High-frequency electromagnetic protection carbon-based wave-absorbing material dynamic regulation system

Abstract

The application relates to the technical field of electromagnetic wave absorbing materials and devices, and particularly discloses a dynamic regulation and control system of a high-frequency electromagnetic protection carbon-based wave-absorbing material. The system comprises an environment sensing layer, an intelligent decision-making layer and a dynamic execution layer. The environment sensing layer captures and analyzes the frequency spectrum characteristics of incident electromagnetic waves in real time; the intelligent decision-making layer calculates optimal regulation and control parameters according to the environment characteristics based on deep reinforcement learning and impedance matching optimization; and the dynamic execution layer converts the optimal parameters into a physical regulation and control field and applies the field to the material to dynamically adjust the electromagnetic parameters of the material. The application constructs a complete closed-loop feedback control chain from environment sensing, intelligent decision-making to dynamic execution. The environment sensing layer captures the accurate frequency spectrum characteristics of incident electromagnetic waves in real time, the intelligent decision-making layer calculates optimal regulation and control parameters in real time based on deep reinforcement learning and impedance matching theory, and the dynamic execution layer accurately generates a regulation and control field to act on the material.

Description

Technical Field

[0001] This invention belongs to the field of electromagnetic wave absorbing materials and devices, specifically relating to a dynamic control system for high-frequency electromagnetic protection carbon-based absorbing materials. Background Technology

[0002] Electromagnetic protection and stealth technology are key supporting technologies in national defense and electronic warfare. Their core lies in developing high-performance absorbing materials capable of effectively attenuating or absorbing electromagnetic wave energy. Carbon-based absorbing materials have become a research hotspot in this field due to their advantages such as lightweight, wide-band absorption potential, and high designability. Among these, dynamic control technology aims to actively adjust the absorption performance of materials at specific frequency bands by changing their electromagnetic parameters through external excitation, in order to cope with complex and ever-changing electromagnetic environments.

[0003] Existing technologies typically use preset control parameters (such as voltage and temperature) to change the dielectric constant or permeability of carbon-based materials in order to achieve dynamic tuning. However, existing dynamic control systems for carbon-based microwave absorbing materials have significant shortcomings: their control behavior is based on preset fixed parameters or simple timing sequences, lacking the ability to sense and respond to the surrounding electromagnetic environment in real time.

[0004] When the external high-frequency electromagnetic environment undergoes abrupt changes, such as encountering radar pulse interference or communication frequency band switching, the system cannot obtain the spectral characteristics of the current incident electromagnetic wave in real time. This causes its control output to become disconnected from the actual electromagnetic wave impedance matching requirements, resulting in a serious impedance mismatch problem. This mismatch directly causes a sharp decline in the absorption efficiency of materials in critical frequency bands, making it impossible to achieve effective dynamic protection or stealth.

[0005] Therefore, how to construct a dynamic control system that can sense the electromagnetic environment in real time and intelligently adjust the absorption characteristics has become an urgent technical problem to be solved in order to improve the survival and combat capabilities of modern electronic equipment in complex electromagnetic environments. Summary of the Invention

[0006] The purpose of this invention is to provide a dynamic control system for high-frequency electromagnetic protection carbon-based absorbing materials, in order to solve the technical problem in the prior art where the lack of real-time electromagnetic environment perception and feedback capability leads to a disconnect between the control output and the impedance matching requirements of the incident electromagnetic wave, resulting in a decrease in absorption efficiency.

[0007] The technical solution of this invention is a dynamic control system for high-frequency electromagnetic protection carbon-based absorbing materials. This system consists of an environmental perception layer, an intelligent decision-making layer, and a dynamic execution layer, and achieves adaptive optimization of absorbing characteristics through a real-time closed-loop feedback mechanism.

[0008] The environmental awareness layer is used to capture and analyze the spectral characteristics of incident electromagnetic waves in real time. This layer includes a broadband electromagnetic wave receiving array and a high-speed spectrum analysis module. The broadband electromagnetic wave receiving array consists of multiple miniature antenna elements arranged in a matrix, covering a frequency range of 2 GHz to 40 GHz, with each antenna element connected to a low-noise amplifier.

[0009] The high-speed spectrum analysis module receives radio frequency signals from the broadband electromagnetic wave receiving array. First, it converts the analog signals into digital signals through an analog-to-digital converter. Then, it uses a fast Fourier transform algorithm to perform real-time spectrum analysis on the digital signals, extracting the energy distribution spectrum, main frequency, and signal bandwidth of the current incident electromagnetic wave in the preset frequency band, and outputting these parameters as environmental feature vectors.

[0010] The intelligent decision-making layer is used to calculate and output the optimal combination of control parameters in real time based on environmental feature vectors. This layer includes a deep reinforcement learning decision engine and an impedance matching optimizer. The deep reinforcement learning decision engine has a built-in pre-trained deep neural network model. The input of this model is the environmental feature vector output by the environmental perception layer, and the output is the preliminary control strategy for the carbon-based absorbing material.

[0011] This deep neural network model is constructed through a combination of offline training and online fine-tuning. During the offline training phase, a supervised learning algorithm is used to initialize and train the network weights using a historical dataset containing a large number of electromagnetic environment scenarios and corresponding optimal control parameters.

[0012] During the online fine-tuning phase, the system continuously collects environmental status, execution actions (i.e., control parameters), and reward signals fed back by the impedance matching optimizer during operation. It uses a near-end strategy optimization algorithm to dynamically update the network model to adapt to the non-stationary changes in the electromagnetic environment.

[0013] The impedance matching optimizer receives the initial control strategy output by the deep reinforcement learning decision engine and performs calculations based on transmission line theory. Its core objective is to solve an optimization problem that minimizes the surface reflection coefficient of the absorbing material. Specifically, based on the predicted electromagnetic parameters of the material corresponding to the initial control strategy, and combined with the dominant frequency in the current environmental feature vector, the theoretical input impedance of the material at that frequency is calculated.

[0014] Furthermore, the theoretical input impedance is compared with the free-space wave impedance to calculate its reflection coefficient magnitude. The impedance matching optimizer incorporates a gradient descent algorithm, using the reflection coefficient magnitude as the loss function and the control parameters as variables for iterative optimization until the reflection coefficient magnitude falls below a preset matching threshold. At this point, the corresponding combination of control parameters is determined as the optimal combination and output.

[0015] The dynamic execution layer converts the optimal combination of control parameters output by the intelligent decision-making layer into a physical control field, which is then applied to the carbon-based absorbing material. This layer comprises a multi-channel power drive module and a field generation array. The multi-channel power drive module receives the optimal combination of control parameters in digital form, which includes at least voltage amplitude, voltage frequency, and duty cycle parameters.

[0016] This module contains multiple independent power channels, each generating a corresponding high-precision analog voltage or current signal according to assigned parameters. The field generation array is electrically connected to each output channel of the multi-channel power drive module, and it consists of microelectrode pairs or microcoils embedded inside a carbon-based absorbing material substrate or closely attached to the surface of the material.

[0017] When a voltage or current signal is applied from the multi-channel power drive module, the field generation array generates a controllable electric or magnetic field in a localized region of the carbon-based absorbing material. This electric or magnetic field alters the carrier mobility and interfacial polarization characteristics of conductive fillers such as carbon nanotubes or graphene in the material, thereby achieving dynamic and continuous adjustment of the real and imaginary parts of the material's complex permittivity, ultimately matching the material's input impedance with the impedance of the current incident electromagnetic wave.

[0018] In one embodiment of the present invention, the miniature antenna element of the broadband electromagnetic wave receiving array adopts a log-periodic antenna structure to ensure stable gain and radiation pattern across a wide bandwidth. The fast Fourier transform processing frame length of the high-speed spectrum analysis module is set to 1024 points, and the sampling rate is set to more than twice the highest frequency to be analyzed according to the Nyquist theorem, so as to achieve accurate analysis of signals below 80 GHz.

[0019] As one embodiment of the present invention, the deep neural network model in the deep reinforcement learning decision engine adopts an architecture that integrates convolutional neural networks and long short-term memory networks. The convolutional neural network is used to process the spectral energy distribution image data in the environmental feature vector to extract spatial features.

[0020] Long Short-Term Memory (LSTM) networks are used to process time-varying parameters such as dominant frequency and bandwidth, extracting time-dependent features. The output features of the two networks are concatenated in a fusion layer, and then mapped to a preliminary control strategy through three fully connected layers.

[0021] In one embodiment of the present invention, the impedance matching optimizer has a preset matching threshold of 0.1, i.e., the reflection coefficient is below -10 dB. The gradient descent algorithm employs an adaptive moment estimation algorithm, with an initial learning rate set to 0.001, and a setting is made so that when the loss function decreases by less than 1× dB for 5 consecutive iterations... When the iteration terminates, the current control parameters are output.

[0022] In one embodiment of the present invention, the multi-channel power drive module has four power supply channels, an output voltage range of ±200 volts, and an output current accuracy in the milliampere range. The micro-electrode pairs in the field generation array adopt an interdigitated electrode structure with an electrode width of 10 micrometers, an electrode spacing of 15 micrometers, and are made of gold. They are fabricated on a flexible polyimide film using photolithography and then integrated with a carbon-based absorbing material.

[0023] In one embodiment of the invention, the system further includes a health monitoring and calibration module. This module periodically applies a set of known test signals to the carbon-based absorbing material and measures its response through an environmental sensing layer. Furthermore, the measured response is compared with the expected response model of the material under standard conditions to calculate the performance drift.

[0024] When the performance drift exceeds the preset tolerance range, the health monitoring and calibration module sends a calibration command to the intelligent decision layer, triggering the deep reinforcement learning decision engine to fine-tune the model online using new test data, or adjust the basic values ​​of material parameters in the impedance matching optimizer to compensate for the performance degradation of the material caused by aging or environmental factors.

[0025] In one embodiment of the present invention, the system adopts a hierarchical distributed control architecture. The environmental perception layer, intelligent decision-making layer, and dynamic execution layer are three independent hardware computing nodes that interact with each other via a high-speed serial bus. Each node uses a multi-core processor to process data in parallel, and the communication protocol between nodes adopts a bus protocol based on a time-triggered mechanism to ensure the determinism and real-time performance of control command transmission in environments with strong electromagnetic interference. The overall system response time is less than 1 millisecond.

[0026] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention constructs a complete closed-loop feedback control chain from environmental perception and intelligent decision-making to dynamic execution. The environmental perception layer captures the precise spectral characteristics of the incident electromagnetic wave in real time, the intelligent decision-making layer calculates the optimal control parameters in real time based on deep reinforcement learning and impedance matching theory, and the dynamic execution layer accurately generates a control field that acts on the material. This closed-loop mechanism ensures that the system can respond to sudden changes in the external electromagnetic environment in real time, always adjusting the input impedance of the material to match the current incident wave, fundamentally solving the problems of impedance mismatch and decreased absorption efficiency caused by the lack of environmental perception, and achieving true adaptive dynamic protection.

[0027] 2. This invention introduces deep reinforcement learning as the core decision engine, enabling the system to learn online and adapt to non-stationary electromagnetic environments. The deep neural network model acquires basic decision-making capabilities through offline training, and then continuously optimizes its strategy through online fine-tuning. The impedance matching optimizer, starting from the physical essence of electromagnetic wave transmission, performs precise mathematical optimization and verification of the decision results. This dual decision-making mechanism, combining data-driven and physical model approaches, ensures both the intelligence and adaptability of the decisions, while also guaranteeing the physical feasibility and optimality of the decision results through physical constraints. This significantly improves the system's robustness and control accuracy in complex and unknown electromagnetic scenarios.

[0028] 3. This invention employs a multi-channel independent programmable drive and a miniaturized field generation array design at the dynamic execution layer. Multi-channel drive allows for the application of differentiated control parameters to different regions of the material, providing a hardware foundation for achieving surface impedance gradient distribution and expanding absorption bandwidth. The miniaturized electrode or coil array is highly integrated with the material, ensuring that the control field can act efficiently and uniformly on the material's microstructure, enabling rapid and continuous adjustment of dielectric parameters. Simultaneously, the system's health monitoring and calibration module automatically diagnoses and compensates for material performance drift. The hierarchical distributed architecture ensures strong real-time performance and anti-interference capabilities, collectively forming a highly reliable and maintainable engineering system solution. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the overall technical solution architecture of the dynamic control system for high-frequency electromagnetic protection carbon-based absorbing materials proposed in this invention; Figure 2 This is a schematic diagram of the core principle framework of the intelligent decision-making layer based on deep reinforcement learning and impedance matching optimization in this invention. Figure 3 This is a flowchart illustrating the logical flow of the environmental perception layer in this invention, from signal capture to feature extraction. Figure 4 This is a schematic diagram illustrating the working principle of the dynamic execution layer in this invention, which converts control parameters into the electromagnetic properties of materials. Figure 5 This is a schematic diagram illustrating the multi-level interaction between system closed-loop feedback and health monitoring calibration in this invention. Detailed Implementation

[0030] Please refer to Figures 1 to 5 This invention provides a dynamic control system for high-frequency electromagnetic shielding carbon-based absorbing materials. This system aims to solve the technical problem of decreased absorption efficiency caused by a lack of real-time electromagnetic environment perception and feedback capabilities, resulting in a disconnect between the control output and the impedance matching requirements of the incident electromagnetic wave.

[0031] The core of the system lies in constructing a complete closed loop from environmental perception and intelligent decision-making to dynamic execution. This allows for the dynamic and continuous adjustment of the complex permittivity of the carbon-based absorbing material, enabling its input impedance to adaptively match the changing incident electromagnetic wave impedance, thereby maintaining and optimizing high-frequency electromagnetic protection performance. Please refer to the appendix. Figure 1 The system consists of an environmental perception layer, an intelligent decision-making layer, and a dynamic execution layer, forming a tightly coupled real-time closed loop via a high-speed data bus. The system also integrates a health monitoring and calibration module and employs a hierarchical distributed control architecture to ensure strong real-time performance and high reliability.

[0032] The environmental perception layer serves as the data input for the entire system. Its core function is to capture and analyze the spectral characteristics of electromagnetic waves incident on the surface of the carbon-based absorbing material in real time. This layer provides accurate and quantifiable environmental state input for subsequent intelligent decision-making. Please refer to the appendix. Figure 3 The environmental perception layer includes a broadband electromagnetic wave receiving array and a high-speed spectrum analysis module. The broadband electromagnetic wave receiving array is responsible for converting electromagnetic wave energy in space into processable electrical signals.

[0033] The array consists of multiple miniature antenna elements arranged in a matrix. These elements are regularly distributed at specific intervals in front of the area to be protected or integrated into the material surface to ensure representative spatial sampling of the incident wave field. Each miniature antenna element employs a log-periodic antenna structure. This design ensures stable gain and radiation pattern characteristics across a wide frequency band from 2 GHz to 40 GHz, avoiding drastic fluctuations in signal reception sensitivity due to frequency variations.

[0034] Each antenna element is directly connected to a low-noise amplifier at its rear end. This amplifier typically has a noise figure below 2 dB and an adjustable gain range of 20 to 40 dB. The primary function of the low-noise amplifier is to initially amplify the weak received signal at the beginning of the signal link, thereby improving the signal-to-noise ratio relative to the inherent noise of subsequent circuits. Its gain can be dynamically adjusted by the system based on the ambient electromagnetic field strength. The RF signals received and initially amplified by all antenna elements are then synchronously transmitted to the high-speed spectrum analysis module via a well-shielded coaxial cable or microwave transmission line.

[0035] The high-speed spectrum analysis module is the core signal processing unit of the environmental perception layer. Its task is to convert analog radio frequency signals into digital spectral characteristic parameters. This module receives parallel analog signals from all channels of the broadband electromagnetic wave receiving array.

[0036] First, the multi-channel analog-to-digital converter array inside the module synchronously samples and quantizes the input analog signal. The sampling rate of the analog-to-digital converter is a key parameter. According to the Nyquist theorem, in order to analyze signals with a maximum frequency of 40 GHz without distortion, the sampling rate must be set to 80 GHz or higher.

[0037] In this embodiment, the sampling rate is configured to 100 GHz, providing ample frequency margin for signal processing. The analog-to-digital converter (ADC) resolution is set to 12 bits to achieve a balance between dynamic range and quantization accuracy. After ADC conversion, the continuous analog voltage signal is converted into a discrete digital sequence.

[0038] Subsequently, the digital signal processor within the high-speed spectrum analysis module applies a Fast Fourier Transform (FFT) algorithm to each frame of the digital signal. The processing frame length for the FFT is fixed at 1024 points, which determines the frequency resolution.

[0039] At a sampling rate of 100 GHz, the frequency resolution corresponding to the 1024-point transform is approximately 97.66 MHz, which is sufficient to distinguish the bandwidth characteristics of most communication or radar signals. The digital signal processor performs preprocessing operations such as windowing and overlapping on each frame of data to reduce spectral leakage, and then calculates its Fast Fourier Transform result to obtain the complex spectrum of the signal in the DC to 50 GHz Nyquist frequency range.

[0040] Based on the calculated complex spectrum, the high-speed spectrum analysis module performs a feature extraction process. This process first calculates the amplitude spectrum or power spectral density of the spectrum. Then, the algorithm scans the entire analysis frequency band and identifies all spectral peaks whose amplitude exceeds a preset noise threshold.

[0041] For each significant spectral peak, its center frequency (dominant frequency) and half-power bandwidth (signal bandwidth) are recorded. Simultaneously, the algorithm integrates or performs segmented statistical analysis on the power spectral density across the entire preset frequency band, such as 2 GHz to 40 GHz, generating an energy distribution spectrum describing the distribution of electromagnetic energy within different sub-bands.

[0042] This energy distribution spectrum can be represented as a vector, where each element represents a normalized energy value for a specific frequency band. Finally, the high-speed spectrum analysis module encapsulates the extracted core parameters into a structured environmental feature vector. This vector contains at least the following fields: The vector represents the energy distribution spectrum data as an array, the main frequency as one or more floating-point numbers, and the signal bandwidth as a corresponding floating-point number. Additionally, the vector may contain metadata such as timestamps and array channel consistency check codes. This environmental feature vector is sent to the intelligent decision-making layer in real time via a high-speed data interface, serving as the primary input to the decision-making process. The entire processing latency from signal acquisition to feature vector output is strictly controlled within 100 microseconds to meet the system's stringent real-time requirements.

[0043] The intelligent decision-making layer is the brain of the system, responsible for calculating and outputting the optimal combination of control parameters that enables the carbon-based absorbing material to achieve the best impedance matching state, based on the real-time environmental feature vectors provided by the environmental perception layer. Please refer to the appendix. Figure 2 The intelligent decision-making layer includes a deep reinforcement learning decision engine and an impedance matching optimizer, which work in series to form a dual decision-making mechanism that combines data-driven decision-making with physical model verification.

[0044] The core of a deep reinforcement learning decision engine is a pre-trained deep neural network model. The input dimension of this model strictly corresponds to the dimension of the environmental feature vector, and the output is a preliminary control strategy for carbon-based absorbing materials. The preliminary control strategy is a parameter vector, whose elements may include the real part offset, the imaginary part offset, or a reference value directly corresponding to the driving signal of the dynamic execution layer.

[0045] To effectively handle different types of features in the environmental feature vector, the deep neural network in this embodiment adopts an architecture that integrates convolutional neural networks and long short-term memory networks. Specifically, the energy distribution spectrum data in the environmental feature vector is rearranged into a two-dimensional matrix form, simulated as a spectrum image, and input into a three-layer convolutional neural network.

[0046] The first convolutional network layer uses 32 3×3 convolutional kernels and a modified linear unit activation function to extract local pattern features in the spectrum, such as narrowband interference and broadband background. The second and third convolutional networks gradually increase the number of convolutional kernels to 64 and 128, respectively, while using max pooling layers to reduce the spatial size of the feature maps in order to extract more abstract and global spectral features.

[0047] On the other hand, scalar parameters such as the dominant frequency and signal bandwidth in the environmental feature vector, along with similar parameters from recent historical moments, are organized into a time series and input into a two-layer long short-term memory network. Each hidden layer of the long short-term memory network contains 128 memory units, whose function is to capture the trends and short-term dependencies of electromagnetic environmental parameters over time, such as the frequency hopping mode of frequency-agile radar or the periodic occurrence of interference signals.

[0048] The flattened feature vector output by the convolutional neural network and the feature vector output at the last time step of the long short-term memory network are concatenated in a fusion layer to form a joint feature vector. This joint feature vector is then subjected to nonlinear transformation and dimensionality reduction through three fully connected layers. The first fully connected layer contains 256 neurons, the second contains 128 neurons, and the final output layer has the same number of neurons as the initial control strategy, using a hyperbolic tangent activation function to limit the output value to the range of -1 to +1. This deep neural network model is constructed and optimized through a combination of offline training and online fine-tuning.

[0049] The offline training phase is completed before system deployment. The network weights are initialized and trained using a historical dataset containing numerous electromagnetic environment scenarios and corresponding optimal control parameters. This historical dataset is obtained through simulation or physical experiments; each data sample includes an environmental feature vector and a label for the optimal control parameter derived from expert knowledge or optimization algorithms.

[0050] Training employs a supervised learning algorithm, specifically an adaptive moment estimation algorithm, with mean squared error as the loss function. This loss function minimizes the gap between the network's predicted control parameters and the true labels. Training continues until the model's prediction error on independent validation sets falls below a preset threshold. At this point, the model possesses the basic mapping ability from environmental features to control strategies.

[0051] The online fine-tuning phase continues throughout system operation and is crucial for the system's adaptation to non-stationary electromagnetic environments. Within each control cycle, the system collects empirical tuples, including the current environmental state (environmental characteristic vector), the executed action (the final applied control parameters), and the reward signal fed back from the impedance matching optimizer. The reward signal is a scalar value calculated based on the optimal reflection coefficient obtained by the impedance matching optimizer; a lower reflection coefficient results in a higher reward value. These empirical tuples are stored in an experience playback buffer.

[0052] Deep reinforcement learning decision engines periodically sample a batch of empirical data from a buffer and use a proximal policy optimization algorithm to update the network model's policy. The proximal policy optimization algorithm improves learning stability while encouraging the model to explore better control policies by constraining the step size of policy updates. Its core objective function is to maximize the expected reward while limiting the difference between the old and new policies.

[0053] The algorithm's update process involves calculating the estimated advantage function and solving an optimization problem for a pruned surrogate objective function. Through continuous online fine-tuning, the deep neural network model can gradually adapt to novel electromagnetic interference patterns or slow drifts in material properties not present in historical data, achieving self-evolution of decision-making capabilities.

[0054] The impedance matching optimizer receives the initial control strategy from the deep reinforcement learning decision engine and performs precise mathematical optimization based on the physical principles of electromagnetic wave propagation. Its core task is to solve an optimization problem aimed at minimizing the reflection coefficient of the absorbing material surface, ensuring the physical optimality of the decision result.

[0055] The impedance matching optimizer internally maintains a material electromagnetic parameter response model. This model describes the mapping relationship between the control parameters of the dynamic execution layer and the effective complex permittivity of the carbon-based absorbing material. Upon receiving the initial control strategy, the optimizer first calculates the predicted values ​​of the real and imaginary parts of the material's complex permittivity based on this model.

[0056] Subsequently, by combining the dominant frequency identified in the current environmental feature vector, the theoretical input impedance of the material at that specific frequency is calculated. The input impedance calculation is based on transmission line theory; for a single-layer absorbing material, the input impedance formula is: ; in, Indicates the input impedance of the material. This represents the intrinsic impedance of the material itself. This indicates that the free-space wave impedance is approximately 377 ohms. It is the imaginary unit. It is the dominant frequency of the current incident electromagnetic wave. It is the thickness of the absorbing material. It is the speed of light in a vacuum. and These are the relative complex permittivity and complex permeability of the material, respectively. For carbon-based materials where electrical losses are dominant, it is usually assumed that... It is approximately 1. In this formula, This is the complex permittivity predicted by the initial control strategy.

[0057] Obtain the theoretical input impedance Then, the impedance matching optimizer calculates its impedance relative to the free-space wave impedance. Reflectance coefficient between The calculation formula is: The goal of optimization is to maximize the magnitude of the reflection coefficient. Minimize. The impedance matching optimizer will Defined as a loss function, the initial control strategy output by the deep reinforcement learning decision engine is used as the initial point, and a numerical optimization algorithm is used for iterative optimization.

[0058] In this embodiment, an adaptive moment estimation algorithm is used as the gradient descent optimizer. The adaptive moment estimation algorithm combines the advantages of momentum method and adaptive learning rate adjustment, enabling efficient handling of non-convex optimization problems. The optimizer sets the initial learning rate to 0.001 and monitors the change in the loss function. In each iteration, the algorithm calculates the gradient of the loss function with respect to the adjustment parameters and updates the estimated values ​​of the parameters. The iterative process continues until one of the following termination conditions is met: the magnitude of the reflection coefficient... A match value below the preset matching threshold of 0.1 corresponds to -10 dB, or the absolute value of the decrease in the loss function over five consecutive iterations is less than 1 × 10⁻⁶. -5 .

[0059] When the iteration terminates, the current combination of control parameters is determined as the optimal combination of control parameters within the current control cycle. This optimal combination of control parameters is a digital vector, the content of which is formatted and sent to the dynamic execution layer in real time via a high-speed communication interface. Simultaneously, the final reflection coefficient magnitude is converted into a reward signal and fed back to the deep reinforcement learning decision engine for its online fine-tuning. The overall computational latency of the intelligent decision layer, from receiving the environmental feature vector to outputting the optimal combination of control parameters, is optimized to within 500 microseconds.

[0060] The dynamic execution layer is the system's execution terminal, responsible for accurately and rapidly converting the optimal combination of control parameters output by the intelligent decision-making layer into a physical control field, which is then applied to the carbon-based absorbing material, thereby actually altering the material's electromagnetic properties. Please refer to the appendix. Figure 4 The dynamic execution layer includes a multi-channel power drive module and a field generation array.

[0061] The multi-channel power drive module receives the optimal combination of control parameters from the intelligent decision-making layer. This parameter combination typically includes multiple independent control quantities; for example, in voltage control mode, it may include multiple parameters such as voltage amplitude, voltage frequency, duty cycle, and phase. The multi-channel power drive module internally contains multiple independent power supply channels.

[0062] In this embodiment, the number of channels is set to four, which allows the system to independently or collaboratively control different regions or polarization directions of the absorbing material. Each power channel is a complete closed-loop control unit, the core of which is a high-precision digital-to-analog converter and power amplifier circuit controlled by a digital signal processor. The digital signal processor analyzes the parameters assigned to this channel and generates a corresponding digital control sequence. The digital-to-analog converter converts this sequence into an analog voltage or current reference signal, typically with a resolution of 16 bits.

[0063] The power amplifier circuit amplifies the reference signal to the required power level. In this embodiment, the power supply channel output voltage range is ±200 volts, the output current capability reaches 100 mA, and the current output accuracy is in the milliampere range. Each channel has overvoltage, overcurrent, and overheat protection functions, and the output status is monitored in real time. A local feedback loop ensures the consistency between the output signal and the command parameters, with a steady-state error of less than 5 / 1000. The start-up, stop, and parameter updates of all channels are controlled by a unified synchronization signal to ensure that the control field applied to the material is precisely aligned in time.

[0064] The field generation array serves as the physical interface between the multi-channel power drive module and the carbon-based absorbing material, converting electrical signals into electric or magnetic fields in space. The field generation array is electrically connected to each output channel of the multi-channel power drive module via low-impedance wires. The array consists of numerous microelectrode pairs or microcoils, these microstructures embedded in a specific pattern within the substrate of the carbon-based absorbing material or tightly adhered to the material surface. In this embodiment, micro-interdigital electrode pairs are used as the field generation units. The electrode material is gold due to its excellent conductivity and chemical stability.

[0065] Electrodes are fabricated on a flexible polyimide film using microfabrication processes such as photolithography, evaporation, and lift-off to form independent polarization modules. Each module contains dozens of pairs of interdigitated electrodes, with each electrode being 10 micrometers wide and the spacing between adjacent electrodes being 15 micrometers. This fine scale allows the generated electric field to be highly localized and deeply influence the surface microstructure of the material. The fabricated polarization modules are then integrated with carbon-based microwave absorbing materials via conductive adhesive bonding or hot pressing, ensuring low contact resistance and good mechanical adhesion between the electrodes and the material.

[0066] When the multi-channel power drive module applies a high-voltage signal to these miniature interdigitated electrode pairs, a high-intensity, non-uniform electric field is generated between the electrodes. This electric field directly acts on the conductive fillers in the carbon-based absorbing material, such as carbon nanotubes, graphene sheets, or carbon black particles. The electric field effect alters the electromagnetic properties of the material through multiple microscopic mechanisms.

[0067] First, a strong electric field can change the mobility of charge carriers inside the conductive filler, affecting its conductive path and thus changing the material's conductivity. This is directly related to the imaginary part of the complex permittivity, i.e., the loss factor.

[0068] Secondly, the electric field induces interfacial polarization at the interface between the filler and the insulating matrix. This polarization relaxation process is an important mechanism for material loss in the microwave band. Adjusting the electric field can change the polarization intensity and relaxation time.

[0069] Furthermore, for oriented fillers such as carbon nanotubes, an electric field can cause minute rotations or changes in orientation, altering the overall anisotropy of the material. The macroscopic manifestation of these microscopic changes is the tunability of the real and imaginary parts of the material's complex permittivity. By changing the intensity and frequency of the applied electric field in real time, the dynamic execution layer can achieve dynamic and continuous adjustment of the material's electromagnetic parameters.

[0070] Ultimately, this adjustment precisely pushes the material's input impedance to a state that matches the current free-space wave impedance, minimizing the reflection coefficient and thus maximizing the absorption of incident electromagnetic waves at that frequency. The dynamic execution layer's response time from receiving digital parameters to establishing a stable control field is less than 200 microseconds.

[0071] To ensure the reliability and performance stability of the system during long-term operation, this invention also includes a health monitoring and calibration module. Please refer to the appendix. Figure 5 This module operates independently of the main control loop, performing diagnostic and calibration tasks periodically at a lower frequency. At the start of a monitoring cycle, the health monitoring and calibration module sends instructions to the dynamic execution layer via the system bus, commanding it to apply a predefined sequence of test signals with known amplitudes and frequencies to the carbon-based absorbing material.

[0072] These test signals cover the material's main operating frequency band and modulation range. Simultaneously, the module notifies the environmental sensing layer to prepare for acquiring the material's response to the test signals. The environmental sensing layer operates in a high-sensitivity mode, measuring changes in the electromagnetic wave signal reflected or transmitted from the material surface when the test signal is applied. The health monitoring and calibration module collects these measurement response data and compares them with a standard response model stored internally.

[0073] The standard response model describes the expected response curve to the same set of test signals when the material is in its factory or calibration state. The comparison process is achieved by calculating the differences between the measured response and the expected response at key characteristic points such as absorption peak frequency, absorption depth, and bandwidth, quantifying one or more performance drift indicators. When the calculated performance drift exceeds the preset tolerance range, it indicates that the material's electromagnetic parameter baseline may have undergone a significant change due to aging, temperature cycling, mechanical stress, or chemical corrosion. At this point, the health monitoring and calibration module sends a calibration instruction package to the intelligent decision layer.

[0074] The instruction package contains trigger event identifiers and relevant test data. Upon receiving the instruction, the intelligent decision layer may take two calibration actions. The first is to trigger the deep reinforcement learning decision engine, using the new test data as additional training samples to accelerate its online fine-tuning process, enabling the model to quickly adapt to the new properties of the material. The second is to directly adjust the fundamental parameter values ​​in the material electromagnetic parameter response model within the impedance matching optimizer, such as updating the coefficients in the relationship between the dielectric constant and the control voltage, thereby compensating for baseline drift in material performance. Through this periodic self-checking and calibration, the system can maintain long-term control accuracy and counteract the negative impacts of time-varying material properties.

[0075] The system's hardware and control architecture adopts a layered distributed design to meet the extreme requirements of determinism, real-time performance, and reliability in environments with strong electromagnetic interference. The environmental perception layer, intelligent decision-making layer, dynamic execution layer, and health monitoring and calibration module each serve as independent hardware computing nodes. Each node is equipped with a dedicated multi-core processor or field-programmable gate array (FPGA) to handle the core algorithms and tasks of its layer, and has independent power supply, clock, and memory.

[0076] Nodes interact with each other via a high-speed serial bus. In this embodiment, a bus system based on the time-triggered Ethernet protocol is used. The time-triggered Ethernet protocol divides the communication cycle into fixed time windows. Each node sends and receives data within a pre-allocated time slot. This mechanism ensures that even in the presence of random electromagnetic interference causing occasional noise, the transmission of critical control commands and data has strict determinism and predictable low latency. The time of all nodes is synchronized by a global clock on the bus, with a synchronization accuracy within 1 microsecond.

[0077] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0078] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A dynamic control system for high-frequency electromagnetic protection carbon-based absorbing materials, characterized in that, include: The environment perception layer is used to capture and analyze the spectral characteristics of incident electromagnetic waves in real time to output an environment feature vector. The environmental perception layer includes a broadband electromagnetic wave receiving array and a high-speed spectrum analysis module. The intelligent decision-making layer is used to calculate and output the optimal combination of control parameters in real time based on the environmental feature vector; The intelligent decision-making layer includes a deep reinforcement learning decision engine and an impedance matching optimizer. A dynamic execution layer is used to convert the optimal combination of control parameters into a physical control field and apply it to the carbon-based microwave absorbing material. The dynamic execution layer includes a multi-channel power drive module and a field generation array; The deep reinforcement learning decision engine has a built-in pre-trained deep neural network model, which is used to receive environmental feature vectors and output preliminary control strategies for carbon-based microwave absorbing materials. The deep neural network model is constructed by combining offline training with online fine-tuning; The impedance matching optimizer is used to receive the preliminary modulation strategy and perform calculations based on transmission line theory; The calculation process is as follows: based on the predicted values ​​of the material's electromagnetic parameters corresponding to the preliminary control strategy, and combined with the dominant frequency in the current environmental feature vector, the theoretical input impedance of the material at that frequency is calculated. The theoretical input impedance is compared with the free-space wave impedance, and its reflection coefficient magnitude is calculated. The reflection coefficient magnitude is used as the loss function, and the control parameters are used as variables for iterative optimization until the reflection coefficient magnitude is lower than the preset matching threshold. At this point, the corresponding combination of control parameters is determined as the optimal combination of control parameters and output.

2. The dynamic control system for high-frequency electromagnetic protection carbon-based absorbing materials according to claim 1, characterized in that, A broadband electromagnetic wave receiving array consists of multiple miniature antenna elements arranged in a matrix, used to receive radio frequency signals; The high-speed spectrum analysis module is used to receive radio frequency signals from the broadband electromagnetic wave receiving array, convert analog signals into digital signals through an analog-to-digital converter, and perform real-time spectrum analysis on the digital signals using a fast Fourier transform algorithm to extract the energy distribution spectrum, dominant frequency, and signal bandwidth of the current incident electromagnetic wave within a preset frequency band, and output these parameters as environmental feature vectors.

3. The dynamic control system for high-frequency electromagnetic protection carbon-based absorbing materials according to claim 1, characterized in that, The multi-channel power drive module is used to receive the optimal combination of control parameters in digital form and generate corresponding high-precision analog voltage or current signals based on the parameters. The field generation array is electrically connected to the multi-channel power drive module and consists of micro-electrode pairs or micro-coils embedded inside the carbon-based absorbing material substrate or closely attached to the surface of the material. It is used to generate a controllable electric field or magnetic field in a local area of ​​the carbon-based absorbing material when a voltage or current signal is applied, so as to change the carrier mobility and interface polarization characteristics of the conductive filler in the material, thereby realizing the dynamic adjustment of the real part and imaginary part of the complex permittivity of the material.

4. The dynamic control system for high-frequency electromagnetic protection carbon-based absorbing materials according to claim 1, characterized in that, The micro-antenna unit of the broadband electromagnetic wave receiving array adopts a log-periodic antenna structure; the fast Fourier transform processing frame length of the high-speed spectrum analysis module is set to 1024 points, and the sampling rate is set to more than twice the highest frequency to be analyzed according to the Nyquist theorem.

5. The dynamic control system for high-frequency electromagnetic protection carbon-based absorbing materials according to claim 1, characterized in that, The deep reinforcement learning decision engine adopts a deep neural network model that integrates convolutional neural networks and long short-term memory networks. The convolutional neural network is used to process the spectral energy distribution image data in the environmental feature vector. The long short-term memory network is used to process the frequency and bandwidth temporal variation parameters of the main frequency point. The output features of the two networks are concatenated in the fusion layer and then mapped to the initial control strategy through three fully connected layers.

6. The dynamic control system for high-frequency electromagnetic protection carbon-based absorbing materials according to claim 3, characterized in that, The impedance matching optimizer has a preset matching threshold of 0.1; the gradient descent algorithm used in the iterative optimization is an adaptive moment estimation algorithm, with an initial learning rate set to 0.001, and a setting that if the loss function decreases by less than 1× for 5 consecutive iterations... When the iteration ends, the iteration is terminated.

7. The dynamic control system for high-frequency electromagnetic protection carbon-based absorbing materials according to claim 1, characterized in that, The multi-channel power drive module includes four independent power supply channels with an output voltage range of ±200 volts and an output current accuracy in the milliampere range. The micro-electrode pairs in the field generation array adopt an interdigitated electrode structure with an electrode width of 10 micrometers, an electrode spacing of 15 micrometers, and gold as the electrode material. They are fabricated on a flexible polyimide film using photolithography.

8. The dynamic control system for high-frequency electromagnetic protection carbon-based absorbing materials according to claim 1, characterized in that, It also includes a health monitoring and calibration module; the health monitoring and calibration module is used to periodically apply a set of known test signals to the carbon-based absorbing material and measure its response through an environmental sensing layer; The measured response is compared with the expected response model of the material under standard conditions to calculate the performance drift. When the performance drift exceeds the preset tolerance range, a calibration command is sent to the intelligent decision layer to trigger the deep reinforcement learning decision engine to perform online fine-tuning or adjust the basic values ​​of the material parameters in the impedance matching optimizer.

9. The dynamic control system for high-frequency electromagnetic protection carbon-based absorbing materials according to claim 1, characterized in that, The system adopts a hierarchical distributed control architecture; the environment perception layer, intelligent decision-making layer and dynamic execution layer are three independent hardware computing nodes that interact with each other through a high-speed serial bus; the communication protocol between nodes adopts a bus protocol based on a time-triggered mechanism, and the overall system response time is less than 1 millisecond.

Images