Acoustic vector sensor network sound source tracking method based on node selection

Abstract

The invention belongs to the technical field of sound source tracking, and discloses an acoustic vector sensor network sound source tracking method based on node selection, and the method comprises the following specific steps: 1, setting N AVS nodes and one sound source in a two-dimensional plane, and forming a node network; 2, K AVS nodes are selected from the node network, DOA and TDOA in the N AVS nodes are fused, and a selection vector w is determined; and 3, continuously updating a selection vector w for the selected AVS node by the EKF model to obtain an observation vector # imgabs0 #, using the observation vector # imgabs1 # as the input of the EKF model, determining a system state Pk by updating parameters in the EKF model, and determining a sound source position through the system state Pk. Through the above mode, the method combines the characteristic that the DOA can be estimated by a single AVS and a node selection technology, so as to achieve the balance between the system cost and the tracking precision, achieve the more efficient and more accurate sound source tracking, reduce the energy consumption, and optimize the network performance. In this way.

Description

Technical Field

[0001] The invention relates to the technical field of sound source tracking, and in particular to a sound source tracking method of an acoustic vector sensor network based on node selection. Background Art

[0002] Sound source tracking technology is one of the important branches of acoustic signal processing. It estimates the trajectory information of the sound source by receiving the acoustic signal. It has wide applications in human-computer interaction, video conferencing, military and other fields. With the rapid development of sensors and intelligent voice technology, traditional acoustic sensor networks are widely used in sound source tracking tasks. Traditional acoustic sensor networks are composed of multiple sensor nodes and have the advantages of strong flexibility and high spatial coverage.

[0003] Acoustic vector sensor (AVS) has gradually become a hot topic in acoustic sensor research because it can obtain more sound field information and adapt to complex scenarios. AV consists of an omnidirectional sound pressure sensor and an orthogonal particle velocity sensor. A single AVS can estimate the direction of arrival (DOA) of the sound source, which is a feature that a single traditional acoustic sensor does not have. Compared with networks based on traditional acoustic sensors, networks based on AVS can capture more sound field information and achieve higher positioning accuracy. However, when there are many nodes in the network, it often leads to problems such as increased communication pressure and energy consumption of the network.

[0004] To address the above problems, node selection technology has been introduced. By selecting some sensor nodes for task processing, it can maintain high positioning accuracy and effectively reduce the energy consumption and communication burden of the network. Existing node selection technology usually takes minimizing the Cramer-Rao lower bound (CRLB) as the optimization goal, and transforms the node selection problem into an integer programming problem with Boolean vectors as optimization variables. In the literature [1], the author proposed a node selection method based on time difference of arrival (TDOA), which completed the node selection task by using semi-positive relaxation. However, the above method only uses traditional acoustic sensor networks for traditional sensor network methods, and only uses TDOA single information, failing to give full play to the potential of other information in the sound field. If the node selection technology is applied to the AVS network, not only TDOA information can be used, but also DOA information can be combined to achieve higher accuracy sound source tracking. However, how to balance tracking performance and system cost through node selection technology in AVS networks is still an important problem to be solved.

[0005] [1] Dai Z, Wang G, Jin X, et al. Nearly optimal sensor selection for TDOA-based source localization in wireless sensor networks[J]. IEEETransactions on Vehicular Technology, 2020, 69(10): 12031-12042. Based on this, the present invention designs a sound source tracking method of an acoustic vector sensor network based on node selection to solve the above problem. Summary of the invention

[0006] In view of the above-mentioned shortcomings of the prior art, the present invention provides a method for tracking sound sources in an acoustic vector sensor network based on node selection.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for tracking a sound source in an acoustic vector sensor network based on node selection, the specific steps are as follows: Step 1: Set up in 2D N AVS nodes and 1 sound source form a node network; Step 2: Select from the node network K AVS nodes, integrated N The DOA and TDOA in each AVS node determine the selection vector w ; Step 3: The EKF model continuously updates the selection vector for the selected AVS node w , and obtain the observation vector , using the observation vector As input to the EKF, the system state is determined by updating the parameters in the EKF model. P k , through the system status P k Locate the source of the sound.

[0008] Furthermore, the coordinates of the i-th AVS in the node network are s i =[ x i y i ], i = 1 , 2 , … , N , the estimated coordinates of the sound source are u =[xy ].

[0009] Furthermore, the specific steps of step 2 are as follows: The specific steps of step 2 are as follows: Step 21: Select in the node network K AVS nodes, the selection vector is set to w .

[0010] (1) Select Vector w middle, w i ∈{0,1} represents the selection state of the i-th node, w i =0 means the i-th node is not selected. w i =1 means the i-th node is selected; A reference node is set in the TDOA information, and the first node selected is used as the reference node. w The first non-zero element in is set to -1, and a new selection vector is constructed. ; For the new selection vector , which means as follows:

[0011] represents the selection state of the i-th node, Indicates that the i-th node is not selected, and when the i-th node is selected, Indicates that it is a reference node, otherwise ; Step 22: According to Construction of TDOA N×(K-1) Dimension selection matrix , according to the selection vector w Constructing DOA N×(K-1) Dimension selection matrix ,pass and Determine the complete selection matrix ; Step 23: N AVS nodes determine the Jacobian matrix H ,according to and H Sure ; Step 24: Right To impose restraints; Step 25: Transform the constrained results; Step 26: Introduce auxiliary variables and constraints into the transformed structure and use the CVX toolbox to solve the selection vector w .

[0012] Further, the complete selection matrix The calculation is as follows: First, use the new selection vector Build , the specific steps are: for the matrix diag( ), set all elements of the row corresponding to its reference node to -1, and then delete all columns corresponding to the reference node and unselected nodes. diag{·} means to construct a diagonal matrix with the specified elements; With the selection vector w The following relations are satisfied: (3) (4) Among them, I K-1 express (K-1)×(K-1) The identity matrix ，1 K-1 express (K-1)× 1's all-1 vector;

[0013] .

[0014] Furthermore, H The matrix is H DOA and H TDOA Composition, in TDOA Medium, speed of sound c is a constant, the Jacobian matrix H is calculated as follows: (7) (8) (9) (10) Among them, the distance measurement noise is .

[0015] Furthermore, The specific calculation is as follows: (11); (12) (13) (14) (15) The DOA measurement noise is , and for and The variance of .

[0016] Furthermore, The specific calculation of the constraints is as follows: ; Here, tr(·) means taking the trace of the matrix.

[0017] Going further, transform the specific operation: first introduce a positive definite matrix Q 0, making Q = Q 0+ αI N ,in α is a positive scalar, Q = Q 0+ αI N Substituting the objective function into formula (16) is:

[0018] Formula (16) can be transformed into: (19) C = H T Q 0 -1 H , B = Q 0 -1 H ; Using the matrix inversion lemma for formula (20), we can obtain: (20) C -1 is a constant, and formula (20) is transformed into: (twenty one).

[0019] Furthermore, the specific operation of step 26 is: first introduce a 2×2 matrix , The following conditions are met: (twenty two) According to Schur complement, formula (22) is changed into the following matrix inequality: (twenty three) Next, we introduce another auxiliary variable W=ww T ,at this time: (twenty four) Then transform (24) into: ; Wherein, Diag(·) is the diagonal element extraction symbol.

[0020] Furthermore, firstly, by linearizing the nonlinear state transfer function and the nonlinear observation function, the system state P k And covariance for prediction; then through the Kalman gain K k Update, using the updated Kalman gain K k and observation information System status P k and covariance update, and then enter the next time step processing. Through continuous parameter iteration, the system state P k The estimation and update of the system status P k Locate the source of the sound.

[0021] Beneficial Effects The present invention combines the characteristic of a single AVS that can estimate DOA with the node selection technology to achieve a balance between system cost and tracking accuracy, so as to achieve more efficient and accurate sound source tracking while reducing energy consumption and optimizing network performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0023] Figure 1 This is a flow chart of sound source tracking based on node selection in the present invention.

[0024] Figure 2 Schematic diagram of the positions of nodes and sound sources at the initial moment in the present invention.

[0025] Figure 3 This is the influence of the DOA measurement standard deviation on the tracking accuracy under different numbers of nodes in the present invention.

[0026] Figure 4 This is the influence of the distance measurement standard deviation on the tracking accuracy under different methods in the present invention. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] The present invention will be further described below in conjunction with the embodiments.

[0029] Embodiment 1 Please refer to the instruction manual Figure 1-2 , a sound source tracking method of an acoustic vector sensor network based on node selection, the specific steps are as follows: Step 1: Set up in 2D N AVS nodes and 1 sound source form a node network; The coordinates of the i-th AVS in the node network are s i =[ x i y i ], i = 1 , 2 , … , N , the estimated coordinates of the sound source are u =[ xy ] Step 2: Select from the node network K AVS nodes, integrated N The DOA and TDOA in each AVS node determine the selection vector w ; The specific steps of step 2 are as follows: The specific steps of step 2 are as follows: Step 21: Select in the node network K AVS nodes, the selection vector is set to w .

[0030] (1) Select Vector w middle,w i ∈{0,1} represents the selection state of the i-th node, w i =0 means the i-th node is not selected. w i =1 means the i-th node is selected; A reference node is set in the TDOA information, and the first node selected is used as the reference node. w The first non-zero element in is set to -1, and a new selection vector is constructed. ; For the new selection vector , which means as follows: (2) represents the selection state of the i-th node, Indicates that the i-th node is not selected, and when the i-th node is selected, Indicates that it is a reference node, otherwise ; Step 22: According to Construction of TDOA N×(K-1) Dimension selection matrix , according to the selection vector w Constructing DOA N×(K-1) Dimension selection matrix ,pass and Determine the complete selection matrix ; Step 23: N AVS nodes determine the Jacobian matrix H ,according to and H Sure ; Step 24: Right To impose restraints; Step 25: Transform the constrained results; Step 26: Introduce auxiliary variables and constraints into the transformed structure and use the CVX toolbox to solve the selection vector w .

[0031] Complete selection matrix The calculation is as follows: First, use the new selection vector Build , the specific steps are: for the matrix diag( ), set all elements of the row corresponding to its reference node to -1, and then delete all columns corresponding to the reference node and unselected nodes. diag{·} means to construct a diagonal matrix with the specified elements; With the selection vector w The following relations are satisfied: (3) (4) Among them, I K-1 express (K-1)×(K-1) The identity matrix ，1 K-1 express (K-1)× 1's all-1 vector;

[0032] .

[0033] H The matrix is H DOA and H TDOA Composition, in TDOA Medium, speed of sound c is a constant, the Jacobian matrix H is calculated as follows: (7) (8) (9) (10) Among them, the distance measurement noise is .

[0034] The specific calculation is as follows: (11); (12) (13) (14) (15) The DOA measurement noise is , and for and The variance of .

[0035] The specific calculation of the constraints is as follows: ;

[0036] Here, tr(·) means taking the trace of the matrix.

[0037] Conversion specific operation: first introduce a positive definite matrix Q 0, making Q = Q 0+ αI N ,in α is a positive scalar, Q = Q 0+ αI N Substituting the objective function into formula (16) is:

[0038] Formula (16) can be transformed into: (19) C = H T Q 0 -1 H , B = Q 0 -1 H ; Using the matrix inversion lemma for formula (20), we can obtain: (20) C -1 is a constant, and formula (20) is transformed into: (twenty one).

[0039] Step 26 Specific operation: First introduce a 2×2 matrix , The following conditions are met: (twenty two) According to Schur complement, formula (22) is changed into the following matrix inequality: (twenty three) Next, we introduce another auxiliary variable W=ww T ,at this time: (twenty four) Then transform (24) into:

[0040] Wherein, Diag(·) is the diagonal element extraction symbol.

[0041] Step 3: The EKF model continuously updates the selection vector for the selected AVS node w , and obtain the observation vector , using the observation vector As input to the EKF, the system state is determined by updating the parameters in the EKF model. P k , through the system status P k Locate the source of the sound.

[0042] First, the system state is linearized by linearizing the nonlinear state transfer function and the nonlinear observation function. P k And covariance for prediction; then through the Kalman gain K k Update, using the updated Kalman gain K k and observation information System status P k and covariance update, and then enter the next time step processing. Through continuous parameter iteration, the system state P k The estimation and update of the system status P k Locate the source of the sound.

[0043] Figure 3 The influence of the DOA measurement standard deviation on the tracking accuracy under different numbers of nodes in the present invention is evaluated by the root mean square error (RMSE / m). K As the DOA measurement standard deviation increases, the tracking accuracy gradually decreases.

[0044] Figure 4 This is the influence of the distance measurement standard deviation on the tracking accuracy under different methods in the present invention.

[0045] (a) K=5 , (b) K=8 , (c) K=10 , we can see that as the measurement standard deviation increases, the tracking accuracy gradually decreases. As the measurement standard deviation increases, the method in the literature [1] that only uses TDOA single information is weaker than the random selection method. Among the methods that use two types of information, the method in this patent maintains the best tracking effect.

[0046] The present invention combines the characteristic of a single AVS that can estimate DOA with the node selection technology to achieve a balance between system cost and tracking accuracy, so as to achieve more efficient and accurate sound source tracking while reducing energy consumption and optimizing network performance.

[0047] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for tracking sound sources in an acoustic vector sensor network based on node selection, characterized in that: The specific steps are as follows: Step 1: Set up in 2D N AVS nodes and 1 sound source form a node network; Step 2: Select from the node network K AVS nodes, integrated N The DOA and TDOA in each AVS node determine the selection vector w ; Step 3: The EKF model continuously updates the selection vector for the selected AVS node w , and obtain the observation vector , using the observation vector As input to the EKF, the system state is determined by updating the parameters in the EKF model. P k , through the system status P k Locate the source of the sound.

2. The method for tracking sound source of acoustic vector sensor network based on node selection according to claim 1, characterized in that: The coordinates of the i-th AVS in the node network are s i =[ x i y i ], i = 1 , 2 , … , N , the estimated coordinates of the sound source are u =[ xy ].

3. The method for tracking sound source of acoustic vector sensor network based on node selection according to claim 2, characterized in that: The specific steps of step 2 are as follows: The specific steps of step 2 are as follows: Step 21: Select in the node network K AVS nodes, the selection vector is set to w； (1) Select Vector w middle, w i ∈{0,1} represents the selection state of the i-th node, w i =0 means the i-th node is not selected. w i =1 means the i-th node is selected; A reference node is set in the TDOA information, and the first node selected is used as the reference node. w The first non-zero element in is set to -1, and a new selection vector is constructed. ; For the new selection vector , which means as follows: ； represents the selection state of the i-th node, Indicates that the i-th node is not selected, and when the i-th node is selected, Indicates that it is a reference node, otherwise ; Step 22: According to Construction of TDOA N×(K-1) Dimension selection matrix , according to the selection vector w Constructing DOA N ×(K-1) Dimension selection matrix ,pass and Determine the complete selection matrix ; Step 23: N AVS nodes determine the Jacobian matrix H ,according to and H Sure ; Step 24: Right To impose restraints; Step 25: Transform the constrained results; Step 26: Introduce auxiliary variables and constraints into the transformed structure and use the CVX toolbox to solve the selection vector w .

4. The method for tracking sound source of acoustic vector sensor network based on node selection according to claim 3 is characterized in that: Complete selection matrix The calculation is as follows: First, use the new selection vector Build , the specific steps are: for the matrix diag( ), set all elements of the row corresponding to its reference node to -1, and then delete all columns corresponding to the reference node and unselected nodes. diag{·} means to construct a diagonal matrix with the specified elements; With the selection vector w The following relations are satisfied: （3） （4） Among them, I K-1 express (K-1)×(K-1) The identity matrix ，1 K-1 express (K-1)× 1's all-1 vector; （5） （6）。 5. The method for tracking sound source in an acoustic vector sensor network based on node selection according to claim 4, characterized in that: H The matrix is H DOA and H TDOA Composition, in TDOA Medium, speed of sound c is a constant, the Jacobian matrix H is calculated as follows: （7） （8） （9） （10） Among them, the distance measurement noise is .

6. The method for tracking sound source of acoustic vector sensor network based on node selection according to claim 5, characterized in that: The specific calculation is as follows: （11）； （12） （13） （14） （15） The DOA measurement noise is , and for and The variance of .

7. The method for tracking sound source of acoustic vector sensor network based on node selection according to claim 6, characterized in that: The specific calculation of the constraints is as follows: ; Here, tr(·) means taking the trace of the matrix.

8. The method for tracking sound source of acoustic vector sensor network based on node selection according to claim 7, characterized in that: Conversion specific operation: first introduce a positive definite matrix Q 0, making Q = Q 0+ αI N ,in α is a positive scalar, Q = Q 0+ αI N Substituting the objective function into formula (16) is: ； Formula (16) can be transformed into: （19） C = H T Q 0 -1 H , B = Q 0 -1 H ; Using the matrix inversion lemma for formula (20), we can obtain: （20） C -1 is a constant, and formula (20) is transformed into: （21）。 9. The method for tracking sound source of acoustic vector sensor network based on node selection according to claim 8, characterized in that: Step 26 Specific operation: First introduce a 2×2 matrix , The following conditions are met: （22） According to Schur complement, formula (22) is changed into the following matrix inequality: （23） Next, we introduce another auxiliary variable W=ww T ,at this time: （24） Then transform (24) into: ; Wherein, Diag(·) is the diagonal element extraction symbol.

10. The method for tracking sound source in an acoustic vector sensor network based on node selection according to claim 9, characterized in that: First, the system state is linearized by linearizing the nonlinear state transfer function and the nonlinear observation function. P k and covariance for prediction; Then through the Kalman gain K k Update, using the updated Kalman gain K k and observation information System status P k and covariance update, and then enter the next time step processing. Through continuous parameter iteration, the system state P k The estimation and update of the system status P k Locate the source of the sound.

Images