Test error analysis method based on high-precision acoustic phased array Doppler log

Abstract

The invention relates to the technical field of precision measurement and instrument testing, in particular to a test error analysis method based on a high-precision acoustic phased array Doppler log, and aims to solve the problems that a traditional Doppler log mostly adopts a black box model driven by pure data, explicit modeling of a sound wave propagation and array receiving physical mechanism is lacked, and the test error analysis accuracy is poor. The problems that the generalization ability is weak under errors such as channel amplitude-phase mismatch and emission nonlinearity, real Doppler frequency shift is difficult to accurately extract, and the credibility of speed estimation cannot be quantified, so that output in a complex underwater environment is unreliable are solved. Physical priori such as an acoustic wave equation, a Doppler effect and array manifold is embedded into a neural network, physical consistent modeling and efficient reasoning of multi-channel echoes are achieved, an uncertainty quantification mechanism based on the evidence theory is further introduced, and the reliability and the likelihood of velocity estimation are dynamically evaluated, so that the reliability and the reliability of velocity estimation are improved. And the robustness and interpretability of the system in a complex underwater environment are effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of precision measurement and instrument testing, more particularly, to a test error analysis method based on a high-precision acoustic phased array Doppler log. BACKGROUND

[0002] The high-precision acoustic phased array Doppler log uses multiple array elements to synchronously transmit and receive sound waves, combines digital beam forming technology, and realizes high directivity and high signal-to-noise ratio speed measurement through Doppler frequency shift inversion, and has been widely used in high-precision navigation scenes such as underwater unmanned platforms and deep-sea probes; however, in actual testing, the speed measurement accuracy is easily affected by various coupling errors, including array element phase and amplitude inconsistency, beam pointing deviation, Doppler frequency shift estimation error, platform micro-vibration, water sound speed profile change, seabed bottom scattering characteristic difference, and uncertainty of the motion state of the test platform.

[0003] The patent application with publication number CN118500392A discloses a DVL speed error correction method for an underwater robot based on an improved extreme learning machine (ELM), aiming to improve the precision and reliability of the SINS / DVL navigation system of the underwater robot, the improved ELM model is applied to the DVL speed error correction method, the whole process includes: collecting the output DVL measurement data of the DVL equipment of the underwater robot motion for preprocessing, generating the training set and test set of the DVL speed correction model, constructing the DVL speed prediction model of the improved ELM hybrid method, training the ELM model under the condition that the GPS signal is effective, and the DVL speed prediction output, the present application can solve the problem of the decline of the underwater navigation precision of the underwater robot caused by the inconsistency of the installation angle of the inertial measurement unit and the DVL in the underwater robot navigation system; However, traditional Doppler logs mostly use pure data-driven black box models, lack explicit modeling of sound wave propagation and array reception physical mechanisms, have weak generalization ability under channel amplitude and phase mismatch, transmission nonlinearity and other errors, are difficult to accurately extract the real Doppler frequency shift, and cannot quantify the reliability of the speed estimation, resulting in unreliable output in complex underwater environments, and are difficult to meet the requirements of explainability and robustness for high safety level autonomous navigation; at the same time, the conventional method is generally based on the assumption of uniform medium, ignores the non-uniformity of sound speed and attenuation in space and time, does not compensate for wavefront distortion and ray bending, causes Doppler deviation, imaging defocus and speed misalignment, especially in dynamic, layered or turbulent ocean environments, significantly reduces the speed measurement accuracy and environmental adaptability, and is difficult to support high-fidelity acoustic perception.

[0004] Therefore, the present application proposes a test error analysis method based on a high-precision acoustic phased array Doppler log to solve the above problems. SUMMARY

[0005] The purpose of the present application is to provide a test error analysis method based on a high-precision acoustic phased array Doppler log, which solves the problem that traditional Doppler logs generally use a pure data-driven black box model, lack explicit modeling of the physical mechanism of sound wave propagation and array reception, have weak generalization ability under errors such as channel amplitude-phase mismatch and transmission nonlinearity, are difficult to accurately extract the real Doppler frequency shift, and cannot quantify the reliability of the speed estimation, resulting in unreliable output in complex underwater environments and difficulty in meeting the requirements of high safety level autonomous navigation for explainability and robustness; at the same time, the conventional method is generally based on the assumption of uniform medium, ignores the non-uniformity of sound speed and attenuation in space and time, does not compensate for wavefront distortion and ray bending, causes Doppler deviation, imaging defocus and speed misalignment, significantly reduces the speed measurement accuracy and environmental adaptability, and is difficult to support high-fidelity acoustic perception.

[0006] The purpose of the present application is achieved by the following technical solutions: The test error analysis method based on a high-precision acoustic phased array Doppler log comprises the following steps: Step one: collect calibration observation data of each channel, pre-process the collected calibration observation data, perform amplitude-phase compensation and time sequence alignment processing on the pre-processed calibration observation data, and output calibrated multi-channel original echo data; Step two: based on the input and output responses of the final power amplifier of each transmission channel, a generalized memory polynomial nonlinear model is constructed, and nonlinear digital predistortion is performed on the main detection pulse signal in the digital domain; Step three: taking the calibrated multi-channel original echo data as input, a neural network embedding the physical constraints of sound wave equation, Doppler effect and array manifold is used to output the three-axis velocity vector of the carrier and uncertainty estimation online; Step four: based on the calibrated multi-channel original echo data, an integral type forward relationship between the echo data and the medium sound speed disturbance and attenuation disturbance is established, a four-dimensional sound speed field and attenuation field are inverted by an iterative reconstruction algorithm, and the reconstructed sound speed field is used to correct the sound wave propagation path and wavefront phase; Step five: based on the three-axis velocity vector of the carrier and the uncertainty estimation, the contribution proportion of the transmission gain correction factor, the reception channel amplitude compensation factor and the reconstructed sound speed field to the speed estimation variance is quantified.

[0007] As a preferred embodiment of the present application, the process of constructing a generalized memory polynomial nonlinear model based on the input and output responses of the final power amplifier of each transmission channel in step two comprises: For each transmit channel, obtain the input response sequence and the output response sequence of the final power amplifier thereof, divide each sample of the output response sequence by the small-signal linear gain of the corresponding transmit channel to obtain a normalized output sequence, and traverse each valid sampling time of the normalized output sequence to extract the input samples at the current time and the preset number of past times therefrom to form a current memory window input set; For each input sample in the current memory window input set, calculate the amplitude thereof, multiply the amplitude by the zeroth power, the first power, the second power, and the highest power corresponding to the preset nonlinear order to generate an instantaneous nonlinear basis function item, multiply the amplitude by each amplitude in the future envelope amplitude set to generate a future envelope modulation basis function item, and multiply the amplitude by each amplitude in the historical envelope amplitude set to generate a historical envelope accumulation basis function item, arrange all the basis function items into a row vector in a preset order, and append the row vector to the end of the feature matrix. After the traversal of all the valid sampling times is completed, extract the samples corresponding to the number of rows of the feature matrix from the normalized output sequence to form a target vector, and solve a coefficient vector by using the least squares method, wherein the obtained coefficient vector is used as the generalized memory polynomial nonlinear model parameter of the corresponding transmit channel.

[0008] As a preferred embodiment of the present application, the process of performing nonlinear digital predistortion on the main probe pulse signal in step two in the digital domain includes: For each transmit channel, obtain the digital baseband signal of the main probe pulse, load the generalized memory polynomial digital predistorter of the corresponding transmit channel, traverse each valid sampling time of the main probe pulse digital baseband signal, and extract the input samples at the current time and within the memory depth range from the input signal to form a current memory window input set. For each input sample in the current memory window input set, calculate the amplitude thereof, multiply the amplitude by the zeroth power, the first power, the second power, and the highest power corresponding to the nonlinear order to generate an instantaneous nonlinear basis function item, multiply the amplitude by each amplitude in the future envelope amplitude set to generate a future envelope modulation basis function item, and multiply the amplitude by each amplitude in the historical envelope amplitude set to generate a historical envelope accumulation basis function item. Arrange all the basis function items into a row vector in a preset order, multiply the row vector by the predistorter coefficient vector of the transmit channel element by element, sum the multiplication results to obtain the predistortion output sample at the current time, and append the predistortion output sample to the end of the predistortion output sequence. For each transmit channel, use the predistortion output sequence as the predistorted digital signal thereof.

[0009] As a preferred embodiment of the present application, the process of using the calibrated multi-channel original echo data as input and processing the data by using a neural network embedded with the acoustic wave equation, the Doppler effect, and the array manifold physical constraint in step three includes: The current time calibrated multi-channel original echo data is acquired, and the current time calibrated multi-channel original echo data is input into a pre-trained neural network, and the neural network performs the following operations in the inference process: based on the second-order change amount of the signal at three adjacent points in the time and space directions, the time curvature and the space curvature are obtained respectively, the phase change rate of adjacent time from the multi-channel phase sequence is extracted, converted into an observed Doppler shift, combined with the current estimated velocity, the sound source position and the array geometry to generate a theoretical Doppler shift, the equivalent sound source direction is inversely deduced according to the current velocity, the theoretical phase difference between any two channels is calculated combined with the known wavelength and the array geometry, and then compared with the actual phase difference extracted from the echo data, the position coordinates and the generalized momentum of the carrier in the three-dimensional space are output, the change rates of the position and the generalized momentum with time are calculated respectively, and the negative value of the partial derivative of the total mechanical energy with respect to the generalized momentum and the partial derivative with respect to the position are compared, the velocity vector is calculated from the generalized momentum and the mass matrix, and the velocity components of the carrier in the three orthogonal directions are output.

[0010] As a preferred embodiment of the present application, the process of outputting the three-axis velocity vector of the carrier and the uncertainty estimation in step three includes: The velocity component at the current time is acquired to form a three-axis velocity vector, and according to the values of the three orthogonal directions, a unique velocity interval containing the corresponding velocity vector is determined in a pre-defined candidate velocity interval set, the evidence energy of each candidate velocity interval is acquired, and for each interval, the basic credibility is calculated by substituting the evidence energy into the formula; All candidate velocity intervals are traversed: if a certain interval is completely contained in the interval where the velocity vector is located, the basic credibility of the interval is added to the confidence degree; if there is an intersection between a certain interval and the interval where the velocity vector is located in any dimension, the basic credibility of the interval is added to the likelihood degree, and the velocity vector at the current time, the confidence degree and the likelihood degree are output.

[0011] As a preferred embodiment of the present application, the process of establishing the integral type forward relationship between the echo data and the medium sound speed disturbance and the attenuation disturbance based on the calibrated multi-channel original echo data in step four includes: The current estimated sound speed disturbance is acquired, which is defined within the spatial range of the imaging area and covers the frequency range of the signal, the background sound field is acquired, the second-order derivative of the background sound field with respect to time is calculated, the constant sound speed in the background medium, the receiving position, and the frequency domain Green function from any spatial position in the imaging area to the receiving position are acquired, the sound speed disturbance, the time second-order derivative of the background sound field, the constant sound speed, the receiving position and the frequency domain Green function are substituted into the formula to calculate the echo disturbance of the receiving position at the current time, and the calculated echo disturbance is output as the forward prediction value under the current medium disturbance state.

[0012] As a preferred embodiment of the present application, the process of inverting the four-dimensional sound speed field and the attenuation field in step four by the iterative reconstruction algorithm comprises: Based on the calibrated multi-channel raw echo data, the sound speed disturbance and the attenuation disturbance in the three-dimensional space and the time dimension are solved by the iterative reconstruction algorithm, and the specific process is as follows: S1: Set the sound speed field as a uniform distribution, and set the attenuation field as zero; S2: Read the current sound speed field and the attenuation field, and call the integral forward relationship to calculate the theoretical echo; S3: Read the measured echo, calculate the difference between the theoretical echo and the measured echo at each receiving position and each time sampling point to form a residual signal; S4: Solve the adjoint equation with the residual signal as the source term to obtain the adjoint field, and use the adjoint field and the background sound field to calculate the full-field sensitivity of the residual to the sound speed disturbance and the attenuation disturbance; S5: Determine the update direction of the sound speed field and the attenuation field based on the sensitivity, and complete the iterative update of the sound speed field and the attenuation field; Repeat S2 to S5 until the norm of the residual signal is lower than a preset threshold or the maximum number of iterations is reached, and output the final four-dimensional sound speed field and four-dimensional attenuation field.

[0013] As a preferred embodiment of the present application, the process of using the reconstructed sound speed field to correct the sound wave propagation path and the wave front phase in step four comprises: Using the reconstructed four-dimensional sound speed field, the fast marching algorithm is used to numerically solve the travel time equation to obtain the travel time distribution from each sound source to each spatial position in the imaging region at each time sampling point. Based on the travel time distribution, the spatial gradient of the travel time is calculated, and the sound wave propagation path is generated by tracing from the receiving point to the sound source in the opposite direction of the gradient; Using the signal angular frequency and the travel time accumulated along the corresponding propagation path, the wave front phase is calculated. For a wideband signal, the dispersion coefficient and the sound speed field are used to accumulate the integral of the ratio of the dispersion coefficient to the sound speed cube along each propagation path. The product of the square of the signal angular frequency and the obtained integral is used as the dispersion phase compensation term, which is added to the wave front phase to output the corrected sound wave propagation path and wave front phase.

[0014] As a preferred embodiment of the present application, the process of quantifying the transmission gain correction factor, the receiving channel amplitude compensation factor, and the contribution proportion of the reconstructed sound speed field to the speed estimation variance based on the carrier three-axis velocity vector and the uncertainty estimation in step five comprises: Call the reference sample set to calculate the corresponding carrier three-axis velocity vector to obtain the reference output, and call the independent sample set to calculate the corresponding carrier three-axis velocity vector to obtain the independent output; The emission gain mixed sample set is called, the corresponding carrier three-axis velocity vector is calculated, the emission gain mixed output is obtained, the receiving channel mixed sample set is called, the corresponding carrier three-axis velocity vector is calculated, the receiving channel mixed output is obtained, and the sound speed field mixed sample set is called, the corresponding carrier three-axis velocity vector is calculated, and the sound speed field mixed output is obtained; The speed estimation variance is calculated based on the reference output, the main contribution proportion and the total contribution proportion of the emission gain correction factor are calculated based on the reference output, the independent output and the emission gain mixed output, the main contribution proportion and the total contribution proportion of the receiving channel amplitude compensation factor are calculated based on the reference output, the independent output and the receiving channel mixed output, and the main contribution proportion and the total contribution proportion of the reconstructed sound speed field are calculated based on the reference output, the independent output and the sound speed field mixed output. The speed estimation variance is calculated based on the reference output, the main contribution proportion and the total contribution proportion of the emission gain correction factor are calculated based on the reference output, the independent output and the emission gain mixed output, the main contribution proportion and the total contribution proportion of the receiving channel amplitude compensation factor are calculated based on the reference output, the independent output and the receiving channel mixed output, and the main contribution proportion and the total contribution proportion of the reconstructed sound speed field are calculated based on the reference output, the independent output and the sound speed field mixed output.

[0015] Compared with the prior art, the advantages of the present application are: (1) In the present application, the physical prior such as the acoustic wave equation, the Doppler effect and the array manifold is embedded into the neural network, physical consistent modeling and efficient reasoning of the multi-channel echo are realized, high-precision three-axis velocity vectors are output in real time, an uncertainty quantification mechanism based on evidence theory is further introduced, the confidence and the likelihood of the speed estimation are dynamically evaluated, the robustness and the interpretability of the system in the complex underwater environment are effectively improved, the advantages of the physical law and the data-driven are combined, the overfitting and the generalization failure of the black box model are avoided, and high-reliability state feedback is provided for underwater autonomous navigation; (2) In the present application, based on the Born approximation physical forward model, the multi-channel echo is strictly associated with the sound speed and the attenuation disturbance, the four-dimensional sound speed and the attenuation field are jointly inverted with high resolution by combining the iteration algorithm driven by the adjoint field, the propagation path and the wavefront phase are accurately corrected by using the reconstructed sound speed field, the ray bending and the phase distortion caused by the medium non-uniformity are effectively compensated, the imaging focusing and the positioning accuracy are significantly improved, the wave motion physics, the full waveform information and the time sequence space constraint are combined, the stability is ensured while the wideband dispersion effect is taken into account, and high-fidelity acoustic perception support is provided for the complex dynamic underwater environment. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 The step flow chart for the test error analysis method of the Doppler log in the present application is shown in the figure. Figure 2 The step flow chart for solving the joint distribution of the sound speed disturbance and the attenuation disturbance in the three-dimensional space and the time dimension is shown in the figure. DETAILED DESCRIPTION

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0018] Example 1: As Figure 1 As shown, the test error analysis method based on a high-precision acoustic phased array Doppler speedometer proposed in this invention includes the following steps: Step 1: Collect calibration observation data for each channel, including the time-domain voltage sequences at the input and output of the final stage power amplifier for each transmit channel, and the digital sampling sequences after the calibration signal is injected into each receive channel. Preprocess the collected calibration observation data, including: for each transmit channel, calculating the complex ratio of the time-domain voltage sequences at the input and output of the final stage power amplifier to obtain the transmit gain correction factor and transmit phase offset; for each receive channel, performing quadrature demodulation on the digital sampling sequences based on the known frequency and initial phase of the injected calibration signal, and comparing the demodulation results with the theoretical complex envelope of the calibration signal. By comparison, the received amplitude compensation factor and received phase correction amount are obtained. The preprocessed calibration observation data are then subjected to amplitude and phase compensation and timing alignment processing. The amplitude and phase compensation and timing alignment processing operations include: using the transmit gain correction factor and transmit phase offset, the amplitude and phase compensation of the main probe pulse signal is performed before transmitting the main probe pulse; using the received amplitude compensation factor and received phase correction amount, the amplitude and phase correction of the raw echo data of each channel is performed after receiving the underwater echo and before digital beamforming; based on a unified clock beat, the sampling data of all channels are time-aligned, and the calibrated multi-channel raw echo data is output. By using gain and phase correction at the transmitter, amplitude compensation and phase correction at the receiver, and timing alignment under a unified clock, the amplitude and phase mismatch and delay deviation between channels are effectively eliminated, significantly improving the amplitude and phase consistency and time synchronization accuracy of the multi-channel system. This not only ensures the fidelity of the transmitted and received detection signals and supports high-precision digital beamforming and coherent processing, but also has good versatility and adaptability, enabling it to cope with environmental changes and device aging, and comprehensively improving the stability, reliability and imaging quality of the underwater detection system.

[0019] Step 2: Based on the input and output responses of the final stage power amplifiers of each transmission channel, construct a generalized memory polynomial nonlinear model, and perform nonlinear digital predistortion on the main probe pulse signal in the digital domain; Step two, which involves constructing a generalized memory polynomial nonlinear model based on the input and output responses of the final stage power amplifiers in each transmission channel, includes: For each transmit channel, obtain the input response sequence and the output response sequence of the final power amplifier thereof, divide each sample of the output response sequence by the small-signal linear gain of the corresponding transmit channel to obtain a normalized output sequence, initialize a feature matrix as empty, traverse each valid sampling time of the normalized output sequence, extract the input samples at the current time and the past preset number of times from the input response sequence to form a current memory window input set; extract the input samples at the preset number of future times and calculate their amplitudes to form a future envelope amplitude set, extract the input samples at the preset number of past times and calculate their amplitudes to form a history envelope amplitude set, for each input sample in the current memory window input set, calculate its amplitude, multiply it by the zeroth power, the first power, the second power, and the preset highest order power of the amplitude to generate instantaneous nonlinear basis function terms, multiply it by the amplitudes in the future envelope amplitude set to generate future envelope modulation basis function terms, and multiply it by the amplitudes in the history envelope amplitude set to generate history envelope cumulative basis function terms, arrange all the basis function terms in a row vector according to a preset order, and append them to the end of the feature matrix; After completing the traversal of all valid sampling times, extract the samples corresponding to the number of rows of the feature matrix from the normalized output sequence to form a target vector, use the least squares method to solve a coefficient vector, minimize the squared error between the product of the feature matrix and the coefficient vector and the target vector, and use the obtained coefficient vector as the generalized memory polynomial nonlinear model parameters of the corresponding transmit channel; The process of nonlinear digital predistortion of the main probe pulse signal in step two in the digital domain includes: For each transmit channel, obtain the digital baseband signal of the main probe pulse, and load the generalized memory polynomial digital predistorter of the corresponding transmit channel, including the predistorter coefficient vector, the memory depth, the nonlinear order, the envelope leading step number, and the envelope lagging step number, initialize the predistortion output sequence as empty, traverse each valid sampling time of the main probe pulse digital baseband signal, extract the input samples at the current time and within the memory depth range from the input signal to form a current memory window input set; extract the input samples within the envelope leading step number range and calculate their amplitudes to form a future envelope amplitude set, extract the input samples within the envelope lagging step number range and calculate their amplitudes to form a history envelope amplitude set, for each input sample in the current memory window input set, calculate its amplitude, multiply it by the zeroth power, the first power, the second power, and the highest power corresponding to the nonlinear order of the amplitude to generate instantaneous nonlinear basis function terms, multiply it by the amplitudes in the future envelope amplitude set to generate future envelope modulation basis function terms, and multiply it by the amplitudes in the history envelope amplitude set to generate history envelope cumulative basis function terms; The all basis function items are arranged as a row vector in a preset order, are multiplied with the predistorter coefficient vector of the current transmitting channel element by element, and are summed to obtain a predistortion output sample at the current moment, and are appended to the end of the predistortion output sequence. By constructing a generalized memory polynomial model that fuses instantaneous nonlinearity, memory effect and envelope dynamics, high-precision digital predistortion is performed on the main detection pulse of each transmitting channel, which effectively suppresses the in-band distortion and spectral regeneration caused by the nonlinearity of the power amplifier, improves the fidelity and spectral purity of the transmitted signal, and balances the adaptability of wideband and high peak-to-average ratio signals, and supports multi-channel independent correction, which significantly enhances the overall linearity and detection performance of the system.

[0020] Step three: taking the calibrated multi-channel raw echo data as input, a neural network embedding the wave equation, Doppler effect and array manifold physical constraints is used to output the three-axis velocity vector of the carrier and uncertainty estimation online; The process of taking the calibrated multi-channel raw echo data as input in step three and processing it through a neural network embedding the wave equation, Doppler effect and array manifold physical constraints includes: Obtain the calibrated multi-channel raw echo data at the current moment, input the calibrated multi-channel raw echo data at the current moment into the pre-trained neural network, and the neural network performs the following operations during inference: based on the second-order change of the signal at three adjacent points in time and space, the time curvature and spatial curvature are obtained respectively, and it is verified that the ratio is equal to the square of the known sound speed, the phase change rate of adjacent moments is extracted from the multi-channel phase sequence, which is converted into an observed Doppler shift, combined with the currently estimated velocity, sound source position and array geometry to generate a theoretical Doppler shift, and the two are compared, the equivalent sound source direction is back calculated according to the current velocity, the theoretical phase difference between any two channels is calculated combined with the known wavelength and array geometry, and then compared with the actual phase difference extracted from the echo data, the position coordinates and generalized momentum of the carrier in three-dimensional space are output, and in this framework, the expression of the total mechanical energy is: wherein is the position coordinates of the carrier in three-dimensional space, is the generalized momentum conjugate to the position, equal to the product of the mass matrix and the velocity, is a symmetric positive definite matrix describing the inertial characteristics of the carrier in each degree of freedom, is a potential energy function only dependent on the position, Given the total mechanical energy in the current state, calculate the rate of change of position and generalized momentum with time, and compare it with the negative values ​​of the partial derivative of the total mechanical energy with respect to the generalized momentum and the partial derivative with respect to the position. Calculate the velocity vector from the generalized momentum and the mass matrix, and output the velocity components of the carrier in the three orthogonal directions. Step three, the process of online output of the carrier's three-axis velocity vector and uncertainty estimation, includes: Obtain the velocity components at the current moment to form a three-axis velocity vector. Based on the values ​​in the three orthogonal directions, determine a unique velocity interval containing the corresponding velocity vector from a predefined set of candidate velocity intervals. Obtain the evidence energy for each candidate velocity interval. For each interval, substitute its evidence energy into the formula to calculate the basic confidence level. ,in Indicates a candidate velocity range. This represents the identification framework composed of all candidate velocity ranges. The neural network is represented as Output evidence energy Indicates assignment to Basic credibility; Traverse all candidate velocity intervals: If an interval is completely contained within the interval containing the velocity vector, its basic confidence is added to the confidence score; if an interval intersects with the interval containing the velocity vector in any dimension, its basic confidence is added to the likelihood score. Output the velocity vector, confidence score, and likelihood score at the current moment, and continuously update them. By deeply embedding physical priors such as acoustic wave equations, Doppler effects, and array manifolds into neural networks, the system achieves physical consistency modeling and efficient inference of multi-channel echo data, outputting high-precision three-axis velocity vectors of the carrier online. Simultaneously, an uncertainty quantification mechanism based on evidence theory is introduced to dynamically evaluate the confidence and likelihood of the velocity estimation. This not only improves the system's robustness and interpretability in complex underwater environments but also effectively integrates the advantages of physical laws and data-driven approaches, avoiding overfitting and generalization failures inherent in pure black-box models. This provides reliable and real-time state feedback for autonomous navigation and motion perception.

[0021] Example 2: The technical solution of this embodiment of the invention differs from that of Example 1 in that: like Figure 1 and Figure 2 As shown, step four: Based on the calibrated multi-channel original echo data, establish the integral forward modeling relationship between the echo data and the medium sound velocity disturbance and attenuation disturbance, invert the four-dimensional sound velocity field and attenuation field through the iterative reconstruction algorithm, and use the reconstructed sound velocity field to correct the sound wave propagation path and wavefront phase. The process of establishing the integral type forward relationship between the echo data and the medium sound velocity disturbance and the attenuation disturbance in step four based on the calibrated multi-channel original echo data includes: obtaining the current estimated sound velocity disturbance, which is defined in the spatial range of the imaging area and covers the frequency range of the signal, obtaining the background sound field, calculating the second-order derivative of the background sound field with respect to time, obtaining the constant sound velocity in the background medium, the receiving position, and the frequency domain Green function from any spatial position in the imaging area to the receiving position, and substituting the sound velocity disturbance, the second-order derivative of the background sound field with respect to time, the constant sound velocity, the receiving position and the frequency domain Green function into the formula to calculate the echo disturbance of the receiving position at the current time: , wherein represents the echo disturbance observed at the receiver position at time t, is the spatial coordinate vector of the receiver, represents the sound velocity disturbance at any position r in the medium and angular frequency w, is the spatial coordinate vector of any point in the imaging area, is the constant sound velocity in the background medium, is the frequency domain Green function, which represents the linear propagation response from the source point r to the receiving point at angular frequency w, is the background sound field, that is, the sound pressure produced by the same sound source at position r and time t assuming that the medium sound velocity is constant , is the angular frequency range contained in the signal, is the spatial volume of the three-dimensional imaging area, which is introduced in parallel in the integral kernel, and the influence of the attenuation disturbance is embodied through the amplitude attenuation factor determined by the total propagation path length from the sound source to the receiver through the scattering point and the local attenuation intensity. The amplitude attenuation factor is multiplied by the sound velocity disturbance term to jointly modulate the integral weight, and the calculated echo disturbance is output as the forward prediction value under the current medium disturbance state; The process of inverting the four-dimensional sound velocity field and the attenuation field through the iterative reconstruction algorithm in step four includes: Based on the calibrated multi-channel original echo data, the sound velocity disturbance and the attenuation disturbance in the three-dimensional space and the time dimension are solved through the iterative reconstruction algorithm, and the specific process is as follows: S1: set the sound velocity field as a uniform distribution and the attenuation field as zero; S2: read the current sound velocity field and the attenuation field, and call the integral type forward relationship to calculate the theoretical echo; S3: read the measured echo, calculate the difference between the theoretical echo and the measured echo at each receiving position and each time sampling point to form a residual signal; S4: solve the adjoint equation with the residual signal as the source term to obtain the adjoint field, and use the adjoint field and the background sound field to calculate the full-field sensitivity of the residual to the sound speed perturbation and the attenuation perturbation; S5: determine the update direction of the sound speed field and the attenuation field based on the sensitivity, and in the update process: for each spatial position, constrain the change amplitude of the sound speed value and the attenuation value at the current time and the adjacent time, and for each time, constrain the coordination between the change of the sound speed in three spatial directions and the change of the attenuation in three spatial directions, and complete the iterative update of the sound speed field and the attenuation field; Repeat S2 to S5 until the norm of the residual signal is below a preset threshold or the maximum number of iterations is reached, and output the final four-dimensional sound speed field and four-dimensional attenuation field; The process of using the reconstructed sound speed field to correct the sound wave propagation path and wave front phase in step four includes: Using the reconstructed four-dimensional sound speed field, the fast marching algorithm is used to numerically solve the travel time equation, wherein the modulus of the travel time spatial gradient is equal to the inverse of the local sound speed, and the travel time distribution from each sound source to each spatial position in the imaging area at each time sampling point is obtained. Based on the travel time distribution, the spatial gradient of the travel time is calculated, and the sound wave propagation path is generated by tracing from the receiving point to the sound source in the opposite direction of the gradient. Using the signal angular frequency and the accumulated travel time along the corresponding propagation path, the wave front phase is calculated. For a wideband signal, the dispersion coefficient and the sound speed field are used to accumulate and integrate the ratio of the dispersion coefficient and the sound speed cube along each propagation path. The product of the square of the signal angular frequency and the obtained integral is taken as the dispersion phase compensation term, which is added to the wave front phase to output the corrected sound wave propagation path and wave front phase. By using the Born approximation-based physical forward model, the multi-channel echo data is strictly associated with the medium sound speed and attenuation perturbation, and combined with the adjoint field-driven iterative reconstruction algorithm, the high-resolution joint inversion of the four-dimensional (three-dimensional space + time) sound speed field and attenuation field is realized. On this basis, the travel time equation is accurately solved using the reconstructed sound speed field and the propagation path and wave front phase are corrected, effectively compensating for the ray bending and phase distortion caused by medium inhomogeneity, significantly improving the imaging focusing accuracy and target positioning accuracy. At the same time, this method combines wave physics, full waveform information and space-time constraints, ensuring the stability of the reconstruction while taking into account the wideband dispersion effect, providing key support for high-fidelity acoustic perception and imaging in complex dynamic underwater environments.

[0022] Step five: based on the carrier three-axis velocity vector and uncertainty estimation, quantifying the contribution of the transmit channel pre-distortion coefficient, the receive channel amplitude compensation factor and the reconstructed sound speed field to the speed estimation variance; The process of quantifying the contribution of the transmit gain correction factor, the receive channel amplitude compensation factor and the reconstructed sound speed field to the speed estimation variance based on the carrier three-axis velocity vector and uncertainty estimation in step five includes: The benchmark sample set is called, the benchmark sample set contains the joint values of the transmission gain correction factor, the receiving channel amplitude compensation factor and the reconstructed sound velocity field, the corresponding carrier three-axis velocity vector is calculated, the benchmark output is obtained, the independent sample set is called, the independent sample set has the same number of samples as the benchmark sample set, the serial number is one-to-one corresponding, and the generation process is independent of each other, the corresponding carrier three-axis velocity vector is calculated, and the independent output is obtained; The transmission gain mixed sample set is called, wherein the transmission gain correction factor is taken from the sample corresponding to the serial number of the independent sample set, and the remaining parameters are taken from the sample corresponding to the serial number of the benchmark sample set, the corresponding carrier three-axis velocity vector is calculated, and the transmission gain mixed output is obtained, the receiving channel mixed sample set is called, wherein the receiving channel amplitude compensation factor is taken from the sample corresponding to the serial number of the independent sample set, and the remaining parameters are taken from the sample corresponding to the serial number of the benchmark sample set, the corresponding carrier three-axis velocity vector is calculated, and the receiving channel mixed output is obtained, the sound velocity field mixed sample set is called, wherein the reconstructed sound velocity field is taken from the sample corresponding to the serial number of the independent sample set, and the remaining parameters are taken from the sample corresponding to the serial number of the benchmark sample set, the corresponding carrier three-axis velocity vector is calculated, and the sound velocity field mixed output is obtained; Based on the benchmark output, the main contribution ratio and the total contribution ratio of the transmission gain correction factor are calculated based on the benchmark output, the independent output and the transmission gain mixed output, and the main contribution ratio and the total contribution ratio of the receiving channel amplitude compensation factor are calculated based on the benchmark output, the independent output and the receiving channel mixed output; Based on the benchmark output, the independent output and the sound velocity field mixed output, the main contribution ratio and the total contribution ratio of the reconstructed sound velocity field are calculated, and the main contribution ratio and the total contribution ratio of the transmission gain correction factor, the receiving channel amplitude compensation factor and the reconstructed sound velocity field are output respectively. By introducing the Sobol global sensitivity analysis method, based on the carrier velocity estimation and its uncertainty, the independent contribution and interaction of the three key links of transmission pre-distortion, receiving amplitude and phase compensation and environment modeling to the velocity estimation variance are quantitatively decoupled and accurately quantified; this method not only reveals the priority of the system error source, provides data-driven basis for resource optimization allocation, but also enhances the explainability and reliability of the whole detection system, supports the closed-loop performance optimization for high-precision navigation and perception.

[0023] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can make equivalent replacement or change according to the technical scheme and improvement concept of the present application within the technical range disclosed by the present application, which should be covered in the protection scope of the present application.

Claims

1. A method for analyzing test errors based on a high-precision acoustic phased-array Doppler log, characterized in that, The method comprises the following steps: Step 1: collecting calibration observation data of each channel, preprocessing the collected calibration observation data, performing amplitude and phase compensation and time sequence alignment on the preprocessed calibration observation data, and outputting calibrated multi-channel original echo data; Step 2: based on the input and output responses of the final power amplifier of each transmitting channel, a generalized memory polynomial nonlinear model is constructed, and a nonlinear digital predistortion is performed on the main detection pulse signal in the digital domain; Step 3: taking the calibrated multi-channel original echo data as input, a neural network embedding the acoustic wave equation, Doppler effect and array manifold physical constraints is used to online output the carrier three-axis velocity vector and uncertainty estimation; Step 4: based on the calibrated multi-channel original echo data, an integral type forward relationship between the echo data and the medium sound speed disturbance and the attenuation disturbance is established, a four-dimensional sound speed field and an attenuation field are inverted through an iterative reconstruction algorithm, and the reconstructed sound speed field is used to correct the sound wave propagation path and the wave front phase; Step 5: based on the carrier three-axis velocity vector and the uncertainty estimation, the contribution proportion of the transmitting gain correction factor, the receiving channel amplitude compensation factor and the reconstructed sound speed field to the velocity estimation variance is quantified.

2. The method of claim 1, wherein, The process of constructing a generalized memory polynomial nonlinear model based on the input and output responses of the final power amplifier of each transmitting channel in step 2 comprises: For each transmitting channel, the input response sequence and the output response sequence of the final power amplifier thereof are obtained, each sample of the output response sequence is divided by the small signal linear gain of the corresponding transmitting channel to obtain a normalized output sequence, and at each valid sampling time of the normalized output sequence, the input samples at the current time and the preset number of past times are extracted from the input response sequence to form a current memory window input set; For each input sample in the current memory window input set, the amplitude thereof is calculated, multiplied by the zeroth power, the first power, the second power and the preset highest order power of the amplitude respectively to generate instantaneous nonlinear basis function items, multiplied by the amplitudes in the future envelope amplitude set to generate future envelope modulation basis function items, and multiplied by the amplitudes in the historical envelope amplitude set to generate historical envelope cumulative basis function items, all the basis function items are arranged in a row vector in a preset order, and appended to the end of the feature matrix; After completing the traversal of all valid sampling times, samples corresponding to the number of rows of the feature matrix are extracted from the normalized output sequence to form a target vector, and a least square method is used to solve a coefficient vector, and the obtained coefficient vector is taken as the generalized memory polynomial nonlinear model parameter of the corresponding transmitting channel.

3. The method of claim 2, wherein the high precision acoustic phased array Doppler log based test error analysis method is characterized by, The process of performing nonlinear digital predistortion on the main detection pulse signal in the digital domain in step 2 comprises: For each transmitting channel, the digital baseband signal of the main detection pulse is obtained, and the generalized memory polynomial digital predistorter of the corresponding transmitting channel is loaded, and at each valid sampling time of the main detection pulse digital baseband signal, the input samples within the memory depth range at the current time and the previous time are extracted from the input signal to form a current memory window input set; For each input sample in the current memory window input set, the amplitude thereof is calculated, multiplied by the zeroth power, the first power, the second power, and the highest power corresponding to the non-linear order of the amplitude respectively, to generate an instantaneous non-linear basis function term, multiplied by each amplitude in the future envelope amplitude set respectively to generate a future envelope modulation basis function term, multiplied by each amplitude in the historical envelope amplitude set respectively to generate a historical envelope cumulative basis function term; All basis function terms are arranged in a row vector in a preset order, multiplied by the pre-distortion coefficient vector of the transmission channel element by element, and then summed to obtain the pre-distortion output sample at the current time, and appended to the end of the pre-distortion output sequence. For each transmission channel, the pre-distortion output sequence is used as the pre-distorted digital signal.

4. The method of claim 1, wherein, The process of step three includes: The current time calibrated multi-channel raw echo data is obtained, and the current time calibrated multi-channel raw echo data is input into the pre-trained neural network. The neural network performs the following operations in the inference process: based on the second order change of the signal in the time and space directions, the time curvature and the space curvature are obtained respectively, the phase change rate of the adjacent time is extracted from the multi-channel phase sequence, which is converted into the observed Doppler shift, combined with the current estimated velocity, the sound source position and the array geometry to generate the theoretical Doppler shift, the equivalent sound source direction is back calculated according to the current velocity, combined with the known wavelength and array geometry to calculate the theoretical phase difference between any two channels, and then compared with the actual phase difference extracted from the echo data, the position coordinates and generalized momentum of the carrier in three-dimensional space are output, the change rates of the position and generalized momentum with time are calculated, and the negative value of the partial derivative of the total mechanical energy with respect to the generalized momentum and the partial derivative with respect to the position are compared, the velocity vector is calculated from the generalized momentum and the mass matrix, and the velocity components of the carrier in three orthogonal directions are output.

5. The method of claim 4, wherein the high precision acoustic phased array Doppler log based test error analysis method further comprises: The process of step three includes: The current time velocity component is obtained to form a three-axis velocity vector, and according to the values of the three orthogonal directions, a unique velocity interval containing the corresponding velocity vector is determined in the pre-defined candidate velocity interval set, and the evidence energy of each candidate velocity interval is obtained. For each interval, the basic credibility is calculated by substituting the evidence energy into the formula; All candidate velocity intervals are traversed: if a certain interval is completely contained in the velocity vector interval, the basic credibility is added to the confidence degree; if there is an intersection between the interval and the velocity vector interval in any dimension, the basic credibility is added to the likelihood degree, and the velocity vector, the confidence degree and the likelihood degree at the current time are output.

6. The method of claim 1, wherein, The process of step four includes: Obtaining a current estimated sound speed disturbance, which is defined within a spatial range of the imaging region and covers a frequency range of the signal, obtaining a background sound field, calculating a second-order derivative of the background sound field with respect to time, obtaining a constant sound speed in the background medium, a receiving position, and a frequency-domain Green function from an arbitrary spatial position within the imaging region to the receiving position, substituting the sound speed disturbance, the second-order derivative of the background sound field with respect to time, the constant sound speed, the receiving position, and the frequency-domain Green function into a formula to calculate an echo disturbance of the receiving position at a current time, and outputting the calculated echo disturbance as a forward prediction value under a current medium disturbance state.

7. The method of claim 6, wherein the high precision acoustic phased array Doppler log based test error analysis method further comprises: The process of inverting the four-dimensional sound speed field and the attenuation field in step four through the iterative reconstruction algorithm comprises: Based on the calibrated multi-channel original echo data, the sound speed disturbance and the attenuation disturbance in the three-dimensional space and the time dimension are solved by the iterative reconstruction algorithm, and the specific process is as follows: S1: set the sound speed field as a uniform distribution and the attenuation field as zero value; S2: read the current sound speed field and the attenuation field, and call the integral type forward relationship to calculate the theoretical echo; S3: read the measured echo, calculate the difference between the theoretical echo and the measured echo at each receiving position and each time sampling point to form a residual signal; S4: solve the adjoint equation with the residual signal as the source term to obtain the adjoint field, and calculate the full-field sensitivity of the residual to the sound speed disturbance and the attenuation disturbance using the adjoint field and the background sound field; S5: determine the update direction of the sound speed field and the attenuation field based on the sensitivity, and complete the iterative update of the sound speed field and the attenuation field; Repeat S2 to S5 until the norm of the residual signal is lower than the preset threshold or the maximum iteration number is reached, and output the final four-dimensional sound speed field and four-dimensional attenuation field.

8. The method of claim 7, wherein the high precision acoustic phased array Doppler log based test error analysis method further comprises: The process of using the reconstructed sound speed field to correct the sound wave propagation path and the wave front phase in step four comprises: Using the reconstructed four-dimensional sound speed field, the walk time equation is numerically solved by using the fast marching algorithm to obtain the walk time distribution from each sound source to each spatial position in the imaging region at each time sampling point, and based on the walk time distribution, the spatial gradient of the walk time is calculated, the sound wave propagation path is generated by tracking from the receiving point to the sound source in the opposite direction of the gradient. Using the signal angular frequency and the accumulated walk time along the corresponding propagation path, the wave front phase is calculated, and for a wideband signal, the ratio of the dispersion coefficient to the sound speed cube is accumulated and integrated along each propagation path using the dispersion coefficient and the sound speed field, and the product of the square of the signal angular frequency and the obtained integral is taken as the dispersion phase compensation term, which is added to the wave front phase, and the corrected sound wave propagation path and wave front phase are output.

9. The method of claim 1, wherein, The process of quantifying the transmission gain correction factor, the receiving channel amplitude compensation factor, and the contribution proportion of the reconstructed sound speed field to the speed estimation variance based on the carrier three-axis velocity vector and the uncertainty estimation in step five comprises: The reference sample set is called to calculate the corresponding carrier three-axis velocity vector to obtain the reference output, and the independent sample set is called to calculate the corresponding carrier three-axis velocity vector to obtain the independent output. The emission gain mixed sample set is called, the corresponding carrier three-axis velocity vector is calculated, the emission gain mixed output is obtained, the receiving channel mixed sample set is called, the corresponding carrier three-axis velocity vector is calculated, the receiving channel mixed output is obtained, the sound speed field mixed sample set is called, the corresponding carrier three-axis velocity vector is calculated, and the sound speed field mixed output is obtained; The speed estimation variance is calculated based on the reference output, the main contribution proportion and the total contribution proportion of the emission gain correction factor are calculated based on the reference output, the independent output and the emission gain mixed output, the main contribution proportion and the total contribution proportion of the receiving channel amplitude compensation factor are calculated based on the reference output, the independent output and the receiving channel mixed output; The main contribution proportion and the total contribution proportion of the reconstructed sound speed field are calculated based on the reference output, the independent output and the sound speed field mixed output, and the main contribution proportion and the total contribution proportion of the emission gain correction factor, the receiving channel amplitude compensation factor and the reconstructed sound speed field are output respectively.

Images