is used. A scalar subtraction operation is performed, subtracting the dynamic fusion weight 0.65 from the constant 1.0 to calculate a residual proportion coefficient of 0.35. To ensure the accuracy of system calculations, the hardware logic or system-level code can call a single-precision floating-point instruction set to generate the proportion parameter in this step. The microprocessor then calls the hardware multiplier to add an advance correction term containing difference and differential trend prediction information, for example, a real-time calculated amplitude of... Multiply by the margin ratio factor of 0.35 to calculate the result. The first weighted value of the amplitude; the current actual output voltage of the RC low-pass filter, for example, after smoothing and noise reduction. Multiplying by the dynamic fusion weight of 0.65, the solution is obtained with... The second weighted value of the amplitude. The microprocessor performs a linear superposition operation, adding the first weighted value... With the second weighted value Summing, we get The zero-point compensation voltage ensures that during periods of severe disturbance, the low-pass filter smoothing value is primarily used to effectively suppress high-frequency noise. However, when the system is stable (i.e., the zero-point drift is in a low-frequency steady state), a higher weight is assigned to the lead compensation term. This allows the term to utilize its small difference and differential trend to sensitively track and compensate for slow zero-point drift, referencing... Figure 2 .

[0049] S7: When the zero-point compensation voltage reaches or exceeds the hardware saturation boundary, the input of the zero-point drift signal is cut off, and the internal traction voltage with the opposite polarity to the saturation boundary and the amplitude with a preset shrinkage amount is used as the new input; the zero-point drift signal input is restored after the amplitude of the zero-point drift signal falls back to the saturation margin boundary.

[0050] Within the microcontroller's timer interrupt service function, a digital logic state machine algorithm based on a Schmitt trigger mechanism is constructed to continuously evaluate the zero-point compensation voltage amplitude. This amplitude is compared with the hardware saturation absolute upper and lower boundary thresholds defined by shrinking the positive and negative power supply rails of the operational amplifier by 0.1V inwards, performing a greater than or less than operation. When it is determined that either boundary has been reached, resulting in the integrator saturation anti-tangle critical state, the HAL_GPIO_WritePin function is used to change the control pin level driving the high-isolation solid-state relay, causing the relay contact connected in series at the signal input front of the RC low-pass filter to pop open, completely cutting off the original zero-point drift signal source input. The state machine performs arithmetic operations based on the polarity of the trigger boundary; if the positive power supply saturation boundary is touched, the positive boundary voltage is inverted and added. The retraction amount is used as the reverse internal traction voltage; conversely, if the negative boundary is touched, the absolute value of the negative boundary voltage is taken and subtracted. The retraction amount is calculated to obtain an internal traction voltage with positive polarity. This traction voltage is output through an idle on-chip analog-to-digital channel and connected to the input of the disconnected RC network, forcing the filter capacitor to discharge in reverse. During the disconnection period, the original zero-point drift signal source is virtually sampled using an analog-to-digital converter, and a virtual zero-point compensation voltage is calculated in the background closed loop. The state machine determines the amplitude of this virtual zero-point compensation voltage through an absolute value function. When the amplitude is within the upper and lower limits of the saturation boundary for 10 consecutive sampling cycles, the retraction is further reduced. When the formed dead zone of the saturation margin boundary is formed, the state machine is triggered to execute the automatic reset logic, which flips the control pin level again to disconnect the internal traction voltage and re-engages the solid-state relay to connect the initial zero-point drift signal input.

[0051] In some embodiments, when the zero-point compensation voltage reaches or exceeds the hardware saturation boundary, the input of the zero-point drift signal is cut off, and an internal traction voltage with opposite polarity to the saturation boundary and a preset amplitude reduction is used as the new input, including: When the comparison detects that the zero-point compensation voltage reaches or exceeds the system's preset hardware saturation boundary voltage, the multiplexer is controlled to disconnect the input channel of the zero-point drift signal. Obtain the reference values ​​of the polarity and amplitude of the hardware saturation boundary voltage, calculate the value with the opposite polarity and amplitude minus the preset shrinkage amount, and use it as the internal traction voltage. The internal traction voltage is used as a new input signal and input to the RC low-pass filter through a closed backup channel.

[0052] Continuously monitor the generated zero-point compensation voltage. When the compensation voltage reaches or exceeds the preset hardware saturation boundary voltage of the operational amplifier and ADC sampling terminals, for example using... Dual-rail power supply, with the hardware saturation boundary voltage selectably set to either forward or reverse. Upon this, the output pin control level will immediately be sent to the high-speed multiplexer analog switch at the front end, disconnecting the zero-point drift signal input channel originally connected to the system main circuit, thus cutting off the continuous integration and accumulation of the abnormal voltage. The polarity and amplitude reference values ​​of the triggered hardware saturation boundary voltage will be extracted; if the current voltage reaches the positive boundary... The microcontroller will extract the opposite negative polarity and move it towards the positive boundary. Inverting yields And add an optional range to the inverse amplitude. to The preset retraction amount, for example, is taken as... The corresponding internal traction voltage value was calculated as follows: The microcontroller controls the internal high-precision digital-to-analog converter to output a forward-biased reference level. This level is then converted and polarity-flipped via an externally configured inverting adder amplifier circuit, resulting in the output. The DC level is used as the new input signal, and the backup channel on the multiplexer is closed via a control command, inverting the negative traction voltage into the RC low-pass filter network. This reverse voltage neutralizes the saturation charge accumulated inside the filter capacitor, causing the output voltage to exit the hardware saturation region. Figure 3 .

[0053] The experimental conditions were set such that the ambient temperature was rapidly increased from room temperature (25°C) to 85°C at a rate of 5°C per second, and a peak-to-peak value was continuously injected into the system test circuit. High-frequency white noise was used as the zero-point drift disturbance source. All test groups were operated on the same 16-bit high-precision analog-to-digital converter and signal acquisition hardware platform, and the reference sampling period was uniformly set to [value missing]. The hardware saturation boundary voltage is set to A single complete test cycle lasts 300 seconds.

[0054] The benchmark group uses a conventional RC low-pass filter with a fixed time constant for zero-point compensation. Experimental data shows that the response delay reaches [value missing]. The maximum residual error of zero-point drift is as high as Due to the influence of high-frequency noise, the compensated output oscillates, with a root mean square error of [value missing]. The ablation group, based on the baseline group, only added a module for instantaneous compensation curves based on temperature and the rate of temperature change. Tests showed that the response delay was reduced to... The maximum residual error decreased to However, the root mean square error still reaches The complete solution employs all mechanisms, including information entropy fusion and advance correction, reducing response latency to [missing information]. The maximum residual error is suppressed at The root mean square error is as low as .

[0055] An embodiment of an adaptive zero-point drift compensation system for a pressure sensor: On the other hand, the present invention also provides an adaptive zero-point drift compensation system for a pressure sensor. For example... Figure 4 An adaptive zero-point drift compensation system for a pressure sensor is shown, comprising a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement an adaptive zero-point drift compensation method for a pressure sensor according to the first aspect of the present invention.

[0056] An adaptive zero-point drift compensation system for a pressure sensor also includes other components well known to those skilled in the art, such as a communication interface, the settings and functions of which are known in the art and will not be described in detail here.

Claims

1. An adaptive zero-point drift compensation method for a pressure sensor, characterized in that, Includes the following steps: The initial zero-point signal of the sensor under a known zero-pressure reference is obtained. The difference between the real-time output of the sensor under no-pressure excitation state and the initial zero-point signal is used as the zero-point drift signal and input to the RC low-pass filter. The information entropy of the zero-point drift signal within the sliding time window is calculated. This information entropy is multiplied by a positive first constant gain and a first fixed bias, then substituted into the hyperbolic tangent function for calculation, and the absolute value is taken to obtain the adjustment coefficient. This adjustment coefficient is multiplied by the system reference time constant to obtain the target time constant. The information entropy is then squared, multiplied by a positive second constant gain, and a second fixed bias is added. This squared entropy is used as an input parameter in the Sigmoid function for calculation, and the output is a dynamic fusion weight with a value range limited to 0 and 1. The time constant of the filter is adjusted to the target time constant by adjusting the digital potentiometer array associated with the RC low-pass filter resistor. When the system is powered on or the temperature change rate of the sensor housing exceeds the predetermined threshold, the current temperature value and the temperature change rate of the sensor housing are discretely segmented and quantized, and then the bitmasks are concatenated according to the high and low address segments to obtain the synthetic addressing instruction parameters. Perform a read operation on the storage area in the storage array that synthesizes addressing instruction parameters with the starting address offset, and extract the offset parameters of five consecutive control points; The offset parameters of five consecutive control points are added bit by bit to the coordinates of five pre-cached reference control points to obtain the actual Bessel compensation control point group; the actual Bessel compensation control point group is input into the polynomial interpolation equation operator to generate the corresponding coordinate sampling mapping numerical set to obtain the instantaneous compensation curve; the initial compensation preset voltage is calculated on the instantaneous compensation curve; and the capacitor voltage of the RC low-pass filter is preset as the initial compensation preset voltage. Calculate the difference between the input voltage and the output voltage of the RC low-pass filter and its first derivative, and use the linear combination of the two as the lead compensation term. Weight the lead compensation term and the output voltage of the RC low-pass filter according to dynamic fusion weights to generate a zero-point compensation voltage. When the zero-point compensation voltage reaches or exceeds the hardware saturation boundary, cut off the input of the zero-point drift signal and use an internal traction voltage with opposite polarity to the saturation boundary and a preset amplitude reduction as the new input. Restore the zero-point drift signal input once the amplitude of the zero-point drift signal falls back to within the saturation margin boundary.

2. The adaptive zero-point drift compensation method for a pressure sensor according to claim 1, characterized in that, Calculating the information entropy of the zero-point drift signal within the sliding time window includes: The zero-point drift signal acquisition values ​​of consecutive preset sampling periods are stored in a one-dimensional buffer array to form a time series within the sliding time window; The frequency of the zero-point drift signal acquisition values ​​in the one-dimensional buffer array within multiple equally spaced numerical intervals is counted; the frequency of each numerical interval is divided by the total number of the preset sampling periods to obtain the probability of each interval. Calculate the natural logarithm of each probability, multiply the natural logarithm by the corresponding probability itself to obtain the product, add up the products of all intervals and take the opposite number to obtain the information entropy within the sliding time window.

3. The adaptive zero-point drift compensation method for a pressure sensor according to claim 2, characterized in that, The preset sampling period is 256, and each sampling period is... to ; The number of the plurality of equally spaced numerical intervals is 16 to 32, and the range of the zero-point drift signal acquisition values ​​is [missing information]. to And it is evenly divided into each equally spaced numerical interval; Assign 1×10 to the interval with a frequency of zero. -6 The offset constant is then used in the calculation of the information entropy.

4. The adaptive zero-point drift compensation method for a pressure sensor according to claim 1, characterized in that, The first constant gain ranges from 0.5 to 2.0, the first fixed bias is a positive value from 0.0 to 1.0, and the system reference time constant is... ; The value range of the second constant gain is 0.1 to 0.5, and the value range of the second fixed bias is -2.0 to 0.

5. The adaptive zero-point drift compensation method for a pressure sensor according to claim 1, characterized in that, Adjusting the digital potentiometer array associated with the resistor of the RC low-pass filter to adjust the time constant of the RC low-pass filter to the target time constant includes: The target series resistance value is calculated by dividing the target time constant by the value of the capacitor fixed in the RC low-pass filter. The target series resistance value is then subtracted from the internal contact bottom resistance value of the digital potentiometer array in the zero-resistance state. If the subtraction result is less than zero, the net resistance value is clamped to zero; otherwise, the subtraction result is used as the net resistance value. The net resistance value is then divided by the single-stage tap step resistance value and rounded down to obtain the instruction control step value. The instruction control step value is written to the digital potentiometer array through the serial peripheral interface bus to control the on / off state of the corresponding multiplexer network to match the series resistance units.

6. The adaptive zero-point drift compensation method for a pressure sensor according to claim 1, characterized in that, Under the same sampling clock, the input voltage and output voltage of the RC low-pass filter are read respectively, and the first difference is obtained by subtracting the input voltage and the output voltage under the current sampling clock. Subtract the corresponding difference from the previous sampling clock from the first difference and divide by the sampling period to obtain the first derivative of the difference. The first difference is multiplied by the first static coefficient to obtain the first product, the first derivative of the difference is multiplied by the second static coefficient with time dimension to obtain the second product, and the first product and the second product are added to obtain the lead compensation term; the margin ratio coefficient is obtained by subtracting the dynamic fusion weight from the constant 1, the lead compensation term is multiplied by the margin ratio coefficient to obtain the first weighted value, the output voltage of the RC low-pass filter is multiplied by the dynamic fusion weight to obtain the second weighted value, and the first weighted value and the second weighted value are added to generate the zero-point compensation voltage.

7. The adaptive zero-point drift compensation method for a pressure sensor according to claim 1, characterized in that, When the zero-point compensation voltage reaches or exceeds the hardware saturation boundary, the input of the zero-point drift signal is cut off, and an internal traction voltage with opposite polarity to the saturation boundary and a preset amplitude reduction is used as the new input, including: When the zero-point compensation voltage is detected to reach or exceed the system's preset hardware saturation boundary voltage, the multiplexer is controlled to disconnect the input channel of the zero-point drift signal; the polarity and amplitude reference values ​​of the hardware saturation boundary voltage are obtained, and the value with the opposite polarity and amplitude minus the preset shrinkage amount is calculated as the internal traction voltage; the internal traction voltage is used as a new input signal and input to the RC low-pass filter through a closed backup channel.

8. An adaptive zero-point drift compensation system for a pressure sensor, characterized in that, It includes a processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement an adaptive zero-point drift compensation method for a pressure sensor as described in any one of claims 1 to 7.

Images