Hardware and software noise reduction method for an airborne acoustic detection device

Abstract

The application relates to a hardware and software noise reduction method of an airborne acoustic detection equipment in the technical field of unmanned aerial vehicle acoustic detection, which comprises a hardware noise reduction part and a software noise reduction part, wherein the hardware noise reduction part mainly reduces noise picked up by a microphone array by adopting noise reduction materials and designing a noise reduction structure, so as to physically reduce noise; the software noise reduction part mainly reduces noise of collected signals through post signal processing. Here, experimental and test results of a digital noise reduction method are mainly described; high-accuracy environmental sound collection is realized by combining hardware noise reduction and software noise reduction.

Description

Technical Field

[0001] This invention relates to a hardware and software noise reduction method for an airborne acoustic testing device, belonging to the technical field of UAV acoustic testing. Background Technology

[0002] Insulators, fittings, and other equipment on overhead line towers are prone to operational defects and safety hazards due to environmental factors. Commonly used drone inspections rely on visible light and infrared methods, which are insufficient for detecting and warning of early-stage defects in equipment that are not damaged or have not yet experienced temperature rise.

[0003] Acoustic testing is one of the research hotspots in nondestructive testing. Currently, there are mature products used in fields such as handheld partial discharge testing, substation abnormal noise detection, and gas leak detection. However, these products are severely affected by drone propeller noise and cannot accurately identify partial discharge noise in airborne environments. Therefore, their application in the field of drones, which can be inspected over a wide area, is a first.

[0004] In response, this invention proposes a noise reduction scheme based on UAV acoustic detection equipment, which combines hardware noise reduction and software noise reduction to achieve highly accurate environmental sound acquisition. Summary of the Invention

[0005] To address the aforementioned problems in the existing technology, this invention provides a hardware and software noise reduction method for airborne acoustic detection equipment.

[0006] The technical solution of the present invention is as follows:

[0007] A hardware noise reduction method for airborne acoustic detection equipment, including

[0008] Adding a shock-absorbing structure can eliminate some of the noise generated by mechanical vibration; and

[0009] By setting up a directional sound receiving structure, noise incident from other directions is reduced, and the signal-to-noise ratio of sound waves from possible forward sound sources is enhanced.

[0010] As a further specific implementation

[0011] The shock absorption structure adopts a steel wire rope structure, which uses multiple arc-shaped steel wire ropes arranged in a ring to form a drum-shaped cage structure, and connects the upper and lower ends of the cage structure between the drone's frame and gimbal.

[0012] As a further specific implementation

[0013] The directional sound receiving structure includes a sound wave conduction cavity of a certain length, with an acoustic detection array at one end of the sound wave conduction cavity, which is a MEMS microphone, and the other end is open as the sound receiving end; wherein, the cavity wall of the sound wave conduction cavity has a loose porous structure, and the cavity is designed with chamfers and smooth corners.

[0014] A software noise reduction method for airborne acoustic testing equipment, including

[0015] Acoustic detection array acquires high-quality signals:

[0016] The acoustic detection array integrates observations from multiple microphones for system noise reduction:

[0017] Based on the fundamental theoretical principles of the array reach model localization algorithm, a vibration model of multiple sound sources superimposed on the array is obtained, and considering the noise vector of the microphone, a complete sound field modeling equation is obtained.

[0018] Time-difference-based sound source localization technology calculates the sound source location by calculating the signal delay and combining it with the microphone position, and then uses the arrival wave model to calculate the sound source location. It also uses beamforming theory to perform sound source localization. After calculating the sound field model and summing the time delays through the above process, the sound source estimation result in the sound field is obtained.

[0019] Noise separation is performed. During the calculation of sound source localization, the properties of the matrix are used to decompose the signal subspace and the noise subspace. The direction of the sound source is estimated by the orthogonal vectors of the noise subspace, thereby reducing irrelevant noise.

[0020] As a further specific implementation

[0021] The acoustic detection array acquires high-quality signals:

[0022] The signal is denoised using wavelet packet denoising and reconstruction algorithms. After denoising, the signal is reconstructed using the optimal wavelet packet basis algorithm, which adopts the principle of minimum entropy. The signal is then reconstructed using the optimal wavelet packet tree, which improves the signal-to-noise ratio and highlights the main features.

[0023] As a further specific implementation

[0024] The acoustic detection array integrates the observation results from multiple microphones to perform system noise reduction:

[0025] The vibration equation of the sound source particle is as follows:

[0026]

[0027] After a certain time delay, the vibration equation of the spatial particle P is obtained as follows:

[0028]

[0029] Where c is the speed of sound and x is the propagation distance;

[0030] The above wave equation solution based on time delay assumes a plane wave, meaning the amplitude is independent of the propagation distance, the frequency remains constant during propagation, and the phase changes with distance. In the spherical wave model, only the amplitude A is attenuated with respect to distance. According to the basic theory of acoustics, the sound pressure amplitude is inversely proportional to the propagation distance, thus the propagation model of spherical waves can be obtained.

[0031] As a further specific implementation

[0032] The Dapo model problem description:

[0033] When there are k sound sources in the sound field of m microphones, assuming m > k, the vibration equation of the k-th sound source has been defined above, and the distance between k and m is r. km The distance from the sound source to the array reference center is r m ;

[0034] Consider a near-field model:

[0035] The vibration generated by sound source k at the reference center is:

[0036]

[0037] The signal generated by sound source k at array element m is:

[0038]

[0039] After dividing the two:

[0040]

[0041] The delay term can be extracted from it:

[0042]

[0043] Therefore, the total signal received by the m-th microphone can be considered as the superposition of all sound sources in space at that point:

[0044]

[0045] Based on the above principles, a simple derivation can be made to obtain the vibration model of m sound sources superimposed at the array, i.e., the sound source arrival wave model:

[0046]

[0047] Each column is an M-dimensional direction vector, where M is the number of microphones, representing the way the sound wave from the k-th sound source propagates to the microphone.

[0048] The first term on the right side of the equation is the arrival matrix, which is an M*K dimensional matrix that describes the mode of action of sound waves in the manifold matrix.

[0049] The above equation can be simplified to some extent, and the microphone noise vector must also be considered:

[0050] Y = AX + N

[0051] Where Y is the observation of the microphone array, A is the arrival matrix, X is the propagation equation, and N is the noise vector of the microphone; thus, the complete sound field modeling equation is obtained.

[0052] As a further specific implementation

[0053] The process of estimating sound sources in the sound field:

[0054] The calculation of signal delay relies on the assumption of stationary random signals. Transformers are relatively stable devices with relatively fixed sound source locations, which can meet the basic assumptions required for calculating delay.

[0055] Assume X(t) and Y(t) are discrete sources of generalized stationary randomness, with mathematical expectations that do not change over time, and their autocorrelation functions are:

[0056] R xx (τ)=E[X(n)X * (n+m)]

[0057] The cross-correlation function is:

[0058] R xx (τ)=E[X(n)Y * (n+m)]

[0059] When the correlation function reaches its maximum value, its time offset is the optimal estimated signal delay under this method:

[0060] τ=ΔTargmaxR XY (m)

[0061] The convolution operation in the time domain is transformed into a multiplication process in the frequency domain; the relationship between PSD and the cross-correlation function is as follows:

[0062]

[0063] Alternatively, a generalized cross-correlation method can be used, employing a frequency domain filter to process the correlation function. The optimal filter H is the reciprocal of the signal's self-power spectrum.

[0064]

[0065] After calculating the sound field model and summing the time delays through the above process, the estimation results of the sound sources in the sound field can be obtained.

[0066] As a further specific implementation

[0067] The noise separation process is performed as follows:

[0068] Based on the description of the noise spatial separation problem, the sound source wave equation at the microphone is:

[0069]

[0070] Therefore, at the m-th microphone, the superposition of the fluctuations from all sound sources can be expressed as:

[0071]

[0072] According to geometric relationships, the theoretical time delay can be obtained from the following formula:

[0073]

[0074] By artificially introducing a noise term into the equation, the wave equation of the microphone array can be expressed in matrix form, where the last term is the introduced noise term:

[0075]

[0076] Expressed in matrix form:

[0077] X(n) = AS(n) + V(n)

[0078] Where X is the vibration data received by the microphone, A is the steering vector, S is the sound source wave model, and V is the noise;

[0079] X represents the observed data, which includes other noise from the scene. It is assumed that the noise of each microphone element is independent and satisfies a variance of... The Gaussian distribution is introduced, and the covariance matrix R is introduced;

[0080] R = E[X(n)X] H (n)]

[0081] Through theoretical derivation:

[0082]

[0083] in:

[0084] R s =E[S(n)S H (n)]

[0085] Perform eigenvalue decomposition on R and calculate based on the properties of eigenvalue decomposition:

[0086] R = UΛU H

[0087] UU H =U H U = I

[0088] U is a unitary matrix composed of eigenvectors, possessing orthogonality; ∧ is a diagonal matrix composed of eigenvalues; based on the properties of eigenvalue diagonal matrices, the covariance matrix is ​​substituted into:

[0089]

[0090] Since Rs is a K-dimensional matrix with at most K eigenvalues, when the number of microphones M exceeds K, the matrix is ​​not a full-rank matrix, and the diagonal elements at the end must be zero. After introducing a noise term, the expression of the eigenvalue matrix becomes:

[0091]

[0092] Based on the fundamental assumption that the intensity of the noise space is less than that of the sound source to be monitored, the diagonal elements in the diagonal matrix produce value jumps, thereby allowing the determination of the signal source component and noise component in the eigenvalue matrix.

[0093] Then, by utilizing the properties of the block matrix, the coupled sound sources in the scene are decomposed into signal subspace and noise subspace. Combined with the noise characteristics of urban substations, the distribution of the signal is further modeled to improve the decomposition accuracy of the noise subspace.

[0094] The present invention has the following beneficial effects:

[0095] This invention discloses a hardware and software noise reduction method for an airborne acoustic detection device, comprising a hardware noise reduction component and a software noise reduction component. The hardware noise reduction component primarily reduces noise picked up by the microphone array through the use of a wire rope vibration damping structure and a special directional sound-collecting structure, as well as noise-reducing materials and structural design. The software noise reduction component primarily reduces noise in the acquired signal through post-processing. This section mainly describes the experimental and test results of the digital noise reduction method; by combining hardware and software noise reduction, highly accurate ambient sound acquisition is achieved. Attached Figure Description

[0096] Figure 1 This is a schematic diagram of the MEMS structure of the present invention;

[0097] Figure 2 This is a schematic diagram of the standard gimbal structure of the present invention;

[0098] Figure 3 A steel wire rope shock absorption structure customized for this invention;

[0099] Figure 4 This is a schematic diagram illustrating the sound absorption principle of the sound-absorbing cotton of the present invention;

[0100] Figure 5 This is a diagram of the wavelet packet decomposition structure of the present invention;

[0101] Figure 6 This invention relates to the principle of maximum entropy tree search.

[0102] Figure 7 This is a schematic diagram of the Dabo model of the present invention;

[0103] Figure 8 This is a description of the noise spatial separation problem in this invention. Detailed Implementation

[0104] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0105] See Figure 1-8 An airborne acoustic detection device, comprising

[0106] UAV mounting platform

[0107] The DJI M300 drone has a fuselage length of 0.81 meters (diagonal propeller spacing), a fuselage width of 0.67 meters (diagonal propeller spacing), and a fuselage height of 0.430 meters (ground to cabin canopy). Its dual-battery system enhances flight safety, providing 45-55 minutes of unloaded flight time, a maximum payload of 2.7 kg, and supports customized interfaces. Its dimensions ensure sufficient space and payload for mounting the acoustic array. Therefore, the M300 is currently the most suitable model for compatibility with airborne acoustic signature detection equipment, and subsequent customized development work will be based on the M300 RTK. This report will focus on the noise characteristics of the DJI M300 drone mounted on an acoustic array. Based on the overall design requirements of the drone, the acoustic signature equipment payload needs to be customized and lightweighted to ensure that the drone can still achieve approximately 20 minutes of flight time under load.

[0108] Acoustic acquisition array

[0109] Microphones are commonly used data acquisition devices, specifically those made using microelectromechanical technology, where a pressure-sensing diaphragm is directly fabricated onto a chip. This chip can also integrate numerous computing units, such as amplifiers and digital-to-digital converters, directly generating digital signals that can be read by algorithms. These microphones are small in size, have stable acoustic performance, and a wide frequency response range.

[0110] Traditional microphones work by utilizing the impact of airflow generated by sound waves on a thin plate, causing deformation and altering the capacitance. This changes the output electrical signal, reflecting variations in the frequency and amplitude of the sound waves at the input. MEMS microphones typically consist of a MEMS micro-capacitive sensor, micro-integrated conversion circuitry, a acoustic cavity, and RF anti-interference circuitry. The MEMS micro-capacitive sensor includes a silicon diaphragm for receiving sound and a silicon back electrode. The silicon diaphragm can directly receive audio signals; this is based on Hooke's Law.

[0111] F = -kx

[0112] Compared to traditional microphones, the single-crystal silicon that makes up the MEMS diaphragm and back electrode is a near-perfect Hooke's material, meaning it exhibits almost no hysteresis when bent, and therefore almost no energy dissipation. This ensures the MEMS's precise response and reliable lifespan. Vibration signals are transmitted to a micro-integrated circuit via the MEMS microcapacitive sensor. The micro-integrated circuit converts and amplifies the high-impedance audio signal into a low-impedance signal, which is then filtered by an RF noise suppression circuit, outputting an electrical signal matched to the preamplifier circuit, thus completing the sound-to-electrical conversion. Sound recognition is achieved by reading this electrical signal. A schematic diagram of the MEMS structure is shown below. Figure 1 As shown.

[0113] Noise Analysis

[0114] The acoustic signals collected by airborne acoustic fingerprint cameras contain not only the target signal from partial discharge but also a large amount of noise, such as wind noise, motor noise, propeller noise, and other environmental noise. Most everyday noise energy is distributed in the low-frequency band, while the energy of partial discharge signals is mainly concentrated in the ultrasonic band. This can be utilized to extract the ultrasonic band through filtering, FFT, and other methods, thus eliminating low-frequency noise interference to some extent. On the other hand, compared to acoustic fingerprint cameras used on the ground and indoors, airborne acoustic fingerprint cameras introduce a large amount of high-energy broadband noise into the scene, such as motor noise, propeller noise, and wind noise. This type of noise also has high energy in the ultrasonic band and is difficult to remove through filtering.

[0115] There are two main ways to reduce noise and improve the signal-to-noise ratio of partial discharge signals:

[0116] 1) Hardware noise reduction. This involves using noise-reducing materials and designing noise-reducing structures to physically reduce noise picked up by the microphone array. This method is direct and effective, and is an important means of signal noise reduction. However, physical noise reduction devices are limited by the equipment's usage scenario and the drone's carrying capacity, making it difficult to completely block noise. Therefore, we also need:

[0117] 2) Software noise reduction. Noise reduction is applied to the acquired signals through post-processing. This section primarily describes the experimental and test results of digital noise reduction methods.

[0118] Hardware noise reduction

[0119] Noise propagated through solid mechanical vibration is directly transmitted to the array. Adding a vibration damping structure can eliminate some of this mechanical vibration. Considering the application scenario of line inspection, the array should be kept as stable as possible. Since the acoustic array must maintain a certain aperture to ensure its spatial directivity meets the requirements of acoustic positioning algorithms—a principle determined by physics—the array must maintain a certain size and weight. Within this size and load, traditional vibration damping structures will cause the array to sway more during flight. Vibration damping ball structures have greater elasticity and will produce a larger amplitude of sway during flight. Vibration damping ball structures, such as… Figure 2 As shown:

[0120] By modifying the gimbal's shock-absorbing ball structure and replacing it with a wire rope structure, the gimbal can withstand greater loads and possess a certain degree of rigidity. This ensures that the acoustic array only shifts with the movement of the drone itself in the air, preventing large-scale swaying. The physical characteristics of the wire rope shock absorber are more effective at canceling high-frequency vibrations, better absorbing high-frequency noise and preventing excessive coupling with the frequency bands prone to partial discharge. Figure 3 As shown;

[0121] Noise propagating through the air can directly strike the microphone from all directions within the field of view. Due to the strong interference from the overhead rotor, sound waves incident from non-forward directions will carry even greater rotor noise. Therefore, a directional sound-collecting structure was added to the structural noise reduction design to weaken noise incident from other directions, thereby enhancing the signal-to-noise ratio of potentially forward-facing sound sources.

[0122] The principle of directional sound receiving structure is as follows Figure 4 As shown, sound waves incident from other directions will first be reflected off the inner wall of the cavity; therefore, the purpose of noise reduction here is to suppress reflected sound. Only sound waves incident from the front can directly reach the microphone array. This achieves the structural reduction of airborne noise. Reflected sound mainly propagates within the cavity; the loose, porous structure allows for more effective structural noise reduction targeting the ultrasonic range.

[0123] Regarding the suppression of projected noise, the main source of noise comes from the strong airflow blowing down from the drone, directly impacting the housing of the acoustic signature detection device. Through fluid dynamic design of the external structure, including chamfered cavity design, smooth corner design, and extending the length of the cavity to extend pressure anomalies away from the array, the acoustic signature detection device maintains a stable attitude during flight, preventing excessive vibration and interference noise.

[0124] Software noise reduction

[0125] To obtain high-quality signals, this project uses wavelet packet denoising and reconstruction algorithms to denoise the signals.

[0126] like Figure 5-6 Wavelet packets can decompose the original signal into multiple levels, with each level retaining all the signal's information. Noise reduction is achieved by calculating dynamic thresholds for the wavelet packet coefficients of each node after decomposition. After noise reduction, the optimal wavelet packet basis algorithm is used, which typically employs the principle of minimum entropy, meaning nodes with lower entropy exhibit greater information regularity. We calculate the Shannon entropy of each wavelet packet node; if the sum of the entropies of child nodes is less than that of the parent node, the child node is retained; otherwise, the child node (and subsequent nodes) is removed, retaining only the parent node. Finally, the signal is reconstructed using the optimal wavelet packet tree, effectively improving the signal-to-noise ratio and highlighting the main features.

[0127] Since acoustic detection consists of an array, in addition to decomposing the signal from one microphone, it can also combine the observation results from multiple microphones to perform system noise reduction.

[0128] The sound wave propagation model describes the relationship between the received signal from the array and the signal from the sound source through superposition and synthesis. This model needs to consider information such as array parameters, time delay, and orientation. The vibration equation of the sound source particle is taken into account.

[0129]

[0130] The vibration equation for a point P in space is the vibration after a certain time delay. Let the speed of sound be c and the propagation distance be x, then the vibration equation for the spatial particle P can be obtained as follows:

[0131]

[0132] This time-delay-based wave equation solution makes the plane wave assumption, meaning the amplitude is independent of the propagation distance, the frequency remains constant during propagation, and the phase changes with distance. In contrast, the spherical wave model only attenuates the amplitude A with respect to distance. According to basic acoustic theory, sound pressure amplitude is inversely proportional to propagation distance, thus yielding the spherical wave propagation model.

[0133] The arrival wave model is the fundamental theory behind localization algorithms. It describes the superposition of sound waves from multiple sources on a single microphone during propagation. The arrival wave model problem is described as follows: When there are k sound sources in the sound field of m microphones, assuming m > k, the vibration equation of the k-th sound source is defined above, and the distance between k and m is r. km The distance from the sound source to the array reference center is r m ,as follows Figure 7 As shown:

[0134] Consider a near-field model:

[0135] The vibration generated by sound source K at the reference center is:

[0136]

[0137] The signal generated by sound source k at array element m is:

[0138]

[0139] After dividing the two:

[0140]

[0141] The delay term can be extracted from it:

[0142]

[0143] Therefore, the total signal received by the m-th microphone can be considered as the superposition of all sound sources in space at that point:

[0144]

[0145] Based on the above principles, a simple derivation can be made to obtain the vibration model of m sound sources superimposed at the array, i.e., the sound source arrival wave model:

[0146]

[0147] Each column represents an M-dimensional direction vector, where M is the number of microphones, indicating how the sound wave from the k-th sound source behaves when it reaches the microphone. The first term on the right-hand side of the equation is the arrival matrix, an M*K dimensional matrix describing the behavior of the sound wave within the manifold matrix. We can simplify the above equation to some extent and also need to consider the microphone noise vector:

[0148] Y = AX + N

[0149] Where Y is the observation of the microphone array, A is the arrival matrix, X is the propagation equation, and N is the noise vector of the microphone. Thus, we have obtained the complete sound field modeling equation.

[0150] Currently, beamforming is the dominant localization algorithm, with different beamforming algorithms exhibiting varying spatial awareness characteristics and sensitivities to different frequency bands. The basic idea is time difference of arrival (TDOA)-based sound source localization, which calculates the signal delay and, combined with the microphone's position, uses an arrival wave model to determine the sound source's location. Based on this fundamental idea, numerous variations of beamforming have emerged to adapt to different localization scenarios, such as ultrasonic localization, multi-source localization, single-source localization, sound source tracking, and low-frequency sound source localization. Adaptations are made to each application scenario, focusing on both array design and algorithm design.

[0151] The difference between near-field and far-field sound source localization models lies in their different incident assumptions. The near-field model assumes the incident sound wave is a spherical wave, while the far-field model approximates it as a plane wave. Therefore, the essential difference between the two algorithms lies in the elements of the arrival matrix. By switching the coefficients of matrix A, the near-field and far-field models can be switched.

[0152] The calculation of signal delay relies on the assumption of a stationary random signal. This means that in a single calculation for sound source localization, the location and intensity of the sound source cannot change significantly; otherwise, the stationary signal assumption would be violated. Transformers are relatively stable devices, with a relatively fixed sound source location. Considering that the transformer load also does not change significantly, the sound intensity is also relatively fixed, which satisfies the basic assumptions required for delay calculation.

[0153] Assume X(t) and Y(t) are discrete sources of generalized stationary randomness, with mathematical expectations that do not change over time, and their autocorrelation functions are:

[0154] R xx (τ)=E[X(n)X * (n+m)]

[0155] The cross-correlation function is:

[0156] R xx (τ)=E[X(n)Y * (n+m)]

[0157] When the correlation function reaches its maximum value, its time offset is the optimal estimated signal delay under this method:

[0158] τ=ΔTargmaxR XY (m)

[0159] However, this method is computationally intensive, requiring repeated time-scale shifts and calculations of the signal. Therefore, the cross-power spectral density (CPDS) method was proposed, transforming the time-domain convolution operation into a frequency-domain multiplication process. The effectiveness of this method is guaranteed by the Wiener-Khinchin theorem; the relationship between PSD and the cross-correlation function is as follows:

[0160]

[0161] In addition, there is the generalized cross-correlation method, which uses a frequency domain filter to process the correlation function. Theoretical studies show that the optimal filter H is the reciprocal of the signal's self-power spectrum.

[0162]

[0163] Beamforming theory is an effective way to locate sound sources. By calculating the sound field model and summing the time delays through the above process, the sound source estimation results in the sound field can be obtained.

[0164] Noise separation technology is a highly effective signal processing method in various scenarios. The calculation process for sound source localization involves numerous matrix operations. Utilizing the properties of matrices, the signal subspace and noise subspace can be decomposed. By using the orthogonal vectors of the noise subspace, the direction of the sound source can be estimated, ultimately achieving the effect of reducing irrelevant noise.

[0165] The problem of noise spatial separation is described as follows: Figure 8 As shown, the wave equation for the sound source at the microphone is:

[0166]

[0167] Therefore, at the m-th microphone, the superposition of the fluctuations from all sound sources can be expressed as:

[0168]

[0169] According to geometric relationships, the theoretical time delay can be obtained from the following formula:

[0170]

[0171] By artificially introducing a noise term into the equation, the wave equation of the microphone array can be expressed in matrix form, where the last term is the introduced noise term:

[0172]

[0173] This can be expressed using a concise matrix form:

[0174] X(n) = AS(n) + V(n)

[0175] Where X is the vibration data received by the microphone, A is the steering vector, S is the sound source wave model, and V is the noise.

[0176] X represents the observed data, which includes other noise from the scene. It should be assumed that the noise of each microphone element is independent and satisfies a variance of . The Gaussian distribution is introduced, and the covariance matrix R is introduced.

[0177] R = E[X(n)X] H (n)]

[0178] Through theoretical derivation:

[0179]

[0180] in:

[0181] R s =E[S(n)S H (n)]

[0182] Perform eigenvalue decomposition on R and calculate based on the properties of eigenvalue decomposition:

[0183] R = UΛU H

[0184] UU H =U H U = I

[0185] Here, U is the unitary matrix formed by the eigenvectors, possessing orthogonality. Λ is the diagonal matrix formed by the eigenvalues. Based on the property of eigenvalue diagonal matrices, the covariance matrix is ​​substituted into:

[0186]

[0187] Since Rs is a K-dimensional matrix with at most K eigenvalues, when the number of microphones M exceeds K, the matrix is ​​not a full-rank matrix, and the diagonal elements at the end must be zero. After introducing a noise term, the expression of the eigenvalue matrix becomes:

[0188]

[0189] Based on the fundamental assumption that the intensity of the noise space is less than that of the sound source to be monitored, the diagonal elements in the diagonal matrix produce value jumps, thereby allowing the determination of the signal source component and noise component in the eigenvalue matrix.

[0190] Then, by utilizing the properties of the block matrix, the coupled sound sources in the scene are decomposed into signal subspace and noise subspace. Further research on urban substation noise can lead to further modeling of signal distribution, thereby improving the accuracy of noise subspace decomposition.

[0191] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A hardware and software noise reduction method for an airborne acoustic detection device, characterized in that: Including hardware noise reduction: Adding a shock-absorbing structure eliminates some of the noise generated by mechanical vibration; and setting up a directional sound-receiving structure weakens noise incident from other directions and enhances the signal-to-noise ratio of possible sound sources in the forward direction. The shock absorption structure adopts a steel wire rope structure, which uses multiple arc-shaped steel wire ropes arranged in a ring to form a drum-shaped cage structure, and connects the upper and lower ends of the cage structure between the frame and the gimbal of the UAV. The directional sound receiving structure includes a sound wave conduction cavity of a certain length, with an acoustic detection array at one end of the sound wave conduction cavity, which is a MEMS microphone, and the other end is open as the sound receiving end; wherein, the cavity wall of the sound wave conduction cavity has a loose porous structure, and the cavity is designed with chamfers and smooth corners. It also includes software noise reduction: Acoustic detection array acquires high-quality signals: The acoustic detection array integrates observations from multiple microphones for system noise reduction: Based on the fundamental theoretical principles of the array reach model localization algorithm, a vibration model of multiple sound sources superimposed on the array is obtained, and considering the noise vector of the microphone, a complete sound field modeling equation is obtained. Time-difference-based sound source localization technology calculates the sound source location by calculating the signal delay and combining it with the microphone position, and then uses the arrival wave model to calculate the sound source location; it also uses beamforming theory to perform sound source localization and obtain sound source estimation results in the sound field. Noise separation is performed. During the calculation of sound source localization, the properties of the matrix are used to decompose the signal subspace and the noise subspace. The direction of the sound source is estimated by the orthogonal vectors of the noise subspace, thereby reducing irrelevant noise. The noise separation process is performed as follows: Based on the description of the noise spatial separation problem, the sound source wave equation at the microphone is: ； Therefore, at the m-th microphone, the superposition of the fluctuations from all sound sources can be expressed as: ; According to geometric relationships, the theoretical time delay of wave propagation between two adjacent microphones can be obtained by the following formula: ； By artificially introducing a noise term into the equation, the wave equation of the microphone array can be expressed in matrix form, where the last term is the introduced noise term: ； Expressed in matrix form: ； Where X is the vibration data received by the microphone, B is the steering vector, S is the sound source wave model, and V is the noise; X represents the vibration data received by the microphone, which contains other noise from the scene. It is assumed that the noise of each microphone element is independent and satisfies a variance of... The Gaussian distribution is introduced, and the covariance matrix R is introduced; ； Through theoretical derivation: ； in: ； Perform eigenvalue decomposition on R and calculate based on the properties of eigenvalue decomposition: ； U is a unitary matrix composed of eigenvectors, possessing orthogonality; ∧ is a diagonal matrix composed of eigenvalues; based on the properties of eigenvalue diagonal matrices, the covariance matrix is ​​substituted into: ； because R s It is a K-dimensional matrix with at most K eigenvalues. When the number of microphones M exceeds K, the matrix is ​​not a full-rank matrix, and the diagonal elements at the end must be zero. After introducing a noise term, the expression of the eigenvalue matrix becomes: ； Based on the fundamental assumption that the intensity of the noise space is less than that of the sound source to be monitored, the diagonal elements in the diagonal matrix produce value jumps, thereby allowing the determination of the signal source component and noise component in the eigenvalue matrix.

2. The hardware and software noise reduction method for an airborne acoustic detection device as described in claim 1, characterized in that: The acoustic detection array acquires high-quality signals: The signal is denoised using wavelet packet denoising and reconstruction algorithms. After denoising, the signal is reconstructed using the optimal wavelet packet basis algorithm, which adopts the principle of minimum entropy. The signal is then reconstructed using the optimal wavelet packet tree, which improves the signal-to-noise ratio and highlights the main features.

3. The hardware and software noise reduction method for an airborne acoustic detection device as described in claim 1, characterized in that: The acoustic detection array integrates the observation results from multiple microphones to perform system noise reduction: The vibration equation of the sound source particle is: ； After a certain time delay, the vibration equation of the spatial particle P is obtained as follows: ； Where c is the speed of sound and x is the propagation distance; The above wave equation solution based on time delay assumes a plane wave, meaning the amplitude is independent of the propagation distance, the frequency remains constant during propagation, and the phase changes with distance. In the spherical wave model, only the amplitude A is attenuated with respect to distance. According to the basic theory of acoustics, the sound pressure amplitude is inversely proportional to the propagation distance, thus the propagation model of spherical waves can be obtained.

4. The hardware and software noise reduction method for an airborne acoustic detection device as described in claim 1, characterized in that: The process of estimating sound sources in the sound field: The calculation of signal delay relies on the assumption of stationary random signals. Transformers are relatively stable devices with relatively fixed sound source locations, which can meet the basic assumptions required for calculating delay. Suppose X(n) and Y(n) are discrete sources of information that are generalized stationary randomness, with mathematical expectations that do not change over time, and their autocorrelation functions are: ； The cross-correlation function is: ； When the correlation function reaches its maximum value, its time offset is the optimal estimated signal delay under this method: ； The convolution operation in the time domain is transformed into a multiplication process in the frequency domain; the relationship between PSD and the cross-correlation function is as follows: ； ； Alternatively, a generalized cross-correlation method can be used, employing a frequency domain filter to process the correlation function. The optimal filter H is the reciprocal of the signal's self-power spectrum. ； This allows us to obtain the estimation results of the sound sources in the sound field.

Images