Radio spectrum sensing based drone active positioning and jamming system

Abstract

This application discloses an active localization and jamming system for unmanned aerial vehicles (UAVs) based on radio spectrum sensing, belonging to the field of UAV technology. It integrates radio spectrum sensing and long-range non-contact sensing to achieve collaborative sensing of the target UAV's launch hardware characteristics and dynamic physical state. This overcomes the limitations of traditional countermeasures that rely solely on signal domain features, enabling the acquisition of real-time physical state information of the target UAV under complex maneuvers. By constructing a collaborative evolution model across physical quantities and performing joint calculations, the system can infer the target UAV's current maneuver state, load distribution state, and hardware operating point, thereby improving the accuracy of target identification and localization. Furthermore, based on these joint inference results, the system can generate hardware-oriented jamming strategies that match the target UAV's real-time weaknesses, effectively avoiding the inefficiency and false positives of general-purpose jamming, and improving the effectiveness and targeting of countermeasures.

Description

Technical Field

[0001] This application relates to the field of unmanned aerial vehicle (UAV) technology, and in particular to an active localization and jamming system for UAVs based on radio spectrum sensing. Background Technology

[0002] Currently, the field of drone countermeasures mainly relies on radio spectrum sensing technology to detect and locate target drones. These technologies typically calculate location based on parameters such as the time difference of arrival and angle of arrival of the drone's transmitted signals, and identify targets by analyzing characteristics such as signal modulation and frequency drift.

[0003] However, the aforementioned technologies are still limited to the analysis and processing of the UAV's transmitted signals, and the extracted features are all derived from within the signal domain. Since signal domain features only reflect the UAV's transmitting hardware characteristics and cannot obtain real-time physical state information of the target during flight, jamming systems struggle to achieve accurate target identification and positioning when facing individual differences among UAVs of the same model, complex maneuvers, and physical layer deception signals. Furthermore, due to the lack of awareness of the target's propulsion system's operating status, existing countermeasures typically employ general-purpose jamming, failing to differentiate attacks based on the UAV's hardware weaknesses and real-time load status, resulting in low jamming efficiency and a high risk of collateral damage.

[0004] To address the aforementioned issues, a method for active localization and jamming of unmanned aerial vehicles (UAVs) that can integrate multi-dimensional physical information and achieve collaborative perception of launch hardware characteristics and dynamic physical state is needed to improve the target recognition accuracy and jamming effectiveness of the jamming system. Summary of the Invention

[0005] This application provides an active localization and jamming system for unmanned aerial vehicles (UAVs) based on radio spectrum sensing. The technical solution is as follows: On the one hand, a UAV active localization and jamming system based on radio spectrum sensing is provided, the system including a processor and a memory, the processor being configured to perform the following steps: By sensing the radio spectrum, the target transmission signal of the target UAV is acquired, and the power amplifier response of the target transmission signal is modeled to extract a first hardware feature set to characterize the inherent nonlinear characteristics of the target UAV's transmission hardware. In parallel with the acquisition of the target's transmitted signals, the target physical field data derived from the operation of the power system during the flight of the target UAV is acquired through remote non-contact sensing. The target physical field data is then subjected to time-frequency domain transformation and mode decomposition to extract a second physical state feature set for characterizing the real-time dynamic physical state of the target UAV. A dynamic coupling analysis across physical quantities is performed on the first hardware feature set and the second physical state feature set to construct a co-evolution model between the launch hardware characteristics and dynamic physical state of the target UAV. Based on the co-evolution model, the first hardware feature set and the second physical state feature set are jointly solved to obtain the joint inference results of the current maneuvering state, current load distribution state and current hardware operating point of the target UAV. Based on the joint inference results, the target's transmitted signal, and the target's physical field data, the target UAV is actively located and a hardware-oriented jamming strategy matching the joint inference results is generated.

[0006] On one hand, a computer device is provided, the computer device including one or more processors and one or more memories, the one or more memories storing at least one computer program, the computer program being loaded and executed by the one or more processors to implement the active localization and jamming method for unmanned aerial vehicles based on radio spectrum sensing.

[0007] On the one hand, a computer-readable storage medium is provided, wherein at least one computer program is stored in the computer-readable storage medium, the computer program being loaded and executed by a processor to implement the active localization and jamming method for unmanned aerial vehicles based on radio spectrum sensing.

[0008] On the one hand, a computer program product or computer program is provided, which includes program code stored in a computer-readable storage medium. The processor of a computer device reads the program code from the computer-readable storage medium and executes the program code, causing the computer device to perform the above-mentioned active localization and jamming method for unmanned aerial vehicles based on radio spectrum sensing. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a schematic diagram of the implementation environment of an active localization and jamming method for unmanned aerial vehicles based on radio spectrum sensing provided in an embodiment of this application; Figure 2 This is a flowchart of an active localization and jamming method for unmanned aerial vehicles based on radio spectrum sensing, provided in an embodiment of this application; Figure 3This is a flowchart of another method for active localization and jamming of unmanned aerial vehicles based on radio spectrum sensing provided in this application embodiment; Figure 4 This is a flowchart of another method for active localization and jamming of unmanned aerial vehicles based on radio spectrum sensing provided in the embodiments of this application; Figure 5 This is a flowchart of another method for active localization and jamming of unmanned aerial vehicles based on radio spectrum sensing provided in the embodiments of this application; Figure 6 This is a schematic diagram of the structure of an active localization and jamming system for unmanned aerial vehicles based on radio spectrum sensing, provided in an embodiment of this application. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0012] In this application, the terms "first," "second," etc., are used to distinguish identical or similar items with essentially the same function. It should be understood that there is no logical or temporal dependency between "first," "second," and "nth," nor are there any restrictions on quantity or execution order.

[0013] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, data stored, data displayed, etc.) and signals involved in this application are all authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0014] Figure 1 This is a schematic diagram illustrating the implementation environment of a UAV active localization and jamming method based on radio spectrum sensing, as provided in an embodiment of this application. See also... Figure 1 The implementation environment may include node 110 and system 140.

[0015] Node 110 is connected to system 140 via a wireless or wired network. Optionally, node 110 can be a smartphone, tablet, laptop, desktop computer, etc., but is not limited to these. Node 110 has an application installed and running that supports active drone localization and jamming based on radio spectrum sensing.

[0016] System 140 is an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery network (CDN), and big data and artificial intelligence platforms.

[0017] Traditional drone countermeasures primarily rely on radio spectrum sensing, analyzing the signal domain characteristics of drone-transmitted signals for detection, location, and identification. However, this method only reflects the characteristics of the transmitting hardware and cannot obtain real-time physical state information of the target drone during flight. This results in insufficient identification and location accuracy when facing individual differences among drones of the same model, complex maneuvers, and physical layer spoofing signals. Furthermore, due to the lack of awareness of the power system's operating status, existing countermeasure strategies are mostly general-purpose jamming, making it difficult to conduct differentiated attacks targeting hardware weaknesses and real-time load states, leading to low jamming efficiency and a high risk of collateral damage.

[0018] To address this, this application proposes an active localization and jamming system for unmanned aerial vehicles (UAVs) based on radio spectrum sensing. The system includes a processor and a memory. See [link to relevant documentation]. Figure 2 The processor is configured to perform the following steps: 201. By sensing the radio spectrum, the target UAV's transmission signal is collected, and the power amplifier response of the target transmission signal is modeled to extract the first hardware feature set used to characterize the inherent nonlinear characteristics of the target UAV's transmission hardware.

[0019] 202. In parallel with the acquisition of the target's transmitted signals, remote non-contact sensing is used to acquire target physical field data derived from the operation of the power system during the flight of the target UAV. Time-frequency domain transformation and mode decomposition are performed on the target physical field data to extract a second physical state feature set to characterize the real-time dynamic physical state of the target UAV.

[0020] 203. Perform dynamic coupling analysis across physical quantities on the first hardware feature set and the second physical state feature set, construct a co-evolution model between the launch hardware characteristics and dynamic physical state of the target UAV, and perform joint calculation on the first hardware feature set and the second physical state feature set according to the co-evolution model to obtain the joint inference results of the target UAV's current maneuvering state, current load distribution state and current hardware operating point.

[0021] 204. Based on the joint inference results, the target's transmitted signal, and the target's physical field data, perform active localization calculations on the target UAV and generate a hardware-oriented jamming strategy that matches the joint inference results.

[0022] For ease of understanding, the following explains some key terms in this embodiment: Radio spectrum sensing refers to the technology of detecting, analyzing, and identifying radio signals within a specific frequency band using receiving equipment. Its purpose is to obtain parameters such as signal frequency, power, and modulation method to understand the radio transmission characteristics of a target device.

[0023] Target-transmitted signals refer to the electromagnetic wave signals radiated outward by the onboard transmitting equipment of a target unmanned aerial vehicle (UAV) when performing tasks such as communication, navigation, or remote control. These signals carry inherent characteristic information about the UAV's transmitting hardware.

[0024] Power amplifier response modeling refers to establishing a mathematical model describing the nonlinear behavior of a UAV power amplifier by analyzing the output characteristics of the target's transmitted signal at different power levels. This model can reveal the inherent characteristics of the power amplifier, such as its compression point, intermodulation distortion, and thermal memory effect.

[0025] The first hardware feature set refers to a set of parameters or indicators extracted from the power amplifier response modeling to characterize the inherent nonlinear characteristics of the target UAV launch hardware. These features reflect the "fingerprint" information of the launch hardware.

[0026] Remote non-contact sensing refers to the technology of collecting target physical field data through long-distance detection methods (such as electric field, acoustic, optical sensors) without physical contact with the target drone.

[0027] Target physics data refers to the data on physical phenomena such as electromagnetic fields, vibrations, and acoustic noise generated by the operation of the target UAV's power system (such as motors and propellers) during flight. This data reflects the real-time physical state of the UAV.

[0028] The second physical state feature set refers to a set of parameters or indicators extracted from the target physical field data after time-frequency domain transformation and mode decomposition, used to characterize the real-time dynamic physical state of the target UAV. These features reflect the real-time "health" and "motion" state of the UAV.

[0029] Dynamic coupling analysis across physical quantities refers to an in-depth analysis of the mutual influence and correlation between the first hardware feature set and the second physical state feature set, in order to reveal the dynamic evolution relationship between different physical quantities.

[0030] A co-evolutionary model is a mathematical model constructed through dynamic coupling analysis across physical quantities, describing how the launch hardware characteristics and dynamic physical state of a target UAV interact and co-evolve. This model can predict and explain the comprehensive behavior of the UAV under different operating conditions.

[0031] The joint inference result refers to the comprehensive judgment on the target UAV's current maneuvering status, current load distribution status, and current hardware operating point obtained through joint calculation.

[0032] Hardware-oriented jamming strategy refers to a customized and targeted jamming scheme generated based on the hardware weaknesses and real-time operating status of the target UAV identified in the joint inference results.

[0033] In this embodiment, the system acquires the target UAV's transmitted signals through radio spectrum sensing and models the power amplifier response of the transmitted signals to extract a first hardware feature set characterizing the inherent nonlinear features of the target UAV's transmitting hardware. Specifically, a broadband receiver can be deployed to continuously monitor and record the spontaneously radiated radio signals of the target UAV during normal flight; these signals are stored for further analysis. Regarding the power amplifier response modeling, the received target transmitted signals can be input to a signal analysis module, where the nonlinear characteristics of the power amplifier are roughly evaluated by measuring basic parameters such as the signal's average power and total harmonic distortion. These basic parameters can be used as part of the first hardware feature set.

[0034] In parallel with acquiring the target's transmitted signals, the system remotely and non-contactly collects target physical field data derived from the operation of the propulsion system during the flight of the target UAV. It then performs time-frequency domain transformation and mode decomposition on the target physical field data to extract a second physical state feature set characterizing the real-time propulsion physical state of the target UAV. For example, one or more general-purpose acoustic sensors can be deployed near the target UAV's flight path to passively collect environmental noise generated by the UAV's propulsion system. This noise data is recorded as target physical field data. When processing the target physical field data, the acquired data can be subjected to Fourier transform to obtain its spectral distribution. The main frequency peaks identified from the spectrum, along with their corresponding frequencies and amplitudes, can be used as part of the second physical state feature set to roughly reflect the operating state of the propulsion system.

[0035] The system performs dynamic coupling analysis across physical quantities on the first hardware feature set and the second physical state feature set, constructing a co-evolution model between the target UAV's launch hardware characteristics and dynamic physical state. Based on the co-evolution model, it jointly solves the first hardware feature set and the second physical state feature set to obtain the joint inference results of the target UAV's current maneuver state, current load distribution state, and current hardware operating point. For example, by performing simple statistical regression analysis on the first hardware feature set and the second physical state feature set, such as calculating their Pearson correlation coefficient, linear correlations between different features can be found. Based on these statistical correlations, a simplified mapping table or set of empirical rules can be constructed as the co-evolution model. During the joint solution, the first hardware feature set and the second physical state feature set collected and processed at the current moment are input into the co-evolution model. The model can directly output the inference results of the current maneuver state, current load distribution state, and current hardware operating point that best match the input feature set, based on a preset lookup table or decision tree.

[0036] Based on the joint inference results, the target's transmitted signal, and the target's physical field data, the system actively locates the target UAV and generates a hardware-oriented jamming strategy that matches the joint inference results. Specifically, multiple receiving stations can be used to measure the time difference of arrival or angle of arrival of the target's transmitted signal, and based on these measurements, the preliminary spatial position of the target UAV is calculated using triangulation or polygonal positioning algorithms. When generating the jamming strategy, based on the current hardware operating point and load distribution identified in the joint inference results, a jamming waveform and power level that roughly matches the current state is selected from a library containing a limited number of general jamming patterns. For example, if the inference results indicate that the power amplifier is under high load, a high-power broadband jamming signal can be selected.

[0037] This embodiment achieves collaborative perception of the target UAV's launch hardware characteristics and dynamic physical state by integrating radio spectrum sensing and remote non-contact sensing. This overcomes the limitations of traditional countermeasures that rely solely on signal domain features, enabling the acquisition of real-time physical state information of the target UAV under complex maneuvers. By constructing a collaborative evolution model across physical quantities and performing joint calculations, the system can infer the target UAV's current maneuver state, load distribution state, and hardware operating point, thereby improving the accuracy of target identification and positioning. Furthermore, based on these joint inference results, the system can generate hardware-oriented jamming strategies that match the target UAV's real-time weaknesses, effectively avoiding the inefficiency and false positives of general-purpose jamming, and improving the effectiveness and targeting of countermeasures.

[0038] In some of the solutions described above in this application, the target emission signals of the target UAV are collected by radio spectrum sensing to extract hardware features. However, in this process, relying solely on spontaneous signals under normal flight conditions may not be able to fully capture the nonlinear response characteristics of the power amplifier at different output power levels, resulting in inaccurate feature extraction.

[0039] In response, this application further proposes a method for acquiring target transmission signals of a target UAV through radio spectrum sensing, see [link to relevant documentation]. Figure 3 It includes: 301. Transmit a detection signal to the target UAV to induce a change in the response of the target UAV's power amplifier at different output power levels.

[0040] 302. After transmitting the detection signal, collect the first set of transmitted signals returned by the target UAV, which carries the nonlinear response characteristics of the power amplifier.

[0041] 303. During the period when the detection signal is not emitted, collect the second set of emission signals spontaneously radiated by the target UAV in normal flight state.

[0042] 304. Perform joint calibration and screening on the first group of transmitted signals and the second group of transmitted signals to obtain the target transmitted signal.

[0043] Specifically, transmitting probe signals to the target drone aims to actively stimulate its power amplifier, causing it to operate in the nonlinear region and thus fully exposing its inherent nonlinear characteristics. This can be achieved in several ways. For example, a series of pulse or continuous wave signals with different power levels, modulation methods, or frequencies can be transmitted to the target drone to simulate different workload conditions and induce diverse responses from the power amplifier. Alternatively, frequency sweep or frequency hopping signals can be used as probe signals to cover a wide frequency range and power dynamic range in a short period of time, thereby more comprehensively stimulating the nonlinear characteristics of the power amplifier.

[0044] After transmitting the probe signal, the first set of transmitted signals returned by the target UAV, carrying the nonlinear response characteristics of the power amplifier, is acquired. The purpose is to capture the target UAV's response to the controlled probe signal stimulus, which directly reflects the nonlinear behavior of the power amplifier. This can be achieved by using a high-sensitivity broadband receiver to acquire the target UAV's transmitted signals in a specific frequency band in real time immediately after the probe signal is transmitted, and then digitally storing the data. Alternatively, a multi-channel receiving system can be used to simultaneously acquire signals at different spatial locations to obtain more comprehensive signal characteristics, and synchronous sampling technology can be used to ensure the temporal consistency of the data.

[0045] During periods when the detection signal is not emitted, a second set of spontaneously emitted signals from the target UAV under normal flight conditions is collected. This serves to obtain a baseline transmission signal from the target UAV under normal, undisturbed conditions for comparison and analysis with signals affected by the detection signal. This can be achieved by continuously monitoring the communication frequency bands likely used by the target UAV, passively receiving its regular communication, navigation, or telemetry signals when no active detection signal is emitted. Alternatively, spectrum monitoring equipment can be used to scan the radio spectrum of the airspace where the target UAV is located within a preset time window to identify and collect its spontaneously emitted signals.

[0046] The target transmitted signal is obtained by jointly calibrating and filtering the first and second sets of transmitted signals. This step aims to integrate signals obtained from two different acquisition methods, eliminating measurement errors through calibration and removing noise and irrelevant signals through filtering, thereby obtaining a high-quality, representative target transmitted signal. Calibration may include time-frequency alignment, power normalization, phase correction, and other processing to eliminate systematic biases caused by different acquisition devices or acquisition points at different times. Filtering may employ methods such as signal-to-noise ratio thresholding and signal integrity assessment to remove signal segments severely affected by environmental interference or with incomplete data. Joint calibration and filtering can also be implemented using machine learning algorithms. For example, a classifier can be trained to identify and remove abnormal signals or noise, while adaptive filtering techniques are used to calibrate the signal to ensure that the obtained target transmitted signal has high fidelity and consistency.

[0047] Through the above technical solution, this application overcomes the limitation that relying solely on spontaneous signals under normal flight conditions cannot fully capture the nonlinear response characteristics of power amplifiers. Specifically, by actively transmitting probe signals to the target UAV, the power amplifier can be induced to produce response changes at different output power levels, thereby actively creating diverse operating conditions. This allows the power amplifier to fully exhibit its inherent nonlinear characteristics under forced conditions, compensating for operating conditions that spontaneous signals may not cover. Collecting the first set of transmitted signals induced by the probe signals provides detailed nonlinear characteristic data of the power amplifier during the dynamic response process. Simultaneously, collecting the second set of transmitted signals during periods without probe signals preserves the baseline information of the target UAV under normal and natural flight conditions, providing a stable reference for subsequent feature analysis. Joint calibration and screening of these two sets of signals effectively eliminates systematic errors and noise interference during the acquisition process, integrating signal data from both forced response and spontaneous emission states, thus obtaining a comprehensive, accurate, and representative target transmitted signal. This method, which combines active induction with multi-state signal acquisition, improves the integrity and reliability of the target's transmitted signal, laying a solid foundation for the subsequent accurate extraction of the inherent nonlinear characteristics of the target UAV's launch hardware, and thus enhancing the ability to identify individual differences among UAVs of the same model.

[0048] In some of the solutions described above in this application, a joint calibration and screening of the first group of transmitted signals and the second group of transmitted signals is proposed to obtain the target transmitted signal. However, in this process, due to the difference in the response of the signal under different states, the hardware features may be inconsistent, affecting the accuracy of feature extraction.

[0049] In response, this application further proposes to jointly calibrate and filter the first set of transmitted signals and the second set of transmitted signals to obtain the target transmitted signal, including: The first set of transmitted signals and the second set of transmitted signals are subjected to time-frequency alignment and power normalization respectively to obtain the calibrated first set of transmitted signals and the calibrated second set of transmitted signals.

[0050] The nonlinear response characteristics of the power amplifier are extracted from the first set of calibrated transmission signals to obtain the first characteristic waveform of the target UAV in the forced response state. The nonlinear response characteristics of the power amplifier are extracted from the second set of calibrated transmission signals to obtain the second characteristic waveform of the target UAV in the autonomous flight state.

[0051] The correlation coefficients between the first characteristic waveform and the second characteristic waveform in terms of power amplifier compression point characteristics, intermodulation distortion characteristics, and thermal memory effect characteristics are determined to obtain a hardware characteristic consistency measure between the first characteristic waveform and the second characteristic waveform.

[0052] Based on the hardware feature consistency metric, signal segments that meet the preset consistency threshold are selected from the first set of calibrated transmission signals and the second set of calibrated transmission signals, and the selected signal segments are merged and reconstructed to obtain the target transmission signal.

[0053] This time-frequency alignment and power normalization process aims to eliminate the time axis offset and power level differences of signals acquired at different time points, ensuring the accuracy of subsequent feature extraction and comparison. Specifically, time-frequency alignment can be achieved by calculating the cross-correlation function of the two sets of signals and calibrating the time axis based on the time delay corresponding to the cross-correlation peak. Alternatively, high-precision timestamps (such as GPS synchronization clocks) can be used to mark the signals during acquisition, and precise alignment can be performed based on the timestamps during processing. Power normalization can be achieved by scaling the signal amplitude to a uniform peak value (e.g., normalizing the maximum amplitude of all signals to 1) or a uniform root mean square value (e.g., normalizing the root mean square power of all signals to 1 watt) to eliminate the impact of power fluctuations on feature extraction.

[0054] The extraction of nonlinear response characteristics of the power amplifier aims to capture the inherent nonlinear behavior patterns of the target UAV launch hardware (especially the power amplifier) ​​under different operating conditions. These nonlinear characteristics are important criteria for identifying specific hardware components. For example, the nonlinear kernel function coefficients can be extracted as characteristic waveforms by modeling the signal using Volterra series. Alternatively, by analyzing the spectral spread, harmonic distortion, and intermodulation products of the signal under different input powers, a feature vector reflecting the nonlinear characteristics of the power amplifier can be constructed, such as the curve of the third-order intermodulation intercept IP3 as a function of power.

[0055] The determination of the correlation coefficient yields a hardware feature consistency metric, designed to quantify the similarity between a first feature waveform acquired under forced response conditions and a second feature waveform acquired under autonomous flight conditions. This metric objectively assesses the stability and consistency of hardware features under different acquisition conditions. For example, the Pearson correlation coefficient can be used to measure the linear correlation between two feature waveform sequences. Alternatively, cosine similarity can be used to assess the directional proximity of two feature vectors, thus reflecting the consistency of hardware features.

[0056] This process involves filtering signal segments that meet a preset consistency threshold, merging and reconstructing them. The aim is to eliminate inconsistencies caused by environmental interference, transient behavior, or acquisition errors from the original acquired signal, retaining only high-quality signal segments that stably reflect the hardware characteristics of the target UAV. For example, a sliding window technique can be used to calculate a hardware feature consistency metric within each window, selecting only the signal segments corresponding to windows with metric values ​​higher than a preset threshold. These filtered signal segments are then spliced ​​together according to their original time sequence, and any gaps at the splicing points are smoothed (e.g., using linear interpolation or spline interpolation) to form a continuous and high-quality target transmission signal.

[0057] Through the above technical solution, this application solves the problem of inconsistent hardware characteristics of UAV transmission signals acquired under different states, improving the accuracy and reliability of target transmission signals. Specifically, by performing time-frequency alignment and power normalization on the first and second sets of transmission signals, timing deviations and power level fluctuations during signal acquisition are eliminated, laying a standardized foundation for subsequent accurate hardware characteristic comparison. Based on this, the first characteristic waveform under forced response state and the second characteristic waveform under autonomous flight state are extracted respectively, enabling a comprehensive capture of the inherent nonlinear characteristics of the target UAV power amplifier under different operating conditions. By quantifying the correlation coefficients of these two characteristic waveforms in compression point features, intermodulation distortion features, and thermal memory effect features, this application can objectively assess signal consistency and identify stable signal segments that truly reflect the hardware characteristics of the UAV. Based on hardware characteristic consistency metrics, signal segments are screened, merged, and reconstructed, ensuring that the obtained target transmission signal contains only highly consistent data. This provides high-quality, high-reliability input for subsequent power amplifier response modeling of the target transmission signal and extraction of the first hardware feature set characterizing the inherent nonlinear characteristics of the target UAV's transmission hardware, greatly improving the accuracy of the first hardware feature set extraction and the overall system recognition and positioning performance.

[0058] In some of the embodiments described above in this application, a first hardware feature set is proposed to characterize the inherent nonlinear characteristics of the target UAV launch hardware. However, in practice, existing methods may not be able to fully and accurately extract key hardware features, such as the compression point of the power amplifier, intermodulation distortion, power-added efficiency, and thermal memory effect, resulting in an insufficient feature set to support accurate cooperative sensing and interference strategy generation.

[0059] To address this, this application proposes a method for power amplifier response modeling of the target's transmitted signal to extract a first hardware feature set characterizing the inherent nonlinear features of the target UAV's launch hardware. See [link to relevant documentation] Figure 4 The method specifically includes the following steps: 401. Logarithmic power curve fitting is performed on the response curves of the target's transmitted signal at different power levels to obtain the power transfer characteristic curve of the target UAV's power amplifier.

[0060] 402. The power transmission characteristic curve is subjected to inflection point detection and nonlinear region division, thereby extracting the compression point characteristics, intermodulation distortion characteristics and power-added efficiency characteristics of the target UAV power amplifier.

[0061] 403. Perform time-frequency analysis on the target's transmitted signal, extract the frequency drift rate of the target UAV power amplifier at different power levels, and construct a thermal memory effect feature to characterize the thermal accumulation and heat dissipation characteristics of the target UAV power amplifier based on the correlation between the frequency drift rate and the power level.

[0062] 404. The compression point feature, the intermodulation distortion feature, the power-added efficiency feature, and the thermal memory effect feature are combined and encoded to obtain the first hardware feature set.

[0063] Logarithmic power curve fitting is performed on the response curves of the target transmitted signal at different power levels to accurately characterize the input-output power relationship of the power amplifier over a wide dynamic range. By converting the power values ​​to logarithmic form (e.g., dBm), the data range can be effectively compressed, allowing the response characteristics across different power regions to be represented uniformly and stably. For example, a polynomial regression method can be used to approximate discrete measurement data points by fitting a high-order polynomial function, thereby obtaining a smooth and continuous power transfer characteristic curve. Alternatively, spline interpolation can be used, interpolating between data points using a piecewise polynomial function to more precisely capture the local variation trends of the curve.

[0064] Inflection point detection and nonlinear region segmentation of the power transfer characteristic curve are performed to identify the key points where the power amplifier transitions from linear to nonlinear operation and to quantify its nonlinear behavior. Inflection point detection can be achieved by analyzing the slope change of the power transfer characteristic curve; for example, when the slope of the curve begins to decrease relative to the linear gain, it can be identified as an inflection point. Nonlinear region segmentation, based on this, divides the power transfer characteristic curve into linear and nonlinear regions. The compression point characteristic typically refers to the output power of the power amplifier when it decreases by a specific value (e.g., 1 dB) relative to the ideal linear output power. The intermodulation distortion characteristic quantifies the intensity of the additional frequency components (intermodulation products) generated due to nonlinear effects when multiple signals are simultaneously input to the power amplifier; for example, the third-order intermodulation intercept (IP3) can be used as its characterization. The power-added efficiency characteristic reflects the efficiency of the power amplifier in converting DC power supply energy into RF output energy and is a key indicator for measuring its energy consumption performance.

[0065] Time-frequency analysis of the target transmitted signal aims to reveal the dynamic characteristics of the power amplifier's output signal frequency changing over time at different power levels. This can be achieved through various time-frequency analysis techniques. For example, Short-Time Fourier Transform (STFT) can obtain a two-dimensional representation of the signal's spectrum over time, thus tracking instantaneous frequency changes. Alternatively, wavelet transform can be used to decompose the signal using wavelet functions of different scales to achieve good resolution in both the time and frequency domains. Extracting the frequency drift rate refers to quantifying the rate of change of the instantaneous frequency over time. Based on the correlation between this frequency drift rate and the power level, a thermal memory effect characteristic is constructed to characterize the performance drift of the power amplifier due to internal temperature accumulation and heat dissipation during prolonged operation or drastic power changes. For example, the rise and fall slopes of the frequency drift rate, as well as the time required for the frequency drift to reach steady state, can be analyzed; these parameters collectively constitute the thermal memory effect characteristic.

[0066] The purpose of combining and encoding the compression point feature, intermodulation distortion feature, power-added efficiency feature, and thermal memory effect feature is to integrate these multi-dimensional hardware characteristics into a unified, structured data representation. For example, these feature values ​​can be directly concatenated into a high-dimensional vector to form a comprehensive feature description. Alternatively, more complex encoding methods can be used, such as principal component analysis (PCA) to reduce the dimensionality of the features and extract their principal components, thereby reducing redundancy and improving feature robustness. This combination and encoding ensures that the first hardware feature set can be fully and efficiently utilized by the subsequent co-evolutionary model.

[0067] Through the above technical solutions, this application can systematically model the response of the launch hardware of the target UAV and comprehensively and accurately extract its inherent nonlinear characteristics. Specifically, logarithmic power curve fitting overcomes the characteristic distortion problem that may occur when traditional methods process signals with a wide dynamic range, providing a stable and accurate foundation for subsequent analysis. Inflection point detection and nonlinear region division, combined with the extraction of compression point features, intermodulation distortion features, and power-added efficiency features, can capture the transition process of the power amplifier from linear to nonlinear and its efficiency changes, making up for the insufficiency of a single feature in fully reflecting hardware weaknesses. In addition, by performing time-frequency analysis on the frequency drift rate and constructing thermal memory effect features, this application can deeply reveal the dynamic behavior of the power amplifier in the process of heat accumulation and heat dissipation, effectively making up for the lack of attention to thermal effects in related technologies. Combining and encoding these multi-dimensional features forms a comprehensive and structured first hardware feature set. This feature set can not only more accurately characterize the inherent nonlinear characteristics of the UAV's launch hardware, but also provide a solid data foundation for the subsequent construction of a co-evolution model with the dynamic physical state. This enables the system to identify targets based on more refined hardware characteristics when facing individual differences, complex maneuvering states, and physical layer spoofing signals of the same model of UAV. This improves the accuracy of active localization calculation and the effectiveness of hardware-oriented interference strategies, avoiding the inefficiency and accidental damage of general-purpose interference.

[0068] In some embodiments described above, the power transfer characteristic curve is extracted to characterize the nonlinear features of the power amplifier. However, during its implementation, real-time physical state information such as battery discharge state may be ignored when calculating the power-added efficiency, leading to inaccurate feature extraction and affecting the accuracy of the interference strategy. To address this, this application further proposes inflection point detection and nonlinear region segmentation of the power transfer characteristic curve to extract the compression point features, intermodulation distortion features, and power-added efficiency features of the target UAV's power amplifier. Specifically, this process includes: The power transfer characteristic curve is piecewise fitted to both the linear and nonlinear regions to detect the inflection point where it transitions from linear to nonlinear growth. The output power value corresponding to this inflection point is used as the compression point feature. This compression point feature is a key indicator of the nonlinear behavior of the power amplifier, indicating the critical point where the amplifier's output power begins to deviate from linear growth. Accurate detection of this inflection point is crucial for understanding the amplifier's operating state. In practice, the least squares method or other regression analysis methods can be used to fit the linear and nonlinear regions of the power transfer characteristic curve separately, and the inflection point can be identified by comparing the slope and intercept of different fitted segments. Alternatively, a preset gain compression threshold can be set, such as a 1dB compression point; the output power value that decreases by 1dB relative to the ideal linear output power is the compression point. Furthermore, first-order or second-order derivative analysis can be performed on the power transfer characteristic curve; the inflection point typically corresponds to the location where the derivative changes.

[0069] Intermodulation distortion (ICD) analysis was performed on the spectral components of the target's transmitted signal within the nonlinear operating region. The relative proportion of the amplitude of the third-order intermodulation component to that of the fundamental component was extracted, and the third-order intermodulation intercept (CDI) was calculated based on this proportion. This CDI was then used as the characteristic of the ICD. Intermodulation distortion is a nonlinear product of power amplifiers when multi-frequency signals are input, particularly the third-order intermodulation component. The relative proportion of its amplitude to the fundamental component, and the calculated third-order CDI, are important indicators of the amplifier's nonlinearity. A higher third-order CDI indicates better linearity of the amplifier. For example, by transmitting a probe signal containing two close frequencies to a target UAV, the power amplifier will generate third-order intermodulation components within the nonlinear operating region. The amplitudes of these components, along with the fundamental component, are measured using a spectrum analyzer, and then the third-order CDI is calculated using a formula. For actual complex target transmission signals, Fourier transform of the signal can be performed to identify the fundamental component and harmonic and intermodulation components generated by nonlinear effects. The power of the third-order intermodulation component can be separated and measured by digital signal processing technology and compared with the fundamental power to calculate the third-order intermodulation cutoff point.

[0070] Furthermore, the input and output power of the target's transmitted signal are obtained, and the characteristic components related to the battery discharge state in the second physical state feature set are acquired. Based on the input power, the output power, and the characteristic components related to the battery discharge state, the DC power consumption of the target UAV's power amplifier at the current operating point is determined. Accurately determining the DC power consumption of the power amplifier is fundamental to calculating its power-added efficiency. Traditional power consumption calculations may only consider input and output power, but ignore the impact of real-time physical factors such as battery discharge state on the amplifier's actual operating point and efficiency. Introducing characteristic components related to the battery discharge state can more accurately reflect the amplifier's energy consumption under actual power supply conditions. These characteristic components related to the battery discharge state can include the battery's real-time voltage, real-time current, remaining charge percentage, or battery internal resistance. By monitoring these parameters, the battery's power supply capacity and internal losses can be inferred. For example, when the battery voltage decreases or the internal resistance increases, even with the same input and output power, the amplifier's actual DC power consumption may change. In addition, the battery's temperature and cycle count also affect its discharge characteristics; these parameters can also be used as characteristic components to further refine the estimation of DC power consumption. The DC power consumption can be determined by establishing a power amplifier power consumption model under different power supply conditions. This model takes the input power, output power, and battery discharge state-related characteristic components as inputs and outputs the DC power consumption.

[0071] Based on the ratio of output power to DC power consumption, the power-added efficiency (PEP) of the target UAV power amplifier at the current operating point is determined, and this PEP is used as a characteristic feature. PEP is an important indicator of power amplifier efficiency; it considers the efficiency of the amplifier in converting DC power to RF output power, subtracting the input RF power loss. A higher PEP indicates a higher efficiency in converting DC power into useful RF power and lower energy consumption. Using it as a characteristic reflects the amplifier's energy efficiency performance at different operating points. PEP can be calculated directly using its defined formula: (Output Power - Input Power) / DC Power Consumption. Alternatively, a lookup table or model of the power amplifier's PEP at different operating points can be pre-established, and the PEP can be calculated from the currently measured input power, output power, and DC power consumption using the lookup table or model.

[0072] Through the above technical solution, this application can accurately identify the inflection point of the power amplifier from linear to nonlinear, avoiding ambiguous operating point judgments and providing an accurate starting point for subsequent nonlinear analysis. Simultaneously, it quantifies the distortion degree of the power amplifier in the nonlinear operating region, providing a key indicator for evaluating its linearity. More importantly, by incorporating the battery discharge state-related feature components from this second physical state feature set into the DC power consumption determination process, the calculation of power-added efficiency more closely reflects the energy consumption of the target UAV under actual flight conditions, solving the problem of inaccurate efficiency evaluation caused by ignoring real-time physical states in traditional methods. This feature extraction method, which combines real-time physical state information, enables the first hardware feature set to more accurately and comprehensively characterize the inherent nonlinear characteristics and actual operating state of the target UAV's power amplifier. This provides a more reliable and refined data foundation for subsequent co-evolution model construction, joint solution, and the generation of hardware-oriented interference strategies, thereby improving the accuracy and effectiveness of the interference strategy and avoiding low interference efficiency or accidental damage due to inaccurate features.

[0073] In some of the embodiments described above in this application, a method for constructing thermal memory effect features to characterize the thermal accumulation and heat dissipation characteristics of the target UAV power amplifier is proposed. However, in its implementation, there is a lack of a precise quantification method for the dynamic changes in frequency drift rate, which makes it impossible to effectively distinguish the specific parameters of thermal accumulation rate and heat dissipation response time. This results in feature extraction that is not comprehensive and accurate enough, thereby affecting the complete characterization of the nonlinear characteristics of the UAV hardware.

[0074] To address this, this application further proposes a thermal memory effect feature based on the correlation between the frequency drift rate and the power level, used to characterize the thermal accumulation and heat dissipation properties of the target UAV's power amplifier. Specifically, this includes: performing time-series analysis on the frequency drift rate of the target's transmitted signal at different power levels, extracting the rising and falling slopes of the frequency drift rate over time, and using the ratio of the rising slope to the falling slope as the thermal accumulation rate feature. Based on the time delay relationship between the power level and the frequency drift rate, determining the response time required for the frequency drift rate to reach a steady state after a change in power level, and using this response time as the thermal time constant feature. Combining the thermal accumulation rate feature and the thermal time constant feature yields the thermal memory effect feature.

[0075] This involves performing time-series analysis on the frequency drift rate of the target's transmitted signal at different power levels. The rising and falling slopes of the frequency drift rate over time are extracted, and the ratio of the rising to falling slopes is used as a thermal accumulation rate characteristic. This involves arranging and processing the frequency drift data of the target's transmitted signal collected at different power levels in chronological order to reveal its dynamic changes. The rising slope characterizes the degree to which the internal temperature of the power amplifier accelerates frequency drift when the workload increases or the ambient temperature rises. The falling slope characterizes the degree to which the internal temperature of the power amplifier slows down frequency drift when the workload decreases or the ambient temperature decreases. Using the ratio of the rising to falling slopes as a thermal accumulation rate characteristic aims to quantify the relative dynamic characteristics of the power amplifier during thermal accumulation and dissipation. One implementation method is to use a sliding window linear regression method to piecewise fit the frequency drift rate time series, identify continuous rising and falling trends, and calculate the slope of each trend segment. For example, a threshold can be set; when the frequency drift rate shows an upward trend for N consecutive sampling points and the slope is greater than the threshold, the average upward slope is calculated. Conversely, the average downward slope is calculated. Another approach is to use signal processing techniques such as wavelet transform or empirical mode decomposition to decompose the frequency drift rate time series into components of different scales, and then estimate the slope of specific components representing heat accumulation and dissipation processes. For example, components reflecting slow-changing trends can be extracted, and their slopes before and after power changes can be calculated.

[0076] Based on the time delay relationship between the power level and the frequency drift rate, the response time required for the frequency drift rate to reach a steady state after a change in power level is determined. This response time is used as a characteristic of the thermal time constant. It refers to the fact that when the operating power level of a power amplifier changes, its internal temperature field and the resulting frequency drift rate do not immediately reach a new steady state, but rather there is a transition process. Response time is an indicator of how fast this transition process is, i.e., the time required from the start of the power level change to the frequency drift rate reaching a new steady state (e.g., reaching 90% or 95% of the steady-state value). Using this response time as a characteristic of the thermal time constant can characterize the magnitude of the inertia and heat dissipation efficiency of the power amplifier's thermal system. One implementation is to monitor the frequency drift rate change curve after a step change in power level and calculate the time required for it to change from the initial value to the steady-state value. This can be determined by fitting an exponential decay or growth curve. Another implementation is to perform cross-correlation analysis on the power level change signal and the frequency drift rate response signal to find the time offset with the maximum correlation between the two. This offset can be used as an approximation of the system response time.

[0077] Combining the heat accumulation rate feature with the thermal time constant feature yields the thermal memory effect feature, aiming to form a more comprehensive and refined thermal memory effect characteristic to fully characterize the thermal behavior of the power amplifier under dynamic operating conditions. The heat accumulation rate feature reflects the relative speed of heat generation and dissipation, while the thermal time constant feature reflects the response speed and thermal inertia of the thermal system to temperature changes. The combination of the two provides a multi-dimensional thermal behavior fingerprint, helping to more accurately identify and predict the nonlinear characteristics of the power amplifier. One implementation is to vectorize and combine these two features to form a two-dimensional or multi-dimensional feature vector. For example, the heat accumulation rate feature value and the thermal time constant feature value can be directly concatenated into a feature pair. Another implementation is to fuse these two features into a single comprehensive index through weighted summation, product, or more complex nonlinear functions, which can reflect the overall thermal stability or thermal sensitivity of the power amplifier.

[0078] By performing time-series analysis on the frequency drift rate of the target transmitted signal at different power levels, this application can accurately extract the rising and falling slopes of the frequency drift rate over time, and use their ratio as a characteristic of the thermal accumulation rate. This technique quantifies the dynamic speed of the power amplifier during thermal accumulation and dissipation, solving the problem that traditional methods cannot effectively distinguish between the dynamics of thermal accumulation and dissipation. Simultaneously, based on the time delay relationship between power level and frequency drift rate, this application can determine the response time required for the frequency drift rate to reach a steady state after a change in power level, and use this as a characteristic of the thermal time constant. This accurately characterizes the inertia and heat dissipation efficiency of the power amplifier's thermal system, overcoming the deficiency that the frequency drift rate alone cannot capture the time response of the thermal system. By combining the thermal accumulation rate characteristic with the thermal time constant characteristic, this application obtains a more comprehensive and refined thermal memory effect characteristic. This combined characteristic integrates the dynamic parameters of thermal accumulation rate and heat dissipation response time, providing a multi-dimensional fingerprint for characterizing the thermal behavior of the power amplifier under dynamic operating conditions, thereby improving the accuracy of the first hardware feature set in depicting the inherent nonlinear characteristics of the target UAV's launch hardware. This more accurate thermal memory effect feature, as an important component of the first hardware feature set, can provide more reliable input for subsequent co-evolution model construction and joint solution, thereby improving the accuracy of inferring the target UAV's current hardware operating point, maneuvering state, and load distribution state, and realizing more accurate active positioning solution and hardware-oriented interference strategy generation.

[0079] In some of the solutions mentioned above in this application, the target physical field data is collected remotely and non-contactly to extract dynamic physical state characteristics. However, in the process of implementation, there are challenges in how to effectively detect and collect various physical field data derived from the operation of the dynamic system, such as electromagnetic field, vibration and noise, to ensure the comprehensiveness and accuracy of the data. Specifically, it is impossible to simultaneously cover multiple dimensions of physical fields such as electrical, mechanical and acoustic fields, and the lack of a synchronization mechanism makes data fusion difficult.

[0080] In response, this application further proposes to collect target physical field data derived from the operation of the propulsion system during the flight of the target UAV through remote non-contact sensing, see [link to relevant documentation]. Figure 5 ,include: 501. Using an electric field sensor array, detect the electromagnetic field radiated by the motor drive circuit of the target UAV's power system during the commutation process, collect the current waveform and voltage waveform of the motor drive circuit, and use the current waveform and voltage waveform as the first physical field data.

[0081] 502. Using a laser vibrometer, a laser beam is emitted toward the surface of the target UAV and the reflected beam is received. Based on the Doppler frequency shift of the reflected beam, the micro-vibration displacement and vibration frequency of the surface of the UAV are calculated, and the micro-vibration displacement and vibration frequency are used as the second physical field data.

[0082] 503. Using a microphone array, beamforming and sound source localization are performed on the mechanical noise generated by the motor operation and the aerodynamic noise generated by the propeller rotation of the target UAV. The spectral characteristics of the mechanical noise and the aerodynamic noise are collected and used as the third physical field data.

[0083] 504. Synchronize and fuse the first physical field data, the second physical field data, and the third physical field data in time to obtain the target physical field data.

[0084] Specifically, this method involves using an electric field sensor array to detect the electromagnetic fields radiated by the motor drive circuit of the target UAV's power system during commutation, and to collect the current and voltage waveforms of the motor drive circuit. The aim is to directly perceive the electromagnetic radiation generated by the UAV's power system (especially the motor drive circuit) during operation. During commutation, the motor generates transient current and voltage changes, which radiate specific electromagnetic field signals. These signals carry information about the motor's operating status, such as load, speed, and faults. By collecting the current and voltage waveforms, the electrical characteristics of the motor drive circuit can be obtained, providing fundamental data for subsequent analysis of the UAV's power and electrical status. For example, an array of high-sensitivity Hall effect current sensors and differential voltage probes can be non-contactly placed near the UAV's flight path to capture the transient current and voltage signals of the motor drive circuit through electromagnetic coupling. These sensors should have wide frequency response characteristics to accurately capture high-frequency harmonic components during commutation. Alternatively, a loop antenna or dipole antenna array, combined with a high-bandwidth oscilloscope or spectrum analyzer, can be used to directly detect the near-field electromagnetic waves radiated by the motor drive circuit. By demodulating and inverting the detected electromagnetic waves, the current and voltage waveforms of the motor drive circuit can be reconstructed.

[0085] A laser vibrometer emits a laser beam towards the surface of a target drone and receives the reflected beam. Based on the Doppler frequency shift of the reflected beam, the micro-vibration displacement and frequency of the drone's surface are calculated, aiming to measure minute vibrations on the drone's surface non-contactly. The drone's propulsion system (such as motors and propellers) generates mechanical vibrations during operation. These vibrations are transmitted to the drone's structure and manifest as specific frequencies and displacement patterns. Measuring these vibrations with a laser vibrometer can reflect the mechanical health, balance, and coupling of the propulsion system with the drone's structure, providing crucial information for analyzing the drone's mechanical dynamic state. For example, a laser Doppler vibrometer based on the heterodyne interferometry principle can be used. This device emits a laser beam towards the target drone's surface; the reflected light interferes with a reference light on a detector. Due to the Doppler frequency shift of the reflected light caused by the drone's vibration, the instantaneous vibration velocity and displacement of the drone's surface can be calculated by analyzing the frequency change of the interference signal. Alternatively, a laser vibration measurement system based on the speckle interferometry principle can also be used. This system can measure the micro-vibration displacement and frequency of the machine surface with high precision by analyzing the dynamic changes of the speckle pattern on the target surface and combining it with digital image correlation technology.

[0086] By using a microphone array, beamforming and sound source localization are performed on the mechanical noise generated by the motor operation and the aerodynamic noise generated by the propeller rotation of the target UAV. The spectral characteristics of the mechanical and aerodynamic noise are also collected, aiming to perceive the operating status of the UAV's propulsion system through acoustic means. The operation of the UAV's motor generates mechanical noise, and the rotation of the propeller generates aerodynamic noise. These noises have specific spectral characteristics that can reflect information such as motor speed, load, propeller thrust, and blade condition. Beamforming and sound source localization using a microphone array can effectively distinguish and extract the characteristics of different sound sources, providing data for analyzing the acoustic dynamic state of the UAV. For example, a planar or three-dimensional array composed of multiple high-sensitivity microphones can be deployed. By processing the multi-channel sound pressure signals received by the array using beamforming algorithms such as delay-and-sum or minimum variance distortion-free response (MVDR), the sound signal in a specific direction can be enhanced, while noise in other directions can be suppressed, thereby achieving the localization and separation of the motor and propeller sound sources. Alternatively, acoustic camera technology can be used, which combines a microphone array and an optical camera. It visualizes the location of the sound source through sound source localization algorithms (such as the generalized cross-correlation method and the CLEAN-SC algorithm) and extracts the sound signal of a specific sound source area for spectral analysis, thereby obtaining the spectral characteristics of mechanical noise and aerodynamic noise.

[0087] The first, second, and third physical field data are synchronized and fused in time to ensure effective integration of multi-source heterogeneous data. Since different sensors (electric field sensor arrays, laser vibrometers, microphone arrays) may have different sampling rates, clock sources, and data transmission delays, time synchronization is essential to ensure accurate alignment of all data points on the time axis. Data fusion, based on time synchronization, integrates data from different physical quantities to form a comprehensive and consistent "target physical field data" set, providing a unified data foundation for subsequent comprehensive analysis and feature extraction. For example, a high-precision GPS timing module or atomic clock can be used as a unified time reference to provide synchronized clock signals for all sensors, ensuring accurate alignment of data acquisition timestamps. During the data fusion stage, methods based on Kalman filtering, particle filtering, or deep learning can be used to weightedly fuse data from different sensors to eliminate noise, fill in missing data, and improve the overall accuracy and robustness of the data. Alternatively, an event-triggered synchronization mechanism can be employed. When a specific event occurs in the drone's propulsion system (such as motor startup, commutation, or sudden load change), all sensors simultaneously record data, using this event as a reference point for time synchronization. Data fusion can employ multimodal data fusion algorithms, such as those based on tensor decomposition, multi-kernel learning, or graph neural networks, to map different physical field data onto a unified feature space for fusion.

[0088] Through the above technical solutions, this application can solve the challenges of multi-dimensional physical field data acquisition, ensuring the comprehensiveness and accuracy of the data. Specifically, by directly detecting the electromagnetic radiation of the motor drive circuit through an electric field sensor array, electrical state information can be captured. The non-contact measurement of micro-vibrations of the aircraft body using a laser vibrometer avoids interference from traditional contact measurements, improving the authenticity and sensitivity of mechanical vibration data. By combining a microphone array with beamforming and sound source localization technology, the spectral characteristics of motor mechanical noise and propeller aerodynamic noise can be effectively separated and extracted, providing rich acoustic perception information. More importantly, by strictly synchronizing and fusing these heterogeneous data, the time deviation between different sensors is eliminated, ensuring precise alignment and integration of multi-dimensional physical field data on the time axis, thereby generating comprehensive, consistent, and highly reliable target physical field data. This lays a solid foundation for the subsequent accurate extraction of the second physical state feature set of the UAV's real-time dynamic physical state, enhances the perception capability of the target UAV's power system operating state, and further enhances the entire system's ability to identify individual differences and complex maneuvering states of UAVs of the same model, providing richer and more reliable input for subsequent accurate positioning and the generation of hardware-oriented interference strategies.

[0089] In some of the embodiments described above in this application, it is proposed to collect micro-vibration displacement and vibration frequency using a laser vibrometer to obtain the dynamic physical state of the target UAV. However, in the process of its implementation, how to accurately calculate the micro-vibration displacement and vibration frequency, and distinguish different frequency sources (such as the motor rotor rotation frequency, motor commutation frequency, and propeller blade passing frequency) to improve data accuracy and reliability is a challenge.

[0090] To address this issue, this application proposes a method for calculating the micro-vibration displacement and vibration frequency of an aircraft surface based on the Doppler frequency shift of a reflected beam. The specific steps include: optically mixing and photoelectric conversion of the reflected beam and a reference beam to generate an interference signal carrying Doppler frequency shift information; orthogonally demodulating the interference signal to extract the in-phase and quadrature components, and calculating the instantaneous displacement waveform of the aircraft surface vibration based on these components; performing a Fourier transform on the instantaneous displacement waveform to obtain the power spectral density distribution of the aircraft surface vibration; identifying the fundamental frequency peak corresponding to the rotor rotation frequency, the harmonic peak corresponding to the commutation frequency, and the modulation sideband peak corresponding to the propeller blade passage frequency from the power spectral density distribution; determining the vibration frequency based on the frequency values ​​corresponding to the fundamental frequency peak, the harmonic peak, and the modulation sideband peak; and calculating the micro-vibration displacement based on the amplitude values ​​corresponding to the fundamental frequency peak, the harmonic peak, and the modulation sideband peak.

[0091] The following will elaborate on each step of the above technical solution.

[0092] The process involves optical mixing and photoelectric conversion of a reflected beam and a reference beam to generate an interference signal carrying Doppler frequency shift information. Optical mixing, also known as heterodyne detection, involves superimposing the signal light from the target (reflected beam) with the local reference light (reference beam) on a photodetector. Utilizing the interference effect of light, the weak Doppler frequency shift information in the signal light is converted into frequency modulation in the electrical signal. Photoelectric conversion refers to the process by which the photodetector converts the received optical signal into an electrical signal. In this way, the minute frequency change (Doppler shift) of the beam caused by tiny vibrations on the surface of the target UAV is amplified and converted into a measurable change in the electrical signal frequency. Specifically, optical mixing can be achieved using a Mach-Zehnder interferometer or a Michelson interferometer. In a Mach-Zehnder interferometer, the laser beam is split into two paths by a beam splitter: one path serves as the reference beam, and the other illuminates the surface of the target UAV and is reflected back as the signal beam. The two beams rejoin at another beam splitter and interfere on the photodetector. The photodetector converts the interfering optical signal into an electrical signal, the frequency of which includes Doppler frequency shift information caused by the target vibration. Another approach is to use an optical fiber coupler to couple the signal light to a reference light, which is then input to a high-speed photodiode for photoelectric conversion, thereby generating the interfering signal.

[0093] The process involves orthogonal demodulation of the interference signal, extracting its in-phase and quadrature components, and then calculating the instantaneous displacement waveform of the body surface vibration based on these components. Quadrature demodulation is a technique for extracting modulation information (in this case, Doppler shift) from a high-frequency carrier signal. By mixing the interference signal with two reference signals (usually the carrier frequency) that are 90 degrees out of phase, in-phase (I) and quadrature (Q) components are obtained. These components contain both amplitude and phase information. Specifically, orthogonal demodulation can be achieved using digital signal processing methods, such as obtaining the analytic signal through the Hilbert transform, and then separating the in-phase and quadrature components. The instantaneous phase of the signal can then be obtained by calculating the arctangent function (arctan(Q / I)). After phase unwrapping of the instantaneous phase and combining this with the laser wavelength, the instantaneous displacement waveform of the body surface vibration can be accurately calculated. Another approach is to use an analog quadrature demodulator to perform signal mixing and low-pass filtering through hardware circuitry, directly outputting the I / Q components, and then converting the analog I / Q signals into digital signals through an analog-to-digital converter (ADC) for further processing.

[0094] Furthermore, regarding the Fourier transform of the instantaneous displacement waveform, the power spectral density distribution of the body surface vibration can be obtained. The Fourier transform is a mathematical tool that converts a time-domain signal into a frequency-domain signal, revealing the various frequency components contained in the signal and their corresponding intensities. Power spectral density (PSD) describes the distribution of signal power at different frequencies, visually displaying the main vibration frequencies and their energy levels. Specifically, the Fast Fourier Transform (FFT) algorithm can be used to process the acquired instantaneous displacement waveform to efficiently obtain its frequency domain representation. Through FFT, the instantaneous displacement waveform of a discrete-time series can be converted into the spectrum of a discrete-frequency series. By calculating the squared magnitude of the spectrum and performing normalization, the power spectral density distribution of the body surface vibration can be obtained. Another approach is to use non-parametric spectral estimation methods such as the Welch method, segmenting the signal, windowing, and averaging to obtain a smoother and more reliable power spectral density estimate.

[0095] The process involves identifying the fundamental frequency peak corresponding to the motor rotor rotation frequency, the harmonic peak corresponding to the motor commutation frequency, and the modulation sideband peak corresponding to the propeller blade passage frequency from the power spectral density distribution. The vibration frequency is determined based on the frequency values ​​corresponding to these peaks, and the micro-vibration displacement is calculated based on their amplitudes. Identifying these specific frequency peaks is crucial for distinguishing different vibration sources. The motor rotor rotation frequency typically represents the fundamental frequency peak, while the motor commutation frequency may generate harmonic peaks relative to the fundamental frequency, and the propeller blade passage frequency may generate modulation sideband peaks near the fundamental frequency. Specifically, peak detection algorithms, such as threshold-based or derivative-based methods, can be used to automatically identify frequency peaks in the power spectral density distribution whose amplitudes exceed a preset threshold. For the identified peaks, comparison with parameters such as the motor speed range, commutation frequency characteristics, and number of propeller blades of known UAV models allows for accurate classification as fundamental frequency peaks, harmonic peaks, or modulation sideband peaks. For example, by detecting the frequency value of the fundamental frequency peak related to the rotational frequency of the motor rotor, the real-time rotational speed of the motor rotor can be directly determined, and thus the vibration frequency can be determined. Simultaneously, the amplitudes corresponding to these peaks directly reflect the energy magnitude of the corresponding vibration source, and the corresponding micro-vibration displacement can be calculated after calibration. Another approach is to use machine learning or deep learning models to train a large amount of UAV vibration spectrum data, enabling it to automatically identify and classify the characteristic peaks of different vibration sources and output the corresponding frequency and amplitude information.

[0096] Through the above technical solutions, this application can accurately calculate the micro-vibration displacement and vibration frequency of the target UAV's surface and effectively distinguish different vibration sources. Specifically, through optical mixing and photoelectric conversion, weak Doppler frequency shift information is efficiently converted into a processable electrical signal, laying the foundation for subsequent accurate analysis. The use of orthogonal demodulation technology ensures the accuracy of instantaneous displacement waveform calculation and avoids errors that may be introduced by single signal processing. Fourier transform converts the time-domain signal into a frequency-domain power spectral density distribution, making different vibration frequency components clearly visible. More importantly, by identifying the fundamental frequency peak corresponding to the motor rotor rotation frequency, the harmonic peak corresponding to the motor commutation frequency, and the modulation sideband peak corresponding to the propeller blade passage frequency, this application can decompose the complex body vibration signal into independent vibration modes generated by different power components, thereby accurately determining the frequency and corresponding micro-vibration displacement of each vibration source. This refined vibration feature extraction capability greatly improves the accuracy and reliability of dynamic physical state data, providing high-quality input for subsequent co-evolution model construction and UAV state inference, thereby improving the accuracy and effectiveness of the entire system in identifying, locating, and interfering with the target UAV.

[0097] In some of the embodiments described above in this application, a method is proposed to extract a second physical state feature set by remotely collecting target physical field data through non-contact sensing. However, in the process of implementation, when collecting the spectral characteristics of mechanical noise and aerodynamic noise, it may be impossible to effectively distinguish and locate different noise sources, resulting in inaccurate feature extraction and affecting the accurate characterization of the subsequent dynamic physical state.

[0098] To address this, this application further proposes using a microphone array to perform beamforming and sound source localization on the mechanical noise generated by the motor operation and the aerodynamic noise generated by the propeller rotation of the target UAV, and to collect the spectral characteristics of the mechanical noise and the aerodynamic noise. Specifically, the process includes: acquiring multi-channel sound pressure signals radiated by the target UAV during flight using a microphone array, and performing delay summation and beamforming processing on the multi-channel sound pressure signals to obtain a first enhanced sound signal pointing to the motor operation position and a second enhanced sound signal pointing to the propeller rotation position. The first enhanced sound signal undergoes time-frequency transformation to extract the first power spectral density at the motor commutation frequency and its harmonics, and the spectral peak distribution in the first power spectral density corresponding to the motor rotor rotation frequency is used as the spectral characteristics of the mechanical noise. The second enhanced sound signal undergoes time-frequency transformation to extract the second power spectral density at the propeller blade passage frequency and its modulation sideband, and the spectral peak distribution in the second power spectral density corresponding to the blade passage frequency and the broadband noise floor related to aerodynamic turbulence are used as the spectral characteristics of the aerodynamic noise.

[0099] This method utilizes a microphone array to perform beamforming and sound source localization on the mechanical noise generated by the motor operation and the aerodynamic noise generated by the propeller rotation of the target drone. The aim is to effectively separate noise sources at different physical locations by leveraging the spatial resolution of the acoustic sensor array. For example, a digital microphone array can be used, and the signals collected by each microphone can be digitally processed to achieve precise sound source pointing. Alternatively, an analog microphone array can be used, and the sound source in a specific direction can be amplified by adjusting the physical position of each microphone and the gain of the signal amplifier.

[0100] Acquiring multi-channel sound pressure signals radiated by the target drone during flight refers to the simultaneous and independent recording of sound wave signals generated by the drone by multiple microphones in a microphone array. This provides the necessary spatial information for subsequent sound source separation and localization. For example, each microphone in the microphone array can independently acquire sound pressure signals and perform synchronous sampling through a multi-channel data acquisition card. Alternatively, a distributed microphone network can be used, with each node wirelessly transmitting the acquired sound pressure signals to a central processing unit for aggregation.

[0101] Delayed summation and beamforming of the multi-channel sound pressure level signal is a technique that utilizes the spatial temporal differences in signals received by multiple microphones. By applying appropriate delays and weighted summation, it enhances sound signals from a specific direction while suppressing noise and interference from other directions. This allows the system to "focus" on specific noise source areas on the UAV. For example, a delayed summation beamforming algorithm can be used to calculate the delay of each channel based on the sound source direction and microphone array geometry, and then superimpose the delayed signals. Alternatively, a minimum variance distortionless response (MVDR) beamforming algorithm can be employed, optimizing array weights to minimize noise and interference from other directions while maintaining signal gain in the desired direction.

[0102] The first enhanced acoustic signal pointing to the motor's operating position and the second enhanced acoustic signal pointing to the propeller's rotating position are the results of beamforming processing, representing the acoustic information mainly originating from the motor and propeller after spatial filtering. This achieves physical separation of mechanical and aerodynamic noise. For example, by adjusting the pointing parameters of the beamforming algorithm, the beam can be directed to the area where the UAV motor is located and the area where the propeller rotates, respectively. Alternatively, multiple parallel beamformers can be used, each preset to point to a different potential noise source location, such as one pointing to the fuselage center (motor) and another pointing to the propeller blade area.

[0103] A time-frequency transformation is performed on the first enhanced acoustic signal to extract its first power spectral density at the motor commutation frequency and its harmonics. This aims to convert the time-domain signal of the motor noise into a frequency-domain representation, focusing on specific frequency components closely related to the motor's operating characteristics. For example, a short-time Fourier transform (STFT) can be used to perform time-frequency analysis on the first enhanced acoustic signal, and then the power spectral density at the motor commutation frequency and its harmonics can be extracted from the resulting spectrum. Alternatively, wavelet transform can be used to perform multi-resolution analysis of the signal to capture the transient characteristics of the motor commutation frequency and its harmonics more precisely.

[0104] Using the distribution of spectral peaks in the first power spectral density corresponding to the rotor rotation frequency as the spectral characteristics of the mechanical noise refers to identifying and quantifying the discrete frequency peaks in the motor noise spectrum related to the rotor rotation frequency. These spectral peaks directly reflect the motor's operating state. For example, a peak detection algorithm can be applied to the first power spectral density to identify spectral peaks higher than the background noise threshold, and based on the correspondence between motor speed and frequency, spectral peaks related to the rotor rotation frequency can be selected. Alternatively, techniques such as cepstral analysis can be used to convert the harmonic structure of the periodic signal into peak values ​​in the cepstral domain, thereby more robustly extracting the rotor rotation frequency.

[0105] A time-frequency transform is performed on the second enhanced acoustic signal to extract its second power spectral density at the propeller blade passage frequency and its modulation sideband. This aims to analyze the frequency characteristics of propeller noise, particularly the frequency components related to blade rotation and aerodynamic effects. For example, a short-time Fourier transform can be used to perform time-frequency analysis on the second enhanced acoustic signal, and then the power spectral density at the propeller blade passage frequency and its modulation sideband can be extracted from the resulting spectrum. Alternatively, a synchronous averaging technique combined with Fourier transform can be used to enhance the frequency components related to the propeller rotation period and suppress random noise.

[0106] Using the distribution of spectral peaks corresponding to the blade passage frequency in the second power spectral density and the broadband noise floor associated with aerodynamic turbulence as the spectral characteristics of the aerodynamic noise means comprehensively characterizing the propeller noise, including both discrete frequency components generated by the periodic motion of the blades and continuous broadband noise components generated by airflow turbulence. For example, discrete spectral peaks associated with the blade passage frequency and its harmonics can be identified from the second power spectral density, and the amplitude and frequency of these peaks can be calculated. Simultaneously, the continuous spectral portion between the peaks is smoothed to obtain the broadband noise floor. Alternatively, spectral line separation techniques can be used to decompose the second power spectral density into discrete spectra (line spectra) and continuous spectra (broadband noise), and their features can be extracted separately.

[0107] Through the aforementioned technical solution, this application utilizes the spatial filtering capability of a microphone array, and through delay summation and beamforming processing, effectively separates the mechanical noise generated by the motor operation of the target UAV from the aerodynamic noise generated by the propeller rotation, generating a first and second enhanced sound signal pointing to specific noise sources. Based on this, targeted time-frequency transformation and feature extraction are performed on the separated signals, focusing on the spectral peaks corresponding to the motor commutation frequency and its harmonics, the motor rotor rotation frequency, the propeller blade passage frequency and its modulation sidebands, and the broadband noise floor related to aerodynamic turbulence. This refined sound source separation and feature extraction method overcomes the problems of noise source aliasing and inaccurate feature extraction in traditional methods, improving the purity and accuracy of the spectral features of mechanical and aerodynamic noise. By providing high-quality third physics field data, this solution lays the foundation for subsequently constructing a more accurate second physics state feature set, thereby making the characterization of the real-time dynamic physical state of the target UAV more accurate, and further improving the performance and effectiveness of the entire system in cross-physical quantity dynamic coupling analysis, joint solution, and active localization and interference strategy generation.

[0108] In some of the embodiments described above in this application, a method is proposed to collect target physical field data remotely and extract a second physical state feature set to characterize the real-time dynamic physical state of the target UAV. However, in this process, the physical field data contains a variety of physical quantities such as current waveforms, voltage waveforms, micro-vibration displacements, vibration frequencies, mechanical noise spectrum characteristics, and aerodynamic noise spectrum characteristics. These data come from diverse sources and are independent of each other. If they are not effectively processed and analyzed, the feature extraction may be inaccurate or incomplete, and the dynamic information of different physical quantities may not be fully integrated. As a result, it is difficult to accurately reflect the real-time dynamic state of the UAV, which affects the construction of subsequent co-evolution models and the generation of interference strategies.

[0109] To this end, this application further proposes to perform time-frequency domain transformation and mode decomposition on the target's physical field data to extract a second physical state feature set for characterizing the real-time dynamic physical state of the target UAV, specifically including: Time-frequency analysis was performed on the current and voltage waveforms in the physical field data to extract the current spike amplitude, duty cycle and modulation frequency of the pulse width modulation signal, and distortion characteristics of the back EMF waveform at the commutation moment of the motor drive circuit. The current spike amplitude, duty cycle, modulation frequency, and distortion characteristics of the back EMF waveform were then used as the power electrical feature set.

[0110] Empirical mode decomposition is performed on the micro-vibration displacement and vibration frequency in the physical field data to obtain multiple intrinsic mode function components. Target mode components associated with the motor rotor rotation frequency, motor commutation frequency, and body structure resonance frequency are selected from these multiple intrinsic mode function components. The instantaneous amplitude and instantaneous frequency of the target mode components are extracted and used as the mechanical vibration feature set.

[0111] The spectral characteristics of mechanical noise and aerodynamic noise in the physical field data are fused and analyzed. The real-time rotational speed of the motor rotor is determined based on the spectral peak distribution in the spectral characteristics of the mechanical noise, and the real-time thrust load of the propeller is determined based on the broadband noise floor in the spectral characteristics of the aerodynamic noise. The real-time rotational speed and the real-time thrust load are used as the acoustic sensing feature set.

[0112] The second physical state feature set is obtained by combining and normalizing the power electrical feature set, the mechanical vibration feature set, and the acoustic sensing feature set through multidimensional feature combination and encoding.

[0113] Specifically, time-frequency domain transformation and mode decomposition are performed on the target's physical field data to transform the raw, complex physical field data into a feature representation that clearly reflects the real-time state of the UAV's dynamic system. Time-frequency domain transformation is a signal processing technique used to analyze the joint distribution of a signal in time and frequency, revealing the frequency components and their intensity that change over time. This can be achieved, for example, through methods such as Short-Time Fourier Transform (STFT), Wavelet Transform, or Hilbert-Huang Transform (HHT). Mode decomposition is a technique that decomposes a complex signal into a series of eigenmode components with specific physical meanings. This can be achieved, for example, through methods such as Empirical Mode Decomposition (EMD), Variational Mode Decomposition (VMD), or Principal Component Analysis (PCA). Through these processes, physically meaningful features can be extracted from the raw data to form this second physical state feature set, which can comprehensively and accurately characterize the real-time dynamic physical state of the target UAV.

[0114] The purpose of performing time-frequency analysis on the current and voltage waveforms in this physical field data is to extract key features reflecting the operating state of the motor drive circuit from its electrical signals. The current and voltage waveforms directly reflect the energy input and output of the motor drive circuit, as well as the operating state of its internal switching devices. Time-frequency analysis can capture the transient changes of these electrical signals under different operating modes. For example, in addition to short-time Fourier transform, continuous wavelet transform or synchronous compression transform can be used to analyze the local time-frequency characteristics of the signals more precisely. The current spike amplitude usually appears during motor commutation or sudden load changes, reflecting the instantaneous current impact intensity of the motor drive circuit and is an important indicator for assessing the motor load and health status. The duty cycle and modulation frequency of the pulse width modulation signal are core parameters of motor drive control, directly determining the average voltage and speed obtained by the motor. Their changes reflect the control strategy and real-time power requirements of the UAV power system. The distortion characteristics of the back EMF waveform can reveal abnormalities in components such as motor windings, magnetic circuits, or bearings, and are a key basis for motor fault diagnosis. Combining these features into a power electrical feature set can comprehensively characterize the electrical operating state of the UAV power system.

[0115] Empirical Mode Decomposition (EMD) is performed on the micro-vibration displacements and frequencies in the physical field data to separate vibration modes related to key components of the UAV's propulsion system from complex mechanical vibration signals. Micro-vibration displacements and frequencies are important physical quantities reflecting the mechanical health and operational status of the UAV's airframe structure and propulsion components (such as motors and propellers). Empirical Mode Decomposition (EMD) is an adaptive signal decomposition method that can decompose nonlinear, non-stationary signals into a series of intrinsic mode functions (IMFs), each representing a vibration mode of the signal at different time scales. Besides EMD, ensemble empirical mode decomposition (EEMD) or variational mode decomposition (VMD) methods can be used to improve the stability and accuracy of the decomposition. The IMF components are physically meaningful single-component signals obtained from the decomposition, each corresponding to a different vibration source or vibration mode. By filtering out target mode components associated with the motor rotor rotation frequency, motor commutation frequency, and airframe structural resonant frequency, we can focus on vibration information crucial to the UAV's propulsion system. For example, these target mode components can be identified by setting a frequency bandpass filter or based on correlation analysis. Extracting the instantaneous amplitude and frequency of these target modal components can dynamically reflect the intensity and frequency changes of these key vibration modes, thereby forming a mechanical vibration feature set to characterize the mechanical vibration state of the UAV.

[0116] The purpose of fusing and analyzing the spectral characteristics of mechanical noise and aerodynamic noise in the physical field data is to infer the real-time operating parameters of the UAV's propulsion system through acoustic information. Mechanical noise mainly originates from the operation of mechanical components such as motors and gearboxes, and its spectral characteristics typically include discrete frequency components related to motor speed. Aerodynamic noise is mainly generated by propeller rotation, and its spectral characteristics include blade passage frequencies and their harmonics, as well as a broadband noise floor caused by turbulence. Fusion analysis combines the characteristics of these two types of noise to provide more comprehensive dynamic state information. The real-time rotor speed of the motor is determined based on the spectral peak distribution of the mechanical noise. Motor speed can be inferred by identifying the fundamental frequency peak and its harmonics in the spectrum. For example, in addition to peak detection and harmonic verification, cepstral analysis or autocorrelation function analysis can be used to determine periodic components. The real-time propeller thrust load is determined based on the broadband noise floor of the aerodynamic noise. This utilizes the physical relationship between turbulent noise intensity and propeller thrust. For example, the broadband noise floor and thrust load can be mapped by establishing an empirical model or based on computational fluid dynamics (CFD) simulation results. Using the real-time rotational speed and the real-time thrust load as an acoustic sensing feature set, the operating status information of the UAV power system can be provided from an acoustic dimension.

[0117] The power electrical feature set, mechanical vibration feature set, and acoustic sensing feature set are combined and normalized using multidimensional features. The aim is to integrate features from different physical quantities into a unified and comparable feature vector. Multidimensional feature combination effectively integrates different types of features to form a more comprehensive feature representation. This can be achieved, for example, through simple concatenation, weighted summation, or machine learning-based feature fusion methods. Normalization encoding involves scaling the combined features to ensure they fall within a similar numerical range, preventing certain features from dominating subsequent analysis due to excessively large numerical ranges. This can be achieved, for example, through min-max scaling or Z-score standardization. In this way, a second physical state feature set is obtained, which comprehensively, accurately, and consistently characterizes the real-time power physical state of the target UAV.

[0118] Through the above technical solutions, this application solves the problem that diverse and independent sources of physical field data lead to inaccurate or incomplete feature extraction, making it difficult to fully integrate the dynamic information of different physical quantities and thus accurately reflect the real-time dynamic state of the UAV. Specifically, by performing time-frequency analysis on current and voltage waveforms, it is possible to capture the current peak amplitude at commutation time of the motor drive circuit, the duty cycle and modulation frequency of the pulse width modulation signal, and the distortion characteristics of the back electromotive force waveform, thereby constructing a refined set of power electrical features that accurately reflects the electrical operating state of the UAV's power system. Simultaneously, empirical mode decomposition is performed on micro-vibration displacement and vibration frequency, and target mode components associated with the motor rotor rotation frequency, motor commutation frequency, and resonant frequency of the airframe structure are selected. Their instantaneous amplitude and instantaneous frequency are extracted, effectively avoiding noise interference and accurately characterizing the mechanical vibration state of the UAV. Furthermore, by fusing and analyzing the spectral characteristics of mechanical and aerodynamic noise, the real-time rotational speed of the motor rotor is determined based on the spectral peak distribution, and the real-time thrust load of the propeller is determined based on the broadband noise floor, providing real-time operating parameters of the UAV's power system from an acoustic perspective. By combining and normalizing these multi-source, multi-dimensional power and electrical feature sets, mechanical vibration feature sets, and acoustic sensing feature sets, a comprehensive, accurate, and highly comparable second physical state feature set is formed. This feature set not only fully integrates the dynamic information of different physical quantities to accurately reflect the real-time dynamic state of the UAV, but also provides high-quality input for the subsequent construction of a co-evolution model between the launch hardware characteristics and the dynamic physical state of the target UAV. This improves the accuracy and robustness of the co-evolution model, thereby making the joint inference results of the target UAV's current maneuvering state, current load distribution state, and current hardware operating point more reliable, and enhancing the accuracy and effectiveness of active localization solutions and hardware-oriented jamming strategies.

[0119] In some of the solutions described above in this application, the real-time rotational speed of the motor rotor is determined based on the spectral peak distribution in the spectral characteristics of mechanical noise in order to extract the real-time dynamic physical state of the target UAV. However, in its implementation, due to the presence of multiple candidate frequency peaks in the spectrum, it is difficult to accurately identify the fundamental frequency of the motor rotor, resulting in inaccurate rotational speed determination.

[0120] In response, this application further proposes a method for determining the real-time rotational speed of the motor rotor based on the spectral peak distribution in the spectral characteristics of the mechanical noise, the steps of which include: Peak detection is performed on the power spectral density distribution in the spectral characteristics of the mechanical noise, and multiple candidate frequency peaks with amplitudes exceeding a preset threshold are extracted from the power spectral density distribution.

[0121] The harmonic relationship of the multiple candidate frequency peaks is verified. Each candidate frequency peak is selected as the assumed fundamental frequency in turn, and the existence of harmonic verification peaks with frequency values ​​of the second and third harmonics of the assumed fundamental frequency is checked.

[0122] The frequency value corresponding to the hypothetical fundamental frequency where the harmonic verification peak exists is taken as the current rotation frequency of the motor rotor, and the real-time speed of the motor rotor is determined according to the correspondence between the current rotation frequency and the number of pole pairs of the motor.

[0123] Specifically, peak detection is performed on the power spectral density distribution of the mechanical noise's spectral characteristics. Multiple candidate frequency peaks with amplitudes exceeding a preset threshold are extracted, aiming to identify energetic frequency components from complex spectral data. Peak detection can be implemented using various algorithms. For example, a local maximum-based method can be used, where points on the power spectral density curve whose values ​​in both their left and right neighborhoods are less than their own are identified as peaks. Alternatively, a derivative-based method can be used, determining peaks by detecting points where the first derivative of the power spectral density curve crosses zero and the second derivative is negative. The preset threshold filters out background noise and interference signals, ensuring sufficient reliability of the frequency peaks processed subsequently. This threshold can be dynamically adjusted according to the signal-to-noise ratio requirements of the actual application scenario, for example, set to a fixed decibel value higher than the average noise level, or set as a percentage of the maximum power spectral density value. This step effectively separates the potential fundamental frequency and its harmonic components from the noise in the spectrum, providing high-quality input for subsequent harmonic verification.

[0124] The system performs harmonic relationship verification on multiple candidate frequency peaks. Each candidate frequency peak is selected sequentially as the assumed fundamental frequency. The system then checks whether any of these candidate frequency peaks contain harmonic verification peaks with frequencies equal to the second and third harmonics of the assumed fundamental frequency. This is crucial for identifying the actual fundamental frequency of the motor rotor. When the motor rotor is running, the mechanical noise it generates typically has clear harmonic characteristics; that is, in addition to the fundamental frequency, it also contains harmonic components with frequencies that are integer multiples of the fundamental frequency (such as second and third harmonics). During harmonic relationship verification, the system iterates through each extracted candidate frequency peak and temporarily sets it as the assumed fundamental frequency. The system then searches among the remaining candidate frequency peaks for frequency peaks that match the second and third harmonics of the assumed fundamental frequency. A small frequency tolerance range is usually set during matching to accommodate minor errors in actual measurements. For example, if the assumed fundamental frequency is f, the system checks for other candidate frequency peaks within the ranges [2f-Δf, 2f+Δf] and [3f-Δf, 3f+Δf]. In this way, it is possible to effectively distinguish the true fundamental frequency and its harmonics generated by the motor rotor from the false frequency peaks caused by other vibration sources or noise.

[0125] The step of determining the motor speed involves using the frequency value corresponding to the hypothetical fundamental frequency where the harmonic verification peak exists as the current rotational frequency of the motor rotor, and determining the real-time rotational speed of the motor rotor based on the correspondence between the current rotational frequency and the number of pole pairs of the motor. Once a candidate frequency peak is successfully verified to have a harmonic relationship with second and third harmonics, that frequency peak is confirmed as the true fundamental frequency of the motor rotor, which is the current rotational frequency. There is a fixed physical relationship between motor speed and rotational frequency, usually determined by the number of pole pairs of the motor. For example, for a two-pole motor (i.e., one number of pole pairs), its rotational frequency and speed (usually expressed in revolutions per second or revolutions per minute) are directly correlated. For a multi-pole motor, the speed needs to be calculated by dividing the rotational frequency by the number of pole pairs. By combining the verified current rotational frequency with the pre-known number of motor pole pairs, the system can accurately calculate the real-time rotational speed of the motor rotor.

[0126] Through the above technical solution, this application solves the problem of inaccurate identification of the motor rotor fundamental frequency due to the presence of multiple candidate frequency peaks in complex spectrum environments. By introducing a harmonic relationship verification mechanism, it is possible to accurately distinguish between the true fundamental frequency and its harmonics generated by the motor rotor and spurious frequency peaks caused by other interferences, thereby improving the reliability and accuracy of the real-time speed determination of the motor rotor. This accurate real-time speed information, as an important component of the acoustic perception feature set in the second physical state feature set, can provide more accurate dynamic physical state input for subsequent cross-physical quantity dynamic coupling analysis, thereby enabling the co-evolution model to more accurately infer the current maneuvering state, current load distribution state, and current hardware operating point of the target UAV, improving the accuracy and effectiveness of active localization calculation and hardware-oriented interference strategy generation. For example, when the UAV performs high-difficulty maneuvers, its motor speed will change drastically. This solution can capture these changes in real time and accurately, thereby providing key decision-making basis for the interference system, enabling the interference strategy to act on the weak points of the UAV more timely and accurately.

[0127] In some of the solutions described above in this application, a broadband noise basis based on the spectral characteristics of aerodynamic noise is proposed to determine the real-time thrust load of the propeller to characterize the real-time dynamic physical state of the target UAV. However, in this process, the spectral characteristics of aerodynamic noise may contain discrete frequency components corresponding to the propeller blade passing frequency and harmonics. These discrete components will interfere with the extraction of the broadband noise basis, resulting in the inability to accurately reflect the true intensity of the turbulent boundary layer pressure fluctuations, thereby affecting the accuracy of determining the real-time thrust load and reducing the accuracy of the perception of the UAV's dynamic state.

[0128] To address this, this application further proposes a method for determining the real-time thrust load of a propeller based on a broadband noise substrate in the spectral characteristics of the aerodynamic noise. The method includes: separating the line spectrum from the continuous spectrum of the aerodynamic noise; removing discrete frequency components corresponding to the propeller blade passing frequency and harmonics; and retaining the continuous frequency components generated by turbulent boundary layer pressure fluctuations as the broadband noise substrate. The average power spectral density of the broadband noise substrate within a preset frequency range is calculated, and the turbulent fluctuation pressure intensity on the propeller blade surface is determined based on the correspondence between the average power spectral density and the propeller diameter, number of blades, and current airspeed. Based on the aeroacoustic mapping relationship between the turbulent fluctuation pressure intensity and the propeller thrust load, the turbulent fluctuation pressure intensity is converted into the real-time thrust load of the propeller.

[0129] Specifically, separating the line spectrum from the continuous spectrum of the aerodynamic noise spectrum involves distinguishing the discrete line spectrum components generated by periodic motion (such as propeller blade rotation) from the continuous broadband spectrum components generated by non-periodic turbulent phenomena (such as turbulent boundary layer pressure fluctuations). This step aims to lay the foundation for subsequent accurate extraction of the broadband noise floor directly related to the thrust load. This separation can be achieved using adaptive filtering methods, such as the Least Mean Square (LMS) algorithm or the Recursive Least Squares (RLS) algorithm, which predict and eliminate periodic line spectrum components to separate the continuous spectrum components. Alternatively, signal decomposition methods, such as Empirical Mode Decomposition (EMD) or wavelet decomposition, can be used to decompose the original spectrum into components at different frequency scales, then identify and remove the components corresponding to the line spectrum, retaining the broadband components.

[0130] After separating the line spectrum from the continuous spectrum, it is necessary to remove the discrete frequency components corresponding to the blade passing frequency and harmonics of the propeller. The blade passing frequency (BPF) and its harmonics are strong periodic noise generated during propeller rotation. They appear as sharp discrete spectral lines in the spectrum and are the main periodic interference components. To accurately obtain the broadband noise floor caused by turbulence, these periodic interference components need to be removed from the spectral characteristics of aerodynamic noise. This process can be achieved by using digital notch filters or comb filters to precisely suppress or eliminate discrete spectral lines and their harmonics at known frequency positions in the spectrum. Alternatively, masking can be performed in the frequency domain. That is, after identifying the frequency range of the discrete frequency components and their harmonics, the spectral amplitudes within these frequency ranges are set to zero or interpolated, thereby removing the discrete components without affecting other frequency components.

[0131] The continuous frequency components generated by turbulent boundary layer pressure fluctuations are retained as the broadband noise substrate. Turbulent boundary layer pressure fluctuations are non-periodic pressure fluctuations generated by the interaction between random vortex structures in the surface flow field and the blade surface during high-speed propeller blade motion; their intensity is closely related to the thrust load generated by the propeller. Therefore, after removing discrete frequency components, the retained continuous frequency components are considered the broadband noise substrate, directly reflecting the turbulent state of the propeller blade surface. This step can directly use the remaining spectral data after line spectrum removal as the broadband noise substrate. Alternatively, the spectral data after line spectrum removal can be smoothed to further remove residual random noise or small fluctuations, thereby obtaining a more stable broadband noise substrate.

[0132] Further, the average power spectral density of the broadband noise substrate within a predetermined frequency range is calculated. The power spectral density (PSD) of the broadband noise substrate describes the distribution of noise energy at different frequencies. To quantify the overall intensity of broadband noise, it is typically necessary to calculate its average power spectral density within a representative frequency range. This predetermined frequency range should be selected in a band where turbulent noise energy is concentrated and less affected by other noise sources to ensure the validity of the calculation results. This calculation can be achieved by integrating the power spectral density value of the broadband noise substrate within the predetermined frequency range and then dividing by the bandwidth of that frequency range. Alternatively, a weighted averaging method can be used, assigning different weights to different frequencies based on their importance in the thrust load mapping, and then calculating the weighted average power spectral density.

[0133] Based on this, the turbulent pulsating pressure intensity on the propeller blade surface is determined according to the correspondence between the average power spectral density and the propeller diameter, number of blades, and current airspeed. The intensity of turbulent noise generated by the propeller is not only directly related to the pressure pulsation on the blade surface, but is also affected by the propeller's own geometric parameters (such as diameter and number of blades) and flight environment parameters (such as current airspeed). These parameters collectively determine the turbulent characteristics of the flow field around the propeller. By establishing the correspondence between the average power spectral density and these physical parameters, the turbulent pulsating pressure intensity on the propeller blade surface can be more accurately retrieved from the acoustic signal. This correspondence can be established in advance through a large amount of experimental data or computational fluid dynamics (CFD) simulations, using empirical formulas or lookup tables to establish the relationship between the average power spectral density and the propeller diameter, number of blades, current airspeed, and turbulent pulsating pressure intensity. Alternatively, a physical model-based method can be used, such as a simplified aeroacoustic theoretical model, to substitute these parameters into the model and calculate the theoretical relationship between the turbulent pulsating pressure intensity and the average power spectral density.

[0134] Based on the aeroacoustic mapping relationship between the turbulent pulsating pressure intensity and the propeller thrust load, the turbulent pulsating pressure intensity is converted into the real-time thrust load of the propeller. The propeller thrust load is a key parameter for generating lift or propulsion during flight, directly reflecting the power output state of the UAV. Turbulent pulsating pressure intensity, as a direct physical source of propeller aerodynamic noise, has a clear aeroacoustic mapping relationship with thrust load. By utilizing this mapping relationship, the real-time thrust load, which is difficult to measure directly, can be inverted from the measurable turbulent pulsating pressure intensity. This mapping relationship can be pre-established through bench tests or flight tests, measuring the turbulent pulsating pressure intensity on the propeller blade surface under different thrust loads, and establishing a calibration curve or regression model between the two. Alternatively, a machine learning-based approach can be used to collect a large amount of turbulent pulsating pressure intensity and corresponding thrust load data, training a regression model (such as support vector regression or neural networks) to achieve a nonlinear mapping from pressure intensity to thrust load.

[0135] Through the above technical solution, this application can solve the problem of interference from discrete frequency components in the aerodynamic noise spectrum on the extraction of broadband noise substrate, thus improving the accuracy of determining the real-time thrust load of the propeller. Specifically, by separating the line spectrum and continuous spectrum of the aerodynamic noise spectrum characteristics and accurately removing discrete frequency components corresponding to the propeller blade passage frequency and harmonics, it ensures that the retained broadband noise substrate can purely reflect the true intensity generated by the pressure pulsation of the turbulent boundary layer, avoiding the confusion of periodic interference signals on the thrust load assessment. On this basis, the average power spectral density of the broadband noise substrate in the preset frequency range is calculated, and combined with the correspondence of key physical parameters such as propeller diameter, number of blades, and current airspeed, the turbulent pulsation pressure intensity on the propeller blade surface can be quantified more accurately, because these parameters directly affect the generation mechanism of turbulent noise. Based on the aeroacoustic mapping relationship between turbulent pulsation pressure intensity and propeller thrust load, the intensity is converted into real-time thrust load, realizing a direct and high-precision inversion from acoustic signal to power output state. This precise real-time thrust load information, as a crucial component of the acoustic sensing feature set within the second physical state feature set, provides a more reliable input for subsequent dynamic coupling analysis across physical quantities. For example, when constructing a co-evolutionary model with the first hardware feature set, more accurate thrust load data allows the model to more precisely capture the co-evolutionary relationship between the UAV's launch hardware characteristics and its dynamic physical state, thereby improving the accuracy of the joint inference results regarding the target UAV's current maneuvering state, current load distribution state, and current hardware operating point. Furthermore, in the active localization calculation and hardware-oriented jamming strategy generation stages, based on more precise thrust load information, the system can more accurately assess the UAV's power output and flight attitude, thereby generating more targeted and efficient jamming strategies and improving the target recognition accuracy and jamming effectiveness of the UAV countermeasure system.

[0136] In some of the solutions described above in this application, a dynamic coupling analysis across physical quantities is proposed to construct a co-evolution model for the first hardware feature set and the second physical state feature set. However, in the implementation process, there are shortcomings in how to accurately determine the time delay and correlation between different features, and how to detect event synchronization to ensure the accuracy of the model. Specifically, existing methods may not be able to effectively correlate the time dynamics of hardware heat accumulation and motor current changes, or accurately capture the synchronization events of power amplifier nonlinear operation and motor heavy load state, resulting in inaccurate model construction and affecting the reliability of subsequent state inference.

[0137] To address this, this application further proposes a method for conducting dynamic coupling analysis across physical quantities on the first hardware feature set and the second physical state feature set, and constructing a co-evolution model of the target UAV between its launch hardware characteristics and dynamic physical state. This method includes: performing time-series alignment and cross-correlation analysis on the thermal memory effect characteristics in the first hardware feature set and the power electrical feature set in the second physical state feature set; determining the time delay and correlation coefficient between the change in frequency drift rate in the thermal memory effect characteristics and the change in current peak amplitude in the power electrical feature set; and establishing a first coupling relationship between the thermal accumulation degree of the target UAV power amplifier and the motor drive current based on the time delay and the correlation coefficient. Furthermore, it involves performing event synchronization analysis on the compression point characteristics in the first hardware feature set and the power electrical feature set in the second physical state feature set; detecting the time alignment relationship between the moment when the output power enters the compression zone in the compression point characteristics and the moment when the current peak amplitude in the power electrical feature set exceeds the load threshold; and establishing a second coupling relationship between the moment when the target UAV power amplifier enters the nonlinear operating region and the moment when the motor enters the heavy load state based on the time alignment relationship. Using the first coupling relationship and the second coupling relationship as initial constraints, the intermodulation distortion characteristics and power-added efficiency characteristics of the first hardware feature set, as well as the mechanical vibration feature set and acoustic sensing feature set of the second physical state feature set, are obtained. The co-evolution model is then constructed using a multivariate state-space modeling method.

[0138] The dynamic coupling analysis across physical quantities refers to a comprehensive and dynamic correlation analysis of feature data from different physical fields (such as radio frequency electronics and mechanical dynamics). It aims to reveal the intrinsic mechanisms of mutual influence and constraint between different physical quantities, thereby providing a more comprehensive understanding of the overall system behavior. This analysis can employ machine learning-based feature fusion methods, such as multimodal deep learning networks, to jointly learn and represent features of different physical quantities. Alternatively, it can use a set of coupled equations based on physical models, analyzing the interactions between different physical quantities by establishing mathematical expressions. The co-evolution model refers to a mathematical or computational model that describes how two or more interrelated systems or feature sets influence and co-evolve over time. It is used to capture the complex dynamic dependencies between the components within a system, thereby enabling the prediction and inference of the system's future state. This model can adopt a framework based on Hidden Markov Models (HMMs), using different physical quantity features as observations and constructing the co-evolution model by learning state transition probabilities and observation probabilities. Alternatively, neural network-based sequence models, such as Long Short-Term Memory (LSTM) networks or Transformer models, can be used to model multimodal time series data in order to capture their co-evolutionary patterns.

[0139] This time series alignment and cross-correlation analysis is used to accurately quantify the temporal lag relationship between two time series and the degree of their linear correlation. One implementation is to use the Dynamic Time Warping (DTW) algorithm, which non-linearly aligns the two time series to find the optimal matching path, thereby eliminating non-linear distortion on the time axis. Another implementation is to calculate the Pearson correlation coefficient of the two time series at different time delays, and determine the optimal time delay by finding the delay corresponding to the maximum correlation coefficient. This time delay refers to the lag or lead of one signal relative to another on the time axis. The correlation coefficient quantifies the strength and direction of the linear relationship between the two variables. Time delay reveals causality or response time, while the correlation coefficient indicates the synchronicity or antisynchronicity of their changes. Time delay can be determined by the peak position of the cross-correlation function; for example, the delay corresponding to the maximum value of the cross-correlation function is the optimal time delay. The correlation coefficient can be obtained by calculating statistics such as the Pearson correlation coefficient, Spearman's rank correlation coefficient, or Kendall's rank correlation coefficient. The first coupling relationship refers to the mutual influence mechanism between the thermal accumulation level of the target UAV's power amplifier and the motor drive current. This relationship reveals how the motor's operating state (such as current magnitude) affects the power amplifier's temperature changes and performance drift, providing crucial physical constraints for subsequent model construction. This relationship can be represented as a mathematical function, such as a linear or nonlinear regression model, taking the motor drive current as input and the power amplifier's thermal accumulation level (such as frequency drift rate) as output. Alternatively, it can be represented as a lookup table, storing the time delay and correlation coefficient of the frequency drift rate changes corresponding to different current peak amplitudes.

[0140] This event synchronization analysis is used to detect whether there is temporal synchronization or close correlation between two or more independent event sequences. One implementation is to use an event overlap metric, which assesses synchronicity by calculating the number of event pairs that occur simultaneously within a specific time window. Another implementation is to use a synchronization detection method based on conditional probability, for example, calculating the probability of another event occurring within a preset time window after one event occurs, to determine whether the two are synchronized. The time alignment relationship refers to the correspondence or matching of specific event occurrence times on the time axis in two or more event sequences. This relationship is used to accurately capture the simultaneous occurrence times of different physical phenomena (such as a power amplifier entering the nonlinear region and a motor being overloaded), providing important synchronization event information for co-evolution models. This relationship can be represented as a series of timestamp pairs, each containing a first event time and a second event time, with the time difference between them satisfying a preset synchronization tolerance. Alternatively, it can be represented as a density distribution of synchronized events, describing the frequency of synchronized events occurring within a specific time period. The second coupling relationship refers to the interrelationship between the target UAV's power amplifier entering the nonlinear operating region and the motor entering a heavy load state. This relationship reveals that when the UAV's power system is under heavy load, the operating point of the power amplifier in the RF transmission link may change, or even enter the nonlinear region, thereby affecting signal quality and hardware characteristics. This relationship can be expressed as a logical rule, such as "if the amplitude of the motor current spike exceeds the threshold, the power amplifier has a high probability of entering the nonlinear operating region." Alternatively, it can be expressed as a probability distribution, describing the conditional probability of the power amplifier entering the nonlinear region under heavy motor load conditions.

[0141] Initial constraints refer to a set of pre-defined limitations or known conditions when constructing a mathematical model or optimization problem. These conditions guide the model's learning process, reduce its degrees of freedom, improve its convergence speed and accuracy, and ensure that the model conforms to actual physical laws. These constraints can manifest as initial values ​​or range limitations for model parameters, such as restricting certain coupling coefficients to a physically reasonable range. Alternatively, they can manifest as structural limitations on the model, such as forcing the model to satisfy certain known physical equations under specific conditions. The multivariate state-space modeling method is a mathematical framework for describing the dynamic behavior of a system. It defines a set of state variables to represent the system's internal state and uses state transition equations and observation equations to describe the evolution of the state and the relationship between the state and observation data. This method can handle multi-input multi-output systems, estimating, predicting, and controlling the system state, and is particularly suitable for dynamic systems with noise and uncertainty. This method can employ Extended Kalman Filter (EKF) or Unscented Kalman Filter (UKF) to handle nonlinear state-space models, estimating the state of the nonlinear system through linearization or sampling approximation. Alternatively, the particle filter method can be used to approximate the posterior probability distribution through Monte Carlo sampling, which is suitable for more complex nonlinear non-Gaussian systems.

[0142] Through the above technical solutions, this application can accurately construct a co-evolution model between the launch hardware characteristics and the dynamic physical state of the target UAV. Specifically, by performing time-series alignment and cross-correlation analysis on the thermal memory effect characteristics and the power electrical feature set, the temporal dynamics and correlation between the thermal accumulation of the power amplifier and the change in the motor drive current can be captured, thereby establishing an accurate first coupling relationship and avoiding model distortion caused by time deviation. Simultaneously, by performing event synchronization analysis on the compression point characteristics and the power electrical feature set, the synchronous events of the power amplifier entering the nonlinear operating region and the motor entering a heavy load state can be identified, ensuring the accuracy of the second coupling relationship and preventing the impact of event misalignment on model consistency. Using these accurately established coupling relationships as initial constraints, and combining them with other key features, a multivariate state-space modeling method is adopted, enabling the constructed co-evolution model to more comprehensively and accurately reflect the complex dynamic dependencies between various physical quantities within the UAV. This improves the reliability of the model's UAV state inference, providing a solid foundation for subsequent active localization calculations and the generation of hardware-oriented interference strategies, thus solving the problem of inaccurate model construction and its impact on the reliability of state inference in related technologies.

[0143] In some embodiments described above in this application, a time-series alignment and cross-correlation analysis is proposed to establish a first coupling relationship by combining the thermal memory effect features in the first hardware feature set with the power electrical feature set in the second physical state feature set. However, during implementation, the existence of acquisition clock deviation may lead to time series misalignment, affecting the accuracy of cross-correlation analysis and thus making it impossible to accurately determine time delay and correlation coefficient.

[0144] In response, this application further proposes a step of performing time-series alignment and cross-correlation analysis on the thermal memory effect characteristics in the first hardware feature set and the power electrical feature set in the second physical state feature set, to determine the time delay and correlation coefficient between the change in frequency drift rate in the thermal memory effect characteristics and the change in current peak amplitude in the power electrical feature set, specifically including: The first time series of frequency drift rate changing with time in the thermal memory effect feature and the second time series of current peak amplitude changing with time in the power electrical feature are time-axis calibrated to eliminate the acquisition clock deviation between the first time series and the second time series, so as to obtain the aligned first time series and the aligned second time series.

[0145] A sliding window cross-correlation calculation is performed on the aligned first time series and the aligned second time series. Within a preset time delay range, the correlation coefficient is calculated sequentially for each time delay value to obtain the cross-correlation function in which the correlation coefficient changes with the delay time.

[0146] Extract the delay time corresponding to the maximum value of the cross-correlation function as the time delay, and use the maximum value of the cross-correlation function as the correlation coefficient.

[0147] The calibration of the time axis involves calibrating the first time series of the frequency drift rate variation over time in the thermal memory effect characteristic and the second time series of the current peak amplitude variation over time in the power electrical characteristic. This aims to eliminate time discrepancies caused by clock asynchrony between different sensors or data acquisition systems, ensuring accurate alignment of the time series of different physical quantities on the time axis. This is a prerequisite for accurate time series analysis (such as cross-correlation). This calibration can be achieved in several ways. For example, it can be based on high-precision timestamp synchronization, where a timestamp generated by a high-precision clock (such as GPS timing or NTP server synchronization) is appended to each data point during data acquisition. During calibration, interpolation or resampling maps the data points of different time series onto a unified time axis. Alternatively, it can be based on characteristic event synchronization, identifying common physical events with specific time points in both time series (e.g., the rising or falling edge of a specific signal), using these events as reference points for time axis translation or stretching calibration.

[0148] A sliding window cross-correlation calculation is performed on the aligned first and second time series. The purpose is to conduct cross-correlation analysis within a finite time window, then slide this window along the time axis to capture correlation changes in local signal regions, avoiding local dynamic features that might be masked by global correlation analysis. This is particularly important for analyzing non-steady-state or dynamically changing systems. In practice, discrete cross-correlation calculation can be used; that is, for two discrete time series, their cross-correlation function is calculated within each sliding window. This calculation is repeated as the window slides. Alternatively, frequency domain cross-correlation calculation can be used. The signal is converted to the frequency domain through Fourier transform, and the cross-correlation function is obtained by calculating the frequency domain product and then performing an inverse Fourier transform. The sliding window corresponds to performing a short-time Fourier transform (STFT) or wavelet transform on the signal.

[0149] Extracting the time delay corresponding to the maximum value of the cross-correlation function as the time delay, and using the maximum value of the cross-correlation function as the correlation coefficient, is a crucial step in quantifying the dynamic relationship between two time series. The cross-correlation function typically has a peak; the time delay corresponding to this peak represents the relative time offset when the strongest correlation between the two signals occurs, while the peak itself (or its normalized value) represents the strength of this strongest correlation. These values ​​can be extracted using peak detection algorithms, which search for local maxima on the cross-correlation function curve and select the largest one as the global maximum. Alternatively, curve fitting (e.g., parabolic fitting) can be performed on the region near the peak of the cross-correlation function to more accurately determine the peak location and magnitude.

[0150] Through the above technical solution, the time axis calibration step directly solves the problem of acquisition clock deviation, ensuring accurate temporal alignment between the frequency drift rate in the thermal memory effect characteristics and the amplitude of the concentrated current spikes in the power electrical characteristics, providing a reliable foundation for subsequent cross-correlation analysis. Sliding window cross-correlation calculation can capture the dynamic and local correlation between two time series, which is crucial for analyzing the complex relationship between unsteady hardware thermal accumulation and power system load changes during UAV flight, avoiding the loss of transient or evolutionary coupling features that global analysis might miss. Accurately extracting the delay time and correlation coefficient corresponding to the maximum value from the cross-correlation function quantifies the strongest temporal relationship and its strength. These precisely determined time delays and correlation coefficients improve the accuracy and robustness of the first coupling relationship. Based on this, the constructed co-evolution model will more accurately reflect the co-evolution law between the target UAV's launch hardware characteristics and its power physics state, making the joint inference results of the target UAV's current maneuvering state, current load distribution state, and current hardware operating point more reliable, thus providing a more accurate basis for active positioning calculation and the generation of hardware-oriented interference strategies.

[0151] In some of the above-mentioned schemes in this application, event synchronization analysis is proposed to establish the coupling relationship between the moment when the target UAV power amplifier enters the nonlinear operating region and the moment when the motor enters the heavy load state. However, in this process, since the events of the output power entering the compression region and the current spike exceeding the load threshold may occur at different times or there may be deviations in the acquisition clock, the time alignment relationship is inaccurate, which affects the accuracy of the construction of the co-evolution model and thus reduces the effectiveness of the interference strategy.

[0152] In response, this application further proposes to perform event synchronization analysis on the compression point features in the first hardware feature set and the power electrical feature set in the second physical state feature set, detecting the time alignment relationship between the moment when the output power enters the compression zone in the compression point features and the moment when the current peak amplitude in the power electrical feature set exceeds the load threshold, specifically including: Time series analysis is performed on the compression point feature to detect every moment when the output power value first drops below the preset decibel level of the expected power value in the linear region, and each of these moments is taken as the first event moment sequence when the output power enters the compression region.

[0153] Threshold detection is performed on the current peak amplitude of the power electrical feature set. The current peak amplitude is detected at every moment when it exceeds the preset load threshold, and each moment is taken as the second event time sequence when the motor enters the heavy load state.

[0154] The first event time sequence and the second event time sequence are matched. For each first event time in the first event time sequence, the second event time that is closest in time and has a time difference less than the preset synchronization tolerance is found in the second event time sequence. The found second event time is then paired with the corresponding first event time to form a synchronization event pair.

[0155] Based on the time difference between the first event time and the second event time in the synchronous event pair, the time offset corresponding to each synchronous event pair is determined, and the time offset and the distribution density of the synchronous event pair in the first event time sequence and the second event time sequence are used as the time alignment relationship between the moment when the output power enters the compression zone in the compression point feature and the moment when the current peak amplitude in the power electrical feature exceeds the load threshold.

[0156] Specifically, time-series analysis is performed on the compression point feature to detect each moment when the output power value first drops below a preset decibel level of the expected power value in the linear region. Each of these moments is considered the first event sequence indicating that the output power has entered the compression region, aiming to accurately identify the moment when the target UAV's power amplifier begins to enter the nonlinear operating region. This is crucial for understanding the actual operating state of the power amplifier and its potential nonlinear distortion. Specifically, methods such as sliding window averaging, Kalman filtering, or wavelet analysis can be used to smooth and denoise the time-series data of the compression point feature to reduce the impact of measurement errors and instantaneous fluctuations. By comparing the output power value at each time point with a pre-established expected power model in the linear region (e.g., obtained through linear regression of the power amplifier's response under low power input), the moment when the deviation first exceeds a preset negative decibel threshold (e.g., -1 dB or -3 dB) is marked as an event moment of "entering the compression region." Alternatively, the first or second derivative of the output power time series can be calculated. When the growth rate of the output power decreases, or when the rate of change of its change rate shows a significant negative inflection point, a dynamic threshold or adaptive algorithm can be used to determine that it has entered the compression zone and record it as the first event time series.

[0157] Threshold detection is performed on the current spike amplitude of the power electrical feature set. Every moment when the current spike amplitude exceeds a preset load threshold is detected, and each moment is recorded as a second event sequence indicating that the motor has entered a heavy load state. This aims to identify the moments when the target UAV's power system (motor) is subjected to heavy loads. This provides physical background information on the nonlinear behavior of the power amplifier, helping to establish a correlation between hardware characteristics and power state. Specifically, one or more fixed current spike amplitude thresholds can be preset based on the UAV motor model, rated power, and typical flight mission data. When the real-time acquired current spike amplitude continuously or instantaneously exceeds these thresholds, the motor is considered to have entered a heavy load state, and the corresponding moment is recorded as a second event sequence. Alternatively, an adaptive threshold algorithm can be used. For example, based on historical data or real-time operating status, the load threshold can be dynamically adjusted. A load threshold more consistent with the current operating conditions can be calculated in real-time using machine learning models or statistical methods based on parameters such as the motor's average operating current, battery voltage, or ambient temperature. When the current spike amplitude exceeds this dynamic threshold, it is marked as a heavy load event.

[0158] Event matching is performed between the first event time sequence and the second event time sequence. For each first event time in the first event time sequence, the closest event time in the second event time sequence with a time difference less than a preset synchronization tolerance is found. This second event time is then paired with its corresponding first event time to establish a synchronization event pair. This aims to establish a direct temporal correlation between the power amplifier entering nonlinear operation and the motor entering heavy load state—two events in different physical domains. This is crucial for subsequent coupling analysis. Specifically, a tree-based data structure (such as a kd-tree) or hash table can be used to store the second event time sequence for fast lookup. For each time in the first event time sequence, a nearest neighbor search is performed in the second event time sequence, while simultaneously checking if the time difference between the found nearest time and the current first event time is less than a preset synchronization tolerance (e.g., milliseconds or microseconds). If the condition is met, the two are paired. Alternatively, a sliding time window can be used. For each event in the first event time sequence, a sliding window centered on that event and with a width equal to the synchronization tolerance is defined. Then, it is checked whether any event in the second event time sequence falls within this window. If multiple events exist, the event that is closest in time within the window can be selected for pairing.

[0159] Based on the time difference between the first and second event moments in the synchronized event pair, the time offset corresponding to each synchronized event pair is determined. This time offset, along with the distribution density of the synchronized event pair in the first and second event moment sequences, is used as the time alignment relationship between the moment when the output power enters the compression zone in the compression point feature and the moment when the current peak amplitude in the power electrical feature exceeds the load threshold. This aims to quantify and characterize the precise time relationship between the power amplifier's nonlinear response and the motor's heavy load state. This not only provides a numerical value for the time difference but also considers the frequency and consistency of event synchronization, providing a more robust input for constructing a co-evolution model. Specifically, for each synchronized event pair, its time difference is calculated as the time offset. Then, statistical analysis is performed on all valid time offsets, such as calculating the mean, median, and standard deviation, to characterize the overall time offset trend and volatility. Simultaneously, the distribution density of synchronized event pairs can be evaluated by calculating the ratio of the number of synchronized event pairs to the total number of events within a specific time window. Alternatively, the first and second event moment sequences can be converted into binary event sequences, and then cross-correlation analysis can be performed on these two binary sequences. The peak value of the cross-correlation function will indicate the most likely time offset, while the height of the peak value can reflect the strength or distribution density of event synchronization.

[0160] Through the above technical solutions, this application can identify the starting point of hardware nonlinear response by performing time series analysis on the compression point features, avoiding misjudgments caused by signal fluctuations. By threshold detection of the current peak amplitude in the power electrical feature set, critical events of power system load changes can be reliably captured, providing a basis for event matching. Through the event matching process, event pairs that are closest in time and have a time difference less than the preset synchronization tolerance are found, effectively associating hardware events and physical state events, solving the time deviation problem. Based on the time difference of the synchronized event pairs, the time offset is determined, and the distribution density is considered to quantify the time alignment relationship, providing accurate input for the cooperative model. These steps work together to accurately detect the time alignment relationship between hardware characteristic events and power physical state events, solving the problem of inaccurate event synchronization, thereby improving the reliability of the cooperative evolution model. This precise coupling relationship enables the system to more accurately understand the hardware response of the UAV under different load conditions, thereby generating more targeted and effective hardware-oriented jamming strategies, improving the target recognition accuracy and jamming effectiveness of the countermeasure system.

[0161] In some embodiments described above in this application, a co-evolutionary model is proposed to describe the co-evolution between the launch hardware characteristics and the dynamic physical state of a target UAV. However, in its implementation, challenges exist in how to effectively integrate hardware features and physical state features into observation variables, generate dynamic state transition equations and observation equations based on the coupling relationship, and accurately estimate model parameters using historical data to ensure that the model can reliably reflect the interaction and evolution law between the hardware operating point and the dynamic load state.

[0162] To address this, this application further proposes a multivariate state-space modeling method to construct the co-evolution model. Specifically, the thermal memory effect, intermodulation distortion, and power-added efficiency features from the first hardware feature set, and the power electrical, mechanical vibration, and acoustic sensing feature sets from the second physical state feature set, are used as observation variables in the state-space model. These observation variables are features that can be directly measured by the system or extracted from the raw data. They serve as inputs to the state-space model to reflect the external performance of the target UAV at different points in time. By using these multi-dimensional, cross-physical-quantity features as observation variables, the nonlinear characteristics of the UAV launch hardware and the real-time physical state of its power system can be comprehensively captured, providing a rich and reliable data foundation for subsequent state estimation. For example, these feature values ​​can be directly used as elements of the observation vector, or before being used as observation variables, these features can be weighted and fused or subjected to dimensionality reduction processing such as principal component analysis (PCA) or independent component analysis (ICA) to eliminate redundant information, highlight key features, and form a more compact observation vector.

[0163] Based on this, and using the first and second coupling relationships, state transition equations are generated for the state vectors in the state-space model between different time steps. These state transition equations describe the mutual influence and co-evolution between the launch hardware operating point and the power load state of the target UAV. The state transition equations are the core of the state-space model, defining how the internal states of the system dynamically evolve over time. By utilizing the previously established coupling relationships between the hardware and the power system, this application can construct a dynamic model that accurately reflects these complex interactions. For example, linear difference equations or linear differential equations can be used to describe the state transitions, where the parameters of the state transition matrix can be parameterized according to the first and second coupling relationships. Considering the inherent nonlinear characteristics of the UAV system, nonlinear functions can also be used to describe the state transitions, such as through neural networks, Gaussian process regression, or other nonlinear regression methods, to more accurately capture the complex co-evolution patterns.

[0164] Simultaneously, based on the physical mapping relationship between the observed variable and the state vector, the observation equations for the state-space model are generated. These observation equations are used to map the state vector to the observation space of the first hardware feature set and the second physical state feature set. The observation equations establish a bridge between the internal, unmeasurable states of the system and the external, observable features, enabling the model to update and correct its estimates of the internal states using actually acquired observation data. For example, linear equations can be used to describe the observation relationship, where the observation matrix can be constructed based on known physical principles or empirical relationships. For more complex physical mapping relationships, such as when there are threshold effects, saturation effects, or polynomial relationships between the observed variable and the state vector, nonlinear observation functions can be used.

[0165] The first hardware feature set sequence and the second physical state feature set sequence of the target UAV during its historical flight period are obtained. The unknown parameters in the state transition equation and the observation equation are estimated to obtain the co-evolution model. Parameter estimation is a crucial step in model construction; it utilizes a large amount of historical data to "train" the model, enabling it to accurately reflect the actual behavior of the target UAV. By analyzing the hardware feature set sequence and the physical state feature set sequence in the historical flight data, methods such as maximum likelihood estimation (MLE), least squares, expectation-maximization (EM) algorithms, or Bayesian estimation can be used to optimize the unknown parameters in the state transition equation and the observation equation. This results in a calibrated co-evolution model that accurately reflects the interaction and co-evolution law between the UAV's launch hardware operating point and power load state.

[0166] Through the above technical solutions, this application addresses the challenges of observation variable selection, dynamic equation generation, and parameter estimation in the construction of co-evolutionary models using a multivariate state-space modeling method, thereby improving the model's accuracy and reliability. By using multi-dimensional hardware and physical state features as observation variables, it ensures that the model input covers multiple dimensions of the UAV's state, avoiding the bias caused by single features and enabling the model to comprehensively reflect the overall state of the UAV. State transition equations are generated based on established coupling relationships, utilizing the correlation between hardware and dynamic states to solve the problem of dynamically modeling mutual influences and enhancing the model's ability to capture complex dynamic evolution. The observation equations establish mappings through physical correlations, solving the challenge of associating hidden states with observation data and maintaining data consistency and verifiability. Parameter calibration using historical data sequences solves the error problem caused by model parameter uncertainty, ensuring that the model is optimized based on real flight data. The obtained co-evolution model can accurately reflect the complex and dynamic co-evolution law between the launch hardware characteristics and dynamic physical state of the target UAV, providing a solid and reliable foundation for the subsequent joint inference of the target UAV's current maneuvering state, current load distribution state and current hardware operating point, thereby supporting more accurate active positioning calculation and the generation of hardware-oriented jamming strategies.

[0167] In some of the solutions described above in this application, a co-evolutionary model can be constructed to describe the co-evolutionary relationship between the launch hardware characteristics and dynamic physical state of the target UAV. However, in actual countermeasure scenarios, there are technical challenges in how to efficiently jointly solve the first hardware feature set and the second physical state feature set acquired in real time using the constructed co-evolutionary model to accurately infer the current hardware operating point, load distribution state, and maneuver state of the target UAV, and integrate these multi-dimensional state information into a unified joint inference result. Specifically, it is necessary to solve problems such as how to fuse high-dimensional feature data observed in real time into a state vector representing the overall state of the system, how to separate physically interpretable hardware operating point parameters and load states from the abstract state vector, and how to predict the future motion trend of the target based on the current state to identify its maneuvering intentions.

[0168] To address this, this application further proposes a method for jointly inferring the current maneuvering state, current load distribution state, and current hardware operating point of a target UAV by jointly solving a first hardware feature set and a second physical state feature set based on a co-evolution model. The method includes the following steps: inputting the first hardware feature set and the second physical state feature set, acquired and processed at the current moment, into the co-evolution model; performing Kalman filtering recursion on the state transition equation and observation equation in the co-evolution model to obtain the system state vector at the current moment; extracting state components corresponding to the nonlinear response characteristics of the power amplifier from the system state vector at the current moment as parameters of the current hardware operating point; and extracting state components corresponding to the electrical and mechanical vibration characteristics of the power system from the system state vector at the current moment as the current load distribution state; predicting the state evolution trajectory of the target UAV within a preset future time domain based on the system state vector at the current moment; and identifying the maneuvering state of the target UAV at the current moment based on the slope and inflection point characteristics of the state evolution trajectory.

[0169] The first hardware feature set is a dataset characterizing the inherent nonlinear characteristics of the target UAV's launch hardware. Its main function is to capture the nonlinear response characteristics of key components in the UAV's RF transmission link, such as power amplifiers, under different operating conditions, including compression point characteristics, intermodulation distortion characteristics, power-added efficiency characteristics, and thermal memory effect characteristics. These characteristics are crucial for UAV individual identification and hardware health status assessment. The second physical state feature set is a dataset characterizing the real-time dynamic physical state of the target UAV. Its main function is to reflect the actual operating status of the UAV's power system (such as motors and propellers) during flight, including sets of power electrical features, mechanical vibration features, and acoustic sensing features. These features reveal the UAV's real-time load, energy consumption, and potential mechanical failures, providing key information for determining the UAV's flight intentions and physical vulnerabilities. The co-evolution model is a mathematical model describing the mutual influence and co-evolution laws between the target UAV's launch hardware characteristics and dynamic physical state. This model is typically constructed based on multivariable state-space modeling methods and includes state transition equations and observation equations. The state transition equations characterize the changes in the system's internal state over time, while the observation equations establish the mapping relationship between the system state and observable characteristics. Its function is to provide a unified framework that integrates the characteristics of multi-source heterogeneous hardware and physical state, and enables dynamic estimation and prediction of system state.

[0170] Kalman filtering recursion is an optimal linear estimation method used to estimate the state of a dynamic system from a series of noisy measurements. Its role is to combine the first set of hardware features and the second set of physical state features (as observations) acquired at the current moment with a co-evolutionary model (containing state transition equations and observation equations) through a predict-update iterative process, thereby obtaining an optimal estimate of the system state vector. One implementation uses a standard Kalman filter, suitable for linear systems. In each iteration, the current state is predicted based on the state estimate and state transition equations from the previous moment, and then the state estimate is updated based on the current observations and observation equations. Another implementation uses an extended Kalman filter (EKF) or an unscented Kalman filter (UKF), suitable for nonlinear systems. EKF approximates the nonlinear function by linearizing it, while UKF more accurately handles nonlinearity by propagating the mean and covariance through deterministic sampling points. The system state vector is a mathematical representation in the co-evolutionary model used to fully describe the current internal state of the target UAV. This vector typically contains multiple components that collectively characterize the operating point of the UAV launch hardware, the load on the propulsion system, energy consumption, mechanical vibration modes, and other potential variables related to flight status. Its purpose is to abstract multi-dimensional, heterogeneous feature data into a unified, mathematically operable internal state representation, providing a foundation for subsequent state decoupling and maneuver state identification.

[0171] Extracting state components corresponding to the nonlinear response characteristics of a power amplifier aims to identify and separate the parts directly related to the nonlinear response characteristics of the power amplifier from the abstract system state vector. Its role is to assign specific physical meaning to the state components in the mathematical model, enabling them to directly reflect the actual operating condition of the power amplifier. One implementation method is to map specific dimensions or combinations thereof in the system state vector to key nonlinear parameters of the power amplifier through a predefined mapping matrix or function, such as compression point margin, intermodulation distortion, and power-added efficiency. Another implementation method is to use machine learning models (such as support vector machines or neural networks) to train the system state vector, enabling it to automatically identify and output state components related to the nonlinear response characteristics of the power amplifier. Current hardware operating point parameters are extracted specific indicators used to describe the actual operating state of launch hardware such as the power amplifier of the target UAV at the current moment. These parameters can quantify the hardware's operating load, performance, and potential degradation trends. For example, they can be the power amplifier's output power, gain compression, third-order intermodulation intercept, and thermal memory effect intensity. These parameters are crucial for assessing the health status of the hardware and identifying its vulnerabilities.

[0172] Extracting state components corresponding to the electrical and mechanical vibration characteristics of the power system aims to identify and separate the parts directly related to these characteristics from the system state vector. Its function is to transform abstract system state components into physical quantities that reflect the actual operating status of the UAV power system. One approach is to perform principal component analysis (PCA) or independent component analysis (ICA) on the system state vector to identify independent components highly correlated with electrical feature sets (such as current spike amplitude and duty cycle) and mechanical vibration feature sets (such as instantaneous amplitude and instantaneous frequency). Another approach is to utilize pre-trained pattern recognition algorithms to classify or regress relevant components in the system state vector to specific power system parameters, such as motor speed, propeller thrust, and battery discharge rate. The current load distribution state is extracted to describe the actual load experienced by the target UAV power system at the current moment. Its function is to quantify the external drag, internal power consumption, and resulting power system stress faced by the UAV during flight. For example, it can be classified into discrete states such as light load, medium load, heavy load, and overload, or represented as continuous values ​​such as specific thrust requirements and power consumption levels. These states are of great significance for judging the UAV's flight intentions and energy reserves.

[0173] Predicting the state evolution trajectory of a target UAV within a predetermined future time domain utilizes the state transition equation in a co-evolutionary model. Starting with the current system state vector, iterative calculations infer the system state sequence of the target UAV over a future period (predetermined time domain). Its purpose is to provide the ability to predict the target's future behavior, transforming static current state information into dynamic future trend information. One implementation method is to use numerical integration methods such as the Euler method or the Runge-Kutta method to discretize and iteratively solve the state transition equation, thereby generating a series of predicted state vectors for future moments. Another implementation method is to use Monte Carlo simulation, considering system noise and model uncertainties, to generate multiple possible future state evolution trajectories to evaluate the robustness of the prediction results. The state evolution trajectory is the predicted sequence of changes in the target UAV's system state vector over time within the predetermined future time domain. This trajectory reflects the expected development trend of the UAV's flight state and hardware performance under the current power and hardware operating point. Its purpose is to provide an intuitive and dynamic view, helping analysts understand the target UAV's impending action patterns.

[0174] Slope and inflection point features refer to the points on the trajectory where the signs of the first and second derivatives change. Slope features reflect the rate and direction of system state change, while inflection point features reflect the turning point in the trend of system state change. Their function is to quantify the dynamic characteristics of the trajectory through differential geometric analysis, thus providing a mathematical basis for identifying maneuver states. One implementation method is to numerically differentiate the trajectory and calculate its first and second derivatives. Slope features can be determined by the sign and magnitude of the first derivative, while inflection point features can be detected by changes in the sign of the second derivative. Another implementation method is to use wavelet transform or Fourier transform-based methods to analyze the frequency components of the trajectory, thereby extracting features related to slope and inflection points. Identifying the current maneuver state of the target UAV involves determining the flight maneuver mode currently being executed by the target UAV based on the slope and inflection point features of the trajectory. Its function is to map the mathematical characteristics of the trajectory to physically meaningful maneuvering behaviors, providing crucial information for countermeasure decisions. One approach is to build a rule base or decision tree that associates different combinations of slope features (e.g., consistently positive, consistently negative, approaching zero) and inflection point features (e.g., presence, location) with predefined maneuver states (e.g., climb, descent, acceleration, deceleration, hovering, constant speed cruising, turning, etc.). Another approach is to utilize a supervised learning model (e.g., support vector machine, random forest, or deep neural network) to train the model using slope and inflection point features as input, automatically classifying and identifying the current maneuver state.

[0175] Through the above technical solution, this application addresses the technical challenge of efficiently jointly solving the first hardware feature set and the second physical state feature set acquired in real time using a constructed co-evolutionary model to accurately infer the current hardware operating point, load distribution state, and maneuver state of the target UAV, and integrating this multi-dimensional state information into a unified joint inference result. Specifically, by inputting the first hardware feature set and the second physical state feature set acquired in real time into the co-evolutionary model and using Kalman filtering recursion, the system state vector that best reflects the current true operating condition of the target can be iteratively estimated from noisy real-time observation data. This solves the problem of integrating discrete, multi-source feature data into a unified, continuous system state representation. Based on this, by extracting the state component corresponding to the nonlinear response characteristics of the power amplifier as the current hardware operating point parameter from the system state vector, and extracting the state component corresponding to the electrical characteristics and mechanical vibration characteristics of the power system as the current load distribution state, the physical semantics of the abstract mathematical vector are decoupled, enabling the joint inference result to directly serve subsequent hardware weakness identification and interference strategy matching. Furthermore, based on the system state vector at the current moment, the system predicts the trajectory of the target UAV's state evolution within a preset future time domain, and identifies its maneuvering state based on the trajectory's slope and inflection point characteristics. This provides the ability to predict the target's future behavior, transforming static state information into dynamic trend information and providing a forward-looking decision-making basis for selecting the timing of interference. Overall, this application achieves a complete transformation from multi-source feature data to multi-dimensional state inference results through robust state estimation using Kalman filtering, accurate decoupling of physical semantics, and effective prediction of dynamic behavior. This provides comprehensive, accurate, and forward-looking input information for subsequent active localization calculations and hardware-oriented interference strategy generation, improving the target recognition accuracy and interference effectiveness of the UAV countermeasure system.

[0176] In some of the solutions described above in this application, a state recursion method using a co-evolution model is proposed to estimate the system state of the target UAV. However, in its implementation, the state estimation may be inaccurate due to the influence of system noise and observation noise, which leads to a decrease in the reliability of the localization and jamming strategies.

[0177] To address this, this application further proposes a Kalman filter recursive method using the state transition equation and observation equation in the aforementioned co-evolution model to obtain the system state vector at the current moment. Specifically, this process includes: acquiring the system state vector and the state covariance matrix from the previous moment. The system state vector represents the comprehensive state of the target UAV at a given moment, potentially containing key information such as its hardware operating point, power load, and maneuvering trends. The state covariance matrix quantifies the uncertainty or error range of these state estimates. Acquiring this information from the previous moment is the foundation for the Kalman filter's recursive prediction, ensuring the continuity of state estimation and the utilization of historical information. This information can be stored in memory as the filtering result of the previous moment or obtained from the previous processing cycle via a real-time data stream interface.

[0178] By inputting the system state vector from the previous moment into the state transition equation of the co-evolutionary model, the predicted state vector for the current moment is obtained. The state transition equation is a core component of the co-evolutionary model, describing the evolution of the system state over time. By inputting the known state from the previous moment into this equation, the possible state the system can reach at the current moment can be predicted based on the system's inherent dynamics. This prediction is a preliminary estimate based on the system's own evolutionary laws. The state transition equation can be a nonlinear function or a linear matrix operation, for example, constructed based on a physical model or a data-driven model. Based on this, the predicted state covariance matrix for the current moment is calculated using the state covariance matrix from the previous moment and the process noise covariance matrix of the state transition equation. The predicted state covariance matrix reflects the uncertainty of the predicted state; it considers not only the uncertainty of the state estimate from the previous moment but also the uncertainty of the system model itself, i.e., process noise. The process noise covariance matrix quantifies the random error introduced by the state transition equation during the prediction process. By combining these two, the reliability of the prediction results can be assessed more accurately. This calculation typically follows the standard formula for Kalman filtering, where the process noise covariance matrix can be determined through statistical analysis or experimental measurement of the system model error.

[0179] The observation equations in the co-evolutionary model are obtained. These equations establish the mathematical relationship between the system state and actual observation data, describing how to derive measurable features (observation vectors) from the system state. Obtaining the observation equations is a prerequisite for the Kalman filter's correction step, as it defines how to compare the actual observations with the predicted state. The observation equations can be functions that map the UAV's hardware operating point, power load, and other states to a first hardware feature set and a second physical state feature set. Then, the Kalman gain matrix is ​​calculated based on the predicted state covariance matrix at the current moment and the observation noise covariance matrix of the observation equations. The Kalman gain matrix is ​​a core parameter of the Kalman filter, determining the weights assigned to the predicted state and actual observation data when updating the system state. The calculation of the gain matrix comprehensively considers the uncertainty of the predicted state and the uncertainty of the observation data. If the observation data is more reliable, the Kalman gain will be larger, making the observation data have a greater impact on the state update. Conversely, if the prediction is more reliable, the Kalman gain will be smaller. The observation noise covariance matrix can be determined through statistical analysis of sensor measurement errors or calibration experiments.

[0180] Furthermore, the first hardware feature set and the second physical state feature set at the current moment are used as the observation vector at the current moment. The observation vector is the data actually measured by the system at the current moment, which contains real-time information about the launch hardware characteristics and dynamic physical state of the target UAV obtained through radio spectrum sensing and remote non-contact sensing. These actual observation data are used as the observation vector as input for Kalman filtering to correct the state based on model prediction. The construction of the observation vector involves appropriately combining and formatting the first hardware feature set and the second physical state feature set. Based on the predicted state vector at the current moment, the Kalman gain matrix, and the residual between the current observation vector and the mapping result of the observation equation to the predicted state vector, the system state vector at the current moment is updated, and the state covariance matrix at the current moment is updated based on the Kalman gain matrix and the predicted state covariance matrix at the current moment. The residual is the difference between the actual observation vector and the "predicted observation" obtained by mapping the predicted state vector through the observation equation, reflecting the inconsistency between prediction and reality. The Kalman gain matrix, weighted by this residual, is used to correct the predicted state vector, resulting in a more accurate and reliable system state vector at the current time step. Updating the state covariance matrix reflects the reduction in uncertainty in the system state estimate after correction using observed data. This update process demonstrates that the understanding of the system state becomes more precise after incorporating actual observation information. The updated covariance matrix serves as the input to the Kalman filter at the next time step, ensuring continuous optimization throughout the recursive process.

[0181] Through the aforementioned technical solution, this application leverages the recursive nature of Kalman filtering to effectively integrate the predictive capabilities of the co-evolutionary model with real-time multi-source observation data (a first hardware feature set and a second physical state feature set), thereby improving the accuracy and reliability of target UAV system state estimation. This method, by precisely quantifying and processing system noise and observation noise, enables accurate inferences about the target UAV's current hardware operating point parameters, current load distribution state, and current maneuver state even in complex and variable environments. This provides a more solid and reliable foundation for subsequent active localization calculations and allows the generated hardware-oriented jamming strategy to more accurately match the target UAV's real-time state, significantly improving the effectiveness and targeting of the jamming and avoiding the inefficiency and collateral damage problems of traditional general-purpose jamming.

[0182] In some of the solutions mentioned above in this application, the system state vector at the current moment is used to predict the state evolution trajectory of the target UAV in a future preset time domain to assist in identifying the maneuver state. However, in this process, the prediction may rely on simple linear extrapolation or static models, which cannot accurately reflect the dynamic coupling relationship between the UAV hardware characteristics and the dynamic physical state, resulting in large deviations in future state prediction and affecting the pertinence and timeliness of the interference strategy.

[0183] To address this, this application further proposes a method for predicting the state evolution trajectory of a target UAV within a future preset time domain based on the system state vector at the current moment. The method includes: obtaining the state transition equation in the cooperative evolution model, and inputting the system state vector at the current moment as the initial state into the state transition equation; performing multi-step iterative recursion through the state transition equation to sequentially calculate the predicted state vector for each future moment until the future preset time domain is covered; and combining the predicted state vectors for each future moment in chronological order to generate the state evolution trajectory of the target UAV within the future preset time domain.

[0184] Specifically, the state transition equation in the co-evolution model is obtained, and the current system state vector is used as the initial state and input into the state transition equation. The state transition equation is the core component of the co-evolution model, describing the evolution of the system state over time. It can predict the current system state based on the previous system state and possible inputs, thus capturing the interaction and co-evolution between the UAV launch hardware operating point and the power load state. This state transition equation can be a nonlinear difference equation system, learned from historical data using machine learning methods (e.g., Recurrent Neural Networks (RNNs) or Long Short-Term Memory Networks (LSTMs), capable of capturing complex temporal dependencies. Alternatively, the state transition equation can be a hybrid model combining physical mechanisms and empirical models, containing physical equations describing power amplifier heat accumulation, battery discharge characteristics, and the relationship between motor speed and thrust, with model parameters determined through system identification methods. The current system state vector is a comprehensive representation of all key state variables of the target UAV at the current time point. It includes information such as the current maneuvering state, current load distribution state, and current hardware operating point obtained from the joint solution, serving as the starting point for future state prediction. The current system state vector can be updated based on observation data (a first hardware feature set and a second physical state feature set) at the current moment using a Kalman filter or other state estimation algorithm, ensuring its accuracy and real-time performance. Alternatively, the current system state vector can be an extended state vector that fuses data from multiple sensors (e.g., inertial measurement unit, GPS, etc.), estimated using a data fusion algorithm (e.g., Extended Kalman Filter (EKF), Unscented Kalman Filter (UKF)) to provide more comprehensive initial state information. Using this current system state vector as the initial state and inputting it into the state transition equation means using the current system state vector as the starting condition or input parameter of the state transition equation to initiate the prediction process for the future state. This can be achieved by directly substituting the vector into the equation for calculation. Alternatively, in a neural network-based model, the current state vector can be used as the initial hidden state of the network or the first element of the input sequence.

[0185] This process involves multi-step iterative recursion using the state transition equation, sequentially calculating the predicted state vector for each future time step until the predetermined future time domain is covered. This multi-step iterative recursion utilizes the state transition equation, using the current prediction result as the input for the next time step, progressively extrapolating forward to calculate the system state for multiple future time steps. This iterative approach can simulate the dynamic evolution of the system over time. Numerical integration methods such as the Euler method and the Runge-Kutta method can be used to solve the state transition equation at discrete time steps, gradually obtaining the future state. Alternatively, for deep learning-based models, sequence-to-sequence prediction can be used, allowing the model to autonomously generate a sequence of states for multiple future time steps given an initial state. The predicted state vector for each future time step refers to the estimated system state value obtained for each discrete future time point during the multi-step iterative recursion. Each predicted state vector contains information such as the UAV's hardware operating point, load distribution, and maneuvering trends at that time. The dimension and composition of the predicted state vector can be consistent with the initial system state vector, ensuring the completeness of the prediction results. Alternatively, depending on the forecasting requirements, only the components in the predicted state vector related to specific points of interest (e.g., power amplifier temperature, motor speed) can be extracted. The preset time domain covering the future refers to the duration of the forecasting process. The length of this time domain needs to be set according to the actual application requirements (e.g., the lead time of the jamming strategy, the UAV maneuver response time). This preset time domain can be a fixed length, such as 5 seconds or 10 seconds in the future. Alternatively, the preset time domain can be dynamically adjusted in real time according to the UAV's current maneuvering state or mission requirements; for example, shortening the forecast time domain during high-speed maneuvers and lengthening it during stable flight.

[0186] The predicted state vectors for each future moment are combined in chronological order to generate the state evolution trajectory of the target UAV within the predetermined future time domain. This chronological combination involves sorting and connecting the discrete predicted state vectors obtained through multi-step iterative recursion based on their corresponding timestamps, forming a continuous or quasi-continuous state sequence. This can be achieved by storing the predicted state vectors for each moment in an array or list and arranging them according to their time indices. Alternatively, interpolation algorithms (e.g., linear interpolation, spline interpolation) can be used to smooth the discrete predicted points, generating a more continuous trajectory. This state evolution trajectory refers to the dynamic curve or sequence of key parameters of the target UAV, such as its hardware operating point, load distribution, and maneuvering state, changing over time within the predetermined future time domain. This trajectory visually demonstrates the possible future behavioral trends of the UAV. The trajectory can be represented as multi-dimensional time series data, with each dimension corresponding to a state variable. Alternatively, the trajectory can be displayed through a visualization interface, such as plotting curves showing the changes in parameters like power amplifier temperature, motor speed, and thrust over time.

[0187] Through the aforementioned technical solution, this application utilizes the state transition equation in the co-evolution model for multi-step iterative recursion, enabling dynamic and accurate prediction of the target UAV's state evolution trajectory within a predetermined future time domain. Specifically, by obtaining the state transition equation from the co-evolution model and using the precisely estimated system state vector at the current moment as the initial input, it ensures that the prediction starting point closely matches the UAV's real-time physical state, avoiding error accumulation caused by inaccurate initial states in traditional prediction methods. The multi-step iterative recursion using the state transition equation fully considers the complex coupling relationship between the UAV's launch hardware characteristics and its dynamic physical state, allowing the prediction results to dynamically reflect the mutual influence and co-evolution between these physical quantities, rather than simple linear extrapolation or static models. This improves the accuracy and reliability of future state prediction, especially when the UAV performs complex maneuvers or experiences load changes. Combining these predicted state vectors in chronological order into a complete state evolution trajectory provides comprehensive and forward-looking data support for subsequent maneuver state identification. This precise and dynamic predictive capability enables the system to anticipate the behavior trends of drones earlier and more accurately, thus providing a key basis for generating more targeted, timely, and effective hardware-oriented jamming strategies and improving the overall performance of the drone countermeasure system.

[0188] In some of the solutions mentioned above in this application, an active positioning solution and hardware-oriented interference strategy is proposed to improve the target recognition accuracy and interference effectiveness. However, in the implementation process, the positioning result may be affected by Doppler frequency shift and body vibration, resulting in errors. The interference strategy may not be able to accurately match the current hardware weaknesses and load status, resulting in inappropriate interference timing or mismatched duration, thereby reducing interference efficiency.

[0189] To address this, this application further proposes a method for actively locating a target UAV based on joint inference results, target transmitted signals, and target physical field data, and generating a hardware-oriented jamming strategy that matches the joint inference results. The implementation of this strategy includes the following steps: Based on the mechanical vibration feature set and acoustic sensing feature set from the physical field data, the fundamental frequency of the motor rotor and the passing frequency of the propeller blades of the target UAV are extracted. The mechanical vibration feature set is a set of parameters characterizing the mechanical vibration state of the target UAV's power system. It originates from the analysis of micro-vibration displacement and vibration frequency on the UAV's surface and reflects the dynamic response of the motor rotor, propeller, and airframe structure during operation. It is used to reveal the operational health status, load conditions, and potential structural resonance modes of the UAV's internal mechanical components. For example, time-frequency analysis can be performed on the micro-vibration signals collected by a laser vibrometer to extract the characteristic frequencies and amplitudes in its power spectral density distribution. Alternatively, an accelerometer array can be used to directly measure the vibration acceleration of various parts of the airframe, and modal analysis can be used to obtain vibration modes and characteristic parameters. The acoustic sensing feature set is a set of parameters characterizing the acoustic radiation characteristics of the target UAV's power system. It originates from the spectral analysis of the mechanical noise generated by the UAV's motor operation and the aerodynamic noise generated by the propeller rotation. It reflects the real-time operating status of the power system and is used to evaluate the UAV's real-time thrust load, motor speed, and aerodynamic efficiency. For example, beamforming and spectral analysis of sound pressure signals using microphone arrays can extract harmonic characteristics of mechanical noise and broadband spectral characteristics of aerodynamic noise. Alternatively, acoustic camera technology can be used to visualize and locate the sound source of the UAV and extract features of the sound signal from a specific source area. The fundamental frequency of the motor rotor refers to the main rotational frequency of the motor rotor in the UAV's power system under normal operating conditions. This frequency is directly related to the number of pole pairs and the rotational speed of the motor, and is a key parameter characterizing the real-time rotational speed of the motor, used to accurately reflect the motor's operating state, and thus infer the power output of the UAV. For example, the main peak in the mechanical noise spectrum can be detected and identified; this main peak usually corresponds to the fundamental frequency of the motor rotor. Alternatively, the rotational frequency of the motor rotor can be calculated by analyzing the periodic components in the current or voltage waveform of the motor drive circuit. The propeller blade passage frequency refers to the periodic signal frequency generated when each blade of the UAV propeller passes through a fixed point during rotation. This frequency is directly related to the propeller's rotational speed and the number of blades, and is an important parameter characterizing the propeller's operating state, used to accurately reflect the propeller's rotational speed and thrust output, and thus infer the flight state of the UAV. For example, discrete spectral lines in the aerodynamic noise spectrum can be detected; these lines typically correspond to the propeller blade passage frequencies and their harmonics. Alternatively, the propeller's rotation period can be directly measured using optical or vibration sensors, and the blade passage frequencies can be calculated. Extracting these frequencies aims to accurately identify frequency information directly related to key operating parameters of the UAV's propulsion system from complex physical field data, providing fundamental data for subsequent positioning corrections.For example, a peak-detection-based algorithm can be used to perform Fourier transform or wavelet transform on the mechanical vibration feature set and the acoustic sensing feature set. Then, the frequency peaks with amplitude can be identified in the resulting power spectral density map, and their attribution can be determined based on their physical meaning. Alternatively, a machine learning model can be used, trained on a large amount of physical field data with known motor speeds and propeller speeds, enabling the model to automatically identify and extract the corresponding fundamental frequency and pass frequency from new physical field data.

[0190] Based on the fundamental frequency of the motor rotor and the passing frequency of the propeller blades, Doppler frequency shift correction and body vibration phase compensation are applied to the preliminary positioning result obtained based on the time difference of arrival of the target's transmitted signal to obtain the active positioning solution for the target UAV. The preliminary positioning result refers to the initial position estimate obtained after processing the target UAV's transmitted signal using traditional radio signal processing methods, such as techniques based on Time Difference of Arrival (TDOA), Angle of Arrival (DOA), or Received Signal Strength Indicator (RSSI). This serves as the basis for subsequent precise positioning, providing a rough range of the UAV's location. For example, multiple receiving stations can be used to simultaneously receive the target's transmitted signal. By calculating the time difference of arrival at each receiving station and combining this with the known positions of the receiving stations, a hyperbolic positioning algorithm can be used for solution. Alternatively, a phased array antenna can be used to measure the angle of arrival of the target's transmitted signal, and positioning can be achieved through triangulation or angle intersection methods. Doppler frequency shift correction aims to eliminate or reduce the frequency shift of the target's transmitted signal caused by the relative motion between the target UAV and the receiving system, thereby improving positioning accuracy. For example, the radial velocity of the UAV can be estimated based on the extracted fundamental frequency of the motor rotor and the passing frequency of the propeller blades, combined with the UAV's flight model and the geometric position of the receiving station. This Doppler frequency shift can then be calculated and subtracted from the frequency term of the preliminary positioning result. Alternatively, a Kalman filter or extended Kalman filter can be used to estimate the Doppler frequency shift as a state variable and incorporated into the positioning calculation process for real-time correction. Airframe vibration phase compensation aims to eliminate or reduce the positioning phase error caused by the periodic modulation of the transmitted signal propagation path length due to minute vibrations generated by the UAV during flight, thereby further improving positioning accuracy. For example, the periodic effect of airframe vibration on the signal propagation path length can be estimated based on the extracted mechanical vibration feature set, particularly micro-vibration displacement and vibration frequency, and converted into a phase shift to compensate for the phase term in the preliminary positioning result. Alternatively, a UAV airframe vibration model can be established, and the impact of vibration on the signal phase can be predicted by real-time monitoring of vibration characteristics, and this prediction can be used as a correction factor in the positioning algorithm.

[0191] The current hardware operating point parameters and current load distribution state are obtained from the joint inference results. Based on these parameters and state, a basic jamming strategy corresponding to the target UAV's current hardware weaknesses is matched from a pre-defined jamming strategy library. The joint inference results are obtained through dynamic coupling analysis across physical quantities using a first hardware feature set and a second physical state feature set, and jointly solved using a collaborative evolution model. This comprehensive information includes the target UAV's current maneuvering state, load distribution state, and current hardware operating point, providing a complete and real-time insight into the target UAV's internal state and external behavior. This is a crucial basis for formulating accurate jamming strategies. The current hardware operating point parameters refer to the operating status indicators of the target UAV's launch hardware (especially power amplifiers) at the current moment, such as power transmission characteristics, nonlinear distortion, and thermal accumulation effects. These are used to identify real-time weaknesses and performance bottlenecks in the UAV's launch hardware, providing a basis for targeted jamming. The current load distribution state refers to the load condition of the target UAV's power system at the current moment, such as motor drive current, mechanical vibration intensity, and propeller thrust. This is used to assess the real-time pressure and energy consumption of the UAV's power system, providing a basis for selecting jamming timing and intensity. The preset jamming strategy library is a database storing various jamming schemes for different UAV hardware operating points and load distribution states. Each scheme includes specific jamming waveform parameters (such as frequency, modulation method, and bandwidth) and jamming power levels, providing a pre-designed and optimized jamming template to quickly select the most effective jamming method based on the real-time perceived UAV state. For example, a multi-dimensional lookup table can be created, with its index dimensions including hardware operating point parameters and load distribution state, and the table entries storing the corresponding jamming waveform parameters and power levels. Alternatively, a rule-based expert system can be used to trigger the corresponding jamming strategy when a specific UAV state is perceived, based on preset logical rules and thresholds. The basic jamming strategy is a preliminary jamming scheme obtained from the preset jamming strategy library based on the current hardware operating point parameters and current load distribution state, targeting the target UAV's current hardware weaknesses and load conditions. It provides the core content of the jamming, including the characteristics and strength of the jamming signal. The matching process aims to select the most effective jamming scheme from the preset jamming strategy library to attack the target UAV's weaknesses based on the target UAV's real-time hardware operating point parameters and load distribution state. For example, a nearest neighbor matching algorithm can be used, taking the current hardware operating point parameters and load distribution status as feature vectors, to find the preset strategy with the smallest Euclidean distance or the highest similarity in the strategy library. Alternatively, fuzzy logic or decision tree models can be used to score each strategy in the strategy library according to preset rules and weights, and select the strategy with the highest score.

[0192] The system obtains the current maneuvering state from the joint inference results and adjusts the timing and duration of interference in the basic interference strategy based on this current maneuvering state, generating a hardware-directed interference strategy that matches the joint inference results. The current maneuvering state refers to the target UAV's flight attitude and motion trend at the current moment, such as climb, descent, acceleration, deceleration, hovering, or constant-speed cruise, used to predict the UAV's behavior patterns over a short period, thereby optimizing the timing and duration of interference. The adjustment process aims to dynamically optimize the application time and duration of the basic interference strategy based on the target UAV's real-time maneuvering state to maximize interference effectiveness and avoid resource waste. For example, the corresponding interference timing and duration adjustment rules can be directly found and applied based on a preset maneuvering state-interference parameter mapping table. Alternatively, reinforcement learning algorithms can be used to train the system in a simulated environment, enabling it to autonomously learn and optimize the timing and duration of interference based on the UAV's maneuvering state. The interference timing refers to the optimal time point or time window for applying the interference signal, ensuring that interference is carried out when the UAV is most vulnerable or most in need of stable communication to achieve the best effect. For example, based on the current maneuvering state and load distribution, the system selects a time point when the UAV's power system is under high load, the communication link is unstable, or energy consumption is severe. The jamming duration refers to the effective length of time the jamming signal is applied, ensuring effective jamming while avoiding unnecessary energy consumption and potential impacts on other equipment. For example, the duration is set to match the specific behavior patterns of the UAV (such as acceleration or regenerative braking) based on the current maneuvering and load distribution. The generated hardware-directed jamming strategy is the result of active positioning calculation after Doppler frequency shift correction and airframe vibration phase compensation, and the jamming scheme obtained after matching and adjusting according to the current hardware operating point parameters, current load distribution, and current maneuvering. This strategy not only targets the inherent weaknesses of the UAV's launch hardware but also considers its real-time dynamic physics and flight behavior, aiming to achieve accurate positioning of the target UAV and efficient, low-collateral damage targeted jamming.

[0193] Through the above technical solution, this application improves the target identification accuracy and interference effectiveness of the UAV countermeasure system by integrating multi-dimensional physical information to actively locate the target UAV and generate a hardware-oriented interference strategy that matches the joint inference results. Specifically, by extracting the fundamental frequency of the motor rotor and the propeller blade passage frequency of the target UAV based on the mechanical vibration feature set and acoustic sensing feature set in the physical field data, and using these frequencies to perform Doppler frequency shift correction and body vibration phase compensation on the preliminary positioning results obtained based on the arrival time difference of the target transmitted signal, this application can effectively overcome the frequency shift and phase error caused by the UAV's own motion and body vibration to the radio signal positioning, thereby obtaining more accurate and robust real-time UAV position information. This avoids the problem of decreased positioning accuracy of traditional radio positioning methods in complex dynamic environments. In addition, by obtaining the current hardware operating point parameters and current load distribution status from the joint inference results, and matching the basic interference strategy corresponding to the current hardware weaknesses of the target UAV from the preset interference strategy library, this application realizes personalized customization of the interference strategy. This approach can target the hardware differences between individual UAVs of the same model and their real-time load changes at different flight phases, selecting the most effective jamming waveform and power level to attack the weak points of their launch hardware. This improves the targeting and efficiency of jamming and avoids the inefficiency and risk of collateral damage associated with general-purpose jamming. Furthermore, by obtaining the current maneuvering state from the joint inference results and dynamically adjusting the timing and duration of jamming in the basic jamming strategy based on this state, this application ensures that jamming is applied at the most vulnerable or critical moment when stable communication is most needed for the UAV, and for the most appropriate duration. For example, during high-load states such as UAV climb or acceleration, its power system and communication link may be under greater pressure, making jamming more effective. In specific states such as regenerative braking, short-term accurate jamming can also effectively utilize the system's characteristics. This dynamic matching mechanism allows the jamming strategy to closely coordinate with the real-time behavior of the UAV, further optimizing the effectiveness of jamming, avoiding the problem of the jamming window being out of sync with the UAV's state, and thus improving the overall system's countermeasure capability. In summary, this application constructs a highly intelligent and adaptive UAV active positioning and jamming system through multi-source data fusion, precise positioning correction, personalized interference matching, and dynamic timing adjustment, solving the problems of insufficient positioning accuracy, mismatched jamming strategies, and low jamming efficiency in related technologies.

[0194] In some of the embodiments described above in this application, Doppler frequency shift correction and body vibration phase compensation are proposed to improve positioning accuracy by performing Doppler frequency shift correction and body vibration phase compensation on the preliminary positioning results obtained based on the arrival time difference of the target transmitted signal. However, in the implementation process, due to the lack of a specific method for accurately calculating the correction and compensation amounts based on the fundamental frequency of the motor rotor and the passing frequency of the propeller blades, the positioning results may still be affected by the residual influence of Doppler frequency shift and vibration phase shift, and the errors caused by motion state and body vibration cannot be completely eliminated, thus limiting the further improvement of positioning accuracy.

[0195] To address this, this application further proposes a method for performing Doppler frequency shift correction and body vibration phase compensation on the preliminary positioning results obtained based on the arrival time difference of the target transmitted signal, using the fundamental frequency of the motor rotor and the passing frequency of the propeller blades, to obtain the active positioning solution for the target UAV. Specifically, this method includes: determining the radial velocity and body vibration period of the target UAV at the current moment based on the fundamental frequency of the motor rotor and the passing frequency of the propeller blades; determining the Doppler frequency shift correction amount based on the radial velocity and the carrier frequency of the target transmitted signal; determining the periodic modulation depth of the target UAV's body vibration on the propagation path length of the transmitted signal based on the body vibration period and the carrier wavelength of the target transmitted signal, and converting the periodic modulation depth into a phase offset compensation amount; correcting the frequency offset term in the preliminary positioning results using the Doppler frequency shift correction amount; and compensating the phase offset term in the preliminary positioning results using the phase offset compensation amount, thus obtaining the active positioning solution for the target UAV.

[0196] Specifically, when determining the radial velocity and vibration period of a target UAV at the current moment, the fundamental frequency of the motor rotor and the propeller blade passing frequency extracted from the operation of the UAV's power system can be used. For example, by analyzing the rate of change of the fundamental frequency of the motor rotor and combining it with the UAV's dynamic model, its radial acceleration can be calculated, and then the radial velocity can be obtained by integration. Simultaneously, the vibration period can be directly determined by the periodicity of the propeller blade passing frequency. Alternatively, a database or machine learning model can be established to map the fundamental frequency of the motor rotor and the propeller blade passing frequency to the UAV's radial velocity and vibration period, allowing for real-time querying or inference to obtain the corresponding parameters.

[0197] When determining the Doppler frequency shift correction, this step aims to quantify the signal frequency shift caused by the relative motion between the target UAV and the receiving device. It can be calculated using the classical Doppler effect formula Δf=(v_r / c)*f_c, where Δf is the Doppler frequency shift correction, v_r is the radial velocity, c is the speed of light, and f_c is the carrier frequency of the target's transmitted signal. Alternatively, a lookup table can be established beforehand, calibrating or simulating the relationship between radial velocity, carrier frequency, and the Doppler frequency shift correction, allowing for real-time lookup of the correction amount.

[0198] When determining the periodic modulation depth of the transmission path length caused by the vibration of the target UAV, this step aims to quantify the periodic impact of the UAV's vibration on the signal propagation path length. For example, the vibration period can be combined with the vibration amplitude (which can be indirectly derived from the vibration period or obtained through other physical field data) to calculate the maximum change in the propagation path length caused by the vibration, which is the periodic modulation depth. Alternatively, a UAV vibration model can be established, and the range of periodic changes in the signal propagation path length under a specific vibration period can be simulated or calculated, thereby determining the modulation depth.

[0199] The periodic modulation depth is converted into a phase offset compensation amount. This conversion is crucial for correcting phase errors introduced by vibration. The conversion can be performed based on the relationship between phase and path length: φ = (2π / λ)*ΔL, where φ is the phase offset compensation amount, λ is the carrier wavelength of the target transmitted signal, and ΔL is the periodic modulation depth. Alternatively, signal processing techniques, such as Hilbert transform or instantaneous phase extraction, can be used to deduce the corresponding phase offset compensation amount from the known modulation depth.

[0200] The frequency offset in the preliminary positioning results is corrected using a Doppler frequency shift correction, and the phase offset is compensated using a phase offset compensation. In the frequency domain processing stage of the preliminary positioning algorithm, the Doppler frequency shift correction can be directly added to or subtracted from the center frequency of the received signal to eliminate frequency deviations caused by motion. Alternatively, in the post-processing stage of the preliminary positioning results, the calculated frequency offset parameters are corrected based on the Doppler frequency shift correction. Similarly, in the phase domain processing stage of the preliminary positioning algorithm, the phase offset compensation can be directly added to or subtracted from the phase of the received signal to eliminate phase deviations caused by vibration. Alternatively, in the post-processing stage of the preliminary positioning results, the calculated phase offset parameters are corrected based on the phase offset compensation.

[0201] Through the above technical solution, this application utilizes the fundamental frequency of the motor rotor and the passing frequency of the propeller blades extracted from the operation of the target UAV's power system to accurately infer the radial velocity and body vibration period of the target UAV at the current moment. Based on these physical parameters, the Doppler frequency shift correction caused by the UAV's motion and the periodic modulation depth of the signal propagation path caused by body vibration can be calculated and converted into phase offset compensation. By applying these accurately calculated correction and compensation amounts to the frequency offset and phase offset terms in the preliminary positioning results, this application can effectively eliminate or reduce the interference caused by the UAV's own motion and body vibration on the positioning accuracy, thereby significantly improving the active positioning accuracy of the target UAV. This method fully utilizes the real-time dynamic information obtained from multi-physics collaborative sensing, making the positioning results more robust and reliable, and providing more accurate target position information for subsequent hardware-oriented jamming strategies.

[0202] In some of the solutions mentioned above in this application, a basic interference strategy is proposed to generate a hardware-oriented interference strategy by matching the current hardware operating point parameters and the current load distribution state. However, in this process, there is a lack of specific adjustment mechanism for different maneuver states, which makes it impossible for the interference strategy to accurately take advantage of the real-time weaknesses of the UAV under specific flight states. The setting of interference timing and duration may not match the actual power system operating state, thereby reducing interference efficiency and increasing resource waste.

[0203] To address this, this application further proposes adjusting the timing and duration of interference in the basic interference strategy based on the current maneuvering state to generate a hardware-directed interference strategy that matches the joint inference results. This includes: when the current maneuvering state is a climb or acceleration state, adjusting the timing of interference in the basic interference strategy to within a load duration window where the current peak amplitude continuously exceeds the heavy load threshold in the current load distribution state, and setting the duration of interference in the basic interference strategy to match the length of the load duration window. When the current maneuvering state is a descent or deceleration state, adjusting the timing of interference in the basic interference strategy to the moment when the motor drive circuit enters energy regenerative braking, and setting the duration of interference in the basic interference strategy to be less than the duration of energy regenerative braking. When the current maneuvering state is a hovering or constant-speed cruising state, adjusting the timing of interference in the basic interference strategy to multiple short-term interference windows distributed according to a pseudo-random time sequence, and setting the duration of interference in the basic interference strategy to the length of each short-term interference window.

[0204] The current maneuvering state refers to the flight attitude and motion trend of the target UAV at the current moment, identified by analyzing the slope and inflection point characteristics of the target UAV's future state evolution trajectory, such as climb, descent, acceleration, deceleration, hovering, or constant speed cruise. This can be achieved by analyzing parameters such as speed, acceleration, and attitude angle change rate of the state evolution trajectory. The basic jamming strategy refers to the initial jamming scheme selected from a pre-set jamming strategy library based on the target UAV's current hardware operating point parameters and current load distribution, targeting its specific hardware weaknesses and load conditions. It typically includes specific jamming waveform parameters and jamming power levels. The jamming timing refers to the specific point in time or time period at which the jamming signal is applied. This can be achieved through a precise time synchronization mechanism, initiating jamming when the UAV is in a specific vulnerable state, or triggering jamming when a specific event occurs. The jamming duration refers to the length of time the jamming signal lasts from the start to the end. This can be achieved through timer control or dynamic adjustment based on changes in the UAV's state. The climb or acceleration state refers to the flight state where the UAV's altitude increases vertically or its speed increases horizontally. In this state, the UAV's power system typically requires higher power output and operates under high load. A sustained load window where the current spike amplitude continuously exceeds the heavy load threshold refers to a period of time during the operation of the UAV's power system (especially the motor drive circuit) where the instantaneous current peak value in its current waveform remains higher than a preset heavy load threshold, indicating that the power system is operating under high load for an extended period. This can be achieved by real-time monitoring of the current spike amplitude in the power electrical characteristic set and setting a threshold and duration detection mechanism. Descent or deceleration refers to the UAV's flight state where its altitude decreases vertically or its speed decreases horizontally. During this time, the UAV's power system may enter energy regenerative braking mode, or the load may decrease. The moment the motor drive circuit enters energy regenerative braking refers to the instant when the UAV motor converts kinetic energy into electrical energy to feed back to the battery or drive circuit during descent or deceleration. At this time, the motor acts as a generator, and the operating mode of the drive circuit changes. This can be determined by monitoring the voltage, current direction, or specific control signals of the motor drive circuit. Hovering or constant speed cruise refers to the UAV maintaining relative stillness in the air or flying at a constant speed and altitude. Multiple short-term interference windows in pseudo-random timing distribution refer to interference signals not being applied continuously, but rather applied multiple times in short intervals with seemingly random but actually controllable time intervals and durations. This can be achieved by using a pseudo-random number generator combined with preset minimum / maximum intervals and durations to generate the interference timing sequence.

[0205] Through the above technical solutions, this application can finely adjust the timing and duration of interference in the basic interference strategy according to the real-time maneuvering state of the target UAV, thereby improving the accuracy and effectiveness of the interference. Specifically, when the target UAV is climbing or accelerating, its power system is usually operating under high load, and the current spike amplitude of the motor drive circuit will continuously exceed the heavy load threshold. Adjusting the interference timing to within this "load duration window" and matching the interference duration with the length of this window can target the vulnerabilities of the UAV's power system under heavy load, such as power amplifier saturation, motor stall, or control system instability, thereby maximizing the interference effect. When the target UAV is descending or decelerating, its motor drive circuit may enter an energy regenerative braking mode. Applying interference at this specific moment and setting the interference duration to be less than the duration of energy regenerative braking can effectively utilize the sensitivity of the system's energy regenerative braking during the braking phase, avoiding excessive interference that leads to ineffectiveness or resource waste, and achieving optimized allocation of interference resources. When the target UAV is hovering or cruising at a constant speed, its flight state is relatively stable, and the control system may have a certain degree of adaptability to continuous and predictable interference. At this point, employing multiple short-term interference windows with pseudo-random temporal distribution introduces unpredictability into the interference. Through intermittent, sudden interference pulses, the closed-loop control system of the UAV is continuously disrupted, forcing it to constantly adjust its attitude or reacquire signals. This effectively prevents the UAV system from adapting to a single interference mode, thus maintaining continuous interference pressure. This adaptive interference strategy based on real-time maneuvering states allows the interference system to more accurately exploit the UAV's real-time dynamic weaknesses, avoiding the inefficiency of general interference strategies under specific flight conditions, and improving the target recognition accuracy and interference effectiveness of the UAV countermeasure system.

[0206] All of the above-mentioned optional technical solutions can be combined in any way to form the optional embodiments of this application, and will not be described in detail here.

[0207] Figure 6This is a schematic diagram of a UAV active positioning and jamming system based on radio spectrum sensing, provided in an embodiment of this application. The UAV active positioning and jamming system 600 based on radio spectrum sensing can vary considerably due to different configurations or performance. It may include one or more Central Processing Units (CPUs) 601 and one or more memories 602. The one or more memories 602 store at least one computer program, which is loaded and executed by the one or more processors 601 to implement the methods provided in the above-described method embodiments. Of course, the UAV active positioning and jamming system 600 based on radio spectrum sensing may also have wired or wireless network interfaces, a keyboard, and input / output interfaces for input and output. The UAV active positioning and jamming system 600 based on radio spectrum sensing may also include other components for implementing device functions, which will not be elaborated here.

[0208] In an exemplary embodiment, a computer-readable storage medium is also provided, such as a memory including a computer program that can be executed by a processor to perform the UAV active localization and jamming method based on radio spectrum sensing in the above embodiments. For example, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, floppy disk, and optical data storage device, etc.

[0209] In an exemplary embodiment, a computer program product or computer program is also provided, which includes program code stored in a computer-readable storage medium. The processor of a computer device reads the program code from the computer-readable storage medium and executes the program code, causing the computer device to perform the above-described method for active localization and jamming of unmanned aerial vehicles based on radio spectrum sensing.

[0210] In some embodiments, the computer program involved in the present application embodiments may be deployed and executed on a computer device, or executed on multiple computer devices located in one location, or executed on multiple computer devices distributed in multiple locations and interconnected through a communication network. Multiple computer devices distributed in multiple locations and interconnected through a communication network may constitute a blockchain system.

[0211] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0212] The above are merely optional embodiments of this application and are not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A UAV active positioning and jamming system based on radio spectrum sensing, characterized in that, The system includes a processor and memory, the processor being configured to perform the following steps: By sensing the radio spectrum, the target transmission signal of the target UAV is acquired, and the power amplifier response of the target transmission signal is modeled to extract a first hardware feature set to characterize the inherent nonlinear characteristics of the target UAV's transmission hardware. In parallel with the acquisition of the target's transmitted signals, the target physical field data derived from the operation of the power system during the flight of the target UAV is acquired through remote non-contact sensing. The target physical field data is then subjected to time-frequency domain transformation and mode decomposition to extract a second physical state feature set for characterizing the real-time dynamic physical state of the target UAV. A dynamic coupling analysis across physical quantities is performed on the first hardware feature set and the second physical state feature set to construct a co-evolution model between the launch hardware characteristics and dynamic physical state of the target UAV. Based on the co-evolution model, the first hardware feature set and the second physical state feature set are jointly solved to obtain the joint inference results of the current maneuvering state, current load distribution state and current hardware operating point of the target UAV. Based on the joint inference results, the target's transmitted signal, and the target's physical field data, the target UAV is actively located and a hardware-oriented jamming strategy matching the joint inference results is generated.

2. The system according to claim 1, characterized in that, The method of acquiring the target UAV's transmission signals through radio spectrum sensing includes: A detection signal is transmitted to the target drone to induce a change in the response of the target drone's power amplifier at different output power levels; After transmitting the detection signal, the first set of transmitted signals returned by the target UAV, carrying the nonlinear response characteristics of the power amplifier, is collected. During the period when the detection signal is not emitted, a second set of emitted signals spontaneously radiated by the target UAV under normal flight conditions is collected; The target transmission signal is obtained by jointly calibrating and filtering the first group of transmission signals and the second group of transmission signals.

3. The system according to claim 1, characterized in that, The step of modeling the power amplifier response of the target transmitted signal and extracting a first hardware feature set to characterize the inherent nonlinear characteristics of the target UAV's launch hardware includes: Logarithmic power curve fitting is performed on the response curves of the target's transmitted signal at different power levels to obtain the power transfer characteristic curves of the target UAV's power amplifier; The power transfer characteristic curve is subjected to inflection point detection and nonlinear region division, and the compression point feature, intermodulation distortion feature and power-added efficiency feature of the target UAV power amplifier are extracted. Time-frequency analysis is performed on the target transmitted signal to extract the frequency drift rate of the target UAV power amplifier at different power levels. Based on the correlation between the frequency drift rate and the power level, a thermal memory effect characteristic is constructed to characterize the thermal accumulation and heat dissipation properties of the target UAV power amplifier. The compression point feature, the intermodulation distortion feature, the power-added efficiency feature, and the thermal memory effect feature are combined and encoded to obtain the first hardware feature set.

4. The system according to claim 3, characterized in that, The process of detecting inflection points and dividing nonlinear regions on the power transfer characteristic curve, and extracting the compression point features, intermodulation distortion features, and power-added efficiency features of the target UAV power amplifier, includes: The power transfer characteristic curve is piecewise fitted in the linear and nonlinear regions to detect the inflection point where the power transfer characteristic curve changes from linear growth to nonlinear growth, and the output power value corresponding to the inflection point is used as the compression point feature. Intermodulation distortion analysis is performed on the spectral components of the target transmitted signal in the nonlinear operating region. The relative ratio between the amplitude of the third-order intermodulation component and the amplitude of the fundamental component is extracted. The third-order intermodulation intercept is calculated based on the relative ratio, and the third-order intermodulation intercept is used as the intermodulation distortion feature. The input power and output power of the target transmitted signal are obtained, and the feature components related to the battery discharge state in the second physical state feature set are obtained. Based on the input power, the output power and the feature components related to the battery discharge state, the DC power consumption of the target UAV power amplifier at the current operating point is determined. Based on the ratio of the output power to the DC power consumption, the power-added efficiency of the target UAV power amplifier at the current operating point is determined, and the power-added efficiency is used as the power-added efficiency characteristic.

5. The system according to claim 1, characterized in that, The method of collecting target physical field data generated by the operation of the power system during the flight of the target UAV through remote non-contact sensing includes: The electromagnetic field radiated by the motor drive circuit of the target UAV power system during the commutation process is detected by an electric field sensor array. The current waveform and voltage waveform of the motor drive circuit are collected and used as the first physical field data. A laser beam is emitted toward the surface of the target UAV using a laser vibrometer and the reflected beam is received. The micro-vibration displacement and vibration frequency of the surface of the UAV are calculated based on the Doppler frequency shift of the reflected beam, and the micro-vibration displacement and vibration frequency are used as the second physical field data. Using a microphone array, beamforming and sound source localization are performed on the mechanical noise generated by the motor operation and the aerodynamic noise generated by the propeller rotation of the target UAV. The spectral characteristics of the mechanical noise and the aerodynamic noise are collected and used as third physical field data. The first physical field data, the second physical field data, and the third physical field data are synchronized in time and fused to obtain the target physical field data.

6. The system according to claim 5, characterized in that, The calculation of the micro-vibration displacement and vibration frequency of the body surface based on the Doppler frequency shift of the reflected beam includes: The reflected beam and the reference beam are optically mixed and photoelectrically converted to generate an interference signal carrying Doppler frequency shift information; The interference signal is orthogonally demodulated to extract the in-phase and quadrature components of the interference signal, and the instantaneous displacement waveform of the vibration of the body surface is calculated based on the in-phase and quadrature components. The power spectral density distribution of the vibration on the body surface is obtained by performing a Fourier transform on the instantaneous displacement waveform. Identify the fundamental frequency peak corresponding to the motor rotor rotation frequency, the harmonic peak corresponding to the motor commutation frequency, and the modulation sideband peak corresponding to the propeller blade passage frequency from the power spectral density distribution. Determine the vibration frequency based on the frequency values ​​corresponding to the fundamental frequency peak, the harmonic peak, and the modulation sideband peak. Calculate the micro-vibration displacement based on the amplitude values ​​corresponding to the fundamental frequency peak, the harmonic peak, and the modulation sideband peak.

7. The system according to claim 1, characterized in that, The step of performing time-frequency domain transformation and mode decomposition on the target physical field data to extract a second physical state feature set for characterizing the real-time dynamic physical state of the target UAV includes: Time-frequency analysis is performed on the current waveform and voltage waveform in the physical field data to extract the current peak amplitude, duty cycle and modulation frequency of the pulse width modulation signal, and distortion characteristics of the back electromotive force waveform at the commutation moment of the motor drive circuit. The current peak amplitude, duty cycle, modulation frequency, and distortion characteristics of the back electromotive force waveform are then used as a power electrical feature set. Empirical mode decomposition is performed on the micro-vibration displacement and vibration frequency in the physical field data to obtain multiple intrinsic mode function components. Target mode components associated with the motor rotor rotation frequency, motor commutation frequency, and body structure resonance frequency are selected from the multiple intrinsic mode function components. The instantaneous amplitude and instantaneous frequency of the target mode components are extracted, and the instantaneous amplitude and instantaneous frequency are used as the mechanical vibration feature set. The spectral characteristics of mechanical noise and aerodynamic noise in the physical field data are fused and analyzed. The real-time rotational speed of the motor rotor is determined based on the peak distribution in the spectral characteristics of the mechanical noise, and the real-time thrust load of the propeller is determined based on the broadband noise floor in the spectral characteristics of the aerodynamic noise. The real-time rotational speed and the real-time thrust load are used as an acoustic sensing feature set. The second physical state feature set is obtained by combining and normalizing the power electrical feature set, the mechanical vibration feature set, and the acoustic sensing feature set through multi-dimensional feature combination and encoding.

8. The system according to claim 1, characterized in that, The step of performing dynamic coupling analysis across physical quantities on the first hardware feature set and the second physical state feature set to construct a co-evolution model of the target UAV between its launch hardware characteristics and dynamic physical state includes: Time alignment and cross-correlation analysis are performed on the thermal memory effect features in the first hardware feature set and the power electrical feature set in the second physical state feature set to determine the time delay and correlation coefficient between the change in frequency drift rate in the thermal memory effect features and the change in current peak amplitude in the power electrical feature set. Based on the time delay and the correlation coefficient, a first coupling relationship between the thermal accumulation degree of the target UAV power amplifier and the motor drive current is established. Event synchronization analysis is performed on the compression point features in the first hardware feature set and the power electrical feature set in the second physical state feature set. The time alignment relationship between the moment when the output power enters the compression zone in the compression point features and the moment when the current peak amplitude in the power electrical feature set exceeds the load threshold is detected. Based on the time alignment relationship, a second coupling relationship is established between the moment when the target UAV power amplifier enters the nonlinear operating region and the moment when the motor enters the heavy load state. Using the first coupling relationship and the second coupling relationship as initial constraints, the intermodulation distortion feature and power-added efficiency feature of the first hardware feature set, as well as the mechanical vibration feature set and acoustic sensing feature set of the second physical state feature set, are obtained. The cooperative evolution model is then constructed using a multivariate state-space modeling method.

9. The system according to claim 1, characterized in that, The step of jointly solving the first hardware feature set and the second physical state feature set according to the cooperative evolution model to obtain the joint inference results of the target UAV's current maneuvering state, current load distribution state, and current hardware operating point includes: The first hardware feature set and the second physical state feature set obtained at the current moment are input into the co-evolution model; the system state vector at the current moment is obtained by Kalman filtering recursion through the state transition equation and observation equation in the co-evolution model. Extract the state components corresponding to the nonlinear response characteristics of the power amplifier from the system state vector at the current moment, and use them as the current hardware operating point parameters. Also extract the state components corresponding to the electrical and mechanical vibration characteristics of the power system from the system state vector at the current moment, and use them as the current load distribution state. Based on the system state vector at the current moment, predict the state evolution trajectory of the target UAV in the future preset time domain, and identify the maneuvering state of the target UAV at the current moment according to the slope characteristics and inflection point characteristics of the state evolution trajectory.

10. The system according to claim 1, characterized in that, The step of actively locating the target UAV based on the joint inference result, the target transmitted signal, and the target physical field data, and generating a hardware-oriented jamming strategy that matches the joint inference result, includes: Based on the mechanical vibration feature set and acoustic sensing feature set in the physical field data, the fundamental frequency of the motor rotor and the passing frequency of the propeller blades of the target UAV are extracted. Based on the fundamental frequency of the motor rotor and the passing frequency of the propeller blades, Doppler frequency shift correction and body vibration phase compensation are performed on the preliminary positioning result obtained based on the arrival time difference of the target transmitted signal to obtain the active positioning solution result of the target UAV. The current hardware operating point parameters and the current load distribution state are obtained from the joint inference results. Based on the current hardware operating point parameters and the current load distribution state, a basic interference strategy corresponding to the current hardware weakness of the target UAV is matched from the preset interference strategy library. The preset interference strategy library stores the interference waveform parameters and interference power levels corresponding to different combinations of hardware operating points and load distribution states. The current maneuver state is obtained from the joint inference result, and the timing and duration of interference in the basic interference strategy are adjusted according to the current maneuver state to generate a hardware-oriented interference strategy that matches the joint inference result.

Images