Vehicle-mounted camera chip dynamic voltage frequency self-adaptive adjusting system and control method

Abstract

The invention discloses a vehicle-mounted camera chip dynamic voltage frequency self-adaptive adjusting system and a control method. The system comprises an image quality evaluation module, an environmental parameter acquisition module, a composite optimization calculation module, a self-adaptive frequency adjusting module and a multi-layer voltage adjusting array. The input end of the image quality evaluation module is connected with an output bus of the image sensor, the image quality evaluation module is provided with a digital noise analysis unit and a dynamic fuzzy detection unit, and the adaptive frequency adjustment module is designed based on an LVDS architecture; by establishing a real-time mutual feedback network of image quality parameters and environment states, the dynamic stability of a vehicle-mounted visual system under complex working conditions is remarkably improved, full-parameter coupling optimization of image processing load fluctuation and chip physical states is achieved based on a dynamic weight distribution strategy of a three-dimensional regulation and control matrix, and the system is suitable for large-scale popularization and application. And the signal-to-noise ratio characteristic of the current scene and the environmental stress load are automatically matched in the voltage frequency adjustment process.

Description

Technical Field

[0001] The present invention belongs to the field of automatic control oscillators, and more specifically, particularly relates to a dynamic voltage and frequency adaptive regulation system for in-vehicle camera chips. At the same time, the present invention also relates to a control method for the dynamic voltage and frequency adaptive regulation system of in-vehicle camera chips. Background Art

[0002] In the field of existing vehicle vision domain controller technologies, such as CN112313601B - Vehicle, Vision Domain Controller, Video Data Processing Method, Device and Medium, this patent mainly improves the image quality by optimizing the inter-frame correlation of video processing algorithms, and its technical solution focuses on image enhancement and target recognition accuracy improvement at the software level. However, this technology does not fully consider the problem of chip-level hardware performance decline caused by physical environmental stress in in-vehicle camera chips under extreme working conditions. For example: Single-dimensional adjustment defect: The existing solutions only improve the image quality through software compensation (such as exposure correction, denoising filtering), without establishing a closed-loop coupling mechanism between hardware power supply parameters and physical environment changes, resulting in a lack of fundamental solutions to problems such as the decrease in the quantum efficiency of image sensors and the deterioration of charge transfer efficiency caused by vibration in high-temperature environments; Limitation of static voltage and frequency configuration: The power supply strategy of in-vehicle camera chips adopts a stepped fixed voltage and frequency combination, which cannot adapt to the dynamically changing image processing load in real time (such as the sudden high frame rate demand in the ADAS scenario), resulting in a contradiction between energy waste at low load and insufficient power noise suppression ability at high load.

[0003] The above is difficult to ensure imaging stability in complex electromagnetic environments. Especially for the ASIL-D functional safety level required by autonomous driving systems, the environmental parameter isolation processing mechanism of existing technologies cannot meet the strict requirements for maintaining safety states under multiple failure modes. Summary of the Invention

[0004] The purpose of the present invention is to solve the disadvantages existing in the prior art, and to propose a dynamic voltage and frequency adaptive regulation system and control method for in-vehicle camera chips. By establishing a real-time mutual feedback network between image quality parameters and environmental states, the dynamic stability of the in-vehicle vision system under complex working conditions is significantly improved. Based on the dynamic weight distribution strategy of the three-dimensional regulation matrix, the full-parameter coupling optimization of the image processing load fluctuation and the chip physical state is realized, enabling the voltage and frequency adjustment process to automatically match the signal-to-noise ratio characteristics and environmental stress load of the current scene.

[0005] To achieve the above purpose, the present invention provides the following technical solutions:

[0006] A dynamic voltage and frequency adaptive regulation system for in-vehicle camera chips, comprising:

[0007] Image quality assessment module, environmental parameter acquisition module, composite optimization calculation module, adaptive frequency adjustment module and multi-layer voltage regulation array;

[0008] The input end of the image quality assessment module is connected to the image sensor output bus and is equipped with a digital noise analysis unit and a dynamic blur detection unit;

[0009] The environmental parameter acquisition module includes a temperature monitoring unit and a vibration sensor array; the composite optimization calculation module has a parameter fusion unit and an LSTM-based prediction control unit, which receives the noise index, fuzziness data and environmental parameter data from the image quality assessment module;

[0010] The adaptive frequency adjustment module is designed based on the LVDS architecture and includes a programmable phase-locked loop array, the output of which is connected to the image sensor core clock tree;

[0011] The multi-layer voltage regulation array is provided with N groups of multi-phase charge pump units, each group including four mutually coupled PWM controllers.

[0012] Preferably, the composite optimization calculation module generates a three-dimensional control matrix through a dynamic weight allocation algorithm, and outputs a frequency compensation vector to the adaptive frequency adjustment module in a time-sharing manner, and simultaneously outputs a voltage gradient control signal to the multi-layer voltage adjustment array, thereby forming a dual negative feedback loop based on the image quality degradation rate and the change of environmental parameters;

[0013] Preferably, the dynamic weight allocation algorithm satisfies the following three-dimensional control matrix generation rules:

[0014] Where M t is the three-dimensional control matrix at time t, α is the dynamic feedback coefficient, and α∈(0,1.5], ΔQ is the gradient change of the image quality degradation rate in the time window, β is the noise sensitivity factor, P env is the normalized environmental parameter vector, W hist It is the historical weight window function output by the LSTM prediction control unit;

[0015] The above formula introduces the constrained dynamic coefficient α and the exponential decay term exp(-β·ΔQ) to suppress the image quality degradation in real time, and the environmental parameter P env With historical weight W hist The tensor product operation is performed to establish a nonlinear mapping relationship between the optical system and the environment. The noise sensitivity factor β is dynamically calibrated through threshold comparison, which solves the overfitting caused by fixed parameters.

[0016] Preferably, the N groups of charge pump units of the multi-layer voltage regulator array operate according to the following interleaved control strategy:

[0017] Wherein, V out is the output voltage, V ref is the reference voltage, n is the number of the current charge pump group, and n ∈ [1, N], A is the dynamic amplitude adjustment variable depending on the frequency compensation vector, f sw is the preset switching frequency band of 250 kHz - 2 MHz, θ n is the phase offset, τ n is the RC time constant adjustment parameter related to the output of the temperature monitoring unit; the above formula is through (-1) n ·A·sgn[sin(2πf sw t + θ n )] with an alternating polarity design, so that the switching transients of adjacent charge pump units cancel each other out, reducing the common-mode noise coupling, The item adaptively adjusts the rise time according to the temperature parameter τ n to avoid current spikes caused by simultaneous switching of multiple units.

[0018] Preferably, each branch of the programmable phase-locked loop array includes a digital phase detector, a loop filter, and 4 groups of parallel voltage-controlled oscillator arrays, and the phase detection reference clock sources of each branch have a programmable division ratio of 1 - 8; the frequency tuning curve of the voltage-controlled oscillator array is divided into a three-segment compensation structure, adopting a linear tuning mode below 1.2 GHz, a quadratic curve compensation mode between 1.2 - 2.4 GHz, and an exponential compensation mode above 2.4 GHz.

[0019] Preferably, the dynamic blur detection unit implements a dual-path processing architecture, including a high-frequency component extraction branch based on the Sobel operator and a time-domain analysis branch based on the optical flow motion vector, and the outputs of the two are weighted and summed to generate blur data; the digital noise analysis unit performs wavelet packet decomposition processing in the spatial domain and adaptive Kalman filtering in the frequency domain, and the final noise index is determined by the geometric mean of the processing results of the two domains.

[0020] Preferably, the memory window length of the LSTM prediction control unit is adjusted in real time according to the ΔQ value. When ΔQ > the threshold Q th the window length is shortened to 1 - ΔQ / Q max times the original value to improve the response speed.

[0021] A control method for a dynamic voltage and frequency adaptive adjustment system of an in-vehicle camera chip, the method is implemented by using the above system, and includes the following steps:

[0022] S1. Obtain the image data stream from the output bus of the image sensor through the image quality evaluation module, and calculate the noise index and blur data by using the digital noise analysis unit and the dynamic blur detection unit;

[0023] S2. The temperature monitoring unit of the environmental parameter acquisition module collects the surface temperature distribution data of the chip in real time, and at the same time detects the three-dimensional axial acceleration parameters through the vibration sensing array;

[0024] S3. The composite optimization calculation module performs normalized fusion on the noise index, ambiguity data and environmental parameters, and generates a three-dimensional regulation matrix based on the dynamic weight allocation algorithm;

[0025] S4. The adaptive frequency adjustment module analyzes the frequency compensation vector in the three-dimensional regulation matrix, and adjusts the phase synchronization characteristics of the image sensor core clock tree through the programmable phase-locked loop array with LVDS architecture;

[0026] S5. The multi-layer voltage regulation array triggers the PWM controllers of N groups of multi-phase charge pump units to enter the coupled working mode according to the voltage gradient control signal output by the three-dimensional regulation matrix;

[0027] S6. Based on the cross-evaluation results of the image quality degradation rate and the environmental parameter change rate, a dual closed-loop negative feedback control chain for the frequency vector and the voltage gradient is established.

[0028] Preferably, the dynamic ambiguity detection unit implements dual-path collaborative processing: performing Sobel operator edge sharpness analysis in the spatial domain, and at the same time using optical flow motion vector estimation in the time domain. Finally, the ambiguity data takes the weighted sum of squares of the results of the two paths; the temperature monitoring unit performs redundant data acquisition: the sampling points of three PT100 sensors are evenly distributed in a 120-degree ring along the chip power supply rail, and the sampling period is synchronized with the switching frequency of the charge pump unit in the multi-layer voltage regulation array at an integer multiple of 1:4.

[0029] The technical effects and advantages of the present invention: The in-vehicle camera chip dynamic voltage and frequency adaptive regulation system and control method provided by the present invention have the following advantages compared with the prior art:

[0030] Environment-adaptive cross-domain collaborative optimization mechanism: Through the real-time fusion and analysis of the image quality parameters and environmental physical quantities by the composite optimization calculation module, a three-dimensional dynamic regulation space under non-steady-state conditions is constructed, breaking through the response bottleneck of traditional single-dimensional threshold control;

[0031] Enhanced dynamic stability of multi-physical field coupling: Utilizing the phase synchronization characteristics of the LVDS architecture and the interleaved regulation of multi-phase charge pumps, the matching offset between the power supply noise spectrum and the image sensor clock jitter is realized, significantly suppressing the cascading amplification effect of common-mode interference;

[0032] Feedforward compensation ability for non - linear hysteresis: Based on the time - varying weight allocation strategy of the LSTM prediction unit, the system has the characteristics of lead compensation in the voltage - frequency linkage response under vibration transient impact, effectively blocking the vicious coupling chain of vibration - temperature - electrical parameters;

[0033] Topological reconstruction characteristics for anti - disturbance robustness: Through the redundant decision - making mechanism of the double negative feedback loop, a double fault - tolerant judgment on the deterioration trend of image quality and the mutation direction of environmental parameters is formed to ensure control continuity under extreme working conditions. Description of the Drawings

[0034] Figure 1 It is the flow chart of the control method of the dynamic voltage - frequency adaptive regulation system for the in - vehicle camera chip of the present invention. Detailed Implementation Modes

[0035] In order to make the objectives, technical solutions and advantages of the present invention more clear and understandable, the following further details the present invention in conjunction with specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0036] The present invention provides a dynamic voltage - frequency adaptive regulation system and control method for in - vehicle camera chips. By establishing a real - time mutual feedback network of image quality parameters and environmental states, the dynamic stability of the in - vehicle vision system under complex working conditions is significantly improved; based on the dynamic weight allocation strategy of the three - dimensional regulation matrix, the full - parameter coupling optimization of the image - processing load fluctuation and the physical state of the chip is realized, enabling the voltage - frequency adjustment process to automatically match the signal - to - noise ratio characteristics and environmental stress loads of the current scene; the interleaved control of the multi - phase charge pump combined with the LVDS phase synchronization technology fundamentally weakens the interference penetration of high - frequency switching noise into the CMOS image sensor clock tree, ensuring pixel sampling consistency during high - speed imaging; and the double - layer closed - loop feedback architecture enables the system to synchronously trigger the double decision - making mechanisms of feedforward compensation and lag correction when encountering temperature gradient mutations or mechanical vibration impacts, avoiding the overshoot oscillation problem in traditional linear control; this solution greatly improves the operation reliability of the in - vehicle camera module in the extreme temperature - vibration composite environment, providing a highly robust visual signal guarantee for the autonomous driving system.

[0037] The dynamic voltage - frequency adaptive regulation system for in - vehicle camera chips includes:

[0038] An image quality evaluation module, an environmental parameter acquisition module, a composite optimization calculation module, an adaptive frequency adjustment module, and a multi - layer voltage regulation array;

[0039] The composite optimization calculation module generates a three-dimensional control matrix using a dynamic weight allocation algorithm. It then outputs frequency compensation vectors to the adaptive frequency adjustment module in a time-sharing manner. It also outputs voltage gradient control signals to the multi-layer voltage regulation array, forming a dual negative feedback loop based on image quality degradation rate and environmental parameter changes. The dynamic weight allocation algorithm satisfies the following three-dimensional control matrix generation rules:

[0040] Where M t is the three-dimensional control matrix at time t, α is the dynamic feedback coefficient, and α∈(0,1.5], ΔQ is the gradient change of the image quality degradation rate in the time window, β is the noise sensitivity factor, P env is the normalized environmental parameter vector, W hist It is the historical weight window function output by the LSTM prediction control unit;

[0041] The above formula introduces the constrained dynamic coefficient α and the exponential decay term exp(-β·ΔQ) to suppress the image quality degradation in real time, and the environmental parameter P env With historical weight W hist The tensor product operation is performed to establish a nonlinear mapping relationship between the optical system and the environment. The noise sensitivity factor β is dynamically calibrated through threshold comparison, which solves the overfitting caused by fixed parameters.

[0042] The memory window length of the LSTM prediction control unit is adjusted in real time according to the ΔQ value. When ΔQ> threshold Q th The window length is shortened to 1-ΔQ / Q of the original value max times to increase response speed.

[0043] The N groups of charge pump units in the multi-layer voltage regulator array operate according to the following interleaved control strategy:

[0044] Where V out is the output voltage, V ref is the reference voltage, n is the number of the current charge pump group, and n∈[1,N], A is the dynamic amplitude adjustment variable depending on the frequency compensation vector, f sw For the preset 250kHz-2MHz switching frequency band, θ n is the phase offset, τ n Adjust the RC time constant parameter related to the temperature monitoring unit output; the above formula is (-1) n ·A·sgn[sin(2πf sw t+θ n )] by using the alternating polarity design, the switching transients of adjacent charge pump units cancel each other out, reducing common-mode noise coupling. The term is based on the temperature parameter τ nAdjust the rise time adaptively to avoid current spikes caused by simultaneous switching of multiple units;

[0045] The memory window length of the LSTM prediction control unit is adjusted in real time according to the ΔQ value. When ΔQ > threshold Q th the window length is shortened to (1 - ΔQ / Q) times the original value to improve the response speed. max times to improve the response speed.

[0046] It should be noted that the frequency compensation vector satisfies the phase synchronization constraint:

[0047]

[0048] Constraint (s.t.):

[0049]

[0050] In the above formula, ω k is the instantaneous angular frequency of the kth LVDS channel, ω0 is the output value of the reference frequency source, and Δω m is the phase difference compensation amount of the mth branch of the programmable phase-locked loop (PLL) array, K v is the voltage-frequency conversion gain coefficient, ΔV dd is the real-time output voltage gradient of the charge pump unit (i.e., the dynamic change rate of the supply voltage), and F comp is the optimized optimal compensation control strategy;

[0051] In the above formula, by minimizing ∑ k |ω k -(ω0 + Δω m )| 2 the deviation of the frequency of each LVDS channel from the reference frequency (after adjustment) is minimized, thereby improving the multi-channel data synchronization accuracy; using least squares optimization, compared with the traditional fixed frequency offset compensation method, it can dynamically adapt to the frequency drift of different channels;

[0052] The lower bound limit (Δω max ) ensures that the adjustment rate of the phase-frequency compensation amount Δω m is not lower than the minimum threshold to avoid the accumulation of phase-locked errors caused by too slow compensation;

[0053] The upper bound limit (K v ·ΔV dd ) is based on the system voltage-frequency conversion characteristic K v and the power supply dynamic change ΔV dd to limit the maximum compensation rate and prevent the PLL from losing lock due to too fast compensation.

[0054] The input end of the image quality evaluation module is connected to the output bus of the image sensor, and is configured with a digital noise analysis unit and a dynamic blur detection unit; the dynamic blur detection unit implements a dual-path processing architecture, including a high-frequency component extraction branch based on the Sobel operator and a time-domain analysis branch based on the optical flow motion vector. The outputs of both are weighted and summed to generate blur data; the digital noise analysis unit performs wavelet packet decomposition processing in the spatial domain and uses adaptive Kalman filtering in the frequency domain. The final noise index is determined by the geometric mean of the processing results in the two domains.

[0055] The environmental parameter acquisition module includes a temperature monitoring unit and a vibration sensing array; the composite optimization calculation module has a parameter fusion unit and an LSTM-based predictive control unit, and receives the noise index, blur data, and environmental parameter data of the image quality evaluation module; the temperature monitoring unit of the environmental parameter acquisition module includes three redundant PT100 sensors, which are arranged in an equilateral triangle along the power supply rail of the image sensor, and the sampling period has an integer multiple relationship with the switching frequency of the charge pump unit;

[0056] The vibration sensing array adopts a hybrid layout of MEMS accelerometers and piezoelectric film sensors, where three axial accelerometers are orthogonally distributed, and the piezoelectric film sensors are arranged on the bottom surface of the image sensor packaging substrate; the memory window length of the LSTM predictive control unit is adjusted in real time according to the ΔQ value. When ΔQ > threshold Q_th, the window length is shortened to (1 - ΔQ / Q_max) times the original value to improve the response speed.

[0057] The adaptive frequency adjustment module is designed based on the LVDS architecture and includes a programmable phase-locked loop array, whose output end is connected to the core clock tree of the image sensor; each branch in the programmable phase-locked loop array includes a digital phase detector, a loop filter, and a 4-group parallel voltage-controlled oscillator array. The phase detection reference clock source of each branch has a programmable 1-8 frequency division ratio. The frequency tuning curve of the voltage-controlled oscillator array is divided into a three-segment compensation structure. Below 1.2 GHz, a linear tuning mode is adopted. Between 1.2 - 2.4 GHz, a quadratic curve compensation mode is adopted. Above 2.4 GHz, an exponential compensation mode is enabled.

[0058] The multi-layer voltage regulation array is provided with N groups of multi-phase charge pump units, each group includes 4 mutually coupled PWM controllers, and the coupling between the PWM controllers adopts a capacitor-inductor hybrid energy transfer structure. There is a fixed 15 ns dead time compensation for the drive signals of adjacent controllers.

[0059] This embodiment also provides Figure 1 The control method of the dynamic voltage and frequency adaptive regulation system of the in-vehicle camera chip as shown, the method is implemented by using the above system, and includes the following steps:

[0060] S1. The image quality evaluation module obtains the image data stream from the output bus of the image sensor, and calculates the noise index and blur degree data by using the digital noise analysis unit and the dynamic blur detection unit;

[0061] Further, the image quality evaluation module intercepts the original Bayer format data stream of the image sensor in real time through the high-speed parallel bus interface. The digital noise analysis unit adopts a spatial-frequency domain joint discrimination strategy: in the spatial domain, it implements abnormal pixel clustering detection based on the local variance threshold, and simultaneously performs wavelet packet decomposition on the image line cycle signal in the frequency domain to extract the energy mutation characteristics of the high-frequency subband. The dynamic blur detection unit constructs a dual decision path: the spatial path uses an improved Sobel operator to perform directional cumulative statistics on the edge gradient modulus value, and the time domain path introduces a pyramid optical flow algorithm to calculate the inter-frame motion vector field. In the microcontroller, the quantization values output by the two paths are orthogonally projected and transformed with dynamic weight coefficients, and finally fused to generate a normalized blur degree parameter. This process synchronously updates the noise index map and marks the coordinate distribution characteristics of the high-noise sensitive areas.

[0062] S2. The temperature monitoring unit of the environmental parameter acquisition module real-time collects the surface temperature distribution data of the chip, and simultaneously detects the three-dimensional axial acceleration parameters through the vibration sensing array;

[0063] Further, the temperature monitoring unit is configured with three platinum resistance PT100 sensor arrays, mounted on the three-dimensional thermal gradient maximum change area of the camera chip substrate, and connected to the ADC acquisition channel by using star topology wiring. The system performs Grubbs outlier test on the data of the three sensors in each sampling window period, and calculates the weighted geometric mean temperature value after removing the mutation interference points. The vibration sensing array is composed of a three-axis MEMS accelerometer and a piezoelectric film sensor. The hardware layer performs frequency domain integration preprocessing: after the output of the accelerometer is filtered by a fourth-order Butterworth low-pass filter, the spectral energy centroid detection is performed on the time domain waveform to extract the characteristic frequency components of the X / Y / Z axes, and the confidence cross-validation is performed with the resonant frequency band data of the piezoelectric sensor. Finally, the vibration intensity index and the main frequency position parameters are generated.

[0064] S3. The composite optimization calculation module normalizes and fuses the noise index, blur degree data and environmental parameters, and generates a three-dimensional regulation matrix based on the dynamic weight allocation algorithm;

[0065] Furthermore, the composite optimization calculation module sets up a multi-source heterogeneous data fusion interface to flatten the image noise index in the spatial dimension and map it into two-dimensional grid data of the same size as the chip temperature distribution matrix. The normalization process introduces adaptive boundary control: dynamically adjusts the normalization scale factor of the fuzzy degree parameter according to the cross-linking characteristics of the current ambient temperature and vibration main frequency. The dynamic weight allocation algorithm implements a three-level decision-making mechanism: the bottom layer of the hardware state layer focuses on the efficiency curve of the charge pump unit, the middle layer of the image quality layer evaluates the SNR decay rate, and the upper layer of the environmental stress layer constructs a temperature-vibration joint action domain. Dynamically adjusts the base dimension weights of the three-dimensional regulation matrix through the coupling factors output by the three-level decision-making, and finally forms a composite control matrix including frequency tuning priority, voltage gradient sensitivity coefficient, and coupling constraint boundary.

[0066] S4. The adaptive frequency adjustment module analyzes the frequency compensation vector in the three-dimensional regulation matrix and adjusts the phase synchronization characteristics of the image sensor core clock tree through the programmable phase-locked loop array of the LVDS architecture;

[0067] Furthermore, the adaptive frequency adjustment module adopts a hierarchical analysis strategy: first extracts the key phase compensation vector of the main clock channel from the three-dimensional regulation matrix, and then obtains the optimal timing offset of the LVDS differential pair through the auxiliary channel matrix decomposition. The programmable phase-locked loop array implements dual-loop control: after the digital phase detector receives the compensation vector data, the main loop controls the frequency compensation range of the voltage-controlled oscillator array, and the auxiliary loop regulates the phase interpolation step accuracy of the delay-locked loop (DLL). For the fan-out buffer at the end of the clock tree, a phase difference dynamic equalization technology is adopted: automatically adjusts the delay compensation value of the driver according to the load capacitance parameters of each branch to ensure that the propagation delay difference of the clock signal in the sensor pixel array is controlled within a 60ps error band.

[0068] S5. The multi-layer voltage regulation array triggers the PWM controllers of N groups of multi-phase charge pump units to enter the coupled working mode according to the voltage gradient control signal output by the three-dimensional regulation matrix;

[0069] Furthermore, the charge pump units of the multi-layer voltage regulation array work in an interleaved multi-phase mode, and each group of PWM controllers receives the phase allocation identifier and amplitude modulation code in the voltage gradient control signal. When triggering the coupled working mode, the controller group implements three-stage coordination: 1) The phase synchronization stage distributes the reference clock signal to all PWM units through a binary tree structure; 2) The ripple cancellation stage automatically configures the switching timing stagger angle of adjacent charge pump units according to the spectrum planning parameters in the gradient control signal; 3) The dynamic compensation stage dynamically adjusts the maximum allowable change rate of the PWM duty cycle by real-time monitoring the magnetic saturation trend of the inductor through a Hall sensor. This process ensures that under 3A-class load transient conditions, the voltage overshoot amplitude of the power supply rail does not exceed 5% of the nominal value.

[0070] S6. Establish a dual closed-loop negative feedback control chain for the frequency vector and the voltage gradient based on the cross-evaluation results of the image quality degradation rate and the environmental parameter change rate;

[0071] Furthermore, the dual closed-loop negative feedback control chain adopts a heterogeneous data cross-triggering mechanism: the image quality degradation rate is quantified by analyzing the singularity change trend of the noise index matrix and the second derivative of the blur parameter, and the environmental parameter change rate is calculated based on the Laplacian eigenvalue of the temperature gradient and the vibration frequency deviation. The minimum beat control strategy is implemented in the frequency vector closed-loop channel, with the phase noise power spectral density as the adjustment objective function; the sliding mode variable structure control is adopted in the voltage gradient closed-loop channel, and a dynamic constraint boundary is set up to prevent the charge pump from overloading. When the degradation evaluation indexes of the two types of parameters exceed the threshold, the system triggers an emergency enhancement mode: simultaneously shrink the adjustment period of the two closed loops to 1 / 8 of the normal state, and enable the prediction compensation buffer data of the LSTM unit to establish a feedforward-feedback composite control channel until the system returns to the steady-state working area.

[0072] In summary, the present invention realizes the following core efficiency improvements through mechanism innovation:

[0073] An environment-adaptive cross-domain collaborative optimization mechanism: through the real-time fusion analysis of the image quality parameters and the environmental physical quantities by the composite optimization calculation module, a three-dimensional dynamic regulation space under non-steady-state conditions is constructed, breaking through the response bottleneck of the traditional single-dimensional threshold control;

[0074] Enhanced dynamic stability of multi-physical field coupling: Utilize the phase synchronization characteristics of the LVDS architecture and the interleaved regulation of the multi-phase charge pump to achieve a matching offset between the power supply noise spectrum and the clock jitter of the image sensor, significantly suppressing the cascading amplification effect of the common-mode interference;

[0075] Nonlinear hysteresis feedforward compensation ability: Based on the time-varying weight distribution strategy of the LSTM prediction unit, the voltage-frequency linkage response of the system under vibration transient impact has an advanced compensation characteristic, effectively blocking the vicious coupling chain of vibration-temperature-electrical parameters;

[0076] Topological reconstruction characteristics of anti-interference robustness: Through the redundant decision-making mechanism of the dual negative feedback loop, a dual fault-tolerant judgment on the image quality degradation trend and the environmental parameter mutation direction is formed to ensure the control continuity under extreme working conditions.

[0077] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing embodiments or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A dynamic voltage and frequency adaptive adjustment system for in-vehicle camera chips, characterized in that, include: Image quality assessment module, environmental parameter acquisition module, composite optimization calculation module, adaptive frequency adjustment module and multi-layer voltage regulation array; The input end of the image quality assessment module is connected to the image sensor output bus and is equipped with a digital noise analysis unit and a dynamic blur detection unit; The environmental parameter acquisition module includes a temperature monitoring unit and a vibration sensor array; The composite optimization calculation module has a parameter fusion unit and an LSTM-based prediction control unit, which receives the noise index, fuzziness data and environmental parameter data of the image quality assessment module; The adaptive frequency adjustment module is designed based on the LVDS architecture and includes a programmable phase-locked loop array, the output of which is connected to the image sensor core clock tree; The multi-layer voltage regulation array is provided with N groups of multi-phase charge pump units, each group including four mutually coupled PWM controllers.

2. The in-vehicle camera chip dynamic voltage and frequency adaptive regulation system according to claim 1, characterized in that The composite optimization calculation module generates a three-dimensional control matrix through a dynamic weight allocation algorithm, and outputs a frequency compensation vector to the adaptive frequency adjustment module in a time-sharing manner. At the same time, it outputs a voltage gradient control signal to the multi-layer voltage regulation array, forming a dual negative feedback loop based on the image quality degradation rate and the changes in environmental parameters.

3. The in-vehicle camera chip dynamic voltage and frequency adaptive regulation system according to claim 1, wherein Each branch of the programmable phase-locked loop array includes a digital phase detector, a loop filter, and four parallel voltage-controlled oscillator arrays. The phase detection reference clock source of each branch has a programmable 1-8 frequency division ratio. The frequency tuning curve of the voltage-controlled oscillator array is divided into a three-stage compensation structure, using a linear tuning mode below 1.2GHz, a quadratic curve compensation mode between 1.2-2.4GHz, and an exponential compensation mode above 2.4GHz.

4. The in-vehicle camera chip dynamic voltage and frequency adaptive regulation system according to claim 1, characterized in that The dynamic blur detection unit implements a dual-path processing architecture, including a high-frequency component extraction branch based on the Sobel operator and a time-domain analysis branch based on the optical flow motion vector, and the outputs of the two are weighted summed to generate blur data; The digital noise analysis unit implements wavelet packet decomposition processing in the spatial domain and adopts adaptive Kalman filtering in the frequency domain. The final noise index is determined by the geometric mean of the processing results of the two domains.

5. The in-vehicle camera chip dynamic voltage and frequency adaptive regulation system according to claim 2, wherein The memory window length of the LSTM prediction control unit is adjusted in real time according to the ΔQ value. When ΔQ > threshold Q th the window length is shortened to (1 - ΔQ / Q) times the original value max to improve the response speed.

6. Control method of dynamic voltage and frequency adaptive adjustment system for in-vehicle camera chip, characterized in that, The method is implemented using the system according to any one of claims 1 to 5, and includes the following steps: S1. Obtain image data stream from the image sensor output bus through the image quality assessment module, and calculate noise index and fuzziness data using the digital noise analysis unit and the dynamic blur detection unit; S2. The temperature monitoring unit of the environmental parameter acquisition module collects the chip surface temperature distribution data in real time, and simultaneously detects the three-dimensional axial acceleration parameters through the vibration sensor array; S3, a composite optimization calculation module normalizes and fuses the noise index, fuzziness data and environmental parameters, and generates a three-dimensional control matrix based on a dynamic weight allocation algorithm; S4. The adaptive frequency adjustment module analyzes the frequency compensation vector in the three-dimensional control matrix and adjusts the phase synchronization characteristics of the image sensor core clock tree through the programmable phase-locked loop array of the LVDS architecture; S5. The multi-layer voltage regulation array triggers the PWM controllers of N groups of multi-phase charge pump units to enter the coupled working mode according to the voltage gradient control signal output by the three-dimensional regulation matrix; S6. Based on the cross-evaluation result of the image quality degradation rate and the environmental parameter change rate, a dual closed-loop negative feedback control chain for the frequency vector and the voltage gradient is established.

7. The control method of the in-vehicle camera chip dynamic voltage and frequency adaptive regulation system according to claim 6, characterized in that, The dynamic blur detection unit performs dual-path collaborative processing: performing Sobel operator edge sharpness analysis in the spatial domain, and at the same time using optical flow motion vector estimation in the time domain. Finally, the blur data takes the weighted sum of squares of the results of the two paths; the temperature monitoring unit performs redundant data acquisition: the sampling points of three PT100 sensors are evenly distributed in a 120-degree ring along the chip power supply rail, and the sampling period is synchronized with the switching frequency of the charge pump unit in the multi-layer voltage regulation array at an integer multiple of 1:4.

Images