TDOA localization method for moving sound sources based on improved neural dynamics solution

Abstract

The present invention discloses a TDOA positioning method for a mobile sound source based on improved neural dynamics solution, which relates to the technical field of TDOA positioning, and includes the following steps: S1, constructing a trajectory equation of an acoustic signal source, and constructing an acoustic sensor coordinate equation based on the trajectory equation; S2, converting the acoustic sensor coordinate equation into a linear equation; S3, constructing an NPGZNN model; S4, constructing an MZND discrete model based on the NPGZNN model; S5, solving the linear equation according to the MZND discrete model to obtain the position of a mobile target. The present invention has the advantages of high precision and low latency, and the MZND model achieves global convergence under different conditions, and is superior to existing solutions in terms of convergence speed, precision and robustness.

Description

Technical Field

[0001] The present invention relates to the field of TDOA positioning technology, and in particular to a TDOA positioning method for a moving sound source based on improved neural dynamics solution. Background Art

[0002] In TDOA positioning estimation, the time difference between the acoustic emission source signal reaching different acoustic emission sensors is mainly used for positioning. If multiple receivers are located on a straight line, there are many optimization processing methods. However, if the receivers are randomly distributed in space, the situation is more complicated, and nonlinear problems will be encountered when solving the hyperbolic equations. Some literatures give an exact solution when the number of measurement parameters is the same as the number of source signal coordinates. However, when the number of measurement parameters is redundant, this method cannot fully utilize the statistical information given by the redundant measurement parameters to improve the positioning accuracy. For the case of redundant measurement parameters, relevant literatures give some closed solutions, but these solutions are not optimal. There are also literatures that give an iterative algorithm using Fourier series. This iteration requires a good initial value, otherwise it is easy to fall into the local minimum point, and convergence cannot be guaranteed. There is often no effective global peak, converges to several local maximum values, and is extremely sensitive to the initial search value. In order to solve this problem, the prior art literature proposes a two-step weighted LS method. When the measurement parameter error is very small, the performance is close to the optimal value, but this method introduces the square term of the measurement parameter. When the measurement error is large, the quadratic term of the noise cannot be ignored, and its performance will deteriorate. At the same time, the increase in computing costs and the huge interference of noise make it difficult to solve the problem in real time. Summary of the invention

[0003] In order to solve the above problems, the present invention proposes a TDOA positioning method for a moving sound source based on an improved neural dynamics solution.

[0004] The technical solution of the present invention is: a TDOA positioning method for a moving sound source based on an improved neural dynamics solution comprises the following steps:

[0005] S1, constructing a trajectory equation of the acoustic signal source, and constructing a coordinate equation of the acoustic sensor based on the trajectory equation;

[0006] S2, converting the coordinate equation of the acoustic sensor into a linear equation;

[0007] S3, construct NPGZNN model;

[0008] S4. Based on the NPGZNN model, the MZND discrete model is constructed;

[0009] S5. According to the MZND discrete model, the linear equation is solved to obtain the moving target position.

[0010] Furthermore, in S1, the trajectory equation of the acoustic signal source is expressed as:

[0011] ;

[0012] In the formula, is the theoretical trajectory of the sound source, t is the time value, is a real matrix with three rows and one column.

[0013] Furthermore, in S1, the method for constructing the coordinate equation of the acoustic sensor is: assigning values ​​to the coordinates of each sensor for receiving the acoustic signal, and determining the coordinate equation of the acoustic sensor.

[0014] Furthermore, the expression for assigning values ​​to the coordinates of each sensor for receiving the acoustic signal is:

[0015] ;

[0016] In the formula, are the coordinates of n acoustic sensors, is the x-axis coordinate of the first sub-sensor, is the y-axis coordinate of the first sub-sensor, is the z-axis coordinate of the first sub-sensor, is the x-axis coordinate of the second sub-sensor, is the y-axis coordinate of the second sub-sensor, is the z-axis coordinate of the second sub-sensor, is the x-axis coordinate of the nth sub-sensor, is the y-axis coordinate of the nth sub-sensor, is the z-axis coordinate of the nth sub-sensor, is a real number matrix with three rows and one column of the nth sub-sensor.

[0017] Furthermore, the coordinate equation of the acoustic sensor is expressed as:

[0018] ;

[0019] In the formula, is the x-axis coordinate of the first sub-sensor, is the z-axis coordinate of the first sub-sensor, is the x-axis coordinate of the second sub-sensor, is the y-axis coordinate of the second sub-sensor, is the z-axis coordinate of the second sub-sensor, is the x-axis coordinate of the nth sub-sensor, is the y-axis coordinate of the nth sub-sensor, is the z-axis coordinate of the nth sub-sensor, is the calculation parameter related to the distance from the second sensor to the origin of the virtual three-dimensional coordinate system, is the calculation parameter related to the distance from the nth sensor to the origin of the virtual three-dimensional coordinate system, Indicates the distance between the sound source and the main sensor, Indicates the distance difference between the sound source and the second sensor and the main sensor. Indicates the distance difference between the sound source and the nth sensor and the main sensor. is the x-axis coordinate of the moving target, is the y-axis coordinate of the moving target, is the z-axis coordinate of the moving target.

[0020] Furthermore, in S2, the expression of the linear equation is:

[0021] ;

[0022] In the formula, is the first calculation formula, is the second calculation formula, is the third calculation formula, and t is the time value.

[0023] for , for , for .

[0024] Furthermore, in S3, the expression of the NPGZNN model is:

[0025] ;

[0026] In the formula, is the first-order derivative of the error function of the linear equation, is the integrand of the time value, is the scale parameter that controls the convergence speed, is the first-order derivative of the energy function, Integral coefficient to improve the stability of the model, t is the time value, is the substitution for the time value, Represents the integral of the substitution of time values.

[0027] Furthermore, in S4, the expression of the MZND discrete model is:

[0028] ;

[0029] In the formula, for hour The value of for hour The value of for hour The value of for hour The value of for hour The first derivative of for hour The first derivative of for hour The first derivative of is the time value replacement, t is the time value, is the x-axis coordinate of the moving target, and k is an integer.

[0030] Furthermore, in S5, the expression for solving the linear equation is:

[0031] ;

[0032] In the formula, for hour The value of for hour The value of for hour The value of for hour The value of for hour The first derivative of for hour The first derivative of for hour The first derivative of for hour The value of for hour The value of for hour The value of for hour The generalized inverse matrix of for hour The generalized inverse matrix of for hour The generalized inverse matrix of for hour The first derivative of for hour The first derivative of for hour The first derivative of for hour The value of for hour The value of for hour The value of for hour The first derivative of for hour The first derivative of for hour The first derivative of is the transpose of a vector or matrix, To improve the integration coefficient of the model stability, is the scale parameter that controls the convergence speed, is the time value replacement, t is the time value, is the x-axis coordinate of the moving target, k is an integer, is the first calculation formula, is the second calculation formula, is the third calculation formula, is the first calculation formula for the i-th sensor, is the second calculation formula for the i-th sensor, is the third calculation formula for the i-th sensor.

[0033] The beneficial effects of the present invention are as follows: the TDOA positioning of mobile sound sources based on improved neural dynamics solution proposed in the present invention adds an integral term in the design of the MZND model, so that the model has better robustness, and at the same time adopts an explicit linear three-step algorithm for discretization, so that the model has faster convergence speed and higher accuracy. In solving nonlinear optimization problems with time-varying equation constraints, especially in the time-varying noise environment in the TDOA scenario, the convergence characteristics of the MZND model are improved while maintaining a high convergence accuracy. It has the advantages of high accuracy and low latency. The MZND model achieves global convergence under different conditions, and is superior to existing solutions in terms of convergence speed, accuracy and robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is a flow chart of the TDOA localization method for moving sound sources based on improved neural dynamics solution;

[0035] Figure 2 It is the logarithmic graph of convergence accuracy of different models under noise-free conditions;

[0036] Figure 3 The logarithmic graph of convergence accuracy of different models under the condition of smaller step size and no noise;

[0037] Figure 4 It is the logarithmic graph of convergence accuracy of different models under constant noise conditions;

[0038] Figure 5 It is the logarithmic graph of convergence accuracy of different models under random noise conditions;

[0039] Figure 6 It is the logarithmic graph of convergence accuracy of different models under time-varying linear noise conditions;

[0040] Figure 7 It is the logarithmic graph of convergence accuracy of the MZND model at different step sizes under noise-free conditions;

[0041] Figure 8 It is a schematic diagram of the trajectory comparison between the target path and the calculated path and the position coordinates of the sensor when the TDOA method is used to predict the displacement of the sound source position;

[0042] Fig. 9 A schematic diagram showing the comparison of the x-axis error, y-axis error and z-axis error in the process of localizing a moving sound source;

[0043] Fig.10 Error plot when solving the TDOA problem for the MZND model. DETAILED DESCRIPTION

[0044] The embodiments of the present invention will be further described below in conjunction with the accompanying drawings.

[0045] like Figure 1 As shown, the present invention provides a TDOA positioning method for a moving sound source based on an improved neural dynamics solution, comprising the following steps:

[0046] S1, constructing a trajectory equation of the acoustic signal source, and constructing a coordinate equation of the acoustic sensor based on the trajectory equation;

[0047] S2, converting the coordinate equation of the acoustic sensor into a linear equation;

[0048] S3, construct NPGZNN model;

[0049] S4. Based on the NPGZNN model, the MZND discrete model is constructed;

[0050] S5. According to the MZND discrete model, the linear equation is solved to obtain the moving target position.

[0051] In the embodiment of the present invention, in S1, the trajectory equation of the acoustic signal source is expressed as:

[0052] ;

[0053] In the formula, is the theoretical trajectory of the sound source, t is the time value, is a real matrix with three rows and one column.

[0054] In the embodiment of the present invention, in S1, the method for constructing the coordinate equation of the acoustic sensor is: assigning values ​​to the coordinates of each sensor for receiving the acoustic signal, and determining the coordinate equation of the acoustic sensor.

[0055] In the embodiment of the present invention, the expression for assigning values ​​to the coordinates of each sensor for receiving the acoustic signal is:

[0056] ;

[0057] In the formula, are the coordinates of n acoustic sensors, is the x-axis coordinate of the first sub-sensor, is the y-axis coordinate of the first sub-sensor, is the z-axis coordinate of the first sub-sensor, is the x-axis coordinate of the second sub-sensor, is the y-axis coordinate of the second sub-sensor, is the z-axis coordinate of the second sub-sensor, is the x-axis coordinate of the nth sub-sensor, is the y-axis coordinate of the nth sub-sensor, is the z-axis coordinate of the nth sub-sensor, is a real number matrix with three rows and one column of the nth sub-sensor.

[0058] In the embodiment of the present invention, the coordinate equation of the acoustic sensor is expressed as:

[0059] ;

[0060] In the formula, is the x-axis coordinate of the first sub-sensor, is the z-axis coordinate of the first sub-sensor, is the x-axis coordinate of the second sub-sensor, is the y-axis coordinate of the second sub-sensor, is the z-axis coordinate of the second sub-sensor, is the x-axis coordinate of the nth sub-sensor, is the y-axis coordinate of the nth sub-sensor, is the z-axis coordinate of the nth sub-sensor, is the calculation parameter related to the distance from the second sensor to the origin of the virtual three-dimensional coordinate system, is the calculation parameter related to the distance from the nth sensor to the origin of the virtual three-dimensional coordinate system, Indicates the distance between the sound source and the main sensor, Indicates the distance difference between the sound source and the second sensor and the main sensor. Indicates the distance difference between the sound source and the nth sensor and the main sensor. is the x-axis coordinate of the moving target, is the y-axis coordinate of the moving target, is the z-axis coordinate of the moving target.

[0061] Calculate using the following formula:

[0062] ;

[0063] in, , is the distance from the main sensor to the origin of the virtual 3D coordinate system, is the distance from the ith sensor to the origin of the virtual three-dimensional coordinate system, Represents the distance difference between the sound source from the i-th sensor and the main sensor.

[0064] In the embodiment of the present invention, in S2, the linear equation is expressed as:

[0065] ;

[0066] In the formula, is the first calculation formula, is the second calculation formula, is the third calculation formula, and t is the time value.

[0067] for , for , for .

[0068] In the embodiment of the present invention, in S3, the expression of the NPGZNN model is:

[0069] ;

[0070] In the formula, is the first-order derivative of the error function of the linear equation, is the integrand of the time value, is the scale parameter that controls the convergence speed, is the first-order derivative of the energy function, Integral coefficient to improve the stability of the model, t is the time value, is the substitution for the time value, Represents the integral of the substitution of time values.

[0071] In the embodiment of the present invention, in S4, the expression of the MZND discrete model is:

[0072] ;

[0073] In the formula, for hour The value of for hour The value of for hour The value of for hour The value of for hour The first derivative of for hour The first derivative of for hour The first derivative of is the time value replacement, t is the time value, is the x-axis coordinate of the moving target, and k is an integer.

[0074] In the embodiment of the present invention, in S5, the expression for solving the linear equation is:

[0075] ;

[0076] In the formula, for hour The value of for hour The value of for hour The value of for hour The value of for hour The first derivative of for hour The first derivative of for hour The first derivative of for hour The value of for hour The value of for hour The value of for hour The generalized inverse matrix of for hour The generalized inverse matrix of for hour The generalized inverse matrix of for hour The first derivative of for hour The first derivative of for hour The first derivative of for hour The value of for hour The value of for hour The value of for hour The first derivative of for hour The first derivative of for hour The first derivative of is the transpose of a vector or matrix, To improve the integration coefficient of the model stability, is the scale parameter that controls the convergence speed, is the time value replacement, t is the time value, is the x-axis coordinate of the moving target, k is an integer, is the first calculation formula, is the second calculation formula, is the third calculation formula, is the first calculation formula for the i-th sensor, is the second calculation formula for the i-th sensor, is the third calculation formula for the i-th sensor. for hour The discretization of for hour The discretization of for hour Discretization of . When 0, 1, or 2, the Euler forward difference algorithm is used for discretization: By taking the first-order derivative of the error, i.e. the gradient, x, y, and z are moved in the direction of the negative gradient, and the result finally reached in this direction is the final position of the moving target.

[0077] For the nonlinear optimization planning problem with time-varying equality constraints, the unified form of the nonlinear optimization problem with time-varying equality constraints is defined as:

[0078] ;

[0079] Among them, the objective function for A second-order differentiable nonlinear convex function when , is a time-varying decision variable, is a non-negative time variable, is a time-varying constraint.

[0080] Considering the need to solve this problem, the Lagrangian operator is introduced and the Lagrangian function is defined as:

[0081] ;

[0082] in, is the Lagrange multiplier vector, Represents the transpose of a vector or matrix.

[0083] Considering the Lagrangian solution method, the solution to the above problem can be expressed as the solution of the following system of equations:

[0084] ;

[0085] Transform the above equations into another form:

[0086] ;

[0087] in, .

[0088] According to the error equation, the original ZND model is defined as:

[0089] ;

[0090] In the formula express of and the derivative of t, is a scale parameter that controls the convergence speed, and the appropriate Make the model perform better.

[0091] The above formula is further changed to:

[0092] ;

[0093] Combining the above two equations, we get the following equation:

[0094] ;

[0095] in, express about The partial derivative of express Partial derivatives with respect to t.

[0096] Furthermore, the energy function is introduced and is set as:

[0097] ;

[0098] Therefore, the GND model can be derived as follows:

[0099] ;

[0100] Then, the ZND model is combined with the GND model to obtain the GZNN model as follows:

[0101] .

[0102] The embodiment of the present application constructs NPGZNN based on the aforementioned original ZND model and GND model. An integral term is added to the constructed NPGZND to enhance the robustness of the model and introduce an integral feedback term. , the original model is modified to:

[0103] ;

[0104] in, is the integral term used to suppress noise, represents the integral coefficient that improves the stability of the model, Indicates time The exchange of express 's points.

[0105] The NPGZNN model is discretized using an explicit linear three-step algorithm to obtain a discrete MZND model. The explicit linear three-step algorithm is:

[0106] ;

[0107] in, 3, 4, 5, .

[0108] when When , the Euler forward difference algorithm is used as follows:

[0109] ;

[0110] in, , represents the step length, , , Indicates the step length.

[0111] Based on the above process, the coordinate equation of the acoustic sensor is converted into a linear equation, which specifically includes:

[0112] Will Expressed as ,Will Expressed as ,Will Expressed as , so the resulting linear equation is: , the error function formed is: in Represents the error function of the preceding linear equation.

[0113] Therefore, in the TDOA positioning problem, the NPGZNN model is first used to reconstruct the error function of the linear equation as follows:

[0114] ;

[0115] in, The first derivative of The first derivative of yes The first derivative of .

[0116] Similarly, the NPGZNN model is used to solve the linear equation and the solution of the target position is obtained as:

[0117] ;

[0118] in, express The generalized inverse matrix of .

[0119] Then, an explicit linear three-step algorithm is used to discretize it and obtain the MZND model.

[0120] In order to verify the effect of the present invention, the present invention further provides a process of designing and executing a verification experiment. In the verification process, the MZND model proposed by the present invention is compared with other existing models, and the quadratic programming of the time-varying equality constraints calculated by the MZND model at different step sizes is given. Finally, the experimental simulation results of the MZND model in solving the TDOA problem are given, and the results show the effectiveness and practicality of the model in solving the TDOA problem.

[0121] The MZND model proposed in this invention is compared with other existing models, such as DTGZNN, DTGNN, DTZNN, FTRNN, EVPRNN, VPCDNN, FIFD-K and FIFD-U models. Figure 2-Figure 6 First, Figure 2 and Figure 3 Together, we show the residual accuracy of different models with different step lengths under noise-free conditions. Figure 2 It can be seen intuitively that the MZND model converges faster than other models and reaches , up to , 10 times faster than the FTRNN model, and faster than the DTZNN and EVPRNN models times faster than FIFD-K and FIFD-U models times faster than DTGZNN, DTGNN and VPCDNN models In terms of accuracy, the proposed model achieves approximately The FTRNN model has reached about The DTZNN and EVPRNN models achieved approximately The FIFD-K and FIFD-U models have reached about The DTGZNN, DTGNN and VPCDNN models have reached about Step, Figure 3 Comparison chart of the accuracy of various models with smaller step sizes.

[0122] Figure 4 and Figure 5 The residual accuracy of different models under constant noise conditions is shown. Compared with the noise-free condition, the convergence speed of the MZND model is basically the same as that under noise-free conditions, about The other models have a certain degree of divergence and their convergence speed is not as fast as that of the MZND model. At the same time, the convergence accuracy of the MZND model is higher than that of the other models. Therefore, under constant noise conditions, the MZND model has better performance than other models.

[0123] Figure 6 The residual accuracy of different models under time-varying linear noise is shown. Figure 6 , it is found that due to the influence of time-varying noise, the convergence accuracy of other models is significantly degraded. It is worth noting that the MZND model still converges to order. Therefore, the MZND model has a denoising effect. In addition, the convergence speed of the MZND model is still faster than that of other models, which is the same as the case under constant noise conditions. In short, the performance of the MZND model is more competitive than other advanced solutions.

[0124] Figure 7 The residual accuracy of the MZND model under noise-free and different step-size conditions is shown.

[0125] The experimental simulation results of using the MZND model to solve the TDOA problem are as follows: Figure 8-Figure 10 As shown. Figure 8 It can be seen from that the target trajectory is highly consistent with the actual trajectory. No matter what the initial value is, the final trajectory will tend to the target trajectory. Then, from Fig. 9 In the figure, we can observe the errors of the x-axis, y-axis, and z-axis, and the system error converges to zero quickly within 0.3s and remains globally stable. Fig.10 It can be seen that the MZND model converges quickly when solving the TDOA problem. It can be seen that the MZND model proposed in the present invention has the advantages of high accuracy and low delay in solving the TDOA problem.

[0126] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A TDOA positioning method for moving sound sources based on improved neural dynamics solution, characterized in that: The following steps are involved: S1, constructing a trajectory equation of the acoustic signal source, and constructing a coordinate equation of the acoustic sensor based on the trajectory equation; S2, converting the coordinate equation of the acoustic sensor into a linear equation; S3, construct NPGZNN model; S4. Based on the NPGZNN model, the MZND discrete model is constructed; S5. Solve the linear equation according to the MZND discrete model to obtain the moving target position; In S1, the trajectory equation of the acoustic signal source is expressed as: ； In the formula, is the theoretical trajectory of the sound source, t is the time value, is a real matrix with three rows and one column; The expression of the acoustic sensor coordinate equation is: ； In the formula, is the x-axis coordinate of the first sub-sensor, is the z-axis coordinate of the first sub-sensor, is the x-axis coordinate of the second sub-sensor, is the y-axis coordinate of the second sub-sensor, is the z-axis coordinate of the second sub-sensor, is the x-axis coordinate of the nth sub-sensor, is the y-axis coordinate of the nth sub-sensor, is the z-axis coordinate of the nth sub-sensor, is the calculation parameter related to the distance from the second sensor to the origin of the virtual three-dimensional coordinate system, is the calculation parameter related to the distance from the nth sensor to the origin of the virtual three-dimensional coordinate system, Indicates the distance between the sound source and the main sensor, Indicates the distance difference between the sound source and the second sensor and the main sensor. Indicates the distance difference between the sound source and the nth sensor and the main sensor. is the x-axis coordinate of the moving target, is the y-axis coordinate of the moving target, is the z-axis coordinate of the moving target; In S2, the linear equation is expressed as: ； In the formula, is the first calculation formula, is the second calculation formula, is the third calculation formula, t is the time value; In S3, the expression of the NPGZNN model is: ； In the formula, is the first derivative of the error function of the linear equation, is the integrand of the time value, is the scale parameter that controls the convergence speed, is the first-order derivative of the energy function, Integral coefficient to improve the stability of the model, t is the time value, is a substitution for the time value, The integral of the substitution representing the time value; In S4, the expression of the MZND discrete model is: ； In the formula, for hour The value of for hour The value of for hour The value of for hour The value of for hour The first derivative of for hour The first derivative of for hour The first derivative of is the time value replacement, t is the time value, is the x-axis coordinate of the moving target, k is an integer; In S5, the expression for solving the linear equation is: ； In the formula, for hour The value of for hour The value of for hour The value of for hour The value of for hour The first derivative of for hour The first derivative of for hour The first derivative of for hour The value of for hour The value of for hour The value of for hour The generalized inverse matrix of for hour The generalized inverse matrix of for hour The generalized inverse matrix of for hour The first derivative of for hour The first derivative of for hour The first derivative of for hour The value of for hour The value of for hour The value of for hour The first derivative of for hour The first derivative of for hour The first derivative of is the transpose of a vector or matrix, To improve the integration coefficient of the model stability, is the scale parameter that controls the convergence speed, is the time value replacement, t is the time value, is the x-axis coordinate of the moving target, k is an integer, is the first calculation formula, is the second calculation formula, is the third calculation formula, is the first calculation formula for the i-th sensor, is the second calculation formula for the i-th sensor, is the third calculation formula for the i-th sensor.

2. The mobile sound source TDOA positioning method based on improved neural dynamics solution according to claim 1 is characterized in that: In S1, the method for constructing the coordinate equation of the acoustic sensor is: assigning values ​​to the coordinates of each sensor for receiving the acoustic signal, and determining the coordinate equation of the acoustic sensor.

3. The mobile sound source TDOA positioning method based on improved neural dynamics solution according to claim 2 is characterized in that: The expression for assigning values ​​to the coordinates of each sensor for receiving the acoustic signal is: ； In the formula, are the coordinates of n acoustic sensors, is the x-axis coordinate of the first sub-sensor, is the y-axis coordinate of the first sub-sensor, is the z-axis coordinate of the first sub-sensor, is the x-axis coordinate of the second sub-sensor, is the y-axis coordinate of the second sub-sensor, is the z-axis coordinate of the second sub-sensor, is the x-axis coordinate of the nth sub-sensor, is the y-axis coordinate of the nth sub-sensor, is the z-axis coordinate of the nth sub-sensor, is a real number matrix with three rows and one column of the nth sub-sensor.

Images