A multi-channel active noise control method, system, terminal and storage medium

Abstract

The application provides a kind of multi-channel active noise control method, system, terminal and storage medium, and noise control method includes the following steps: establishing distributed wave domain ANC network model;The global cost function of wave domain ANC problem is split into local cost function and the form of the sum, and a distributed optimization problem is established;Multi-channel distributed wave domain ANC problem is solved using diffusion LMS algorithm, to obtain the estimation of loudspeaker weight;Secondary sound field is generated and residual noise signal is cancelled out.The application also provides a kind of multi-channel active noise control system, terminal and computer readable storage medium.The application uses diffusion strategy to obtain multi-channel distributed wave domain ANC algorithm, so that the calculation burden is dispersed among nodes, compared with centralized scheme, the application makes the expansibility of wave domain ANC system stronger, more robust, more suitable for large-scale application occasions.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of signal processing, and particularly relates to a multi-channel active noise control method, system, terminal and storage medium. BACKGROUND

[0002] Active noise control (ANC) is an electro-acoustic technology based on the principle of superposition, and a secondary sound source generates a signal with the same amplitude but opposite phase to cancel the unwanted noise (primary sound source) in a certain area, so as to achieve the purpose of noise reduction.

[0003] In recent years, active noise control in aircraft and vehicles has attracted extensive attention from researchers, and such applications usually require noise cancellation in a large area, while traditional multi-channel ANC algorithms can only reduce noise at multiple observation points, but the consistency in continuous spatial areas is not high. At present, scholars have proved that the wave domain (spatial Fourier transform) active noise control can perform noise reduction tasks in a larger spatial area, which minimizes the sum of the square harmonic coefficients (related to the energy of the entire control area) to achieve noise reduction in continuous space. In addition, since the noise field is usually time-varying and unknown, it is necessary to use an adaptive algorithm to iteratively calculate the secondary sound source driving signal to generate a secondary sound field. Most existing solutions need to collect error signals at all microphones for centralized processing, but this centralized strategy has high computational complexity and heavy communication burden, making it difficult to implement large-scale deployment. In order to solve this problem, some scholars have proposed a distributed solution to the single-channel wave domain ANC problem based on the diffusion LMS adaptive strategy, which uses a wireless acoustic sensor network to implement a wave domain ANC system, and each node in the network is equipped with only one microphone and one speaker. In fact, in the centralized wave domain ANC algorithm, the number of microphones and speakers can be different. Therefore, designing a generalized scenario with multiple microphones and speakers is crucial to perfecting the distributed wave domain ANC algorithm based on the diffusion strategy. In addition, each node in the distributed ANC network needs to be equipped with a processor, and if a single-channel node network is used, a large number of processors are needed, which may cause waste of resources. SUMMARY

[0004] The purpose of the present application is to overcome the above-mentioned deficiencies, and to provide a multi-channel active noise control method, system, terminal and storage medium, which realizes multi-channel distributed wave domain ANC based on the diffusion LMS strategy, fills the gap of distributed optimization in multi-channel spatial active noise control applications, and achieves similar performance to centralized methods in terms of noise reduction level in both free field and reverberation field environments, while dispersing the computational and communication burden to each node, greatly improving the robustness, scalability and practicality of the wave domain ANC system.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A multi-channel active noise control method comprises the following steps:

[0007] Establish a multi-channel distributed wave-domain ANC network model;

[0008] The global cost function of the multi-channel distributed wave-domain ANC network model is split into the sum of local cost functions to establish a distributed optimization model.

[0009] The distributed optimization model is solved using the diffuse LMS algorithm to obtain an estimate of the loudspeaker weights.

[0010] The secondary sound field generated by the multi-channel distributed wave-domain ANC network model is offset from the residual noise signal.

[0011] The multi-channel distributed wave domain ANC network model is a network model of N-node distributed wave domain ANC connected by topology;

[0012] The specific method for establishing a multi-channel distributed wave-domain ANC network model is as follows:

[0013] Step 1.1: Define a circular control area to complete the noise elimination task. The radius of the area is set to R1, and the noise source is set outside this area.

[0014] Step 1.2: Arrange a uniform circular microphone array with N elements on the boundary of the control area, and evenly arrange N loudspeakers on a circular ring with a radius of R2 outside the area to generate a secondary sound field;

[0015] Step 1.3: Establish a distributed network of N nodes, where each node k has L k microphone and Q k Speakers, and there are L microphones and Q speakers in the entire network. Each node is equipped with a processor with communication and computing capabilities, and all nodes together form a topology.

[0016] The global cost function of the multi-channel distributed wave-domain ANC network model is split into the sum of local cost functions. The specific method of establishing a distributed optimization model is as follows:

[0017] Step 2.1: The optimization objective of the multi-channel distributed wave-domain ANC network model is to minimize the sum of the squares of the residual sound field harmonic coefficients in the control area, that is, J(d) = α H α; where α = B -1 E m e is the harmonic coefficient vector of the residual sound field in the control area, and the matrices B and E mUsed to convert the error signal e into the wave domain coefficient α;

[0018] Step 2.2: The relationship between the error signal e and the speaker weight d is obtained as follows: Where v is the main noise field, T is the sound field transfer function, and the speaker weight d is a Q×1 parameter vector to be estimated;

[0019] Step 2.3: Split the cost function J(d) of the wave domain ANC problem into In the form of, the local cost function on node k (k = 1, ..., N) is obtained as in It is E m No. Row, by The error signal measured by the microphone

[0020] The specific method of using the diffuse LMS algorithm to solve the distributed optimization model and obtain the estimated speaker weight is as follows:

[0021] Step 3.1: At time n, let the cost function J at node k (k = 1, ..., N) be k (d) Derivative the speaker weight d to obtain the gradient vector

[0022] Step 3.2: d k，n represents the local version of the speaker weight d estimated by each node k at iteration n, which contains the weights of all Q speakers, that is, For k=1,…,N, By each node on The speaker weights are composed of d k，n Divided into group, a positive combination weight is assigned to d k，n and with For groups estimating the same speaker weight, a combined weight of zero is assigned to d k，n The rest of the group;

[0023] Step 3.3: Select an appropriate step size μ0, and node k (k=1, ..., N) uses the obtained gradient vector to estimate the speaker weight d k，n Update to the intermediate value ψ k，n+1 ;

[0024] Step 3.4: Choose the appropriate combination coefficient Node k (k=1,…,N) updates its estimate of the speaker weights by combining the intermediate estimates of its neighbor nodes

[0025] The specific steps of step 3.1 are as follows:

[0026] Cost Function Taking the derivative, we get:

[0027]

[0028] Make the gradient vector of the local cost function of the kth node only related to the error signal observed at the node;

[0029] The specific steps of step 3.2 are as follows:

[0030] d k，n represents the local version of the speaker weight d estimated by each node k at iteration n, d k，n Contains the weights of all Q speakers, that is For k=1,…,N, By each node on The speaker weights are composed of , in order to reduce the local estimation d k，n The redundant information in For index collection 's grouping, i.e.

[0031] If m≠m′

[0032] And make Indicated by Index d k，n The subvector of is the set of node k and its neighbor nodes, For collection The number of elements in , and d k，n Divided into group, assigning positive combined weights to d k，n and with For groups estimating the same speaker weight, a combined weight of zero is assigned to d k，n The rest of the group;

[0033] The specific method of step 3.3 is:

[0034] Node k uses the error signal it collects to estimate the local k，n Update to get the intermediate value ψ k，n+1 :

[0035]

[0036] The specific steps of step 3.4 are as follows:

[0037] Selected combination coefficient The following conditions are met:

[0038] like or

[0039] The update expression for the speaker estimation is: in Presentation Elements In vector The index in .

[0040] The specific method for canceling the secondary sound field generated by the multi-channel distributed wave-domain ANC network model and the residual noise signal is as follows:

[0041] Node k (k=1, ..., N) will estimate For each node Estimation of loudspeaker weights at Send to the corresponding speaker to generate a driving signal;

[0042] The loudspeaker generates a secondary sound field based on the obtained driving signal to cancel out the residual noise signal.

[0043] A multi-channel active noise control system, comprising:

[0044] Network model building module, used to build a multi-channel distributed wave domain ANC network model;

[0045] The cost function splitting module splits the global cost function of the multi-channel distributed wave-domain ANC network model into the sum of local cost functions and establishes a distributed optimization model;

[0046] The speaker weight estimation module is used to solve the distributed optimization model using the diffusion LMS algorithm to obtain the estimation of the speaker weight;

[0047] The noise cancellation module is used to cancel the secondary sound field generated by the multi-channel distributed wave domain ANC network model and the residual noise signal.

[0048] A terminal device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of a multi-channel active noise control method when executing the computer program.

[0049] .A computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the multi-channel active noise control method are implemented.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] This paper decomposes the global cost function of a multi-channel distributed wave-domain ANC network model into a sum of local cost functions. This approach, different from traditional time-frequency domain ANC problems, ensures that the gradient vector of the local cost function at each node only requires the error signal observed at that node. The present invention utilizes a diffuse LMS algorithm to solve the distributed optimization model and estimate the loudspeaker weights. This extends the diffuse LMS-based distributed wave-domain ANC algorithm to multi-channel node networks, enabling each node in the network to be equipped with a different number of microphones and loudspeakers. Furthermore, a low-computational-complexity strategy reduces the storage and computational complexity of each node, thereby enhancing the robustness, scalability, and practicality of the distributed wave-domain ANC system. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 Schematic diagram of the construction principle of the multi-channel active noise control method of the present invention;

[0053] Figure 2 A schematic diagram of low-complexity parameter estimation according to the present invention;

[0054] Figure 3 is a flow chart of the multi-channel active noise control method of the present invention;

[0055] Figure 4 Schematic diagram of the simulation results of the present invention in a free field environment;

[0056] Figure 5 Schematic diagram of the simulation results of the present invention in a reverberant field environment. DETAILED DESCRIPTION

[0057] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0058] See also Figure 1 , Figure 2 , Figure 3 The multi-channel active noise control method proposed in this invention first utilizes wireless acoustic sensor network theory to establish a distributed network model for wave-domain active noise control. It then decomposes the global cost function of the wave-domain ANC problem into a sum of local cost functions, establishing a distributed optimization problem. Finally, using the diffusion LMS algorithm, a distributed solution is obtained that requires only local information interaction to estimate speaker weights, thereby reducing spatial noise. This invention utilizes a diffusion strategy to develop a multi-channel distributed wave-domain ANC algorithm, distributing the computational burden across nodes. Compared to centralized solutions, this invention makes the wave-domain ANC system more scalable and robust, making it more suitable for large-scale applications.

[0059] A multi-channel active noise control method, an embodiment of which includes the following steps:

[0060] Step 1: Define the network model of wave-domain ANC, which is a distributed network connected in a certain topology.

[0061] Specifically:

[0062] Step 1.1: Define a circular area with a radius of R1 = 0.5m as the control area where noise elimination is required. The noise source is a 2D point sound source at (2, 0°) outside the control area, with a frequency of 500Hz and an amplitude of 15.

[0063] Step 1.2: Place N = 11 microphone arrays evenly on the border of the control area to capture the residual sound field, and place N = 11 loudspeakers evenly on a ring with a radius of R2 = 1.5m outside the control area to generate the secondary sound field;

[0064] Step 1.3: Establish three systems, where system 1 is a network of 5 nodes, where each node has two or three microphones and two or three speakers; system 2 is a network of 8 nodes, where each node has one or two microphones and one or two speakers; system 3 is a network of 11 nodes, where each node has one microphone and one speaker (i.e., a single-channel node network), where each node of each system has a processor with communication and computing capabilities, and the network is connected in a certain topology.

[0065] Step 2: Decompose the cost function J(d) of the wave domain ANC problem into In the form of, the local cost function at node k is obtained as:

[0066] Step 3: At time n, let the cost function J at node k be k (d) Taking the derivative of the speaker weight d, the gradient vector is:

[0067]

[0068] Step 4: Solve the multi-channel distributed wave domain problem using a diffusion strategy Node k uses the resulting gradient vector to calculate its estimate of the speaker weight d k，n Update to the intermediate value ψ k，n+1 , the update method is:

[0069]

[0070] And select the step size μ0=8.

[0071] Step 5: Node k updates its speaker estimate d by combining the intermediate estimates of its neighbor nodes k，n , the update method is:

[0072]

[0073] Step 6: Node k sends its estimate of the weight to the corresponding speaker to generate the driving signal; Step 7: The speaker generates the secondary sound field according to the driving signal to cancel the residual noise signal, thus achieving the reduction of noise in the control area.

[0074] Step 7: The speaker generates the secondary sound field according to the driving signal to cancel the residual noise signal, thus achieving the reduction of noise in the control area.

[0075] By measuring the residual signal at L = 1296 points uniformly distributed in the control area, the present application reduces the noise in the control area by defined as:

[0076]

[0077] where e l (n) represents the residual signal at the lth microphone in the area at time n, and e l (0) represents the primary noise signal measured at the lth point in the area.

[0078] To better verify the spatial noise reduction effect of the multi-channel distributed algorithm, the present application simulates in free field and reverberation field environments respectively. The simulation starts from collecting time domain signals, with a sampling rate of 8KHz and a window length of 2048. The microphone signals on each node are added with Gaussian white noise with a signal-to-noise ratio (SNR) of 40dB.

[0079] Figure 4 The simulation results in the free field environment are given, and compared with the centralized algorithm. It can be seen that under steady state, the three distributed systems can achieve similar noise reduction level as the centralized algorithm. In addition, system 1 can make the algorithm converge faster than the other two distributed systems, because system 1 uses more microphone signals to perform the adaptive step in the diffusion strategy.

[0080] Figure 5 The simulation results in the reverberation field environment are given, and compared with the centralized algorithm. To simulate the reverberation field environment, the present application chooses to use the image source model to establish a rectangular room with a size of 6m x 6m, with a perfectly absorbing ceiling and floor, and all side walls with a reflection coefficient of 0.5. It can be seen that in the reverberation field environment, the three distributed systems of the present application can still achieve similar attenuation as the centralized algorithm. Similarly, system 1 can make the algorithm converge faster than the other two distributed systems.

[0081] ​The application further provides a multi-channel active noise control system, comprising:

[0082] A network model establishing module is configured to establish a distributed wave domain ANC network model;

[0083] A cost function splitting module is configured to split a global cost function of the wave domain ANC problem into a form of local cost functions and to establish a distributed optimization problem;

[0084] A loudspeaker weight estimating module is configured to solve the distributed wave domain ANC problem to obtain an estimate of the loudspeaker weight;

[0085] A noise cancellation module is configured to generate a secondary sound field to cancel the residual noise signal.

[0086] The application further provides a terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the active noise control method of the application when executing the computer program.

[0087] The application further provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program implements the steps of the active noise control method of the application when executed by a processor.

[0088] The computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the method of the application.

[0089] The terminal device can be a desktop computer, a notebook computer, a palm computer, a cloud server, or other computing devices, and can also be a processor and a memory. The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The memory can be used to store computer programs and / or modules, and the processor can realize various functions of the active noise control system by running or executing the computer programs and / or modules stored in the memory, and by calling data stored in the memory.

[0090] The above merely describes the preferred embodiments of the present application, and is not intended to limit the technical solutions of the present application in any way. Those skilled in the art should understand that, without departing from the spirit and principle of the present application, the technical solutions can also be modified and replaced in several simple ways, and these modifications and replacements also all belong to the protection scope covered by the claims.

Claims

1. A multi-channel active noise control method, characterized in that: The following steps are involved: Establish a multi-channel distributed wave-domain ANC network model; The global cost function of the multi-channel distributed wave-domain ANC network model is split into the sum of local cost functions to establish a distributed optimization model. The specific method is as follows: Step 2.1: The optimization objective of the multi-channel distributed wave-domain ANC network model is to minimize the sum of the squares of the residual sound field harmonic coefficients in the control area, that is, ;in is the harmonic coefficient vector of the residual sound field in the control area, and the matrix and For the error signal Convert to wave domain coefficient ; Step 2.2: Get the error signal With speaker weight The relationship is ,in is the main noise field, is the sound field transfer function, the speaker weight For one The parameter vector to be estimated; Step 2.3: Transform the cost function of the wave domain ANC problem into Split into In the form of, get the node The local cost function on ,in yes No. Row, by The error signal measured by the microphone ; The distributed optimization model is solved using the diffuse LMS algorithm to obtain an estimate of the speaker weights. The specific method is as follows: Step 3.1: At this moment, let the node The cost function at Speaker weighting Derivative, get the gradient vector , as follows: For the cost function Taking the derivative, we get: Make the The gradient vector of the local cost function of a node is only related to the error signal observed at the node; Step 3.2: Represents each node In iteration Speaker weight estimated at the moment A local version of The weight of each speaker, ,for , By each node on The speaker weights are composed of Divided into group, the positive combination weight is assigned to and with For groups estimating the same speaker weight, a combined weight of zero is assigned to The rest of the group; Step 3.3: Choose an appropriate step size ,node Using the resulting gradient vector, we estimate the speaker weights Update to the middle value ; Step 3.4: Choose the appropriate combination coefficient ,node Update its estimate of speaker weights by combining the intermediate estimates of its neighboring nodes ; The secondary sound field generated by the multi-channel distributed wave-domain ANC network model is offset from the residual noise signal.

2. A multi-channel active noise control method according to claim 1, characterized in that: The multi-channel distributed wave domain ANC network model is connected by topology Network model of node-distributed wave-domain ANC; The specific method for establishing a multi-channel distributed wave-domain ANC network model is as follows: Step 1.1: Define a circular control area to complete the noise removal task, and the area radius is set to , and set the noise source outside this area; Step 1.2: Arrange The uniform circular microphone array of array elements is on the boundary of the control area, and the radius outside this area is Evenly distributed on the ring speakers to produce the secondary sound field; Step 1.3: Build a A distributed network of nodes, each node have microphones and speakers, and the entire network has microphones and Each node is equipped with a processor with communication and computing capabilities, and all nodes together form a topology.

3. The multi-channel active noise control method according to claim 1, characterized in that: The specific steps of step 3.1 are as follows: For the cost function Taking the derivative, we get: Make the The gradient vector of the local cost function of a node is only related to the error signal observed at the node; The specific steps of step 3.2 are as follows: Represents each node In iteration Speaker weight estimated at the moment The local version of Includes all The weight of each speaker, ,for , By each node on The speaker weights are composed of The redundant information in For index collection 's grouping, i.e. , ,like And make Indicated by Indexed The subvector of For nodes The set of its neighbor nodes, For collection The number of elements in Divided into group, assigning positive combined weights to and with For groups estimating the same speaker weight, a combined weight of zero is assigned to The rest of the group; The specific method of step 3.3 is: node Use the error signal collected to estimate the local Update to get the intermediate value : 。 4. The multi-channel active noise control method according to claim 1, characterized in that: The specific steps of step 3.4 are as follows: Selected combination coefficient The following conditions are met: like or The update expression for the speaker estimation is: ,in Presentation Elements In vector The index in .

5. The multi-channel active noise control method according to claim 1, characterized in that: The specific method for canceling the secondary sound field generated by the multi-channel distributed wave-domain ANC network model and the residual noise signal is as follows: node will estimate For each node Estimation of loudspeaker weights at Send to the corresponding speaker to generate a driving signal; The loudspeaker generates a secondary sound field based on the obtained driving signal to cancel out the residual noise signal.

6. A multi-channel active noise control system, characterized in that: A multi-channel active noise control method according to any one of claims 1 to 5, comprising: Network model building module, used to build a multi-channel distributed wave domain ANC network model; The cost function splitting module splits the global cost function of the multi-channel distributed wave-domain ANC network model into the sum of local cost functions and establishes a distributed optimization model; The speaker weight estimation module is used to solve the distributed optimization model using the diffusion LMS algorithm to obtain the estimation of the speaker weight; The noise cancellation module is used to cancel the secondary sound field generated by the multi-channel distributed wave domain ANC network model and the residual noise signal.

7. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the multi-channel active noise control method according to any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the multi-channel active noise control method according to any one of claims 1 to 5 are implemented.

Images