Method for collecting underwater acoustic signals

Abstract

The present application relates to the technical field of acoustic signal collection, and particularly relates to a kind of underwater acoustic signal collection methods.The present application includes the following steps: arranging sensor, dynamic fault detection is carried out according to the signal collected by sensor, whether the sensor is judged to be faulty;When judging that the sensor is faulty, the signal of the sensor is compensated and repaired;The sensor signal obtained by step S1 is amplified;The amplified signal enters ADC analog-digital conversion module, according to the requirement of collection task and environmental change, the sampling frequency of ADC analog-digital conversion module is dynamically adjusted using sampling frequency adaptive control mechanism, and the signal is collected by ADC analog-digital conversion module at the dynamically adjusted sampling frequency.The present application breaks through the limitation of traditional redundancy backup, and can still guarantee signal continuity and precision under extreme failure, significantly improves the reliability and environmental adaptability of underwater acoustic signal collection, and greatly improves the detection and collection precision of underwater acoustic signal.

Description

Technical Field

[0001] This invention relates to the field of acoustic signal acquisition technology, and in particular to a method for underwater acoustic signal acquisition. Background Technology

[0002] With the increasing demands for precision in marine resource exploration, underwater target monitoring, and military defense, for example, in the study of marine mammals such as whales, images and acoustic signals captured on the seabed can be combined to conduct daily monitoring and study the habits of whales. Furthermore, in recent years, intelligent reconnaissance devices deployed by other countries have frequently appeared in my country's waters, posing a serious threat to underwater security. These intelligent underwater reconnaissance devices can be detected by deploying underwater acoustic signal sensors at multiple points and collecting underwater acoustic signals in real time.

[0003] Underwater acoustic signal acquisition faces significant technical bottlenecks.

[0004] First, in terms of system fault tolerance, existing underwater acoustic signal acquisition technologies mainly rely on single-axis redundant backup or static calibration strategies. When a coaxial sensor fails, dynamic switching cannot be achieved, and data cannot be recovered through cross-axis physical field reconstruction in the scenario where both sensors fail simultaneously, resulting in a serious lack of system robustness.

[0005] Second, in terms of sampling control, existing underwater acoustic signal acquisition methods use preset fixed sampling frequencies, which are difficult to adapt to complex environmental changes such as sudden changes in underwater sound velocity profiles and turbulent interference, resulting in signal-to-noise ratio fluctuations exceeding ±3dB. Furthermore, the high-energy-consuming static sampling mode leads to insufficient effective data density and wasted storage resources.

[0006] The aforementioned deficiencies collectively restrict the reliability and accuracy of underwater acoustic detection and underwater acoustic signal acquisition in complex deep-sea environments, becoming a key issue that urgently needs to be addressed in this field. Summary of the Invention

[0007] The purpose of this invention is to overcome the above-mentioned defects in the existing technology and propose an underwater acoustic signal acquisition method that breaks through the limitations of traditional redundancy backup. It can still guarantee signal continuity and accuracy under extreme failure conditions, significantly improve the reliability and environmental adaptability of underwater acoustic signal acquisition, and greatly improve the detection and acquisition accuracy of underwater acoustic signals.

[0008] The technical solution of this invention is: a method for acquiring underwater acoustic signals, comprising the following steps:

[0009] S1. Deploy sensors and perform dynamic fault detection based on the signals collected by the sensors to determine whether the sensors have malfunctioned.

[0010] When a sensor malfunction is detected, the sensor signal is compensated and repaired.

[0011] S2. Amplify the sensor signal obtained after processing in step S1;

[0012] S3. The amplified signal is fed into the ADC analog-to-digital converter module. According to the requirements of the acquisition task and environmental changes, the sampling frequency of the ADC analog-to-digital converter module is dynamically adjusted using the sampling frequency adaptive control mechanism. The ADC analog-to-digital converter module acquires the signal at the dynamically adjusted sampling frequency.

[0013] In this invention, step S1 is implemented through the following specific steps:

[0014] S1.1 Sensor Arrangement: Sensors for signal acquisition are set up on the orthogonal three axes, and the sound pressure sensor and vibration velocity sensor on each axis are set up with the same type of sensor pair. The spacing between two adjacent sensors on the same axis satisfies the following formula:

[0015] ;

[0016] in, Indicates the speed of sound. Indicates the highest detection frequency;

[0017] S1.2, the signals collected by coaxial sensors of the same type , Input the dynamic fault detection module to calculate the cross-correlation function. :

[0018] ,

[0019] Where N=1024; This represents the average signal value of the first sensor of the same type located on the same axis. This represents the average signal value of a second, similar sensor located on the same axis.

[0020] At that time, the axis sensor was not faulty; At that time, the single sensor on that axis fails; At that time, both sensors of the same type on that axis failed.

[0021] S1.3, when When this occurs, it indicates that a sound pressure sensor or vibration velocity sensor in a certain axis has failed, and coaxial redundancy seamless switching is initiated. At this time, another sensor of the same type in the coaxial direction is activated. The activated sensor establishes a new signal path, acquires historical noise data, and calculates adaptive weights based on noise characteristics. The adaptive weights are then superimposed in the frequency domain, and the inverse transformation outputs the compensated sensor signal.

[0022] when When this occurs, it indicates that two sensors of the same type on a certain axis have failed simultaneously. This triggers a cross-axis physical field reconstruction, which regenerates the sensor signal of that axis using data from sensors of the same type on the other two orthogonal axes.

[0023] When calculating the adaptive weights, the noise features are first quantized, and the temporal features are extracted using the following formula:

[0024] ,

[0025] Where x represents the xth independent sensor; This represents the mean of the x-th sensor signal sequence across A sampling points; This indicates the specific time when the x-th sensor signal is sampled for the i-th time;

[0026] Frequency features can be extracted using the following formula:

[0027] ,

[0028] in, Indicates the lower limit frequency; Indicates the upper limit frequency; Indicates the first Fourier transform of sensor signals;

[0029] The extracted frequency domain and time domain features are used to dynamically calculate the adaptive weights. The calculation formula is as follows:

[0030] ,

[0031] in, The preweight of the z-axis coordinate system where the failed sensor is located; The time-domain characteristics of the z-axis coordinate system where the failed sensor is located; This represents the time-domain characteristics of the other two orthogonal axes;

[0032] The calculated adaptive weights Frequency domain superposition and inverse transformation are performed to output the final sensor signal.

[0033] when If two identical sensors of the same type fail along a certain axis, the specific steps for repairing the sensor signal are as follows:

[0034] S1.3.2.1 Sound pressure field intensity reconstruction:

[0035] When two sound pressure sensors fail along a certain axis, sensors of the same type on the other two orthogonal axes are activated. The activated sound pressure sensors reconstruct the sound pressure field strength using a sound pressure signal reconstruction algorithm. The formula for sound pressure signal reconstruction is:

[0036] ,

[0037] Where z represents the coordinate axis where the two failed sensors are located; m and n represent the other two orthogonal axes besides the z-axis; This represents the reconstructed sound pressure field intensity along the z-axis. This represents the sound field propagation attenuation compensation coefficient. ; This represents the average value of the instantaneous signal collected in real time by the two sound pressure sensors along the m-axis. This represents the average value of the instantaneous signal collected in real time by the two sound pressure sensors along the n-axis. This represents the spatial vector from the m-axis sensor to the z-axis sensor; This represents the spatial vector from the sound source to the z-axis sensor; This represents the spatial vector from the n-axis sensor to the z-axis sensing axis;

[0038] S1.3.2.2, Vibration velocity vector reconstruction:

[0039] When two sound pressure sensors fail along a certain axis, the velocity sensors on the other two orthogonal axes are activated. The activated velocity sensors use the Euler equation to discretize and calculate the velocity vector, eliminating the velocity data loss caused by the failed z-axis. The calculation formula for the velocity signal reconstruction algorithm is as follows:

[0040] ,

[0041] in, This represents the reconstructed z-axis vibration velocity signal; M represents the number of segments within the time window used to calculate the vibration velocity vector; This represents the time window difference vibration velocity along the m-axis. ; This represents the time-varying transfer coefficient between the z-axis and the m-axis; This represents the time window difference vibration velocity along the n-axis. ; This represents the time-varying transfer coefficient between the z-axis and the n-axis;

[0042] S1.3.2.3, Environmental Disturbance Compensation:

[0043] A compensation model is constructed by integrating the three elements of sound speed, turbulence, and density. The compensation equation is as follows:

[0044] ,

[0045] in, The velocity of sound profile is represented by the following formula:

[0046] ,

[0047] Where T represents seawater temperature and S represents seawater salinity;

[0048] The turbulence intensity is expressed by the following formula:

[0049] ,

[0050] in, Indicates the first Instantaneous sound pressure value at each sampling point; This represents the average sound pressure level across all sampling points;

[0051] The density perturbation is represented by the following formula:

[0052] .

[0053] The specific implementation steps of the sampling frequency adaptive control mechanism include:

[0054] S3.1. Presample and buffer the signal data amplified in step S2;

[0055] After receiving the amplified raw data from the sensor, the system first performs time-domain smoothing to eliminate high-frequency noise interference, and then uses a dynamic threshold detection algorithm to identify valid data segments.

[0056] S3.2 Sampling Feature Analysis;

[0057] The algorithm utilizes multidimensional feature extraction and dynamic weighted decision tree to achieve dynamic evaluation and processing strategy generation for input signals.

[0058] S3.3 Receive the feature parameters calculated in the previous step, and calculate the optimal sampling frequency of the ADC analog-to-digital conversion module based on the multi-mode frequency adjustment strategy.

[0059] The dynamic threshold detection algorithm in step S3.1 includes the following specific steps:

[0060] S3.1.2.1 Calculate the dynamic baseline statistics, including the window mean, window standard deviation, and interquartile range:

[0061] Window mean The calculation formula is:

[0062] ,

[0063] in, This represents the original signal sample value in the time series; D represents the sliding window length; d represents the time index of the current moment.

[0064] The current local baseline of the signal is established by using the window mean, which eliminates short-term fluctuation interference and serves as the reference coordinate for threshold calculation.

[0065] Window standard deviation The calculation formula is:

[0066] ,

[0067] The signal fluctuation intensity is quantified by calculating the window standard deviation, dynamically reflecting the environmental noise level. Increase noise level and automatically widen the threshold boundary;

[0068] Interquartile range The calculation formula is:

[0069] ,

[0070] in, This indicates a value located at the 75th percentile. This indicates a value located at the 25% mark.

[0071] S3.1.2.2 Calculate the adaptive threshold boundary:

[0072] Upper threshold for:

[0073] ,

[0074] in, This represents a dynamic scaling function; Represents the time-domain gradient operator; Indicates the drift compensation weight. =0.15; Indicates the cumulative drift amount;

[0075] lower threshold for:

[0076] ,

[0077] in, Indicates the 10th percentile; Indicates the noise attenuation coefficient; Indicates the system noise reference value;

[0078] S3.1.2.3, Perform anomaly identification and judgment:

[0079] The triggering condition for anomaly detection and judgment is:

[0080] ,

[0081] in, A timestamp indicating the time of detection; Indicates the start time of the detection window; Indicates the end time of the detection window; express The original sampled value at time;

[0082] The valid outlier segments are: ;

[0083] When data exceeds the threshold range, it is immediately marked as abnormal; after the judgment is completed, the data repair mechanism is automatically triggered, the abnormal information is recorded synchronously for fault diagnosis, and the threshold parameters are dynamically adjusted to improve the accuracy of subsequent monitoring.

[0084] The specific implementation process of step S3.2 is as follows:

[0085] S3.2.1 Multidimensional feature extraction: feature extraction is performed from frequency, time domain, time-frequency domain and nonlinearity respectively;

[0086] During frequency domain feature extraction, the amplitude spectral density, the proportion of energy in the main frequency band, and the spectral entropy of the signal are extracted:

[0087] Extracting amplitude spectral density The calculation formula is:

[0088] ;

[0089] in, B represents the actual frequency value corresponding to the discrete frequency point index; B represents the analysis window length. Represents a discretized time-domain signal sequence; Indicates the system sampling frequency; Symbol for the imaginary unit; Indicates the frequency index number;

[0090] Extracting the main frequency band energy percentage The calculation formula is:

[0091] ;

[0092] in, Indicates the index number of the starting frequency point of the target main frequency band; Indicates the index number of the end frequency point of the target main frequency band;

[0093] Extracting spectral entropy The calculation formula is:

[0094] ,

[0095] ;

[0096] When extracting time-domain features, it is necessary to extract the signal's peak value, zero-crossing rate, and waveform factors.

[0097] The formula for extracting the peak Kurt is:

[0098] ,

[0099] in, This represents the arithmetic mean of the signals within the analysis window; This represents the standard deviation of the signal within the analysis window;

[0100] When Kurt=3, the noise is normal; when Kurt>3, the signal exhibits abnormal impulses.

[0101] Extracting zero-crossing rate The calculation formula is:

[0102] ,

[0103] Waveform factor The calculation formula is:

[0104] ,

[0105] ;

[0106] in, This represents the root mean square value of the signal within the analysis window;

[0107] When extracting from the time-frequency domain, it is necessary to extract the sub-band energy entropy of the signal. The calculation formula is as follows:

[0108] ,

[0109] ,

[0110] ,

[0111] in, This represents the energy of the p-th subband; This represents the energy percentage of the p-th subband; Indicates the p-th subband. Wavelet packet decomposition coefficients at discrete points;

[0112] When extracting nonlinear features, it is necessary to extract the recurrence rate and determinism:

[0113] Recursion rate The extraction formula is:

[0114] ,

[0115] in, Indicates the recursive distance threshold; Represents the reference state vector; Represents the current state vector;

[0116] Certainty The calculation formula is:

[0117] ,

[0118] in, This represents the length of the diagonal segment in the recursive graph; This represents the minimum length threshold for a valid line segment. Indicates length is The frequency of occurrence of line segments; This represents the recursion point count;

[0119] S3.2.2 The dynamic weighted decision tree algorithm is used to analyze the multidimensional feature space extracted in the above steps, and the feature importance is ranked by the real-time updated confidence weight coefficients.

[0120] The formula for the weighted decision function of the dynamic weighted decision tree algorithm is:

[0121] ,

[0122] in, for The feature vector includes the features in the frequency domain, time domain, time-frequency domain, and nonlinear features obtained through the above steps;

[0123] The formula for calculating the feature confidence weights in dynamic updates is as follows:

[0124] ,

[0125] in, Representation of features The conditional probability of target y; This represents the KL divergence penalty coefficient; Representation of features The baseline probability distribution; Representation of features The real-time probability distribution; Representation of features The dynamic correction factor; Representation of features The initial baseline weights;

[0126] The characteristic standardization function is calculated using the following formula:

[0127] ;

[0128] in, This represents the baseline mean value under normal operating conditions. This represents the standard deviation under normal operating conditions.

[0129] The multi-mode frequency adjustment strategy in step S3.3 includes the calculation of the optimal sampling frequency of the ADC analog-to-digital converter module under steady-state conditions and the calculation of the optimal sampling frequency of the ADC analog-to-digital converter module under periodic interference signals.

[0130] Under steady-state conditions, a PID algorithm is used to achieve high-precision frequency tracking of ±0.1%.

[0131] The PID algorithm formula is:

[0132] ,

[0133] in, This represents the output of the PID controller; Indicates the proportional gain coefficient; Indicates the error at the current time; Indicates the integral gain coefficient; Represents the historical error sequence; Indicates the discrete time step; represents the differential gain coefficient; c represents the discrete-time index of the current moment;

[0134] Feature parameters through weights Adjustment , , And other relevant parameters: rise, Increase; waveform factor >3.5, Decrease; spectral entropy rise, Decrease;

[0135] When a sudden change in transient signal is detected, the fuzzy logic controller completes the frequency jump decision:

[0136] The fuzzy logic controller first performs fuzzification: the formula for calculating the triangular function in the fuzzification process is as follows:

[0137] ,

[0138] in, This indicates the membership degree of the input variable g to a low-degree fuzzy set, and its value range is [0,1].

[0139] The formula for calculating the Sigmoid function is:

[0140] ,

[0141] in, This indicates the degree of membership of the input variable g to a high-degree fuzzy set; ;

[0142] After fuzzification, the rules follow a fuzzy rule base. The rules in the fuzzy rule base are in the form of conditional statements, which describe the logical relationship between input variables and output actions in natural language.

[0143] Finally, after deblurring, the output formula is:

[0144] ,

[0145] in, This represents the precise frequency control value of the deblurred output; The universe of discourse variable representing the set of fuzzy actions; This represents the aggregation membership function.

[0146] For periodic interference signals, the MPC algorithm is used for forward optimization. The signal characteristic parameters and system state obtained in the current sampling period are used as initial conditions to predict the signal evolution trend in the future finite time domain.

[0147] First, a state equation is constructed to predict the signal evolution trend. Then, the optimal sampling rate sequence is solved by minimizing the objective function. Finally, the sampling window is adjusted in advance.

[0148] The mathematical expression of the MPC algorithm is:

[0149] ,

[0150] in, Represents the dynamic predicted value of the real-time sampling frequency; Indicates the length of the control time domain; Indicates the target reference sampling frequency;

[0151] Its constraints are:

[0152] ,

[0153] in, Indicates the minimum allowed sampling frequency of the system; This indicates the Nyquist sampling frequency.

[0154] The beneficial effects of this invention are:

[0155] (1) When the dynamic fault detection module determines that a single sensor fails through the cross-correlation function ρ, i.e., 0.3≤ρ<0.7, it immediately activates the coaxial redundant sensor and calculates the adaptive weight based on the noise characteristics. The weight generates the final signal through frequency domain superposition and inverse transformation, realizing seamless switching of the faulty sensor and avoiding data interruption or distortion caused by single-point failure. When both coaxial sensors fail simultaneously, i.e., ρ<0.3, the system starts cross-axis physical field reconstruction: using the sound pressure sensor data of the other two orthogonal axes, the sound pressure field strength reconstruction algorithm is combined with spatial vector projection and attenuation compensation coefficient to regenerate the sound pressure signal of the failed axis. At the same time, the vibration velocity vector is calculated based on the Euler equation to eliminate the missing vibration velocity data. This process is combined with the dynamic compensation model of the three elements of sound velocity-turbulence-density to correct environmental interference in the reconstructed data. It breaks through the limitations of traditional redundant backup and can still ensure signal continuity and accuracy under extreme fault conditions, significantly improving the reliability and environmental adaptability of the system.

[0156] (2) By combining cross-axis physical field reconstruction with sound speed-turbulence-density dynamic compensation model, lossless signal recovery with error <3% is achieved in extreme fault scenarios where both sensors fail simultaneously, increasing the system's continuous operation time from 0 hours in the traditional redundancy scheme to the entire mission cycle; at the same time, combined with real-time correction of environmental interference, the signal-to-noise ratio fluctuation range of complex waters is compressed from ±3dB in the background technology to ±0.5dB, significantly improving the system's comprehensive performance in terms of fault tolerance, data continuity and environmental adaptability;

[0157] (3) Through the adaptive sampling frequency control mechanism, the ADC analog-to-digital conversion module is driven by four-dimensional feature analysis to dynamically adjust the sampling rate in real time. While completely eliminating undersampling distortion, the effective data capture rate is increased to 99.3%. The invalid signal segment is intelligently identified through dynamic threshold detection, so that the sampling rate can be changed as needed, reducing the amount of redundant data by more than 40%. This mechanism, combined with the 2ms frequency jump response of the fuzzy controller and the MPC rolling optimization prediction capability, compresses the high-speed target tracking delay to <8ms. Under the premise of ensuring ±0.5dB signal-to-noise ratio stability, it can achieve high-precision continuous detection throughout the entire mission cycle.

[0158] (4) The underwater acoustic signal sensor can be deployed at multiple points underwater by the method described in this application. The sensor fault judgment and sensor signal compensation and repair in this method can ensure and improve the accuracy of sensor signal acquisition. In the process of signal conversion by the ADC analog-to-digital converter module, the ADC analog-to-digital converter module first needs to acquire the sensor signal. Through the sampling frequency adaptive control mechanism proposed in this application, the sampling frequency of the ADC analog-to-digital converter module can be changed according to the frequency of the target being detected, thereby improving the acquisition efficiency of the ADC analog-to-digital converter module for the sensor signal.

[0159] In summary, the method described in this application enables multi-point deployment of underwater acoustic signal sensors. After deployment, the sensors can automatically detect, correct, and compensate for the signals they retrieve, improving the accuracy of the sensor signals. During the acquisition of sensor signals using the ADC analog-to-digital converter module, the acquisition efficiency is improved, ensuring the accuracy of underwater acoustic signal acquisition. This is beneficial for the identification, surveying, monitoring, and observation of underwater target materials. Furthermore, this method is low-cost and enables multi-point deployment of underwater sensors, providing a relatively reliable approach for monitoring vast sea areas. Attached Figure Description

[0160] Figure 1 This is a flowchart of the sensor fault detection process in this application;

[0161] Figure 2 This is a flowchart of the signal compensation and repair process when a single sensor fails, as described in this application.

[0162] Figure 3 This is a flowchart of signal compensation and repair when two coaxial sensors of the same type fail;

[0163] Figure 4 This is a flowchart of the sampling frequency adaptive control mechanism. Detailed Implementation

[0164] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0165] Specific details are set forth in the following description to provide a full understanding of the invention. However, the invention can be practiced in many ways other than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0166] The underwater acoustic signal acquisition method described in this application specifically includes the following steps.

[0167] The first step is to deploy sensors and perform dynamic fault detection based on the signals collected by the sensors to determine if any sensor malfunctions. When a sensor malfunction is detected, the sensor signal is compensated and repaired. The specific implementation process is as follows.

[0168] First, sound pressure sensors, vibration velocity sensors, temperature sensors, humidity sensors, and attitude sensors are respectively set on the three orthogonal axes, namely the X-axis, Y-axis, and Z-axis, to collect signals. Two sets of the same type of sensor are set on each axis for sound pressure sensors and vibration velocity sensors. The sound pressure sensors and vibration velocity sensors on the X-axis, Y-axis, and Z-axis constitute a 3-axis × 2-set 6-sensor array.

[0169] The spacing between two adjacent coaxial sensors satisfies the following formula:

[0170] ,

[0171] in, Indicates the speed of sound; This indicates the highest detection frequency, which in this embodiment is 8kHz. By setting the spacing between adjacent sensors, wavelength suppression can be effectively achieved. The above are phase interference errors.

[0172] Second, initiate dynamic fault detection to determine if the sensor has malfunctioned.

[0173] Signals collected by coaxial sensors of the same type , Input the dynamic fault detection module and calculate the cross-correlation function using the following formula. :

[0174] ,

[0175] Where N=1024. This represents the average signal value of the first sensor of the same type located on the same axis. This represents the average signal value of a second sensor of the same type located on the same axis.

[0176] Based on the above cross-correlation function The calculation results are used to determine system faults, and the fault determination threshold is: If this occurs, it indicates that the sensor is not faulty; When this occurs, it indicates that a single sensor has failed; This indicates that both coaxial and similar sensors have failed.

[0177] Third, when a sensor malfunctions or fails, it is necessary to compensate for and repair the sensor signal.

[0178] when When this occurs, it indicates that a single sensor, such as a sound pressure sensor or a vibration velocity sensor, has failed along a certain axis. At this time, coaxial redundancy seamless switching is initiated. Another sensor of the same type on the coaxial axis is activated. The activated sensor establishes a new signal path, acquires historical noise data, and calculates adaptive weights based on noise characteristics.

[0179] During adaptive weight calculation, noise features are first quantized, and temporal features are extracted using the following formula:

[0180] ;

[0181] Where x represents the xth independent sensor; This represents the mean of the x-th sensor signal sequence across A sampling points; This indicates the specific time when the x-th sensor signal is sampled for the i-th time.

[0182] Frequency features can be extracted using the following formula:

[0183] ;

[0184] in, Indicates the lower limit frequency; Indicates the upper limit frequency; Indicates the first Fourier transform of sensor signals.

[0185] After the frequency and time domain features are extracted, adaptive weights are dynamically calculated, and the calculation formula is as follows:

[0186] ;

[0187] in, Indicates the location of the failed sensor Preweighting of the coordinate axes; The time-domain characteristics of the coordinate axis where the failed sensor is located; This represents the time-domain characteristics of the other two orthogonal axes.

[0188] In the above formula, Time-domain weighting is used to suppress the influence of high-noise sensors; Frequency domain enhancement. Adaptive weight data after calculation. Frequency domain superposition and inverse transform are performed to output the final signal. The flowchart is as follows: Figure 2 As shown.

[0189] when When two sensors of the same type on the same axis fail simultaneously, cross-axis physical field reconstruction is initiated, and the sensor signal of the axis is regenerated by data from sensors of the same type on the other two orthogonal axes.

[0190] When two sound pressure sensors of the same type on a certain coordinate axis fail, sound pressure sensors of the same type on the other two orthogonal axes are activated. The activated sound pressure sensors use a sound pressure signal reconstruction algorithm to reconstruct the sound pressure field strength. The formula for the sound pressure signal reconstruction algorithm is as follows.

[0191] ,

[0192] Where z represents the coordinate axis where the two failed sensors are located; m and n represent the other two orthogonal axes besides the z-axis; This represents the reconstructed sound pressure field intensity along the z-axis. This represents the sound field propagation attenuation compensation coefficient. ; This represents the average value of the instantaneous signal collected in real time by the two sound pressure sensors along the m-axis. This represents the average value of the instantaneous signal collected in real time by the two sound pressure sensors along the n-axis. This represents the spatial vector from the m-axis sensor to the z-axis sensor, which is provided in real time by the attitude sensor. This represents the spatial vector from the sound source to the z-axis sensor; This represents the spatial vector from the n-axis sensor to the z-axis sensor, provided in real time by the attitude sensor.

[0193] In this embodiment, when the X-axis dual sensors fail, the Y-axis and Z-axis sensors are activated, as shown in the flowchart below. Figure 3 As shown. The formula for the sound pressure signal reconstruction algorithm is as follows:

[0194] ;

[0195] Based on the law of conservation of energy of sound waves in isotropic media, this algorithm can effectively repair signal loss or interference caused by the failure axis by using sensor signals from other coordinate axes for compensation and recovery.

[0196] When two sound pressure sensors on a certain axis fail, the vibration velocity sensors on the other two orthogonal axes are activated. The activated vibration velocity sensors use the discretization solution based on the Euler equation to calculate the vibration velocity vector, eliminating the missing vibration velocity data caused by the failed z-axis coordinate. The calculation formula of the vibration velocity signal reconstruction algorithm is shown below.

[0197] ,

[0198] in, This represents the reconstructed z-axis vibration velocity signal; M represents the number of segments within the time window used to calculate the vibration velocity vector; This represents the time window difference vibration velocity along the m-axis. ; This represents the time-varying transfer coefficient between the z-axis and the m-axis; This represents the time window difference vibration velocity along the n-axis. ; This represents the time-varying transfer coefficient between the z-axis and the n-axis;

[0199] In this embodiment, when the vibration velocity data is missing due to sensor failure on the X-axis, the calculation formula is as follows:

[0200] .

[0201] To analyze and compensate for disturbances caused by environmental factors and further ensure that the system is unaffected by external environmental fluctuations, this application integrates the three elements of sound speed, turbulence, and density to construct a compensation model. The compensation equations are shown below:

[0202] ,

[0203] in, The velocity of sound profile is represented by the following formula:

[0204] ,

[0205] Where T represents seawater temperature and S represents seawater salinity.

[0206] The turbulence intensity is expressed by the following formula:

[0207] ;

[0208] in, Indicates the first Instantaneous sound pressure value at each sampling point; This represents the average sound pressure level across all sampling points.

[0209] The density perturbation is represented by the following formula:

[0210] .

[0211] When both the coaxial dual sound pressure sensor and dual vibration velocity sensor malfunction, the sensor signal of the sound pressure sensor is compensated and repaired first, and then the sensor signal of the vibration velocity sensor is compensated and modified.

[0212] Finally, all the compensated and reconstructed data is output to the next processing step for subsequent processing and analysis. In the event of biaxial sensor failure, the above process ensures stable sensor operation and maintains the accuracy and reliability of the signal data acquired by the sensor.

[0213] when When the sensor is functioning correctly, the signal it collects can be sent to the next processing step.

[0214] The second step is to amplify the sensor signal obtained in the first step.

[0215] In this application, an amplifier circuit module is used to amplify the sensor signal.

[0216] The amplifier circuit module includes the AD620 instrumentation amplifier, which contains an amplifier circuit whose gain can be set by a resistor connected between its input and output. This circuit amplifies the positive change of the input signal, with a positive amplification factor. The total gain of the instrumentation amplifier is typically determined by the following formula:

[0217] ;

[0218] in, Indicates the gain adjustment resistor; This indicates the gain feedback resistor.

[0219] The main features of this amplifier circuit module are:

[0220] (1) Gain greater than 1: by selecting an appropriate and The gain can be set to be greater than 1, thereby amplifying the signal;

[0221] (2) Phase relationship: The output signal is in phase with the input signal;

[0222] (3) High input impedance: It usually has high input impedance and is suitable for connection with high impedance signal sources.

[0223] The multiple signals collected by the sensor are transmitted in parallel and amplified simultaneously by a multi-channel amplifier to increase the intensity of weak sound wave signals and reduce the time delay of acquisition and processing.

[0224] In this application, an appropriate resistor and capacitor are added in parallel to form an RC filter at the top of the amplifier circuit to perform preliminary filtering of the acquired signal, removing unwanted noise. Its cutoff frequency... for:

[0225] .

[0226] The third step is to amplify the signal and then input it into the ADC (Analog-to-Digital Converter) module, which will then acquire the signal.

[0227] In this application, the ADC analog-to-digital conversion module includes an ADC circuit and a power supply. The ADC circuit is built using a dual-channel analog-to-digital converter AD7606, providing high-resolution data conversion, enabling accurate measurement and analysis of weak electrical signals. Taking a vibration velocity sensor as an example, each pair of amplifiers is connected to one ADC circuit. An external clock circuit and trigger circuit are connected to the ADC circuit to receive information and trigger it.

[0228] To achieve optimal data acquisition results, this application proposes a sampling frequency adaptive control mechanism, which dynamically adjusts the sampling frequency of the ADC analog-to-digital conversion module according to the requirements of the acquisition task and environmental changes. This mechanism not only ensures the stable operation of the underwater acoustic signal acquisition method under different working conditions, but also maximizes the saving of computing resources and storage space while guaranteeing the quality of the acquired data.

[0229] The flowchart of the sampling frequency adaptive control mechanism is as follows: Figure 4 As shown. The specific implementation process is as follows.

[0230] First, the signal data amplified in the second step is presampled and buffered.

[0231] The presampling buffer is a key preprocessing unit in the data acquisition process. Its core function is to stabilize and preprocess the data stream by temporarily storing the continuous signal stream input from the sensor.

[0232] The presampling buffer in this application adopts a first-in-first-out buffering mechanism. After receiving the amplified raw data from the sensor, it first performs time-domain smoothing to eliminate high-frequency noise interference, and then identifies valid data segments through a dynamic threshold detection algorithm.

[0233] The time-domain smoothing process is as follows: The original underwater acoustic signal is amplified and stored in a buffer queue using a first-in, first-out (FIFO) mechanism. The system performs low-pass filtering on consecutive sampling points within the current window, employing a sliding weighted average method or a digital finite impulse response (FIR) filter. By designing a smoothing window function with a cutoff frequency, each sampling point within the window is assigned a centrally symmetric, decreasing weight coefficient, and the signal amplitude is weighted and averaged. This process suppresses the energy peak of high-frequency noise to within ±0.5σ' through convolution operations, while preserving the low-frequency principal components of the target sound source. Here, σ' represents the standard deviation of the high-frequency noise.

[0234] The dynamic threshold detection algorithm includes the following steps.

[0235] (a) Calculate dynamic baseline statistics, including the window mean, window standard deviation, and interquartile range.

[0236] Window mean The calculation formula is:

[0237] ;

[0238] in, This represents the original signal sample value in the time series; D represents the sliding window length; and d represents the time index of the current moment.

[0239] The current local baseline of the signal is established by using the window mean, which eliminates short-term fluctuation interference and serves as the reference coordinate for threshold calculation.

[0240] Window standard deviation The calculation formula is:

[0241] .

[0242] The signal fluctuation intensity is quantified by calculating the window standard deviation, dynamically reflecting the environmental noise level. Increase noise, automatically widen threshold boundaries.

[0243] Interquartile range The calculation formula is:

[0244] ;

[0245] in, This indicates a value located at the 75th percentile. This indicates the value at the 25% position.

[0246] Under strong pulse interference, the calculation of interquartile range is more robust to characterizing the degree of dispersion.

[0247] (ii) Calculate the adaptive threshold boundary.

[0248] Adaptive threshold boundaries play a crucial role in intelligent decision-making within dynamic threshold detection algorithms.

[0249] Upper threshold for:

[0250] ;

[0251] in, This represents a dynamic scaling function; Represents the time-domain gradient operator; In this embodiment, the drift compensation weight is represented. =0.15; This indicates the cumulative drift amount.

[0252] lower threshold for:

[0253] ;

[0254] in, Indicates the 10th percentile; Indicates the noise attenuation coefficient; This represents the system noise baseline value.

[0255] (iii) Perform anomaly identification and judgment.

[0256] The triggering condition for anomaly detection and judgment is:

[0257] ;

[0258] in, A timestamp indicating the time of detection; Indicates the start time of the detection window; Indicates the end time of the detection window; express The original sampled value at time.

[0259] The valid outlier segments are: By using dynamic threshold detection, the system can avoid the impact of outliers on subsequent analysis.

[0260] The core purpose of setting trigger conditions is to monitor anomalies in the sensor data stream in real time. The judgment process is based on a dynamic time window. Every moment within Sample values Whether the preset upper and lower thresholds are exceeded: When the data exceeds the threshold range, it is immediately marked as abnormal; after the judgment is completed, the system will automatically trigger the data repair mechanism, synchronously record the abnormal information for fault diagnosis, and dynamically adjust the threshold parameters to improve the accuracy of subsequent monitoring, thereby ensuring data reliability and maintaining the stable operation of the system.

[0261] Second, sampling feature analysis. This step is the core decision-making step, which uses multi-dimensional feature extraction and dynamic weighted decision tree algorithm to dynamically evaluate the input signal and generate processing strategies.

[0262] (a) Multidimensional feature extraction: feature extraction is performed from frequency, time domain, time-frequency domain and nonlinearity respectively.

[0263] When extracting frequency domain features, the amplitude spectral density, the energy proportion of the main frequency band, and the spectral entropy of the signal are extracted.

[0264] Extracting amplitude spectral density Used to quantize signals at frequency The energy distribution intensity at a given location is calculated using the following formula:

[0265] ;

[0266] in, B represents the actual frequency value corresponding to the discrete frequency point index; B represents the analysis window length. Represents a discretized time-domain signal sequence; Indicates the system sampling frequency; Symbol for the imaginary unit; Indicates the frequency index number.

[0267] Extracting the main frequency band energy percentage The formula used to calculate the proportion of energy in the key frequency band to the total energy is as follows:

[0268] ;

[0269] in, Indicates the index number of the starting frequency point of the target main frequency band; Indicates the index number of the end frequency point of the target main frequency band.

[0270] Extracting spectral entropy The formula used to determine the degree of disorder in the spectral energy distribution is as follows:

[0271] ,

[0272] .

[0273] When extracting time-domain features, it is necessary to extract the peak value, zero-crossing rate, and waveform factor of the signal.

[0274] The formula for extracting the peak Kurt is:

[0275] ;

[0276] in, This represents the arithmetic mean of the signals within the analysis window; This represents the standard deviation of the signal within the analysis window.

[0277] The sharpness of the signal pulse components is determined by peak quantization: when Kurt=3, it is normal distributed noise; when Kurt>3, the signal has abnormal impulses.

[0278] Extracting zero-crossing rate The calculation formula is:

[0279] .

[0280] Waveform factor characterizes the ratio of peak value to RMS value of a signal; waveform factor extraction The calculation formula is:

[0281] ,

[0282] ;

[0283] in, This represents the root mean square value of the signal within the analysis window.

[0284] When extracting from the time-frequency domain, it is necessary to extract the sub-band energy entropy of the signal. It measures the complexity of the energy distribution of a signal in different frequency subbands, and the calculation formula is:

[0285] ,

[0286] ,

[0287] ;

[0288] in, This represents the energy of the p-th subband; This represents the energy percentage of the p-th subband; Indicates the p-th subband. Wavelet packet decomposition coefficients at discrete points.

[0289] When extracting nonlinear features, it is necessary to extract the recurrence rate and determinism.

[0290] Recursion rate It calculates the cluster density of state points in phase space, and its extraction formula is:

[0291] ;

[0292] in, Indicates the recursive distance threshold; Represents the reference state vector; This represents the current state vector.

[0293] Certainty The physical meaning of is the proportion of the length of the predictable trajectory, and the calculation formula is:

[0294] ;

[0295] in, This represents the length of the diagonal segment in the recursive graph; This represents the minimum length threshold for a valid line segment. Indicates length is The frequency of occurrence of line segments; This represents the recursion point count.

[0296] (ii) A dynamic weighted decision tree algorithm is used to analyze the multidimensional feature space extracted in the above steps, and the feature importance is ranked by the confidence weight coefficients updated in real time. The purpose of dynamic weight allocation is to distinguish the credibility of effective signals by quantifying the differences in signal quality in real time.

[0297] The formula for the weighted decision function of the dynamic weighted decision tree algorithm is:

[0298] ;

[0299] in, for The feature vector includes the features in the frequency domain, time domain, time-frequency domain, and nonlinear features obtained through the above steps.

[0300] The formula for calculating the feature confidence weights in dynamic updates is as follows:

[0301] ;

[0302] in, Representation of features The conditional probability of target y; This represents the KL divergence penalty coefficient; Representation of features The baseline probability distribution; Representation of features The real-time probability distribution; Representation of features The dynamic correction factor; Representation of features The initial baseline weights.

[0303] The characteristic standardization function is calculated using the following formula:

[0304] ;

[0305] in, This represents the baseline mean value under normal operating conditions. This represents the standard deviation under normal operating conditions.

[0306] Third, receive the feature parameters calculated in the previous step, and calculate the optimal sampling frequency of the ADC analog-to-digital conversion module based on the multi-modal frequency adjustment strategy.

[0307] (i) Under steady-state conditions, a PID algorithm is used to achieve high-precision frequency tracking of ±0.1%.

[0308] The PID algorithm formula is:

[0309] ;

[0310] in, This represents the output of the PID controller; Indicates the proportional gain coefficient; Indicates the error at the current time; Indicates the integral gain coefficient; Represents the historical error sequence; Indicates the discrete time step; represents the differential gain coefficient; c represents the discrete-time index of the current moment.

[0311] Feature parameters through weights Adjustment , , And other related parameters. rise, Increasing the waveform factor improves the system's response speed; >3.5, Reduce and suppress pulse-induced oscillations; spectral entropy rise, Reduce the integral effect to avoid noise accumulation.

[0312] When a transient signal change is detected, the fuzzy logic controller can make a frequency jump decision within 2 milliseconds.

[0313] The fuzzy logic controller transforms precise input variables into membership degrees of fuzzy sets, performing fuzzification as the first step in operation. The fuzzification process can follow either the triangular function or the sigmoid function. The formula for calculating the triangular function is as follows:

[0314] ;

[0315] in, This represents the membership degree of the input variable g to a "low-degree" fuzzy set, with a value range of [0,1].

[0316] The formula for calculating the Sigmoid function is:

[0317] ,

[0318] in, This indicates the degree of membership of the input variable g to a "high-degree" fuzzy set; .

[0319] After fuzzification, the rules follow a fuzzy rule base. The rules in the fuzzy rule base are in the form of conditional statements, which describe the logical relationship between input variables and output actions in natural language, such as "if the line spectrum intensity is high and the signal-to-noise ratio is excellent, then the sampling rate will be greatly improved".

[0320] Finally, defuzzification is the process of converting the fuzzy action set output by the inference engine into precise control quantities. The output formula is calculated as follows:

[0321] ;

[0322] in, This represents the precise frequency control value of the deblurred output; The universe of discourse variable representing the set of fuzzy actions; This represents the aggregation membership function.

[0323] (II) For periodic interference signals, Model Predictive Control (MPC) algorithm is used for forward-looking optimization. This algorithm constructs a dynamic state-space model of the underwater acoustic signal, using the signal characteristic parameters and system state obtained in the current sampling period as initial conditions, and predicts the signal evolution trend in the future finite time domain. Based on this, an optimization problem with feature-weighted signal fidelity as the objective is solved online to obtain the optimal sampling frequency sequence for the next few sampling periods. Finally, only the first optimal control variable in the sequence is executed. In the next period, the above "prediction-optimization-execution" process is repeated to achieve adaptive control from signal characteristic perception to closed-loop tuning of sampling frequency, thereby adjusting the sampling window in advance and effectively suppressing periodic interference.

[0324] First, a state equation is constructed to predict the signal evolution trend. Then, the optimal sampling rate sequence is solved by minimizing the objective function. Finally, the sampling window is adjusted in advance to achieve interference suppression and signal fidelity.

[0325] The mathematical expression of the MPC algorithm is:

[0326] ,

[0327] in, Represents the dynamic predicted value of the real-time sampling frequency; Indicates the length of the control time domain; Indicates the target reference sampling frequency.

[0328] Its constraints are:

[0329] ;

[0330] in, Indicates the minimum allowed sampling frequency of the system; This indicates the Nyquist sampling frequency.

[0331] In this embodiment, the entire module interacts deeply with the AD7606 chip via the SPI bus, and its internal CFGR register parameters can be dynamically configured, thereby realizing the dynamic adjustment of the sampling frequency of the ADC analog-to-digital conversion module.

[0332] The ADC analog-to-digital converter module acquires signals at the sampling frequency calculated through the above steps and transmits the acquired signals to the water surface monitoring equipment. By dynamically adjusting the sampling frequency of the ADC module, the amount of invalid data is reduced by more than 40% while ensuring signal integrity, significantly reducing storage and transmission load.

[0333] In this application, the core central control module uses an FPGA to control the method described herein. The Zynq7020 is a high-performance SoC chip from Xilinx, offering significant advantages. It integrates a dual-core ARM Cortex-A9 processor with a high clock speed and powerful computing capabilities, enabling efficient handling of complex control tasks and data processing. The main functions of the central control module are: ① controlling the vector hydrophone to acquire signals; ② coordinating the operation of various sensors and modules to ensure stable system operation; ③ storing the data acquired by the vector hydrophone in memory and interacting with the host computer.

[0334] Performance tests were conducted on the sensor fault tolerance and reconstruction proposed in this application. A triaxial six-sensor array was arranged in an anechoic water tank, and sensor faults were artificially simulated. The signal collected by a high-precision standard hydrophone was used as the reference truth. Scenarios of single sensor failure (with the Z-axis sound pressure sensor short-circuited) and simultaneous dual sensor failure (with both Z-axis sound pressure sensors short-circuited simultaneously) were simulated, and multiple experiments were performed, with the average value taken to evaluate the fault detection accuracy and signal reconstruction precision of the method described in this application.

[0335] Table 1. Test Results of Sensor Fault Detection Accuracy

[0336]

[0337] As shown in Table 1, the sensor fault detection accuracy test results demonstrate that the dynamic fault detection method proposed in this application has a rapid response and high accuracy. The method for compensating and repairing sensor signals proposed in this application can effectively recover signals under extreme fault conditions, with the reconstruction error strictly controlled within 3%.

[0338] This application also conducted performance tests on the adaptive sampling mechanism. In a shallow sea test site, the performance of the adaptive sampling mechanism proposed in this application was compared with that of a traditional underwater acoustic data acquisition device with a fixed sampling rate of 50 kHz. The performance comparison is shown in Table 2. The target simulated signal consisted of a frequency-jumping signal with intermittent 20 kHz and 35 kHz pulses and a low signal-to-noise ratio continuous wave signal (frequency 15 kHz, initial SNR ≈ -2 dB).

[0339] Table 2. Performance Comparison of Adaptive Frequency Sampling and Fixed Frequency Sampling

[0340]

[0341] Similarly, a marine test was conducted in the deep-sea area of ​​the South China Sea at a depth of approximately 1000 meters using the method proposed in this application. During the test, the calls of three different individuals of Bryde's whales were identified, and the confidence level of the voiceprint recognition in the signal reconstruction segment reached 92%, achieving the monitoring of marine mammals. The maximum detection distance for simulated AUVs reached 8.5 kilometers, realizing the underwater safety early warning function. The high-resolution seabed strata reflection signals collected by this method, after processing, improved the strata resolution to 0.3 meters, enabling certain marine geological exploration functions.

[0342] In summary, the method described in this application successfully solves the key technical problem of "reliability and accuracy of exploration signals in complex deep-sea environments," providing an efficient and reliable solution for the field of marine exploration technology.

[0343] The underwater acoustic signal acquisition method provided by this invention has been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this invention without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of this invention. The above description of the disclosed embodiments enables those skilled in the art to implement or use this invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of this invention. Therefore, this invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for acquiring underwater acoustic signals, characterized in that, Includes the following steps: S1. Deploy sensors and perform dynamic fault detection based on the signals collected by the sensors to determine whether the sensors have malfunctioned. When a sensor malfunction is detected, the sensor signal is compensated and repaired. S2. Amplify the sensor signal obtained after processing in step S1; S3. The amplified signal enters the ADC analog-to-digital converter module. According to the requirements of the acquisition task and changes in the environment, the sampling frequency of the ADC analog-to-digital converter module is dynamically adjusted using the sampling frequency adaptive control mechanism. The ADC analog-to-digital converter module acquires the signal at the dynamically adjusted sampling frequency. Step S1 is achieved through the following specific steps: S1.1 Sensor Arrangement: Sensors for signal acquisition are set up on the orthogonal three axes, and the sound pressure sensor and vibration velocity sensor on each axis are set up with the same type of sensor pair. The spacing between two adjacent sensors on the same axis satisfies the following formula: ； in, Indicates the speed of sound. Indicates the highest detection frequency; S1.2, the signals collected by coaxial sensors of the same type , Input the dynamic fault detection module to calculate the cross-correlation function. : ， Where N=1024; This represents the average signal value of the first sensor of the same type located on the same axis. This represents the average signal value of a second, similar sensor located on the same axis. At that time, the axis sensor was not faulty; At that time, the single sensor on that axis fails; At that time, both sensors of the same type on that axis failed. S1.3, when When this occurs, it indicates that a sound pressure sensor or vibration velocity sensor in a certain axis has failed, and coaxial redundancy seamless switching is initiated. At this time, another sensor of the same type in the coaxial direction is activated. The activated sensor establishes a new signal path, acquires historical noise data, and calculates adaptive weights based on noise characteristics. The adaptive weights are then superimposed in the frequency domain, and the inverse transformation outputs the compensated sensor signal. when When this occurs, it indicates that two sensors of the same type on a certain axis have failed simultaneously. This triggers a cross-axis physical field reconstruction, which regenerates the sensor signal of that axis using data from sensors of the same type on the other two orthogonal axes.

2. The underwater acoustic signal acquisition method according to claim 1, characterized in that, In step S1.3, when calculating the adaptive weights, the noise features are first quantized, and the temporal features are extracted using the following formula: ， Where x represents the xth independent sensor; This represents the mean of the x-th sensor signal sequence across A sampling points; This indicates the specific time when the x-th sensor signal is sampled for the i-th time; Frequency features can be extracted using the following formula: ， in, Indicates the lower limit frequency; Indicates the upper limit frequency; Indicates the first Fourier transform of sensor signals; The extracted frequency domain and time domain features are used to dynamically calculate the adaptive weights. The calculation formula is as follows: ， in, The preweight of the z-axis coordinate system where the failed sensor is located; The time-domain characteristics of the z-axis coordinate system where the failed sensor is located; This represents the time-domain characteristics of the other two orthogonal axes; The calculated adaptive weights Frequency domain superposition and inverse transform are performed to output the final sensor signal.

3. The underwater acoustic signal acquisition method according to claim 1, characterized in that, In step S1.3, when If two identical sensors of the same type fail along a certain axis, the specific steps for repairing the sensor signal are as follows: S1.3.2.1 Sound pressure field intensity reconstruction: When two sound pressure sensors along a certain axis fail, the sound pressure sensors on the other two orthogonal axes are activated. The activated sound pressure sensors reconstruct the sound pressure field strength using a sound pressure signal reconstruction algorithm. The formula for sound pressure signal reconstruction is as follows: , Where z represents the coordinate axis where the two failed sensors are located; m and n represent the other two orthogonal axes besides the z-axis; This represents the reconstructed sound pressure field intensity along the z-axis. This represents the sound field propagation attenuation compensation coefficient. ; This represents the average value of the instantaneous signal collected in real time by the two sound pressure sensors along the m-axis. This represents the average value of the instantaneous signal collected in real time by the two sound pressure sensors along the n-axis. This represents the spatial vector from the m-axis sensor to the z-axis sensor; This represents the spatial vector from the sound source to the z-axis sensor; This represents the spatial vector from the n-axis sensor to the z-axis sensing axis; S1.3.2.2, Vibration velocity vector reconstruction: When two sound pressure sensors fail along a certain axis, the velocity sensors on the other two orthogonal axes are activated. The activated velocity sensors use the Euler equation to discretize and calculate the velocity vector, eliminating the velocity data loss caused by the failed z-axis. The calculation formula for the velocity signal reconstruction algorithm is as follows: , in, This represents the reconstructed z-axis vibration velocity signal; M represents the number of segments within the time window used to calculate the vibration velocity vector; This represents the time window difference vibration velocity along the m-axis. ; This represents the time-varying transfer coefficient between the z-axis and the m-axis; This represents the time window difference vibration velocity along the n-axis. ; This represents the time-varying transfer coefficient between the z-axis and the n-axis; S1.3.2.3, Environmental Disturbance Compensation: A compensation model is constructed by integrating the three elements of sound speed, turbulence, and density. The compensation equation is as follows: , in, The velocity of sound profile is represented by the following formula: , Where T represents seawater temperature and S represents seawater salinity; The turbulence intensity is expressed by the following formula: ， in, Indicates the first Instantaneous sound pressure value at each sampling point; This represents the average sound pressure level across all sampling points. The density perturbation is represented by the following formula: 。 4. The underwater acoustic signal acquisition method according to claim 1, characterized in that, In step S3, the specific implementation steps of the sampling frequency adaptive control mechanism include: S3.1 Pre-sample and buffer the signal data amplified in step S2; After receiving the amplified raw data from the sensor, the system first performs time-domain smoothing to eliminate high-frequency noise interference, and then uses a dynamic threshold detection algorithm to identify valid data segments. S3.2 Sampling Feature Analysis; The algorithm utilizes multidimensional feature extraction and dynamic weighted decision tree to achieve dynamic evaluation and processing strategy generation for input signals. S3.3 Receive the feature parameters calculated in the previous step, and calculate the optimal sampling frequency of the ADC analog-to-digital conversion module based on the multi-mode frequency adjustment strategy.

5. The underwater acoustic signal acquisition method according to claim 4, characterized in that, The dynamic threshold detection algorithm in step S3.1 includes the following specific steps: S3.1.2.1 Calculate the dynamic baseline statistics, including the window mean, window standard deviation, and interquartile range: Window mean The calculation formula is: ， in, This represents the original signal sample value in the time series; D represents the sliding window length; d represents the time index of the current moment. The current local baseline of the signal is established by using the window mean, which eliminates short-term fluctuation interference and serves as the reference coordinate for threshold calculation. Window standard deviation The calculation formula is: ， The signal fluctuation intensity is quantified by calculating the window standard deviation, dynamically reflecting the environmental noise level. Increase noise level and automatically widen the threshold boundary; Interquartile range The calculation formula is: ， in, This indicates a value located at the 75th percentile. This indicates a value located at the 25% mark. S3.1.2.2 Calculate the adaptive threshold boundary: Upper threshold for: ， in, This represents a dynamic scaling function; Represents the time-domain gradient operator; Indicates the drift compensation weight. =0.15; Indicates the cumulative drift amount; lower threshold for: ， in, Indicates the 10th percentile; Indicates the noise attenuation coefficient; Indicates the system noise reference value; S3.1.2.3, Perform anomaly identification and judgment: The triggering condition for anomaly detection and judgment is: ， in, A timestamp indicating the time of detection; Indicates the start time of the detection window; Indicates the end time of the detection window; express The original sampled value at time; The valid outlier segments are: ; When data exceeds the threshold range, it is immediately marked as abnormal; after the judgment is completed, the data repair mechanism is automatically triggered, the abnormal information is recorded synchronously for fault diagnosis, and the threshold parameters are dynamically adjusted to improve the accuracy of subsequent monitoring.

6. The underwater acoustic signal acquisition method according to claim 4, characterized in that, The specific implementation process of step S3.2 is as follows: S3.2.1 Multidimensional feature extraction: feature extraction is performed from frequency, time domain, time-frequency domain and nonlinearity respectively; During frequency domain feature extraction, the amplitude spectral density, the proportion of energy in the main frequency band, and the spectral entropy of the signal are extracted: Extracting amplitude spectral density The calculation formula is: ； in, B represents the actual frequency value corresponding to the discrete frequency point index; B represents the analysis window length. Represents a discretized time-domain signal sequence; Indicates the system sampling frequency; Symbol for the imaginary unit; Indicates the frequency index number; Extracting the main frequency band energy percentage The calculation formula is: ； in, Indicates the index number of the starting frequency point of the target main frequency band; Indicates the index number of the end frequency point of the target main frequency band; Extracting spectral entropy The calculation formula is: , ； When extracting time-domain features, it is necessary to extract the signal's peak value, zero-crossing rate, and waveform factors. The formula for extracting the peak Kurt is: ， in, This represents the arithmetic mean of the signals within the analysis window; This represents the standard deviation of the signal within the analysis window; When Kurt=3, the noise is normal; when Kurt>3, the signal exhibits abnormal impulses. Extracting zero-crossing rate The calculation formula is: ， Waveform factor The calculation formula is: ， ； in, This represents the root mean square value of the signal within the analysis window; When extracting from the time-frequency domain, it is necessary to extract the sub-band energy entropy of the signal. The calculation formula is as follows: , , ， in, This represents the energy of the p-th subband; This represents the energy percentage of the p-th subband; Indicates the p-th subband. Wavelet packet decomposition coefficients at discrete points; When extracting nonlinear features, it is necessary to extract the recurrence rate and determinism: Recursion rate The extraction formula is: ， in, Indicates the recursive distance threshold; Represents the reference state vector; Represents the current state vector; Certainty The calculation formula is: ， in, This represents the length of the diagonal segment in the recursive graph; This represents the minimum length threshold for a valid line segment. Indicates length is The frequency of occurrence of line segments; This represents the recursion point count; S3.2.2 The dynamic weighted decision tree algorithm is used to analyze the multidimensional feature space extracted in the above steps, and the feature importance is ranked by the real-time updated confidence weight coefficients. The formula for the weighted decision function of the dynamic weighted decision tree algorithm is: ， in, for The feature vector includes the features in the frequency domain, time domain, time-frequency domain, and nonlinear features obtained through the above steps; The formula for calculating the feature confidence weights in dynamic updates is as follows: ， in, Representation of features The conditional probability of target y; This represents the KL divergence penalty coefficient; Representation of features The baseline probability distribution; Representation of features The real-time probability distribution; Representation of features The dynamic correction factor; Representation of features The initial baseline weights; The characteristic standardization function is calculated using the following formula: ； in, This represents the baseline average value under normal operating conditions. This represents the standard deviation under normal operating conditions.

7. The underwater acoustic signal acquisition method according to claim 4, characterized in that, The multi-mode frequency adjustment strategy in step S3.3 includes the calculation of the optimal sampling frequency of the ADC analog-to-digital converter module under steady-state conditions and the calculation of the optimal sampling frequency of the ADC analog-to-digital converter module under periodic interference signals.

8. The underwater acoustic signal acquisition method according to claim 7, characterized in that, Under steady-state conditions, a PID algorithm is used to achieve high-precision frequency tracking of ±0.1%. The PID algorithm formula is: ， in, This represents the output of the PID controller; Indicates the proportional gain coefficient; Indicates the error at the current time; Indicates the integral gain coefficient; Represents the historical error sequence; Indicates the discrete time step; represents the differential gain coefficient; c represents the discrete-time index of the current moment; Feature parameters are weighted Adjustment , , : rise, Increase; waveform factor >3.5, Decrease; spectral entropy rise, Decrease; When a sudden change in transient signal is detected, the fuzzy logic controller completes the frequency jump decision: The fuzzy logic controller first performs fuzzification: the formula for calculating the triangular function in the fuzzification process is as follows: ， in, This indicates the membership degree of the input variable g to a low-degree fuzzy set, and its value range is [0,1]. The formula for calculating the Sigmoid function is: ， in, This indicates the degree of membership of the input variable g to a high-degree fuzzy set; ; After fuzzification, the rules follow a fuzzy rule base. The rules in the fuzzy rule base are in the form of conditional statements, which describe the logical relationship between input variables and output actions in natural language. Finally, after deblurring, the output formula is: ， in, This represents the precise frequency control value of the deblurred output; The universe of discourse variable representing the set of fuzzy actions; This represents the aggregation membership function.

9. The underwater acoustic signal acquisition method according to claim 7, characterized in that, For periodic interference signals, the MPC algorithm is used for forward optimization. The signal characteristic parameters and system state obtained in the current sampling period are used as initial conditions to predict the signal evolution trend in the future finite time domain. First, a state equation is constructed to predict the signal evolution trend. Then, the optimal sampling rate sequence is solved by minimizing the objective function. Finally, the sampling window is adjusted in advance. The mathematical expression of the MPC algorithm is: , in, Represents the dynamic predicted value of the real-time sampling frequency; Indicates the length of the control time domain; Indicates the target reference sampling frequency; Its constraints are: ， in, Indicates the minimum allowed sampling frequency of the system; This indicates the Nyquist sampling frequency.

Images