Sound source positioning method and device in incomplete observation scene and electronic device

Abstract

The application relates to a sound source positioning method, device and electronic device under an incomplete observation scene, the sound source positioning method comprising: acquiring actual measurement data collected by a linear array; the actual measurement data comprising actual measurement data of inter-element direction arrival angle and actual measurement data of arrival time; based on the actual measurement data, a mixed parameter positioning model of multiple error sources is constructed; the multiple error sources comprising inter-element direction arrival angle measurement noise, arrival time measurement noise and linear array reference element position error; based on the mixed parameter positioning model, an optimization objective function for estimating the sound source position is constructed; a matrix estimation of the gradient of the optimization objective function is calculated, and bias correction is performed to obtain a corrected matrix estimation; based on the corrected matrix estimation, the estimated value of the sound source position is iteratively updated until a preset convergence condition is met, and the final sound source position is obtained. The sound source is accurately positioned by constructing the mixed parameter positioning model and combining the matrix estimation.

Description

Technical Field

[0001] This application relates to the technical field of sound source location, and in particular to sound source location methods, apparatus and electronic devices in scenarios with incomplete observation. Background Technology

[0002] In the field of sound source localization, the two-step localization method has become an important implementation scheme in engineering applications due to its significant advantages such as low equipment requirements, low communication resource consumption, and high computational efficiency. This method generally utilizes signal processing techniques to extract localization parameters related to the sound source location from the signals received by each linear array, and then constructs a mathematical model based on these parameters and estimates the sound source location through optimization. To further improve localization accuracy, fusing multiple localization parameters has become an effective approach.

[0003] Currently, most fusion-based sound source localization algorithms in related technologies are based on the assumption of "complete observation," meaning that each linear array is assumed to simultaneously acquire all angle and time measurement information. In complex real-world environments, factors such as equipment malfunctions, noise interference, and clock deviations often prevent the effective acquisition of some key localization parameters (such as time of arrival), resulting in an "incomplete observation" scenario. Consequently, the performance of fusion-based localization methods in related technologies suffers severe degradation due to model mismatch, leading to a significant decrease in sound source localization accuracy and insufficient robustness, exhibiting the problem of low accuracy in sound source localization under incomplete observation conditions.

[0004] There is currently no effective solution to the problem of low accuracy in sound source localization in related technologies under incomplete observation conditions. Summary of the Invention

[0005] This embodiment provides a sound source localization method, apparatus, and electronic device for scenarios with incomplete observation, in order to solve the problem of low accuracy in sound source localization in related technologies.

[0006] Firstly, this embodiment provides a sound source localization method for scenarios with incomplete observation, including:

[0007] Acquire actual measurement data from the linear array; the actual measurement data includes actual measurement data of the angle of arrival between array elements and actual measurement data of the time of arrival.

[0008] Based on the actual measurement data, a unified hybrid parameter positioning model with multiple error sources is constructed; wherein, the multiple error sources include inter-element direction angle of arrival measurement noise, arrival time measurement noise, and linear array reference element position error;

[0009] The hybrid parameter localization model characterizes the deviation between the actual value of the sound source location and the estimated value of the sound source location as equal to the sum of the linear combinations of the multiple error sources;

[0010] Based on the hybrid parameter localization model, an optimization objective function is constructed for estimating the location of the sound source;

[0011] Calculate the moment estimator of the gradient of the optimization objective function, and perform bias correction to obtain the corrected moment estimator;

[0012] Based on the corrected moment estimate, the estimated value of the sound source location is iteratively updated until the preset convergence condition is met, and the final sound source location is obtained.

[0013] In some embodiments, acquiring the actual measurement data collected by the linear array includes:

[0014] Obtain the position of the linear array reference element and the linear array orientation;

[0015] For each of the linear arrays, determine whether the linear array has the ability to observe the angle of arrival between the array elements and whether it has the ability to observe the time of arrival, and obtain the determination result;

[0016] Based on the determination result, valid observation values ​​are collected from linear arrays that have the ability to observe the arrival angle between array elements and the ability to observe the arrival time, respectively, to obtain the actual measurement data.

[0017] In some embodiments, based on the determination result, valid observation values ​​are collected from linear arrays with the ability to observe the inter-element directional angle of arrival and the ability to observe the arrival time, respectively, to obtain the actual measurement data, including:

[0018] Based on the determination result, an observation capability identifier for the inter-element direction arrival angle and an observation capability identifier for the arrival time are generated for each linear array with observation capability.

[0019] Valid observation values ​​are collected from the observation capability identifiers of the inter-element directional arrival angle and the arrival time to obtain the actual measurement data of the inter-element directional arrival angle and the actual measurement data of the arrival time corresponding to the linear array.

[0020] In some embodiments, constructing a unified hybrid parameter localization model based on the actual measurement data includes:

[0021] Determine the actual observation value vector based on the actual measurement data;

[0022] Establish the residual relationship between the actual observation vector and the vector to be estimated; wherein, the vector to be estimated includes the sound source location, the square of the sound source location modulus, and variables of a portion of the linear array;

[0023] Construct a linear combination of the inter-element directional arrival angle measurement noise, the arrival time measurement noise, and the linear array reference element position error;

[0024] Based on the residual relationship and the linear combination, the unified hybrid parameter localization model for multiple error sources is obtained.

[0025] In some embodiments, constructing an optimization objective function for estimating the sound source location based on the hybrid parameter localization model includes:

[0026] Based on the residual relationship expression defined by the hybrid parameter localization model, a weighted least squares form optimization objective function is constructed.

[0027] In some embodiments, the step of calculating the moment estimator of the gradient of the objective function and performing bias correction to obtain the corrected moment estimator includes:

[0028] Calculate the first and second moments of the gradient of the objective function.

[0029] By introducing corrected first-order moment estimators and corrected second-order moment estimators, deviation corrections are applied to the first-order moment and the second-order moment to obtain corrected first-order moment estimators and second-order moment estimators.

[0030] In some embodiments, iteratively updating the estimated location of the sound source based on the corrected moment estimator until a convergence condition is met includes:

[0031] Based on the corrected first-order moment estimate and second-order moment estimate, the step size for iterative updates is calculated;

[0032] The sound source location estimate is updated based on the step size of the iterative update and the corrected first-order moment estimate until the convergence condition is met; wherein, the convergence condition means that the change in the sound source location estimate in two adjacent iterations is less than a preset location change threshold.

[0033] Secondly, this embodiment provides a sound source localization device for scenarios with incomplete observation, including: a data acquisition module, a function construction module, and an iterative update module; wherein:

[0034] The data acquisition module is used to acquire the actual measurement data collected by the linear array; the actual measurement data includes the actual measurement data of the angle of arrival between array elements and the actual measurement data of the time of arrival.

[0035] The function construction module is used to construct a unified hybrid parameter localization model with multiple error sources based on the actual measurement data; wherein, the multiple error sources include inter-element direction angle of arrival measurement noise, time of arrival measurement noise, and linear array reference element position error; the hybrid parameter localization model characterizes that the deviation between the actual value of the sound source position and the estimated value of the sound source position is equal to the sum of the linear combination of the multiple error sources; based on the hybrid parameter localization model, an optimization objective function for estimating the sound source position is constructed;

[0036] The iterative update module is used to calculate the moment estimate of the gradient of the optimization objective function and perform bias correction to obtain the corrected moment estimate; based on the corrected moment estimate, iteratively update the estimated value of the sound source location until the preset convergence condition is met to obtain the final sound source location.

[0037] Thirdly, this embodiment provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the sound source localization method under incomplete observation scenarios described in the first aspect above.

[0038] Fourthly, this embodiment provides a storage medium storing a computer program that, when executed by a processor, implements the steps of the sound source localization method in the incomplete observation scenario described in the first aspect above.

[0039] Compared with related technologies, this embodiment provides a sound source localization method, apparatus, and electronic device for incomplete observation scenarios. The sound source localization method for incomplete observation scenarios first acquires actual measurement data collected by the linear array; the actual measurement data includes actual measurement data of the inter-element directional angle of arrival and actual measurement data of the arrival time; secondly, based on the actual measurement data, a unified hybrid parameter localization model with multiple error sources is constructed; wherein, the multiple error sources include inter-element directional angle of arrival measurement noise, arrival time measurement noise, and linear array reference element position error; the hybrid parameter localization model characterizes that the deviation between the actual value of the sound source position and the estimated value of the sound source position is equal to the sum of the linear combinations of the multiple error sources; further, based on the hybrid parameter localization model, an optimization objective function for estimating the sound source position is constructed; then, the moment estimate of the gradient of the optimization objective function is calculated, and deviation correction is performed to obtain the corrected moment estimate; finally, based on the corrected moment estimate, the estimated value of the sound source position is iteratively updated until a preset convergence condition is met, thus obtaining the final sound source position. By constructing a hybrid parameter localization model with multiple error sources and combining it with a moment estimation method for deviation correction, it achieves accurate localization of sound sources under incomplete observation conditions.

[0040] Details of one or more embodiments of this application are set forth in the following drawings and description to make other features, objects and advantages of this application more readily apparent. Attached Figure Description

[0041] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0042] Figure 1 This is a flowchart of a sound source localization method in an incomplete observation scenario according to an embodiment of this application;

[0043] Figure 2 This is a simulation layout diagram of the positioning scene used in the simulation of one embodiment of this application;

[0044] Figure 3 This is a performance comparison chart of the adaptive moment estimation algorithm of one embodiment of this application with other algorithms under the influence of direction arrival angle noise;

[0045] Figure 4 This is a performance comparison chart of the adaptive moment estimation algorithm of one embodiment of this application with other algorithms under the influence of arrival time;

[0046] Figure 5 This is a structural block diagram of a sound source localization device in an incomplete observation scenario according to an embodiment of this application. Detailed Implementation

[0047] To better understand the purpose, technical solution, and advantages of this application, the application is described and illustrated below in conjunction with the accompanying drawings and embodiments.

[0048] Unless otherwise defined, the technical or scientific terms used in this application shall have the general meaning understood by one of ordinary skill in the art to which this application pertains. Words such as “a,” “an,” “an,” “the,” “the,” and “these” used in this application do not indicate quantitative limitation and may be singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that comprises a series of steps or modules (units) is not limited to the listed steps or modules (units) but may include steps or modules (units) not listed, or may include other steps or modules (units) inherent to these processes, methods, products, or devices. Words such as “connected,” “linked,” and “coupled” used in this application are not limited to physical or mechanical connections but may include electrical connections, whether direct or indirect. “Multiple” used in this application refers to two or more. “And / or” describes the relationship between related objects, indicating that three relationships may exist; for example, “A and / or B” can represent: A alone, A and B simultaneously, and B alone. Normally, the character " / " indicates that the objects before and after it are in an "or" relationship. The terms "first," "second," "third," etc., used in this application are merely to distinguish similar objects and do not represent a specific order of objects.

[0049] This embodiment provides a sound source localization method in an incomplete observation scenario. Figure 1 This is a flowchart of the sound source localization method in the incomplete observation scenario of this embodiment, as follows: Figure 1 As shown, the process includes the following steps:

[0050] Step S110: Obtain the actual measurement data acquired by the linear array; the actual measurement data includes the actual measurement data of the angle of arrival between array elements and the actual measurement data of the time of arrival.

[0051] In this embodiment, a three-dimensional sound source localization system is constructed, consisting of multiple linear arrays, which measures the inter-element directional angle of arrival (DOA) and time of arrival (TOA). Its localization geometry is based on the spatial intersection of a cone determined by the one-dimensional angle of arrival (1DAOA) and a sphere determined by the time of arrival (TOA). Here, a linear array refers to an array composed of multiple acoustic sensors arranged along a straight line, whose geometric configuration and spatial orientation are known prior information. First, actual measurement data is acquired. This data is obtained by synchronously or asynchronously receiving sound source signals from the sound source through each linear array. Array signal processing technology is used to extract two key localization parameters from the sound source signals. The first is the inter-element directional angle of arrival, defined as the angle between the wavefront of the sound source signal and the normal plane of the array axis when the wavefront arrives at the linear array, reflecting the azimuth information of the sound source relative to the linear array. The second is the time of arrival, defined as the absolute time required for the sound wave emitted by the sound source to propagate to the reference element of the linear array, containing the relative distance information between the sound source and the array. The actual measurement data consists of the two types of observations provided by each linear array.

[0052] Step S120: Based on actual measurement data, construct a unified hybrid parameter localization model with multiple error sources; wherein, the multiple error sources include the measurement noise of the angle of arrival between array elements, the measurement noise of the time of arrival, and the position error of the linear array reference element; the hybrid parameter localization model characterizes that the deviation between the actual value of the sound source position and the estimated value of the sound source position is equal to the sum of the linear combination of the multiple error sources.

[0053] The actual observation data, containing various errors, is transformed into a hybrid parameter localization model through mathematical modeling. The construction of the hybrid parameter localization model follows the logic of "from residual definition to error decomposition." Specifically, firstly, the residual relationship is defined as the difference between the actual observed vector and the vector to be estimated; then, the residual relationship is structurally decomposed into three linearly superimposed error components, corresponding to the inter-element direction angle of arrival measurement noise, the arrival time measurement noise, and the linear array reference element position error, respectively. Among these, the inter-element direction angle of arrival measurement noise reflects the random fluctuations in azimuth estimation, the arrival time measurement noise reflects the random fluctuations in time delay estimation, and the linear array reference element position error reflects the systematic deviation of the geometric reference. Therefore, the finally constructed hybrid parameter localization model is a model that encapsulates the statistical characteristics and geometric coupling relationships of all error sources, with the deviation between the actual sound source position and the estimated sound source position equal to the sum of linear combinations of different error sources, providing an optimization objective for subsequent maximum likelihood estimation.

[0054] Step S130: Based on the hybrid parameter localization model, construct an optimized objective function for estimating the location of the sound source.

[0055] Based on the unified hybrid parameter localization model for multiple error sources constructed in step S120, an optimization objective function for estimating the sound source location is constructed. Minimizing the residual relationship in the hybrid parameter localization model is transformed into a mathematical problem that can be directly solved through numerical optimization. Specifically, in one embodiment, based on the maximum likelihood estimation criterion, and assuming that each measurement noise and position error follows a zero-mean Gaussian distribution, the residual relationship defined by the hybrid parameter localization model is transformed into a scalar objective function. This objective function essentially reflects the likelihood of obtaining actual observation data under the current assumption of the sound source location; the smaller its value, the more the assumption matches the reality. Therefore, the sound source localization problem is explicitly transformed into a numerical optimization problem with the sound source location (including possible array element position deviations) as the optimization variable and minimizing this objective function as the criterion.

[0056] Step S140: Calculate the moment estimator of the gradient of the objective function and perform bias correction to obtain the corrected moment estimator.

[0057] Specifically, the gradient of the objective function is the derivative vector of the objective function with respect to the source position parameters. It quantifies the instantaneous rate of change and the direction of fastest ascent of the objective function value at different positions. During optimization, the gradient of the objective function indicates the direction of deviation between the current estimated position and the optimal solution, providing fundamental information for guiding iterative searches (such as the steepest descent method). However, in complex models containing noise, the directly calculated gradient expectation is usually not zero; its mathematical expectation is itself a function of the parameters. Therefore, the moment estimator of the gradient is essentially the parameter estimate corresponding to the expectation of the gradient vector of the objective function being zero, thus transforming the optimization problem into a problem of solving moment condition equations.

[0058] However, due to the propagation of observation noise through the nonlinear model and the potential finite sample effect, the moment estimator obtained by direct solution will have statistical bias, causing the estimated location to deviate from the true value. Bias correction involves calculating a compensation term based on the statistical characteristics of the noise and the geometric structure of the model, and subtracting this correction bias from the original moment estimator to restore the unbiasedness or asymptotic unbiasedness of the estimator. Ultimately, the corrected moment estimator is a parameter estimator with smaller bias, lower mean square error, and is calibrated.

[0059] Step S150: Based on the corrected moment estimator, iteratively update the estimated value of the sound source location until the preset convergence condition is met, and obtain the final sound source location.

[0060] Based on the corrected moment estimate, an adaptive iterative optimization algorithm is used to cyclically update the sound source location estimate until a preset convergence condition is met, ultimately outputting the sound source localization result. Specifically, the iterative update first uses the corrected moment estimate as the search direction for the current iteration. Simultaneously, it combines the exponential moving averages of the first and second moments to dynamically calculate an update step size that incorporates both momentum and an adaptive learning rate. The current estimate of the sound source location is then corrected based on this update step size to obtain a new round of sound source location estimates. When the convergence condition is met, the iteration automatically terminates, and the sound source location estimate corresponding to the current iteration is determined as the final localization value.

[0061] Compared to related technologies, existing sound source fusion localization techniques rely on the idealized assumption of complete observation, which is difficult to apply in real-world engineering scenarios. Complex underwater or industrial environments with strong noise interference, sensor malfunctions, inaccurate node clock synchronization, and abnormal attenuation or multipath effects during signal propagation can all lead to the complete absence or distortion of key parameters on one or more linear arrays, resulting in a typical "incomplete observation" situation. In this case, the mathematical model upon which traditional algorithms rely suffers from a severe mismatch between input data and model assumptions, introducing systematic biases into the optimization process and potentially preventing convergence. This manifests as a significantly increased localization error, extreme sensitivity to noise and outliers, and the potential for complete system failure when some sensors malfunction. Therefore, there is a problem of low accuracy in sound source localization under incomplete observation conditions.

[0062] Steps S110 to S150 above involve: First, acquiring actual measurement data from the linear array; this data includes actual measurements of the angle of arrival between array elements and the time of arrival. Second, based on the actual measurement data, constructing a unified hybrid parameter localization model with multiple error sources; these sources include measurement noise of the angle of arrival between array elements, measurement noise of the time of arrival, and position error of the reference array elements. The hybrid parameter localization model characterizes the deviation between the actual sound source position and the estimated sound source position as equal to the sum of the linear combinations of the multiple error sources. Further, based on the hybrid parameter localization model, constructing an optimization objective function for estimating the sound source position; then, calculating the moment estimate of the gradient of the optimization objective function and performing deviation correction to obtain the corrected moment estimate; finally, based on the corrected moment estimate, iteratively updating the estimated sound source position until a preset convergence condition is met, thus obtaining the final sound source position. By constructing a hybrid parameter localization model with multiple error sources and combining it with a moment estimation method for deviation correction, accurate sound source localization under incomplete observation conditions is achieved.

[0063] Optionally, in one embodiment, acquiring the actual measurement data collected by the linear array includes: acquiring the position of the linear array reference element and the linear array attitude; for each linear array, determining whether the linear array has the ability to observe the angle of arrival between elements and whether it has the ability to observe the time of arrival, and obtaining the determination result; based on the determination result, acquiring valid observation values ​​from the linear arrays that have the ability to observe the angle of arrival between elements and the ability to observe the time of arrival, respectively, to obtain the actual measurement data.

[0064] First, the 3D sound source localization system loads the known prior information of each linear array, including the nominal positions of the reference elements in the global coordinate system and the spatial attitude angles of the linear array itself. Then, it performs an observation capability diagnosis on each linear array in the 3D sound source localization system to determine whether the array can effectively estimate the inter-element directional angles of arrival and reliably measure the time of arrival in its current state. This determination is automatically made using preset hardware self-check flags, whether the real-time signal-to-noise ratio exceeds a set threshold, and statistical results of the validity of similar observation data from recent historical periods. After obtaining the determination result, the 3D sound source localization system performs directional data acquisition based on this result: it requests and receives measurement values ​​only from linear arrays determined to have the capability to observe inter-element directional angles of arrival; similarly, it acquires measurement values ​​only from linear arrays determined to have the capability to observe the time of arrival. Finally, all the acquired valid inter-element directional angles of arrival and time of arrival observations are aggregated to form the actual measurement data used for subsequent localization calculations.

[0065] For example, the nominal position of the linear array reference element in the global coordinate system is obtained. and the spatial orientation of the linear array itself Where M is the number of reference elements in the linear array. Subsequently, the three-dimensional sound source localization system binary-identifies the observation capabilities of each linear array, defining a vector representing the directional angle of arrival observation capabilities between elements. and arrival time observation capability identifier vector ,at the same time It also represents the total number of 1DAOA observations. This also represents the total number of TOA observations. For the i-th linear array, a value of 1 indicates it has the corresponding observation capability, and a value of 0 indicates it does not; i ranges from 1 to M. Based on this identifier, the 3D sound source localization system counts the number of linear arrays that can simultaneously observe 1DAOA and TOA. And the number of linear arrays that can only observe 1DAOA .

[0066] Furthermore, in one embodiment, based on the determination result, valid observation values ​​are collected from linear arrays with observation capabilities for inter-element directional arrival angles and arrival times, respectively, to obtain actual measurement data, including:

[0067] Based on the determination results, an observation capability identifier for the inter-element directional arrival angle and an observation capability identifier for the arrival time are generated for each linear array with observation capability. Valid observation values ​​are collected from the observation capability identifiers for the inter-element directional arrival angle and the arrival time to obtain the actual measurement data of the inter-element directional arrival angle and the actual measurement data of the arrival time for the corresponding linear array.

[0068] First, a digital identifier for the observation capability of each linear array is generated. This involves generating a corresponding binary observation capability identifier for the inter-element directional angle of arrival and an independent binary observation capability identifier for the time of arrival. Each identifier clearly records whether the linear array is effective for a specific observation type, forming a complete observation capability profile. Subsequently, directional data collection is performed based on this identifier system: only linear arrays marked as "effective" in the 1DAOA observation capability identifier are queried and read to obtain their inter-element directional angle of arrival measurements; similarly, only linear arrays marked as "effective" in the TOA observation capability identifier are collected for their time of arrival measurements.

[0069] The three-dimensional sound source localization system performs directional data acquisition and structured storage based on the inter-element directional angle of arrival observation capability identifier vector and the time of arrival observation capability identifier vector. Specifically, the three-dimensional sound source localization system only acquires and stores directional data based on the parameters that meet the requirements of the array elements. The linear array collects its 1DAOA observations, only from those that meet the requirements. The linear array acquires its TOA observations. For linear arrays where TOA cannot be observed (i.e.... ), and its corresponding distance-related variables These will be initialized as parameters to be estimated. All collected valid observations are arranged in order of category, constituting the actual measurement data used for subsequent positioning calculations, thus providing structured input for unified error modeling in the "incomplete observation" scenario.

[0070] In one embodiment, a unified hybrid parameter localization model for multiple error sources is constructed based on actual measurement data, including: determining the actual observation vector based on the actual measurement data; establishing the residual relationship between the actual observation vector and the vector to be estimated; wherein the vector to be estimated includes the sound source position, the square of the sound source position modulus, and variables of a portion of the linear array; constructing a linear combination of the direction arrival angle measurement noise, arrival time measurement noise, and linear array reference element position error; and obtaining the unified hybrid parameter localization model for multiple error sources based on the residual relationship and the linear combination.

[0071] Specifically, the residual relationship refers to the difference vector between the actual observation vector composed of the effective observations of each linear array and a vector to be estimated. The vector to be estimated is obtained based on the current sound source location, the squared distance to the sound source, and some specific variables of the linear array. Specifically, the vector to be estimated can be expressed as:

[0072] ;

[0073] in, This indicates the current location of the sound source. It is expressed as the square of the modulus at the sound source location. Let be the distance variable to be estimated for the j-th linear array for which only the angle of arrival between array elements can be observed, but the time of arrival cannot be observed. The total dimension of this vector is . , where N is the dimension of the sound source for localization.

[0074] After obtaining the residual vector, the key step is to perform causal decomposition. The reason for representing this residual relationship as a linear combination of three error sources is that, in practice, the deviation between observed and theoretical values ​​mainly stems from three physically independent and distinct types of errors: inter-element azimuth measurement noise (random error in azimuth estimation due to environmental interference and algorithm limitations); time-of-arrival measurement noise (random error in time delay estimation introduced by signal propagation distortion and clock synchronization problems); and linear array reference element position error (systematic deviation between the actual position of the array elements and the nominal value used, caused by installation deviations or platform deformation). Therefore, the final hybrid parameter positioning model is a mathematical model that explicitly expresses the above structured residuals as a linear superposition of these three types of errors.

[0075] In one embodiment, by combining and linearizing the observation equations for the inter-element directional angle of arrival and time of arrival, the unified matrix form observation equations for the 1DAOA-TOA hybrid positioning model are finally derived as follows:

[0076] ;

[0077] The formulas for each coefficient matrix are as follows:

[0078] ;

[0079] Where, the left side of the equation The observation residual vector is constructed from the actual measurements, h1 is the observation vector, and its sub-blocks h 1a and h 1t These correspond to the observation data for 1DAOA and TOA, respectively. G1 is the geometric Jacobian matrix, which augments the state vector. (Including sound source location, squared distance, and specific variables) linearly mapped to the observation space. G 1a and G 1t The theoretical impacts of state variables on angle and time delay observations are described separately. Therefore, the left side of the equation as a whole represents the gap between the actual observed data and the current model predictions. The right side of the equation structurally decomposes the total residual into a linear superposition of three independent error sources. Represents arrival time measurement noise The contribution of time noise is described by its coefficient matrix B1, which describes how time noise affects the observation residuals. noise in the direction of arrival between array elements . contributions. Represents the position error of the reference element of the linear array. The contribution of the error sources. When the difference between the left and right sides of the equation is less than a preset fixed threshold, the equation holds, and the observed residual vector is equal to the linear superposition of the error sources.

[0080] Specifically, in one embodiment, for the TOA observation equation, when The observation vector constructed from the actual measured values ​​is defined as follows:

[0081] ;

[0082] Where c is the speed of sound, t i For the measured arrival time, s i Let G1 be the reference element position of the linear array. Its geometric Jacobian matrix G1 is defined as:

[0083] ;

[0084] The negative gradient component is related to the element position, while the constant component is related to a specific variable. correspond.

[0085] For the 1DAOA observation equation, the elements of the observation vector constructed from actual measurements and the geometric Jacobian matrix differ depending on whether the linear array has TOA observation capability. Sometimes When, where k r This represents the linear array corresponding to having 1DAOA estimate but no TOA estimate:

[0086] ;

[0087] when hour:

[0088] ;

[0089] The other coefficient matrices are as follows:

[0090] ;

[0091] ;

[0092] ;

[0093] Where, n t n is the number of receivers with TOA measurement.a Number of receivers with 1DAOA measurement; N T (i) indicates that the i-th sensor has a TOA measurement, N a (i) indicates that the i-th sensor has 1DAOA measurement. The sub-block B1 in the coefficient matrix... 1a The structure is somewhat unique. Because the number of observations of the angle of arrival (TOA) between array elements is usually not equal to the number of observations of the time of arrival (TOA), and only when a linear array simultaneously possesses both observation capabilities will its TOA measurement noise pass through B. 1a The corresponding row in the equation affects the 1DAOA observation residual equation, which leads to B 1a It is a matrix with a specific sparse pattern or projection relationship, rather than a full matrix or a simple diagonal matrix. Similarly, the sub-blocks D in the D1 matrix... 1a With D 1t The position errors of the linear array reference elements are described respectively. The linear effects on 1DAOA and TOA residuals. Since azimuth observations primarily depend on the direction vector, while time-of-arrival (TOA) observations directly depend on the propagation distance, the geometric sensitivity of the same positional error differs for these two types of observations, leading to D... 1a With D 1t The corresponding coefficient values ​​are not entirely consistent. This difference is precisely preserved in the modeling to ensure accurate estimation of systematic errors.

[0094] In one embodiment, an optimization objective function for estimating the location of a sound source is constructed based on a hybrid parameter localization model, including: constructing a weighted least squares form optimization objective function based on the residual relationship expression defined by the hybrid parameter localization model.

[0095] Based on the 1DAOA-TOA hybrid parameter localization model, an optimization objective function for estimating the sound source location is constructed. First, using the structured residual relationship defined by the aforementioned observation equation as the core mathematical object, and based on the maximum likelihood estimation principle, assuming that the arrival time measurement noise, the inter-element direction angle of arrival measurement noise, and the position error of the linear array reference element all follow a joint zero-mean Gaussian distribution, the corresponding negative log-likelihood function is derived. This function ultimately manifests as a scalar optimization objective function in the form of weighted least squares, with the following specific form:

[0096] ;

[0097] in, To optimize the objective function, Under ideal error-free conditions (i.e., reference element position error) The predicted observation vector; It is the joint covariance matrix of observation noise and reference element position error, and its structure is a block diagonal matrix:

[0098] ;

[0099] in, Corresponding angle observation noise covariance, Corresponding time observation noise covariance. Here... Let be the incomplete observation vector, where t represents the direction of arrival angle observation, and t represents the arrival time observation.

[0100] Accordingly, the constructed optimization objective function can transform the complex problem of "incomplete observation" localization into a numerical optimization problem with clear statistical significance and operability.

[0101] In one embodiment, the moment estimator of the gradient of the objective function is calculated and biased to obtain the corrected moment estimator, including: calculating the first and second moments of the gradient of the objective function; introducing the corrected first moment estimator and the corrected second moment estimator, and performing biased correction on the first and second moments to obtain the corrected first moment estimator and second moment estimator.

[0102] In each iteration, the gradient vector of the objective function at the current estimated sound source location is calculated. Subsequently, the first and second moments of this gradient are calculated: the first moment is obtained by calculating the exponential moving average of the gradient vector, which is used to estimate the average direction (momentum) of the gradient to accelerate convergence and suppress oscillations; the second moment is obtained by calculating the exponential moving average of the squares of each gradient component, which is used to estimate the magnitude of gradient changes in each direction to achieve adaptive step size adjustment of parameters.

[0103] Because the initial bias in the estimation of the first and second moments exists in the early stages of iteration, the update step size is inaccurate. Therefore, this embodiment further introduces corrected first-moment and corrected second-moment estimates: by scaling the original first and second moments respectively, specifically by dividing them by a compensation factor related to the decay hyperparameter and the current iteration number, the underestimation bias caused by initialization is effectively eliminated. After this bias correction, the obtained corrected first-moment estimate can more realistically reflect the long-term average direction of the gradient, and the corrected second-moment estimate can more accurately characterize the scale of gradient changes in each dimension.

[0104] For example, in one embodiment, the gradient of the optimization objective function is calculated. First moment and second moment Where k is the number of iterations, ,right and Update the value:

[0105] ;

[0106] in, It is the gradient vector of the (k+1)th iteration, representing the objective function of optimizing the sound source position at the current location. Gradient estimation at [location]. , and To estimate the decay rate hyperparameter for the control moment, we typically take... , .in, It is the partial derivative matrix of the theoretical observation value with respect to the sound source position vector, reflecting the degree of influence of small changes in the sound source position on the theoretical angle prediction value.

[0107] Next, the corrected first-order moment estimator is introduced. and corrected second-order moment estimator To correct for the deviation of the moment estimate:

[0108] ;

[0109] in, and They represent and The (k+1)th power.

[0110] Furthermore, in one embodiment, the estimated value of the sound source location is iteratively updated based on the corrected moment estimate until the convergence condition is met, including: calculating the step size of the iterative update based on the corrected first-order moment estimate and the second-order moment estimate; updating the estimated value of the sound source location according to the step size of the iterative update and the corrected first-order moment estimate until the convergence condition is met; wherein, the convergence condition may be that the change in the estimated value of the sound source location in two adjacent iterations is less than a preset location change threshold.

[0111] Specifically, the adaptive iterative update step size is first calculated using the bias-corrected first-moment and second-moment estimates to construct a dynamically adjusted step size. This step size is calculated as the product of the corrected first-moment estimate and a small constant, divided by the sum of the square root of the corrected second-moment estimate and another small constant. This design allows the step size to adaptively change across different parameter dimensions. Larger step sizes are used to accelerate progress in dimensions with stable gradient directions (large first-moment) and small historical fluctuations (small second-moment); while in dimensions with uncertain gradient directions or drastic historical fluctuations, the step size is automatically reduced to maintain stability.

[0112] Subsequently, parameter updates are performed. The calculated adaptive step size is combined with the corrected first-order moment estimate to generate the complete update vector for this iteration. This vector is then used to correct the current source location estimate, thus obtaining the initial estimate for the next iteration. Based on the corrected moment estimate, the source location is updated:

[0113] ;

[0114] in and It is a very small constant, and in one embodiment it can be set to... It is 0.001. 10 -8 .

[0115] Finally, convergence is checked: After each update, the Euclidean distance between the newly obtained sound source location estimate and the previous iteration estimate is calculated. The change is compared with a preset location change threshold: if the change is less than the threshold, the algorithm is considered to have converged sufficiently, the iteration loop terminates, and the current sound source location estimate is the final localization result; if the change is still greater than or equal to the threshold, the updated estimate is used as the new starting point, and the algorithm returns to the first step to continue the next round of iteration calculations until the convergence condition is met or the preset maximum number of iterations is reached.

[0116] Figure 2 This is a simulation layout diagram of the positioning scene used in an embodiment of this application. Figure 2 As shown in the figure, a three-dimensional spatial coordinate system is constructed using x, y, and z, with the corresponding unit being meters. The x-axis and y-axis together define a horizontal reference plane, used to characterize the planar orientation and distance between the sound source and the linear array. The z-axis, perpendicular to the horizontal plane, represents the height or depth direction, used to indicate the vertical position of the sound source. Linear array reference elements with M=6 are labeled, and their corresponding positions are... to , , , , , , and the location of the sound source The spatial distribution of the data is shown, and the linear array observation attitude of 1DAOA is indicated by a thick black line. Combined with scene parameters, this figure assumes... and There are a total of 4 1DAOA observations and 4 TOA observations. Unless otherwise specified, the 1DAOA noise is assumed to be... TOA noise is The spatial layout and observation configuration required for verifying the sound source localization algorithm can be intuitively presented based on the linear array reference elements and the attitude of the 1DAOA observation linear array shown in the figure.

[0117] Figure 3 This is a performance comparison chart of an adaptive moment estimation algorithm according to one embodiment of this application with other algorithms under the influence of direction of arrival noise. Figure 3 As shown, the horizontal axis is... The vertical axis represents the noise intensity of 1DAOA, and the root mean square error (RMSE) represents the positioning error. RMSE is a commonly used indicator to measure the deviation between predicted and actual observations; it is the ratio of the square root of the number of observation points to the sum of the squares of the deviations between predicted and actual values. Figure 3 The circle "○" corresponds to the Gauss-Newton algorithm. "Corresponding to the adaptive moment estimation algorithm, the black solid line represents the Cramer-Rao lower bound, which is the theoretical minimum variance of the estimation error of any unbiased estimator under a data probability model that satisfies the regularity condition. Given the observation model and noise statistics, the variance of any unbiased estimator cannot fall below this lower bound. As 1DAOA noise increases, the positioning errors of both the Gauss-Newton algorithm and the adaptive moment estimation algorithm show an upward trend, and eventually both gradually approach the Cramer-Rao lower bound."

[0118] Figure 4 This is a performance comparison chart of an adaptive moment estimation algorithm according to one embodiment of this application and other algorithms under the influence of arrival time. Figure 4 As shown, the horizontal axis is... The vertical axis represents the noise intensity of TOA, and the root mean square error (RMSE) represents the positioning error. Figure 4 The circle "○" corresponds to the Gauss-Newton algorithm. "Corresponding to the adaptive moment estimation algorithm, the black solid line represents the Cramer-Rao lower bound. As TOA noise increases, the positioning errors of both the Gauss-Newton algorithm and the adaptive moment estimation algorithm show an upward trend. Under high TOA noise, the adaptive moment estimation algorithm has higher positioning accuracy. Therefore, in this embodiment, the adaptive moment estimation algorithm performs well in terms of noise resistance, providing an effective algorithm for underwater acoustic 3D positioning research under incomplete observation of mixed positioning parameters."

[0119] Figure 5 This is a structural block diagram of the sound source localization device 50 in the incomplete observation scenario of this embodiment, as shown below. Figure 5As shown, the sound source localization device 50 in the incomplete observation scenario includes: a data acquisition module 52, a function construction module 54, and an iterative update module 56; wherein: the data acquisition module 52 is used to acquire the actual measurement data collected by the linear array; the actual measurement data includes the actual measurement data of the angle of arrival between array elements and the actual measurement data of the arrival time; the function construction module 54 is used to construct a unified hybrid parameter localization model with multiple error sources based on the actual measurement data; wherein, the multiple error sources include the measurement noise of the angle of arrival between array elements, the measurement noise of the arrival time, and the position error of the reference array element of the linear array; the hybrid parameter localization model represents that the deviation between the actual value of the sound source position and the estimated value of the sound source position is equal to the sum of the linear combination of the multiple error sources; based on the hybrid parameter localization model, an optimization objective function for estimating the sound source position is constructed; the iterative update module 56 is used to calculate the moment estimate of the gradient of the optimization objective function and perform deviation correction to obtain the corrected moment estimate; based on the corrected moment estimate, the estimated value of the sound source position is iteratively updated until the preset convergence condition is met to obtain the final sound source position.

[0120] It should be noted that the above modules can be functional modules or program modules, and can be implemented through software or hardware. For modules implemented through hardware, the above modules can reside in the same processor; or the above modules can be located in different processors in any combination.

[0121] This embodiment also provides an electronic device including a memory and a processor, the memory storing a computer program and the processor being configured to run the computer program to perform the steps in any of the above method embodiments.

[0122] Optionally, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor and the input / output device is connected to the processor.

[0123] Optionally, in this embodiment, the processor can be configured to perform the following steps via a computer program:

[0124] S1, acquire the actual measurement data collected by the linear array; the actual measurement data includes the actual measurement data of the angle of arrival between array elements and the actual measurement data of the time of arrival;

[0125] S2. Based on actual measurement data, a unified hybrid parameter localization model with multiple error sources is constructed. Among them, the multiple error sources include the noise of the angle of arrival measurement between array elements, the noise of the time of arrival measurement, and the position error of the linear array reference element. The hybrid parameter localization model represents that the deviation between the actual value of the sound source position and the estimated value of the sound source position is equal to the sum of the linear combination of the multiple error sources.

[0126] S3, based on the hybrid parameter localization model, constructs an optimized objective function for estimating the location of the sound source;

[0127] S4. Calculate the moment estimator of the gradient of the objective function and perform bias correction to obtain the corrected moment estimator.

[0128] S5. Based on the corrected moment estimator, iteratively update the estimated value of the sound source location until the preset convergence condition is met, and obtain the final sound source location.

[0129] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementations, and will not be repeated in this embodiment.

[0130] Furthermore, in conjunction with the sound source localization method for incomplete observation scenarios provided in the above embodiments, this embodiment can also provide a storage medium for implementation. This storage medium stores a computer program; when executed by a processor, the computer program implements any of the sound source localization methods for incomplete observation scenarios described in the above embodiments.

[0131] It should be understood that the specific embodiments described herein are merely illustrative of the application and not intended to limit it. All other embodiments derived by those skilled in the art based on the embodiments provided in this application without inventive effort are within the scope of protection of this application.

[0132] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0133] Obviously, the accompanying drawings are merely some examples or embodiments of this application. Those skilled in the art can apply this application to other similar situations based on these drawings without any creative effort. Furthermore, it is understood that although the work done in this development process may be complex and lengthy, for those skilled in the art, certain design, manufacturing, or production modifications made based on the technical content disclosed in this application are merely conventional technical means and should not be considered as insufficient disclosure of this application.

[0134] The term "embodiment" in this application refers to a specific feature, structure, or characteristic described in connection with an embodiment that may be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily imply the same embodiment, nor does it imply that it is mutually exclusive with or independent of other embodiments. It will be clearly or implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0135] The above embodiments merely illustrate several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of patent protection. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the appended claims.

Claims

1. A method for sound source localization in an incomplete observation scenario, characterized in that, include: Acquire the actual measurement data collected by the linear array; The actual measurement data includes the actual measurement data of the angle of arrival between array elements and the actual measurement data of the time of arrival; Based on the actual measurement data, a unified hybrid parameter localization model with multiple error sources is constructed; wherein, the multiple error sources include inter-element direction angle of arrival measurement noise, arrival time measurement noise, and linear array reference element position error; the hybrid parameter localization model characterizes that the deviation between the actual value of the sound source position and the estimated value of the sound source position is equal to the sum of the linear combination of the multiple error sources; Based on the hybrid parameter localization model, an optimization objective function is constructed for estimating the location of the sound source; Calculate the moment estimator of the gradient of the optimization objective function, and perform bias correction to obtain the corrected moment estimator; Based on the corrected moment estimate, the estimated value of the sound source location is iteratively updated until the preset convergence condition is met, and the final sound source location is obtained.

2. The sound source localization method in an incomplete observation scenario according to claim 1, characterized in that, The acquisition of actual measurement data collected by the linear array includes: Obtain the position of the linear array reference element and the linear array orientation; For each of the linear arrays, determine whether the linear array has the ability to observe the angle of arrival between the array elements and whether it has the ability to observe the time of arrival, and obtain the determination result; Based on the determination result, valid observation values ​​are collected from linear arrays that have the ability to observe the arrival angle between array elements and the ability to observe the arrival time, respectively, to obtain the actual measurement data.

3. The sound source localization method in an incomplete observation scenario according to claim 2, characterized in that, Based on the determination result, valid observation values ​​are collected from linear arrays that have the ability to observe the inter-element directional arrival angle and the arrival time, respectively, to obtain the actual measurement data, including: Based on the determination result, an observation capability identifier for the inter-element direction arrival angle and an observation capability identifier for the arrival time are generated for each linear array with observation capability. Valid observation values ​​are collected from the observation capability identifiers of the inter-element directional arrival angle and the arrival time to obtain the actual measurement data of the inter-element directional arrival angle and the actual measurement data of the arrival time corresponding to the linear array.

4. The sound source localization method in an incomplete observation scenario according to claim 1, characterized in that, The construction of a unified hybrid parameter positioning model based on the actual measurement data includes: Determine the actual observation value vector based on the actual measurement data; Establish the residual relationship between the actual observation vector and the vector to be estimated; wherein, the vector to be estimated includes the sound source location, the square of the sound source location modulus, and variables of a portion of the linear array; Construct a linear combination of the inter-element directional arrival angle measurement noise, the arrival time measurement noise, and the linear array reference element position error; Based on the residual relationship and the linear combination, the unified hybrid parameter localization model for multiple error sources is obtained.

5. The sound source localization method in an incomplete observation scenario according to claim 4, characterized in that, The step of constructing an optimization objective function for estimating the sound source location based on the hybrid parameter localization model includes: Based on the residual relationship expression defined by the hybrid parameter localization model, a weighted least squares form optimization objective function is constructed.

6. The sound source localization method in an incomplete observation scenario according to claim 1, characterized in that, The calculation of the moment estimate of the gradient of the objective function and the subsequent bias correction to obtain the corrected moment estimate include: Calculate the first and second moments of the gradient of the objective function. By introducing corrected first-order moment estimators and corrected second-order moment estimators, deviation corrections are applied to the first-order moment and the second-order moment to obtain corrected first-order moment estimators and second-order moment estimators.

7. The sound source localization method in an incomplete observation scenario according to claim 6, characterized in that, The step of iteratively updating the estimated value of the sound source location based on the corrected moment estimate until the convergence condition is met includes: Based on the corrected first-order moment estimate and second-order moment estimate, the step size for iterative updates is calculated; The estimated value of the sound source location is updated according to the step size of the iterative update and the corrected first-order moment estimate until the convergence condition is met; wherein, the convergence condition means that the change in the estimated value of the sound source location in two adjacent iterations is less than a preset position change threshold.

8. A sound source localization device for incomplete observation scenarios, characterized in that, include: The module comprises a data acquisition module, a function construction module, and an iterative update module; among which: The data acquisition module is used to acquire the actual measurement data collected by the linear array; the actual measurement data includes the actual measurement data of the angle of arrival between array elements and the actual measurement data of the time of arrival. The function construction module is used to construct a unified hybrid parameter localization model with multiple error sources based on the actual measurement data; wherein, the multiple error sources include inter-element direction angle of arrival measurement noise, time of arrival measurement noise, and linear array reference element position error; the hybrid parameter localization model characterizes that the deviation between the actual value of the sound source position and the estimated value of the sound source position is equal to the sum of the linear combination of the multiple error sources; based on the hybrid parameter localization model, an optimization objective function for estimating the sound source position is constructed; The iterative update module is used to calculate the moment estimate of the gradient of the optimization objective function and perform bias correction to obtain the corrected moment estimate; based on the corrected moment estimate, iteratively update the estimated value of the sound source location until the preset convergence condition is met to obtain the final sound source location.

9. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the sound source localization method in the incomplete observation scenario according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the sound source localization method in the incomplete observation scenario as described in any one of claims 1 to 7.

Images