Signal frequency analysis method and system for racing movement of unmanned aerial vehicle

Abstract

The invention relates to the technical field of wireless communication, in particular to a signal frequency analysis method and system for unmanned aerial vehicle racing sports, and the method comprises the steps: building a three-dimensional coordinate system of a racing track, and collecting historical data of multiple times of flight practice to build a historical window set; dividing the historical windows with similar characteristics into the same area according to the motion response characteristics of each historical window in the historical window set; traversing a candidate wavelet decomposition layer number for each region and calculating a denoising effect score to determine an optimal decomposition layer number of each region; collecting the flight data of the current window in real time, matching the area, calling the corresponding optimal decomposition layer number to carry out adaptive wavelet denoising, and extracting the instantaneous frequency to complete the signal frequency analysis. According to the method, the accuracy and adaptability of instantaneous frequency extraction are improved, and the problem of interference caused by signal non-stable change in a complex channel environment is solved.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology. More specifically, this invention relates to a method and system for analyzing the signal frequency of unmanned aerial vehicle (UAV) racing. Background Technology

[0002] Drone racing is a high-speed, highly dynamic competitive sport. When drones travel at high speeds on complex tracks, the received signals are significantly affected by the Doppler effect, and the instantaneous frequency of the signal is closely related to the drone's real-time speed. Therefore, frequency analysis of the received signals to extract Doppler frequency shift information is a key technical means to reconstruct the instantaneous speed and position of the drone.

[0003] Existing signal frequency analysis methods mostly employ wavelet transform to denoise the received signal, followed by Hilbert transform to extract the instantaneous frequency. However, the motion characteristics and channel environment vary significantly across different areas of a drone racing track. Existing methods use a fixed number of wavelet decomposition levels to uniformly process the signal across the entire track during signal frequency analysis, making it difficult to balance the accuracy and stability of frequency extraction.

[0004] Therefore, how to improve the adaptive capability of signal frequency analysis in drone racing is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] To address the aforementioned technical problem of how to improve the adaptive capability of signal frequency analysis in drone racing, this invention provides solutions in the following aspects.

[0006] In the first aspect, a method for analyzing the signal frequency of unmanned aerial vehicle (UAV) racing events includes: Establish a three-dimensional coordinate system for the track and collect historical data from multiple practice flights to construct a historical window set; Based on the motion response characteristics of each historical window in the historical window set, historical windows with similar motion response characteristics are divided into the same region; For each region, the candidate wavelet decomposition levels are traversed and the denoising effect score is calculated to determine the optimal decomposition level for each region. Real-time acquisition of flight data in the current window and matching of regions are performed by calling the corresponding optimal decomposition level for adaptive wavelet denoising and extracting instantaneous frequencies to complete signal frequency analysis. The optimal number of decomposition layers is determined as follows: For each region, for each candidate wavelet decomposition level, the received signals of each historical window in the region are denoised using wavelet decomposition level to restore the instantaneous speed of the UAV. The denoising effect score of each candidate wavelet decomposition level is calculated based on the statistical results of the speed restoration error between the restored instantaneous speed and the true speed. The candidate wavelet decomposition level with the highest denoising effect score is determined as the optimal decomposition level for the region.

[0007] Optionally, collecting historical data from multiple practice flights to construct a historical window set includes: Set the sampling frequency and simultaneously collect the actual flight speed sequence of the UAV, the received signal sequence of the ground receiver, the three-axis angular velocity sequence, and the three-axis acceleration sequence during multiple practice flights, and normalize the collected data. Set the sliding window length and sliding step size, and slide to capture multiple historical windows on the time sequence of each practice flight to construct a historical window set.

[0008] Optionally, dividing historical windows with similar characteristics into the same region includes: For each historical window, extract its three-axis angular velocity sequence and three-axis acceleration sequence, calculate the mean and standard deviation of the angular velocity sequence of each axis and the mean and standard deviation of the acceleration sequence of each axis, and concatenate the calculated statistical features to form the motion feature vector of the historical window. Using the motion feature vectors of all historical windows as samples, a clustering algorithm is used for classification, and historical windows with similar motion response characteristics are grouped into the same cluster, with each cluster corresponding to a type of track area.

[0009] Optionally, the calculation of the denoising effect score includes: The received signal sequence of the historical window is decomposed using the candidate wavelet decomposition level to obtain low-frequency components and high-frequency components of each level. The energy disorder of the received signal is calculated based on the energy proportion of each component. The decision threshold is determined based on the energy disorder. The decision threshold is used to perform hard thresholding on the high-frequency components of each layer. The processed high-frequency components and the original low-frequency components are subjected to wavelet inverse transform to obtain the denoised received signal sequence. The instantaneous frequency is extracted by Hilbert transform and converted into the restored instantaneous velocity. The velocity restoration error between the restored instantaneous velocity and the true velocity is calculated. The mean and standard deviation of the velocity restoration error of all historical windows within the statistical region at the corresponding candidate wavelet decomposition level are used to calculate the denoising effect score.

[0010] Optionally, determining the judgment threshold based on energy disorder includes: Determine the lower and upper limits of the judgment threshold. The lower limit of the judgment threshold is set to the value corresponding to the low quantile of the absolute value of all high-frequency components, and the upper limit of the judgment threshold is set to the value corresponding to the high quantile of the absolute value of all high-frequency components. The judgment threshold is the larger of the lower limit of the judgment threshold and the upper limit of the judgment threshold multiplied by the negative energy disorder degree raised to the power of the natural constant.

[0011] Optionally, real-time acquisition of flight data for the current window and matching of regions include: The system acquires the triaxial angular velocity sequence, triaxial acceleration sequence, and received signal sequence of the current window, which consists of continuous time intervals ending at the current time. Motion feature vectors are extracted based on the triaxial angular velocity sequence and triaxial acceleration sequence of the current window. The distance between the extracted motion feature vectors and the center vectors of each region is calculated, and the region with the smallest distance is selected as the matching region of the current window.

[0012] Optionally, the candidate wavelet decomposition level is generated after determining the maximum decomposable level based on the number of sampling points within the historical window, and the maximum decomposable level is the logarithm of the number of sampling points to the base 2, rounded down.

[0013] In a second aspect, a signal frequency analysis system for drone racing includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the signal frequency analysis method for drone racing described in any one of the claims is implemented.

[0014] The beneficial effects of this invention are: 1. This invention effectively addresses the inherent limitation of fixed decomposition levels in UAV racing, which struggle to achieve comprehensive signal processing across the entire track, by introducing a region-adaptive wavelet denoising mechanism. Based on historical flight data, the track is divided into different regions with similar motion response characteristics, and the optimal decomposition level is independently optimized for each region. This allows the denoising strategy to dynamically adjust according to the UAV's environment, significantly improving the accuracy and adaptability of instantaneous frequency extraction and overcoming interference problems caused by non-stationary signal variations in complex channel environments.

[0015] 2. This invention comprehensively considers the accuracy and stability of velocity reconstruction when optimizing the number of decomposition layers. By constructing an evaluation index that takes into account both the mean and standard deviation of the reconstruction error, it ensures that the selected parameters can effectively suppress noise while avoiding signal distortion caused by excessive smoothing. This mechanism significantly improves the smoothness and continuity of the instantaneous frequency extraction results, reduces the risk of jitter and abrupt changes in velocity output, and provides more reliable and stable velocity and position information support for the UAV flight control system, thereby improving the reliability of navigation and timing judgment in racing sports. Attached Figure Description

[0016] Figure 1This is a flowchart of steps S1-S4 in a signal frequency analysis method for drone racing according to an embodiment of the present invention. Detailed Implementation

[0017] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0018] Reference Figure 1 A signal frequency analysis method for drone racing includes steps S1-S4, as detailed below: S1: Establish a three-dimensional coordinate system for the track and collect historical data from multiple practice flights to construct a historical window set.

[0019] Drone racing tracks have a complex three-dimensional spatial structure, with significant differences in obstacle distribution and spatial morphology at different locations. To adaptively adjust signal processing parameters based on the drone's spatial position, the track space must first be digitized, and raw data reflecting the drone's motion state and the channel environment must be collected. By establishing a unified three-dimensional coordinate system, the drone's position can be precisely correlated with the data; and by collecting historical data from multiple practice flights, sufficient samples can be provided for subsequent analysis of the drone's motion behavior at different locations.

[0020] Specifically, a right-handed Cartesian coordinate system is established with the official starting point of the track as the origin. The X-axis extends horizontally along the main flight direction of the track, the Y-axis extends horizontally perpendicular to the X-axis, and the Z-axis extends vertically upwards. The coordinate unit is set to meters, with an accuracy of 0.01 meters. This coordinate system provides a unified mathematical benchmark for subsequently associating data with spatial location.

[0021] Furthermore, the sampling frequency was set to 100Hz (a frequency sufficient to capture the dynamic changes of the drone's high-speed flight), and the drone was allowed to perform multiple practice flights on the track, simultaneously collecting the following four types of historical data during each flight: The actual flight speed sequence of the drone, which comes from the airborne high-precision speed detection module; the received signal sequence of the ground receiver, which comes from the signal acquisition module of the receiver; the three-axis angular velocity sequence, which comes from the airborne gyroscope; and the three-axis acceleration sequence, which comes from the airborne accelerometer.

[0022] All the collected data were aligned using a unified timestamp to ensure that each sensor reading corresponds to a unique sampling time. The collected data were then subjected to Min-Max normalization, mapping their value ranges to... In order to eliminate the influence of different physical dimensions.

[0023] Furthermore, the sliding window length and sliding step size are set. The window length is determined based on the maximum speed of the drone and the size of the smallest feature region of the track, for example, 1 second, so that the window space span is smaller than the size of the smallest feature region, ensuring that the data within the window has relatively consistent regional characteristics; the sliding step size is set to 0.1 seconds to ensure sufficient overlap between windows and avoid information loss.

[0024] Furthermore, based on the above operations, multiple historical windows are obtained by sliding and capturing the complete time sequence of each practice flight to construct a historical window set, with the data in each historical window being multidimensional data.

[0025] S2: Based on the motion response characteristics of each historical window in the historical window set, historical windows with similar motion response characteristics are divided into the same region.

[0026] When a drone flies on a racing track, the physical characteristics of different areas (such as straightaways, curves, and areas with dense obstacles) will cause the drone to produce different motion responses. These responses are directly reflected in the changes in angular velocity and acceleration. Since angular velocity and acceleration data are not affected by channel interference, they can accurately reflect the environmental characteristics of the area.

[0027] By quantitatively analyzing the motion response of each historical window in the historical window set, historical windows with similar motion characteristics can be grouped into the same category, thereby dividing the continuous track space into several regions with typical characteristics, providing a basis for subsequent differentiated signal processing strategies for different region types.

[0028] Specifically, for any given historical window, its angular velocity and acceleration sequences are extracted, and statistical characteristics for each axis are calculated. Taking the X-axis as an example, the mean and standard deviation of the angular velocity along this axis are calculated. Similarly, the mean and standard deviation of the angular velocities along the Y and Z axes, as well as the mean and standard deviation of the accelerations along the three axes, are calculated. These 12 statistical characteristics are concatenated in a fixed order (XYZ axes) to form the motion feature vector for that historical window. This motion feature vector numerically characterizes the motion response characteristics of the UAV within that historical window period: the mean describes the average motion intensity, and the standard deviation describes the intensity of the motion changes.

[0029] Furthermore, using the motion feature vectors of all historical windows as samples, unsupervised classification is performed using the K-means clustering algorithm.

[0030] Specifically, Euclidean distance is used to calculate the distance between any two historical window motion feature vectors.

[0031] In addition, the silhouette coefficient method is used to determine the optimal number of clusters K: for candidate K values ​​(e.g., from 2 to 10), K-means clustering is performed respectively, the average silhouette coefficient of all samples is calculated, and the K value with the largest average value is selected as the final number of clusters.

[0032] After determining the value of K, the K-means algorithm is executed to obtain K clusters and the center vector of each cluster (i.e., the mean of all samples in that cluster). Each cluster corresponds to a type of track region (hereinafter referred to as region) with similar motion response characteristics, such as open straight track, sharp turn, or obstacle-dense track.

[0033] Through the above clustering, there is no need to manually label what type each region belongs to. The algorithm automatically merges windows with similar motion patterns based on the actual flight response data of the drone.

[0034] S3: Iterate through the candidate wavelet decomposition layers for each region and calculate the denoising effect score to determine the optimal decomposition layer for each region.

[0035] Historical windows within the same region exhibit similar motion characteristics and noise environments, thus allowing for the application of uniform denoising parameters. Among numerous denoising methods, wavelet decomposition is particularly suitable for UAV racing scenarios due to its multi-resolution analysis capabilities: on one hand, the received signal exhibits non-stationarity, and wavelet transform can simultaneously provide localized information in both the time and frequency domains, effectively separating signals from noise in different frequency bands; on the other hand, by selecting different decomposition levels, the fineness of frequency domain partitioning can be flexibly controlled to adapt to the noise characteristics of different regions. The number of wavelet decomposition levels is a key parameter affecting the wavelet denoising effect: too few levels will fail to effectively separate noise from the signal in the frequency domain; too many levels may lead to over-smoothing of useful components in the signal or the introduction of artifacts. Therefore, for each region, it is necessary to utilize historical window data for that region, experiment with different decomposition levels, evaluate the denoising effect, and find the decomposition level that yields the most accurate and stable speed restoration, thereby establishing a parameter knowledge base for subsequent real-time processing.

[0036] Specifically, for each region, perform the following operations in sequence: First, let N be the number of sampling points within the historical window of the region (i.e., the product of the sampling frequency and the historical window length). According to the theory of wavelet decomposition, the maximum number of decomposable layers is... for: Then the set of candidate wavelet decomposition levels is: .

[0037] Then, any candidate wavelet decomposition level is selected as the target decomposition level, and all historical windows within the region are processed as follows: The first step involves using the Db4 wavelet as the wavelet basis (this wavelet possesses orthogonality and compact support, making it suitable for processing non-stationary signals) to perform target-level wavelet decomposition on the received signal sequence within the historical window. This yields the low-frequency components of the last level and the high-frequency components of each level. The low-frequency components represent the overall trend of the signal, while the high-frequency components represent detailed information at different scales.

[0038] The second step is to calculate the energy of each component according to the existing calculation formula, and then calculate the total energy, which is the sum of the energy of the low-frequency component and the average energy of all high-frequency components. Then, calculate the energy ratio of each component, which is the energy of the low-frequency component divided by the total energy, and the energy of each high-frequency component divided by the total energy.

[0039] Furthermore, based on the calculated energy proportions of each component, the energy disorder of the received signal within the historical window is calculated using the existing Shannon entropy calculation formula. When the signal is concentrated in a few components, such as when a pure signal is mainly concentrated in low frequencies, the proportions of each component differ greatly, resulting in a lower energy disorder. When the signal is dispersed across multiple components, such as when it is subject to complex noise interference, the proportions of each component are evenly distributed, resulting in a higher energy disorder.

[0040] The third step is to use a hard thresholding method to process the high-frequency components of each layer. That is, a judgment threshold is set based on the energy disorder to determine whether each high-frequency component is a "valid signal" or "noise".

[0041] First, determine the upper and lower limits of the judgment threshold: the lower limit is set to the value corresponding to the low quantile (e.g., 10%) of the absolute values ​​of all high-frequency components, and the upper limit is set to the value corresponding to the high quantile (e.g., 90%) of the absolute values ​​of all high-frequency components. This setting method ensures that the upper and lower limits of the judgment threshold match the actual amplitude of the currently received signal, avoiding insufficient adaptability caused by a fixed threshold.

[0042] The decision threshold is calculated based on the upper and lower limits of the decision threshold and the energy disorder, satisfying the following relationship: In the formula, To determine the threshold, To determine the lower limit of the threshold, To determine the upper limit of the threshold, The energy disorder of the signal received during the historical window.

[0043] Based on the above calculation method, an adaptive strategy of "relaxed denoising for complex interference and strict denoising for simple interference" is adopted. When the energy disorder is large, it indicates that the energy distribution of the received signal in the historical window is scattered and the interference is complex. When the energy disorder is low, the judgment threshold is lowered accordingly to retain more possible useful details. This is because in complex interference environments, the useful signal itself is also dispersed across multiple frequency bands. If the threshold is too high, useful components will be mistakenly deleted, leading to distortion in speed restoration. Therefore, a relaxed denoising strategy is adopted to achieve "relaxed denoising for complex interference". When the energy disorder is low, it indicates that the energy of the received signal in the historical window is concentrated and the interference is simple. The judgment threshold is then raised to completely remove noise components. This is because in simple interference environments, the useful signal is concentrated in a few components with large amplitudes. Raising the threshold can significantly remove noise without losing the useful signal, making the output speed sequence smoother and more stable, thus achieving "strict denoising for simple interference".

[0044] Furthermore, high-frequency components with absolute values ​​less than the judgment threshold in each layer are set to zero, while high-frequency components with absolute values ​​greater than or equal to the judgment threshold in each layer are retained, thus obtaining the processed high-frequency components.

[0045] The fourth step is to perform inverse wavelet transform on the processed high-frequency components and the original low-frequency components to obtain the denoised received signal sequence. For each sampling moment in the denoised received signal sequence, the instantaneous frequency is extracted by Hilbert transform, and then the instantaneous frequency is converted into the restored instantaneous velocity according to the Doppler frequency shift formula.

[0046] Furthermore, by comparing the calculated restored instantaneous velocity with the true velocity within the historical window, the velocity restoration error of the historical window at the target decomposition level is defined as the sum of the absolute values ​​of the differences between the instantaneous velocity and the true velocity at each sampling moment. The smaller this velocity restoration error, the better the denoising effect of the historical window, and the closer the restored velocity is to the true value.

[0047] The fifth step is to statistically analyze the velocity restoration error of all historical windows within the region at the target decomposition level, and obtain the mean and standard deviation of the velocity restoration error. The mean reflects the average accuracy of velocity restoration at that level, and the standard deviation reflects the stability of the restoration result (the smaller the fluctuation, the better the stability).

[0048] Furthermore, to comprehensively measure accuracy and stability, the mean and standard deviation of the velocity restoration error are divided by their respective maximum values ​​to achieve dimensionless scaling and eliminate magnitude differences. Then, a weighted sum is performed to obtain the comprehensive cost. Finally, a negative exponential function is used to map the comprehensive cost to... The interval is then used to obtain the denoising score. The weights are all set to 0.5, indicating equal emphasis on accuracy and stability.

[0049] The smaller the overall cost (i.e. the smaller the mean and standard deviation), the closer the denoising effect score is to 1, and the better the denoising effect. This denoising effect score takes into account both the accuracy and stability of speed restoration, ensuring that the selected decomposition layers can accurately restore the speed while avoiding flight control misjudgments caused by speed jumps.

[0050] Based on steps one through five above, traverse all candidate wavelet decomposition levels, compare the final calculated denoising effect scores, and select the candidate wavelet decomposition level with the highest denoising effect score as the optimal decomposition level for the corresponding region.

[0051] Then repeat the above operation to obtain the optimal number of decomposition layers for each region.

[0052] S4: Real-time acquisition of flight data in the current window and matching of regions, followed by calling the corresponding optimal decomposition level for adaptive wavelet denoising, and extraction of instantaneous frequencies to complete signal frequency analysis.

[0053] In official competitions, it is necessary to process signals collected by the ground receiver in real time and quickly and accurately extract the current instantaneous speed and position of the UAV. Since the optimal decomposition layer corresponding to different regions has been obtained through the offline learning of S1-S3 mentioned above, it is now only necessary to determine which type of region the UAV is in based on its current motion state (angular velocity, acceleration), and then call the pre-optimized optimal decomposition layer for that type of region for wavelet denoising, thereby achieving accurate analysis of the frequency components in the received signal.

[0054] Specifically, using the same sampling frequency and window length as the offline phase (obtained from S1 above), the three-axis angular velocity sequence, three-axis acceleration sequence, and received signal sequence of the current window, which consists of continuous time periods ending at the current moment, are acquired in real time.

[0055] The triaxial angular velocity sequence and triaxial acceleration sequence were normalized respectively. The normalization parameters were the global minimum and maximum values ​​of each axis saved in the offline stage.

[0056] Further, the motion feature vector of the current window is calculated according to the operation of S2 above. Then, the Euclidean distance between the motion feature vector of the current window and the center vector of each region in S2 above is calculated, and the region with the smallest Euclidean distance is selected as the matching region of the current window.

[0057] The optimal decomposition layer of the matching region (obtained from S3 above) is further used to perform the wavelet inverse transform operations from "step one" to "step four" in S3 above on the received signal sequence of the current window to obtain the denoised received signal sequence. Then, the instantaneous frequency of the current window is extracted by Hilbert transform. This instantaneous frequency is the result of the superposition of the Doppler frequency shift caused by the relative motion between the UAV and the ground receiver and the transmitted carrier frequency, which is the key input for subsequent state restoration.

[0058] Based on the instantaneous frequency of the current window obtained from the above precise analysis, it is converted into velocity and position using the Doppler frequency shift formula. Compared to directly processing the raw received signal, the result is closer to the actual motion state of the UAV. Smooth and stable velocity output helps the flight control system make correct decisions, while accurate position output can be used for navigation, timing, or competition judgment.

[0059] The present invention also provides a signal frequency analysis system for drone racing. The system includes a processor and a memory, the memory storing computer program instructions, which, when executed by the processor, implement the signal frequency analysis method for drone racing according to the first aspect of the present invention.

[0060] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0061] It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept, and these all fall within the scope of protection of this invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for analyzing the signal frequency of unmanned aerial vehicle (UAV) racing, characterized in that, include: Establish a three-dimensional coordinate system for the track and collect historical data from multiple practice flights to construct a historical window set; Based on the motion response characteristics of each historical window in the historical window set, historical windows with similar motion response characteristics are divided into the same region; For each region, the candidate wavelet decomposition levels are traversed and the denoising effect score is calculated to determine the optimal decomposition level for each region. Real-time acquisition of flight data in the current window and matching of regions are performed by calling the corresponding optimal decomposition level for adaptive wavelet denoising and extracting instantaneous frequencies to complete signal frequency analysis. The optimal number of decomposition layers is determined as follows: For each region, for each candidate wavelet decomposition level, the received signals of each historical window in the region are denoised using wavelet decomposition level to restore the instantaneous speed of the UAV. The denoising effect score of each candidate wavelet decomposition level is calculated based on the statistical results of the speed restoration error between the restored instantaneous speed and the true speed. The candidate wavelet decomposition level with the highest denoising effect score is determined as the optimal decomposition level for that region.

2. The signal frequency analysis method for unmanned aerial vehicle (UAV) racing as described in claim 1, characterized in that, Collecting historical data from multiple practice flights to construct a historical window set includes: Set the sampling frequency and simultaneously collect the actual flight speed sequence of the UAV, the received signal sequence of the ground receiver, the three-axis angular velocity sequence, and the three-axis acceleration sequence during multiple practice flights, and normalize the collected data. Set the sliding window length and sliding step size, and slide to capture multiple historical windows on the time sequence of each practice flight to construct a historical window set.

3. The signal frequency analysis method for unmanned aerial vehicle (UAV) racing according to claim 2, characterized in that, Dividing historical windows with similar characteristics into the same region includes: For each historical window, extract its three-axis angular velocity sequence and three-axis acceleration sequence, calculate the mean and standard deviation of the angular velocity sequence of each axis and the mean and standard deviation of the acceleration sequence of each axis, and concatenate the calculated statistical features to form the motion feature vector of the historical window. Using the motion feature vectors of all historical windows as samples, a clustering algorithm is used for classification, and historical windows with similar motion response characteristics are grouped into the same cluster, with each cluster corresponding to a type of track area.

4. The signal frequency analysis method for unmanned aerial vehicle (UAV) racing according to claim 3, characterized in that, The calculation of the denoising effect score includes: The received signal sequence of the historical window is decomposed using the candidate wavelet decomposition level to obtain low-frequency components and high-frequency components of each level. The energy disorder of the received signal is calculated based on the energy proportion of each component. The decision threshold is determined based on the energy disorder. The decision threshold is used to perform hard thresholding on the high-frequency components of each layer. The processed high-frequency components and the original low-frequency components are subjected to wavelet inverse transform to obtain the denoised received signal sequence. The instantaneous frequency is extracted by Hilbert transform and converted into the restored instantaneous velocity. The velocity restoration error between the restored instantaneous velocity and the true velocity is calculated. The mean and standard deviation of the velocity restoration error of all historical windows within the statistical region at the corresponding candidate wavelet decomposition level are used to calculate the denoising effect score.

5. The signal frequency analysis method for unmanned aerial vehicle (UAV) racing according to claim 4, characterized in that, Determining the judgment threshold based on energy disorder includes: Determine the lower and upper limits of the judgment threshold. The lower limit of the judgment threshold is set to the value corresponding to the low quantile of the absolute value of all high-frequency components, and the upper limit of the judgment threshold is set to the value corresponding to the high quantile of the absolute value of all high-frequency components. The judgment threshold is the larger of the lower limit of the judgment threshold and the upper limit of the judgment threshold multiplied by the negative energy disorder degree raised to the power of the natural constant.

6. The signal frequency analysis method for unmanned aerial vehicle (UAV) racing according to claim 3, characterized in that, Real-time acquisition of flight data in the current window and matching of regions including: The system acquires the triaxial angular velocity sequence, triaxial acceleration sequence, and received signal sequence of the current window, which consists of continuous time intervals ending at the current time. Motion feature vectors are extracted based on the triaxial angular velocity sequence and triaxial acceleration sequence of the current window. The distance between the extracted motion feature vectors and the center vectors of each region is calculated, and the region with the smallest distance is selected as the matching region of the current window.

7. The signal frequency analysis method for unmanned aerial vehicle (UAV) racing according to claim 1, characterized in that, The candidate wavelet decomposition level is generated after determining the maximum decomposable level based on the number of sampling points within the historical window. The maximum decomposable level is the logarithm of the number of sampling points to the base 2, rounded down.

8. A signal frequency analysis system for unmanned aerial vehicle (UAV) racing, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement the signal frequency analysis method for unmanned aerial vehicle racing according to any one of claims 1-7.

Images