A TDOA sound source localization method based on AZTND model

Abstract

This invention discloses a TDOA sound source localization method based on the AZTND model. This method, which belongs to the field of system localization, aims to address the shortcomings of traditional technologies in noise suppression, real-time performance, and positioning accuracy. The model optimizes performance by constructing a dynamic matrix equation and combining a piecewise adaptive coefficient function with a negative feedback regulation mechanism. The piecewise adaptive function dynamically adjusts the gain coefficient in stages to achieve rapid convergence, stable tracking, and steady-state maintenance. The activation function selectively adjusts the error direction and, in conjunction with the integral term, suppresses noise interference. Experiments demonstrate that this method converges to a positioning error of 10-4 meters in three-dimensional space and maintains an accuracy of 10-1 meters under time-varying noise. This method combines high real-time performance, strong noise immunity, and low hardware dependence, making it suitable for scenarios such as intelligent monitoring and robotic navigation in dynamic environments.

Description

Technical Field

[0001] The present invention relates to the field of system positioning, and in particular to a TDOA sound source positioning method based on an AZTND model. Background Art

[0002] Sound source localization technology determines the location of a sound source by analyzing the time difference between the arrival of a sound at multiple sensors. It is widely used in intelligent surveillance, robotic navigation, drones, and speech enhancement. In intelligent surveillance, sound source localization can be used to detect and track specific sound events. In robotic navigation, it helps robots identify sound sources in their surroundings and make appropriate navigation decisions. In speech enhancement, sound source localization helps improve speech recognition accuracy and call quality. The TDOA method used is a localization method that measures the time difference between a sound source and different sensors. The basic principle is that the sound emitted by a sound source arrives at different sensors at different times. These time difference data can be used to estimate the source's location using mathematical algorithms and signal processing techniques. The TDOA method has attracted much attention for its high accuracy and real-time performance, making it particularly suitable for applications requiring rapid response. Sound source localization generally uses TDOA or phase differences to establish geometric relationships to calculate the source's location. Traditional sound source localization methods primarily rely on time delay estimation and geometric localization algorithms, such as generalized cross correlation (GCC) and its improved algorithm (GCC-PHAT). These methods meet application requirements to a certain extent, but they have significant limitations in terms of noise interference, real-time performance, and localization accuracy. Especially in dynamic environments, when the location of the sound source changes rapidly, existing algorithms may not be able to update the positioning results in a timely and accurate manner. In recent years, neural dynamic models (such as annihilation neural networks, ZNN) have been applied to sound source localization, and the sound source location is solved in real time by dynamically adjusting the network state. However, existing models still have shortcomings in noise interference, real-time performance, and positioning accuracy, making it difficult to meet the needs of practical applications. The present invention aims to address these shortcomings of the prior art and proposes a TDOA sound source localization method based on the activation-type annihilation neural dynamics (AZTND) model. By simultaneously introducing segmented adaptive coefficients and negative feedback regulation, the positioning accuracy and noise resistance are significantly improved, while the hardware cost is reduced. It is suitable for sound source localization in dynamic environments.

[0003] Traditional sound source localization methods primarily rely on time delay estimation and geometric positioning algorithms. While these methods meet application requirements to a certain extent, they suffer from significant deficiencies in noise interference, real-time performance, and positioning accuracy. Especially in dynamic environments, where the location of a sound source changes rapidly, existing algorithms may not be able to accurately update positioning results in a timely manner. Furthermore, their high dependence on hardware devices limits their widespread adoption in practical applications. With advances in sensor technology and the development of neural dynamics algorithms, neural network-based sound source localization methods have gradually become a research hotspot. Summary of the Invention

[0004] In view of the above-mentioned deficiencies in the prior art, the present invention provides a TDOA sound source localization method based on the AZTND model.

[0005] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:

[0006] 1. A TDOA sound source localization method based on the AZTND model, comprising the following steps:

[0007] S1. Construct TDOA dynamic equations based on TDOA sound source localization problem;

[0008] S2. Convert the dynamic equations into linear matrix equations, establish the error function of the TDOA sound source localization problem, and establish an annihilation neural dynamics model based on the error function. Optimize the constructed annihilation neural dynamics model using the piecewise adaptive coefficient function, activation function, and error integral term to obtain an activated annihilation neural dynamics model.

[0009] S3. Initialize the initial position of the sound source, solve the linear matrix equation using the activation-type zeroing neural dynamics model, and compare the actual trajectory of the sound source with the simulated trajectory to generate a residual comparison graph;

[0010] S4. Based on the solution, the specific location of the sound source is obtained, and a calculation trajectory of the sound source movement is generated.

[0011] Furthermore, the TDOA dynamic equations in S1 are expressed as:

[0012]

[0013] Where, is the three-dimensional space coordinate of the sensor that receives the sound source signal, represents the three-dimensional spatial coordinates of the main sensor that receives the sound source signal, n is the total number of sensors that receive the sound source signal, is the distance from the moving target to the main sensor that receives the sound source signal, is the distance difference between the sensor receiving the sound source signal and the main sensor, They are The three-dimensional spatial coordinates of the sound source at time , is the intermediate amount, , .

[0014] Furthermore, the piecewise adaptive function in S2 is expressed as:

[0015] ;

[0016] Where, To adjust the gain in stages, is the error function, is the norm of the error function, For time.

[0017] Furthermore, the activation function in S2 is expressed as:

[0018] ;

[0019] Where, is the activation function of the error function.

[0020] Furthermore, the activation-type zeroing neural dynamics model is expressed as:

[0021]

[0022] Where, To adjust the gain in stages, is the error function, is the norm of the error function, For time, is the activation function of the error function, is a design parameter greater than 0, is the error integral term.

[0023] The present invention has the following beneficial effects:

[0024] Compared to existing TDOA sound source localization technology, the present invention demonstrates significant improvements and advantages in several key areas. First, in terms of noise suppression, the present invention introduces a piecewise adaptive coefficient and a negative feedback adjustment mechanism to effectively reduce the impact of noise on sound source localization accuracy. This design enables the system to maintain high localization accuracy even in noisy environments, whereas traditional methods often suffer from positioning errors in the presence of noise. Second, in terms of real-time performance, the present invention's activation-based annihilation neural dynamics (AZTND) model rapidly responds to changes in sound source position and updates localization results in real time, which is particularly important for sound source localization in dynamic environments. In contrast, traditional methods often experience significant lag when processing rapidly changing sound source positions, making them unable to provide accurate localization information in a timely manner. Finally, in terms of localization accuracy, the present invention significantly improves localization accuracy and stability through an optimized neural network structure and dynamic adjustment mechanism. This not only enhances the overall performance of sound source localization but also reduces the reliance on complex hardware, thereby reducing system cost and implementation difficulty. In summary, the advantages of the present invention in noise suppression, real-time performance, and positioning accuracy give it a wider application prospect in the fields of intelligent monitoring, robot navigation, speech enhancement, etc., and can better meet the high requirements of sound source localization technology in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 The figure is a flow chart of the TDOA sound source localization method based on the AZTND model of the present invention.

[0026] Figure 2 This is a three-dimensional comparison diagram of the sound source calculated trajectory obtained by the AZTND model and the actual trajectory of the sound source in an embodiment of the present invention.

[0027] Figure 3 This is a polar coordinate comparison diagram of the sound source calculated trajectory obtained by the AZTND model and the actual trajectory of the sound source in an embodiment of the present invention.

[0028] Figure 4 This is a graph showing the variation of error over time in three-dimensional space according to an embodiment of the present invention.

[0029] Figure 5 Schematic diagram of the noise resistance of the TDOA sound source localization method based on the AZTND model in an embodiment of the present invention. DETAILED DESCRIPTION

[0030] The specific embodiments of the present invention are described below to facilitate understanding of the present invention by those skilled in the art. However, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the appended claims, these changes are obvious, and all inventions and creations utilizing the concepts of the present invention are protected.

[0031] A TDOA sound source localization method based on the AZTND model, such as Figure 1 As shown, the following steps are included:

[0032] S1. Construct TDOA dynamic equations based on TDOA sound source localization problem;

[0033] According to the three-dimensional TDOA sound source localization problem, this embodiment first determines that A sensor that receives the sound source signal, and the three-dimensional spatial coordinates of the sensor are To express. When , it indicates that it is the main sensor, that is, represents the three-dimensional spatial coordinates of the main sensor. The square of the Euclidean distance of each sub-sensor is expressed as The time difference between the arrival time of each sub-sensor and the main sensor is recorded as To calculate the distance difference between each sub-sensor and the main sensor, this paper uses the time difference The speed of sound The product of is represented by .

[0034] It is worth noting that the information provided establishes the basic principles and equations of the TDOA method. This paper considers a simplified scenario with five acoustic sensors. To more fully understand and apply the TDOA method with different numbers of sensors, the TDOA sound source localization problem can be expressed as the following dynamic equations:

[0035] in is the distance from the moving target to the main sensor, . Transform its dynamic equations into simple linear matrix equations .

[0036] S2. Convert the dynamic equations into linear matrix equations, establish the error function of the TDOA sound source localization problem, and establish an annihilation neural dynamics model based on the error function. Optimize the constructed annihilation neural dynamics model using the piecewise adaptive coefficient function, activation function, and error integral term to obtain an activated annihilation neural dynamics model.

[0037] Build the error function for the problem , establish a zeroing neural dynamics model based on the error function , where. The performance of the model is improved by designing a piecewise adaptive coefficient function. The expression of the piecewise adaptive coefficient function is as follows , where the design parameters .

[0038] This embodiment designs an activation function suitable for this function to improve the adaptive ability of the model. The activation function is expressed as follows .

[0039] Finally, we get a zeroed neural dynamics model with adaptive capabilities. .

[0040] In response to external noise disturbances, the model introduces the integral term of the error from the perspective of control theory To achieve negative feedback regulation, thus improving the noise resistance of the model. Therefore, the activation-type zeroing neural dynamics model (AZTND) is further obtained, and the expression is as follows .in, is a design parameter greater than 0.

[0041] S3. Initialize the initial position of the sound source, solve the linear matrix equation using the activation-type zeroing neural dynamics model, and compare the actual trajectory of the sound source with the simulated trajectory to generate a residual comparison graph;

[0042] Initialize and adjust the sound source position. In this embodiment, the initial position of the sound source is set to a random value, and the actual three-dimensional trajectory of the sound source is given as The actual trajectory of the sound source will be compared with the calculated trajectory calculated by the model, and a residual comparison graph will be generated.

[0043] S4. Based on the solution, the specific location of the sound source is obtained, and a calculation trajectory of the sound source movement is generated.

[0044] The proposed activation-type zeroing neural dynamics model (AZTND) is used to solve the sound source position and initialize all parameters: This example first substituted the matrix generated from the sound source localization problem into the proposed Activation-Type Zeroing Neural Dynamics (AZTND) model. This matrix contains key data such as the sensor's coordinates and the distance difference between the sound source and the sensor. These data are elements of the linear matrix equation obtained by transforming the dynamic equations constructed from the TDOA sound source localization problem.

[0045] The AZTND model equations are then solved using an ODE solver. During the solution process, the piecewise adaptive coefficient function in the AZTND model dynamically adjusts the gain based on the norm of the error function to achieve rapid convergence, stable tracking, and steady-state maintenance. Simultaneously, the activation function selectively adjusts the error direction and, in conjunction with the error integral term, suppresses noise interference, enabling accurate sound source location determination even in complex noisy environments. Specifically, when the norm of the error function is large, the piecewise adaptive coefficient function increases the gain, enabling rapid model convergence; when the norm of the error function is small, the gain gradually decreases to maintain the model's steady state. The activation function adjusts based on the positive and negative direction of the error to ensure that the error is reduced in the correct direction, while the error integral term accumulates the error, further enhancing the model's noise immunity.

[0046] Through the above-mentioned detailed solution process, the specific location information of the sound source is finally obtained, and the calculation trajectory of the sound source movement is generated based on this.

[0047] First, the matrix generated by the sound source localization problem is substituted into the proposed AZTND model, and the corresponding solution is generated by the ode solver, and finally the calculation trajectory of the sound source movement is generated. The sound source localization results obtained by the proposed AZTND model are as follows Figure 2 shown.

[0048] exist Figure 2In the figure, the blue circle represents the location of the sound sensor, the black solid line represents the actual trajectory of the moving target, and the red dashed line with diamonds represents the calculated trajectory of the moving target using the AZTND model. As can be seen from the figure, in three dimensions, the actual trajectory of the moving target and the calculated trajectory using the AZTND model are basically consistent, indicating that the TDOA sound source localization method based on the AZTND model has high three-dimensional positioning accuracy.

[0049] exist Figure 3 In the figure, the blue circle represents the location of the sound sensor, the black solid line represents the actual trajectory of the moving target, and the red dashed line with diamonds represents the calculated trajectory of the moving target using the AZTND model. As can be seen from the figure, in two dimensions, the actual trajectory of the moving target and the calculated trajectory using the AZTND model are basically consistent, indicating that the TDOA sound source localization method based on the AZTND model has high two-dimensional spatial positioning accuracy.

[0050] exist Figure 4 In the figure, the pink solid line is the error on the x-axis, the green solid line is the error on the y-axis, and the blue dashed line is the error on the z-axis. This figure shows the errors of a moving target on the x, y, and z-axes. As can be seen, as the moving target moves in three-dimensional space, the TDOA sound source localization method based on the AZTND model converges the system error in just over a hundred milliseconds. Furthermore, the positioning errors in all three dimensions converge and remain at the 10^-4 meter level, demonstrating that the TDOA sound source localization method based on the AZTND model has extremely fast convergence speed and high positioning accuracy.

[0051] exist Figure 5 In this example, noise interference is applied to the TDOA sound source localization method based on the AZTND model to verify the method's noise resistance. The blue solid line represents the case without noise interference, the pink solid line represents the case with constant noise interference, and the green solid line represents the case with time-varying noise interference. As can be seen from the figure, the system error converges rapidly to zero in both the case with and without noise interference, and the positioning accuracy reaches 10^−3 in both the case with and without noise interference. Although the system's positioning accuracy is slightly inferior in the case of time-varying noise, the TDOA sound source localization method based on the AZTND model can achieve a positioning accuracy of 10^−1 under strong noise interference, fully demonstrating the strong noise resistance of the TDOA sound source localization method based on the AZTND model in solving the problem of dynamic sound source localization.

[0052] Specific embodiments are used in the present invention to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core ideas. At the same time, for those skilled in the art, according to the ideas of the present invention, there may be changes in the specific implementation methods and application scopes. In summary, the contents of this specification should not be understood as limiting the present invention.

[0053] 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 it should be understood that the scope of protection of the present invention is not limited to such specific descriptions and embodiments. Those skilled in the art can make various other specific variations and combinations based on the technical teachings disclosed in the present invention without departing from the essence of the present invention, and such variations and combinations are still within the scope of protection of the present invention.

Claims

1. A TDOA sound source localization method based on the AZTND model, characterized in that: The steps include: S1. Construct TDOA dynamic equations based on TDOA sound source localization problem; S2. Convert the dynamic equations into linear matrix equations, establish the error function of the TDOA sound source localization problem, and establish the annihilation neural dynamics model based on the error function. Optimize the constructed annihilation neural dynamics model using the piecewise adaptive coefficient function, activation function, and error integral term to obtain the activated annihilation neural dynamics model, namely the AZTND model, where the piecewise adaptive coefficient function is expressed as: ； Where, To adjust the gain in stages; is the error function; is the norm of the error function, For time, the piecewise adaptive coefficient function dynamically adjusts the gain according to the norm of the error function. Specifically, when the norm of the error function is large, the piecewise adaptive coefficient function will increase the gain to make the model converge quickly; when the norm of the error function is small, the gain will gradually decrease to maintain the steady state of the model; The activation function is expressed as: ； Where, It is the activation function of the error function. The activation function is adjusted according to the positive and negative directions of the error to ensure that the error can be reduced in the right direction. The activation-type zeroing neural dynamics model is expressed as: ； Where, is a design parameter greater than 0, is the error integral term, which accumulates the error and further enhances the model's anti-noise ability; S3. Initialize the initial position of the sound source, solve the linear matrix equation using the activation-type zeroing neural dynamics model, and compare the actual trajectory of the sound source with the simulated trajectory to generate a residual comparison graph; S4. Based on the solution, the specific location of the sound source is obtained, and a calculation trajectory of the sound source movement is generated.

2. The TDOA sound source localization method based on the AZTND model according to claim 1, characterized in that: The TDOA dynamic equations in S1 are expressed as: ； Where, is the three-dimensional spatial coordinate of the sub-sensor that receives the sound source signal, represents the three-dimensional spatial coordinates of the main sensor that receives the sound source signal, n is the total number of sensors that receive the sound source signal, is the distance from the moving target to the main sensor that receives the sound source signal, is the distance difference between the sub-sensor receiving the sound source signal and the main sensor, They are The three-dimensional spatial coordinates of the sound source at time , is the intermediate amount, , is the Euclidean distance between each sub-sensor.

Images