Collaborative optimization method for earphone hybrid active-passive noise reduction

Abstract

The invention relates to the technical field of earphone noise reduction methods, in particular to a collaborative optimization method for earphone hybrid active-passive noise reduction, which comprises the following steps: firstly, constructing a passive acoustic unit database; secondly, establishing a transfer function neural network model of the passive acoustic unit based on the database; thirdly, constructing a multi-objective optimization problem of the passive acoustic unit structure; then, constructing a hybrid active and passive noise reduction system; and finally, based on the constructed linear matrix model, targets inside and outside the noise frequency band in the hybrid noise reduction system are determined, multi-target collaborative optimization is completed, and respective advantages of active noise reduction and passive noise reduction are fully exerted. The optimization of the passive noise reduction structure improves the sound insulation effect of middle and high frequency bands, and makes up for the deficiency of active noise reduction in the frequency band; the optimization of the active noise reduction algorithm further enhances the noise reduction effect of the low-frequency band, and the active noise reduction algorithm and the low-frequency band form good complementation; through collaborative optimization, the overall noise reduction performance is far better than the effect of a single technology.

Description

Technical Field

[0001] The present invention relates to the technical field of headphone noise reduction methods, and in particular to a collaborative optimization method for hybrid active-passive noise reduction of headphones. Background Art

[0002] In recent years, as people's demands for sound quality and user experience continue to rise, headphone noise reduction technology has also continued to advance. Currently, noise-cancelling headphones on the market primarily utilize two technologies: active noise reduction and passive noise reduction. Active noise reduction cancels out ambient noise by generating sound waves in phase opposite to the ambient noise, while passive noise reduction relies on physical structures to block external noise. Each technology has its advantages, but also its limitations.

[0003] While traditional active noise cancellation technology excels at suppressing low-frequency noise, its effectiveness in mid- and high-frequency bands is often unsatisfactory. Furthermore, the introduction of active noise cancellation systems can also result in a loss of sound quality, especially in complex acoustic environments. On the other hand, while passive noise cancellation technology doesn't introduce additional electronic noise, its effectiveness is often limited by the physical structure of the headphones, making it difficult to adapt to changing noise environments.

[0004] Currently, the industry's commonly adopted approach is to simply combine active and passive noise reduction technologies, but this approach often involves independent optimization of each, lacking systematic collaborative design. This leads to several key issues: First, it's difficult to strike a good balance between noise reduction and sound quality, with improving noise reduction performance often sacrificing sound quality. Second, existing methods lack flexibility and are difficult to adapt to different usage scenarios and personal preferences. Third, they are energy inefficient, with enabling noise reduction often significantly reducing battery life.

[0005] Furthermore, existing technologies also have shortcomings in modeling and optimization. Traditional acoustic modeling methods are often oversimplified and fail to accurately describe complex acoustic systems. Optimization strategies typically focus on a single objective or a simple combination of multiple objectives, failing to fully consider multiple interrelated factors, such as noise reduction effectiveness, sound quality, and comfort.

[0006] In view of the above problems, there is an urgent need for a method that can systematically and collaboratively optimize active noise reduction and passive noise reduction technologies to achieve comprehensive improvements in noise reduction effects, sound quality performance, and user experience. This invention is an innovative solution to this technical need. Summary of the Invention

[0007] The collaborative optimization method for hybrid active-passive noise reduction in headphones proposed in this paper aims to address existing issues such as the difficulty in balancing noise reduction effectiveness and sound quality, poor adaptability, and low energy efficiency. By constructing a precise acoustic model, employing a multi-objective optimization strategy, and introducing an adaptive algorithm, this method achieves the organic integration and collaborative optimization of active and passive noise reduction technologies.

[0008] The application provides a synergistic optimization method for earphone hybrid active-passive noise reduction, which comprises the following steps: first, constructing a passive acoustic unit database, specifically constructing different forms of passive acoustic units by adjusting the shape and size of the material, and then extracting the transfer function characteristics of these units by modal analysis and finite element simulation, taking the structural parameters of the unit as the input and the transfer function characteristics as the output; second, establishing a transfer function neural network model of the passive acoustic unit based on the above database; third, constructing a multi-objective optimization problem of the passive acoustic unit structure, and optimizing the design of the passive acoustic unit with acoustic reciprocity and considering the noise reduction effect in the noise frequency band and the sound quality performance outside the noise frequency band for the target earphone structure; fourth, constructing a hybrid active-passive noise reduction system, combining the feedforward active noise reduction algorithm with the passive acoustic unit, introducing a multi-input multi-output matrix, a feedforward filter, a secondary sound source and a weight matrix to construct a linear matrix model;

[0009] Finally, based on the linear matrix model constructed above, the target in the noise frequency band and outside the noise frequency band in the hybrid noise reduction system is determined, and the multi-objective synergistic optimization is completed, wherein the synergistic optimization adopts the following objective function:

[0010]

[0011] Wherein, J(ω) is the overall objective function, ω is the frequency; TL(ω) is the transmission loss function, which represents the noise reduction effect in the noise frequency band; L(ω) is the linearity function, which represents the sound quality performance outside the noise frequency band; C(ω) is the clarity function; SWR(ω) is the standing wave ratio function; EP(ω) is the ear pressure function; α, β, γ, δ, ∈ are the weight coefficients of each target, and satisfy α+β+γ+δ+∈=1.

[0012] Preferably, in the step of constructing the passive acoustic unit database, the method of adjusting the material to construct the passive acoustic unit comprises: removing the hole structure with different volume fractions in the passive acoustic unit while keeping the mass unchanged, so as to realize different shapes of the unit; or adding the hole structure with different volume fractions in the passive acoustic unit while increasing the mass by the same amount, so as to realize different sizes of the unit.

[0013] Preferably, in the step of constructing the database of passive acoustic units, the transfer function modeling method of the passive acoustic unit comprises the following sub-steps: first, according to the modal analysis result, extracting the modal shape, natural frequency and transfer function of the passive acoustic unit; second, establishing the frequency response of the unit according to the above-mentioned transfer function; then, establishing a modal identification model of the unit transfer function, taking the natural frequency and modal shape of the unit as input and the transfer function as output, and establishing a nonlinear model of the two; finally, testing the model according to the model training result, if the test requirement is not met, re-extracting the natural frequency and modal shape, if the test requirement is met, obtaining the transfer function output of the unit, wherein the modal identification model adopts the following function:

[0014]

[0015] wherein H(ω) is the transfer function, ω is the frequency, A i is the participation factor of the i-th modal, is the i-th modal shape, ω i is the i-th natural frequency, ζ i is the i-th modal damping ratio, and j is the imaginary unit.

[0016] Preferably, in the step of establishing the transfer function neural network model of the passive acoustic unit, the neural network structure adopted comprises:

[0017] Input layer: composed of one input node, inputting the extracted unit parameters into the network;

[0018] Output layer: composed of one output node, representing the structure transfer function;

[0019] Hidden layer: composed of multiple hidden nodes, using an activation function for nonlinear fitting;

[0020] wherein the connection weights from the input layer to the hidden layer and from the hidden layer to the output layer are random variables, and the connection weights from the input layer to the hidden layer and from the hidden layer to the output layer are parameters,

[0021] The mathematical expression of the neural network model is:

[0022] y = f(∑w i ·x i +b)

[0023] wherein y is the output value, i.e. the predicted transfer function; f is the activation function, using the ReLU function f(x) = max(0, x); w i is the connection weight; x i is the input parameter; and b is the bias term.

[0024] Preferably, in the step of establishing the transfer function neural network model of the passive acoustic unit, the weight adjustment method of the neural network parameters comprises the following sub-steps: firstly, initializing the initial weights and thresholds of the neural network, randomly generating the weights and thresholds of the input layer to the hidden layer and the hidden layer to the output layer, and setting the maximum iteration number and the minimum error threshold; secondly, inputting the parameter data of the unit into the neural network, taking the error between the network output data and the target data as the objective function E i ; then, judging the size of E i , if E i is less than or equal to the minimum error threshold, entering the next step, if E i is greater than the minimum error threshold, performing the next step; then, adjusting the weights of the connection structure of each layer in the network through the back propagation algorithm; subsequently, judging the iteration number, if the iteration number is greater than or equal to the maximum iteration number, entering the next step, otherwise, re-performing the parameter data input step; finally, obtaining the weights and thresholds of each layer of the neural network, wherein the back propagation algorithm adopts the gradient descent method, and the weight update formula is:

[0025]

[0026] wherein w(t) is the weight value of the tthiteration, η is the learning rate, E is the error function, is the partial derivative of the error function with respect to the weight value.

[0027] Preferably, in the step of constructing the multi-objective optimization problem of the structure of the passive acoustic unit, a two-level multi-objective optimization structure is adopted.

[0028] The first-level multi-objective optimization is to select the structure parameters of the passive acoustic unit, the input of which is the parameters of the shape and size of the passive acoustic unit, and the output of which is the noise frequency band internal component transmission loss T n (ω) and the noise frequency band external linearity Linearity(ω), wherein ω represents frequency; the second-level multi-objective optimization is to select the unit number of the target earphone, the input of which is the output result of the first-level optimization, and the output parameters of which are the in-ear sound pressure amplitude P maxI , the difference between the in-ear sound pressure amplitude and the average in-ear sound pressure amplitude ΔP maxI , the difference between the in-ear and out-ear sound pressure amplitude ΔP inT , and the human ear comfort P E ; the optimization parameter is the noise reduction frequency band impedance value Z(ω); wherein the particle swarm algorithm is adopted in both levels of optimization, and the velocity and position update formula of the algorithm is:

[0029]

[0030] x(t+1)=x(t)+v(t+1)

[0031] where v is the particle velocity, x is the particle position, w is the inertial weight, c1 and c2 are the acceleration constants, r1 and r2 are random numbers between 0 and 1, and p best is the individual optimal position, g best is the global optimal position.

[0032] Preferably, in the step of constructing the hybrid active and passive noise reduction system, the method for constructing the linear matrix model comprises the following steps: first, a system block diagram is established, wherein the feed-forward secondary sound source is an audio unit, the input is an expected signal of an active secondary front-end audio amplifier, and the output is an echo signal played by a feed-forward secondary loudspeaker; the primary sound source is an in-ear secondary sound source audio signal, the input is a secondary error signal picked up by an earphone microphone, and the output is a superimposed signal of the feed-forward secondary signal and the primary secondary signal output by the secondary loudspeaker; the secondary sound source output is a target signal of a target space, the input is the superimposed signal output by the primary secondary loudspeaker, and the output is a superimposed signal of the active and passive noise reduction in the target space;

[0033] Then, according to the system block diagram, a linear matrix model is established to obtain the relationship between the active secondary front-end audio signal, the in-ear expected signal, the secondary loudspeaker error signal, the superimposed signal and the target space signal,

[0034] The linear matrix model is represented as Y=HX+N, wherein Y is an output signal vector, H is a system transfer function matrix, X is an input signal vector, and N is a noise vector.

[0035] Preferably, the step of completing the collaborative optimization of multiple targets comprises the following sub-steps:

[0036] First, the noise reduction targets of the hybrid noise reduction system in the noise frequency band are defined, including the total number of units, the in-ear sound pressure amplitude and the secondary sound source output amplitude;

[0037] Second, the noise reduction targets of the hybrid noise reduction system outside the noise frequency band are defined, including linearity, clarity, standing wave ratio and ear pressure;

[0038] Then, according to the linear matrix model, the initial values of each target function are determined;

[0039] Next, the secondary loudspeaker error signal, the superimposed signal and the target space signal are taken as inputs to determine the linearity Linearity(ω);

[0040] Subsequently, it is determined whether the linearity meets the requirements, if yes, the next step is performed, otherwise, the linearity Linearity(ω) is taken as the optimization target, and the step of constructing the hybrid active and passive noise reduction system is returned to update the weight;

[0041] Then, three noise reduction targets other than the linearity Linearity(ω) are taken as optimization targets to judge the solving condition, if within the error threshold range, optimization is completed, otherwise, the impedance value Z(ω) of the noise reduction frequency band is taken as a variable to perform weight correction;

[0042] Finally, returning to the step of establishing the transfer function neural network model of the passive acoustic unit, it is judged whether the active secondary front-end audio signal and the in-ear desired signal meet the requirements, if meeting the requirements, optimization is completed, otherwise, returning to the previous step, wherein the weight correction adopts an adaptive algorithm, and the weight update formula is:

[0043] w(n+1)=w(n)+μ·e(n)·x(n)

[0044] wherein w(n) is the weight value of the n th iteration, μ is a step factor, e(n) is an error signal, and x(n) is an input signal.

[0045] Preferably, the method for determining the linearity Linearity(ω) comprises:

[0046] Firstly, according to the linear matrix model, an active secondary front-end audio signal is input to obtain an in-ear sound pressure L e ; secondly, an in-ear desired signal is input to obtain an in-ear sound pressure T e ; then, the in-ear sound pressure linearity is obtained according to the following formula:

[0047]

[0048] wherein L e (ω) and T e (ω) are dimensionless quantities with a unit of dB.

[0049] Preferably, in the step of taking three noise reduction targets other than the linearity Linearity(ω) as optimization targets, the linear matrix model is linearized in the noise frequency band to obtain the following formula:

[0050] Y(ω)=A1(ω)X(ω)+B1(ω)Z(ω)+C1(ω)

[0051] wherein A1(ω), B1(ω) and C1(ω) are transformation matrices, and Z(ω) represents the frequency domain impedance of the secondary loudspeaker; the optimization weight value is defined, the above transformation matrices are brought into the optimization weight value, and three noise reduction targets are solved by the following formula:

[0052]

[0053] wherein ε I , ε ωT and ε nTerror terms representing the total number of units in the noise band, the amplitude of the sound pressure in the ear, and the difference between the amplitude of the sound pressure in and out of the ear, respectively, the superscript H represents the Hermitian transpose, f is the objective function, Z(ω) is the variable to be optimized, W ω ωT ωp respectively represent the optimization weights of the three objectives.

[0054] The beneficial effects of the present application mainly manifest in the following aspects:

[0055] The method of the present application first provides a rich data basis for subsequent optimization by constructing a detailed passive acoustic unit database. Then, the transfer function under different structural parameters is accurately predicted through the neural network model, greatly improving the accuracy and efficiency of modeling. On this basis, a two-level multi-objective optimization structure is adopted, which not only considers the performance of a single acoustic unit, but also takes into account the overall performance of the entire headphone system. Finally, by establishing a linear matrix model of the hybrid active and passive noise reduction system, the overall accurate description and control of the system are realized.

[0056] The method of the present application performs well in solving the contradiction between algorithm mechanism. For example, in traditional methods, improving the noise reduction effect often leads to a decline in sound quality, while the present application successfully realizes the coordinated improvement of both through a multi-objective optimization strategy, which considers both the transmission loss (noise reduction effect) and the frequency response linearity (sound quality performance) in the objective function. At the same time, by introducing indicators such as human ear comfort, the present application also solves the contradiction between noise reduction performance and wearing comfort.

[0057] In terms of complementary and additive effects, the present application fully utilizes the respective advantages of active noise reduction and passive noise reduction. The optimization of passive noise reduction structure improves the sound insulation effect in the medium and high frequency band, making up for the deficiency of active noise reduction in this frequency band. While the optimization of the active noise reduction algorithm further enhances the noise reduction effect in the low frequency band, forming a good complementarity. More importantly, through collaborative optimization, the present application realizes the superposition of the effects of the two technologies, making the overall noise reduction performance far superior to that of a single technology.

[0058] The synergistic effect of the present application also manifests in the improvement of energy efficiency. By optimizing the passive noise reduction structure, the dependence on active noise reduction is reduced, thereby reducing the power consumption of the system. This not only prolongs the battery life, but also reduces the electronic noise that may be introduced by the active noise reduction system, further improving the sound quality.

[0059] ​​Overall, the earphone hybrid active-passive noise reduction collaborative optimization method proposed by the present application successfully solves many problems existing in the prior art through systematic modeling, multi-objective optimization and collaborative control. It not only significantly improves the noise reduction effect and sound quality performance, but also improves the user experience and energy efficiency. This comprehensive performance improvement provides a new technical path for the development of high-quality noise reduction earphones, and has important practical value and broad application prospects. BRIEF DESCRIPTION OF DRAWINGS

[0060] Figure 1 The overall method logic diagram of the present application.

[0061] Figure 2 The passive acoustic unit database construction logic diagram of the present application.

[0062] Figure 3 The neural network model establishment logic diagram of the present application.

[0063] Figure 4 The multi-objective optimization problem construction logic diagram of the present application.

[0064] Figure 5 The hybrid active-passive noise reduction system construction logic diagram of the present application.

[0065] Figure 6 The multi-objective collaborative optimization logic diagram of the present application. DETAILED DESCRIPTION

[0066] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined invention purpose, the specific implementation, structure, features and effects of the preferred embodiments of the present application are described in detail below in combination with the drawings. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0067] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0068] Example 1

[0069] Reference Figures 1-6 The present application relates to an earphone hybrid active-passive noise reduction collaborative optimization method, in particular to a method that combines active noise reduction and passive noise reduction technology to achieve simultaneous improvement of noise reduction effect and sound quality performance through multi-objective optimization.

[0070] Firstly, the earphone hybrid active-passive noise reduction collaborative optimization method provided by the application comprises the following steps: firstly, a passive acoustic unit database is constructed, specifically, different forms of passive acoustic units are constructed by adjusting the shape and size of the material, and then the transfer function characteristics of these units are extracted by modal analysis and finite element simulation, taking the structural parameters of the unit as the input and the transfer function characteristics as the output; secondly, a transfer function neural network model of the passive acoustic unit is established based on the above database; thirdly, a multi-objective optimization problem of the passive acoustic unit structure is constructed, and the passive acoustic unit with acoustic reciprocity and considering the noise reduction effect in the noise frequency band and the sound quality performance outside the noise frequency band is optimized and designed for the target earphone structure; then, a hybrid active-passive noise reduction system is constructed, the feedforward active noise reduction algorithm is combined with the passive acoustic unit, a multi-input multi-output matrix, a feedforward filter, a secondary sound source and a weight matrix are introduced to construct a linear matrix model;

[0071] Finally, based on the linear matrix model constructed above, the target in the noise frequency band and outside the noise frequency band in the hybrid noise reduction system is determined, and the multi-objective collaborative optimization is completed, wherein the collaborative optimization adopts the following objective function:

[0072]

[0073] Wherein, J(ω) is the overall objective function, ω is the frequency; TL(ω) is the transmission loss function, which represents the noise reduction effect in the noise frequency band; L(ω) is the linearity function, which represents the sound quality performance outside the noise frequency band; C(ω) is the clarity function; SWR(ω) is the standing wave ratio function; EP(ω) is the ear pressure function; α, β, γ, δ, ∈ are weight coefficients of each target, and satisfy α+β+γ+δ+∈=1.

[0074] In an embodiment of the application, the material adjustment of the passive acoustic unit comprises two ways: one is to remove the hole structure with different volume fractions in the passive acoustic unit while keeping the mass unchanged, to realize different shaped units; the other is to increase the hole structure with different volume fractions in the passive acoustic unit while increasing the mass by the same amount, to realize different sized units. This flexible adjustment method enables us to maximize the use of material characteristics to optimize the acoustic performance while keeping the weight.

[0075] Next, the present application proposes a transfer function modeling method of a passive acoustic unit. This method first extracts the modal shape, natural frequency and transfer function of the passive acoustic unit according to the modal analysis results. Then, the frequency response of the unit is established according to these transfer functions. After that, a modal identification model of the unit transfer function is established, taking the natural frequency and modal shape of the unit as input and the transfer function as output, to establish a nonlinear model between the two. Finally, the model is tested according to the model training results, and if the test requirements are not met, the natural frequency and modal shape are re-extracted until the requirements are met.

[0076] Preferably, the modal identification model function used by the present application is:

[0077]

[0078] where H(ω) is the transfer function, ω is the frequency, A i is the participation factor of the i-th mode, is the i-th modal shape, ω i is the i-th natural frequency, ζ i is the i-th modal damping ratio, and j is the imaginary unit. This model can accurately describe the acoustic characteristics of the passive acoustic unit, laying a foundation for subsequent optimization.

[0079] When establishing the transfer function neural network model of the passive acoustic unit, the present application adopts a special neural network structure. This neural network structure includes:

[0080] Input layer: composed of one input node, the extracted unit parameters are input into the network;

[0081] Output layer: composed of one output node, representing the structure transfer function;

[0082] Hidden layer: composed of multiple hidden nodes, using an activation function for nonlinear fitting;

[0083] where the connection weights from the input layer to the hidden layer and from the hidden layer to the output layer are random variables, and the connection weights from the input layer to the hidden layer and from the hidden layer to the output layer are parameters,

[0084] The mathematical expression of this neural network model is:

[0085] y = f(∑w i ·x i +b)

[0086] where y is the output value, i.e. the predicted transfer function; f is the activation function, using the ReLU function f(x) = max(0, x); w i is the connection weight; x i ​where a is the input parameter; b is the bias term. This neural network model can effectively handle complex nonlinear relationships, has good generalization ability, and is particularly suitable for predicting the transfer function under different structural parameters.

[0087] In order to improve the accuracy of the neural network model, the present application also proposes a weight adjustment method. In the step of establishing the transfer function neural network model of the passive acoustic unit, the weight adjustment method of the neural network parameters includes the following sub-steps: first, initialize the initial weight and threshold of the neural network, randomly generate the weight and threshold of the input layer to the hidden layer and the hidden layer to the output layer, and set the maximum iteration number and the minimum error threshold; second, input the parameter data of the unit into the neural network, and take the error between the network output data and the target data as the objective function E i ; then, judge the size of E i , if E i is less than or equal to the minimum error threshold, then enter the next step, if E i is greater than the minimum error threshold, then proceed to the next step; then, adjust the weight of each layer connection structure in the network through the back propagation algorithm; then, judge the iteration number, if the iteration number is greater than or equal to the maximum iteration number, then enter the next step, otherwise, re-input the parameter data; finally, obtain the weight and threshold of each layer of the neural network, wherein the back propagation algorithm adopts the gradient descent method, and the weight update formula is:

[0088]

[0089] where w(t) is the weight of the tth iteration, η is the learning rate, E is the error function, is the partial derivative of the error function with respect to the weight. This method can effectively improve the training effect of the neural network, so that the model can more accurately predict the transfer function.

[0090] In the multi-objective optimization problem of constructing the passive acoustic unit structure, the present application adopts a two-level multi-objective optimization structure. The first level multi-objective optimization is to select the structural parameters of the passive acoustic unit, the input is the parameters of the shape and size of the passive acoustic unit, and the output is the noise frequency band internal component transmission loss and the noise frequency band external linearity. The second level multi-objective optimization is to select the unit quantity of the target earphone, and the output parameters include the in-ear sound pressure amplitude, the difference between the in-ear sound pressure amplitude and the average, the difference between the in-ear and out-ear sound pressure amplitude, and the human ear comfort.

[0091] In an embodiment of the present application, both levels of optimization adopt the particle swarm algorithm, and the velocity and position update formula of the algorithm is:

[0092]

[0093] x(t+1)=x(t)+v(t+1)

[0094] where v is the particle velocity, x is the particle position, w is the inertial weight, c1 and c2 are acceleration constants, r1 and r2 are random numbers between 0 and 1, p best is the individual optimal position, g best is the global optimal position. This multi-level optimization strategy can better balance local optimization and global optimization, and is conducive to obtaining a better overall solution.

[0095] Next, the application proposes a method for constructing a hybrid active-passive noise reduction system. This method first establishes a system block diagram, including a feedforward secondary sound source, a primary sound source and a secondary sound source. Then, according to the system block diagram, a linear matrix model is established to obtain the relationship between various signals.

[0096] Preferably, the linear matrix model used by the application is represented as:

[0097] Y=HX+N

[0098] where Y is the output signal vector, H is the system transfer function matrix, X is the input signal vector, and N is the noise vector. This model can clearly describe the role of each component in the hybrid noise reduction system, providing a theoretical basis for subsequent optimization.

[0099] Finally, the application proposes a multi-objective collaborative optimization method. This method first defines the noise reduction targets of the hybrid noise reduction system within and outside the noise frequency band, and then determines the initial values of each objective function according to the linear matrix model. Next, through iterative optimization, the system parameters are continuously adjusted until the optimal performance is achieved.

[0100] In an embodiment of the application, the steps for completing multi-objective collaborative optimization include the following sub-steps:

[0101] First, define the noise reduction targets of the hybrid noise reduction system within the noise frequency band, including the total number of units, the in-ear sound pressure amplitude, and the secondary sound source output amplitude;

[0102] Second, define the noise reduction targets of the hybrid noise reduction system outside the noise frequency band, including linearity, clarity, standing wave ratio and ear pressure;

[0103] Then, according to the linear matrix model, determine the initial values of each objective function;

[0104] Next, take the secondary loudspeaker error signal, the superimposed signal and the target space signal as input to determine the linearity Linearity(ω);

[0105] Then, it is judged whether the linearity meets the requirement, if yes, the next step is performed, otherwise, the linearity Linearity (ω) is taken as an optimization target, the step of constructing the hybrid active and passive noise reduction system is returned, and the weight is updated.

[0106] Then, three noise reduction targets except the linearity Linearity (ω) are taken as optimization targets, it is judged whether the solution is within an error threshold, if yes, the optimization is completed, otherwise, the impedance value Z (ω) in the noise reduction frequency band is taken as a variable, and the weight correction is performed.

[0107] Finally, the step of returning to the establishment of the transfer function neural network model of the passive acoustic unit is performed, it is judged whether the active secondary front-end audio signal and the in-ear desired signal meet the requirement, if yes, the optimization is completed, otherwise, the previous step is returned, wherein the weight correction adopts an adaptive algorithm, and the weight update formula is:

[0108] w (n + 1) = w (n) + mu * e (n) * x (n)

[0109] Wherein, w (n) is the weight of the n-th iteration, mu is a step factor, e (n) is an error signal, and x (n) is an input signal.

[0110] In an embodiment of the present application, the method for determining the linearity Linearity (ω) comprises:

[0111] Firstly, according to the linear matrix model, the active secondary front-end audio signal is inputted to obtain the in-ear sound pressure L e ; secondly, the in-ear desired signal is inputted to obtain the in-ear sound pressure T e ; then, the in-ear sound pressure linearity is obtained according to the following formula:

[0112]

[0113] Wherein, L e (ω), T e (ω) are dimensionless quantities, and the unit is dB.

[0114] In the step of taking three noise reduction targets except the linearity Linearity (ω) as optimization targets, the linear matrix model is linearized in the noise frequency band to obtain the following formula:

[0115] Y (ω) = A1 (ω) X (ω) + B1 (ω) Z (ω) + C1 (ω)

[0116] Wherein, A1 (ω), B1 (ω), C1 (ω) are transformation matrices, and Z (ω) represents the frequency domain impedance of the secondary loudspeaker; the optimization weight is defined, the above transformation matrix is brought into the optimization weight, and three noise reduction targets are solved through the following formula:

[0117]

[0118] where ε I , ε ωT , and ε nT are error terms representing the total number of units within the noise band, the difference between the in-ear sound pressure amplitude and the out-ear sound pressure amplitude, respectively, with the superscript H representing the Hermitian transpose, f is the objective function, Z(ω) is the variable to be optimized, W ω , W ωT , and W ωp are the optimization weights for the three objectives, respectively. This comprehensive objective function takes into account multiple factors such as noise reduction effect, sound quality performance, and comfort, achieving overall performance improvement.

[0119] In summary, the earphone hybrid active-passive noise reduction collaborative optimization method proposed by the present application achieves comprehensive improvement of noise reduction effect and sound quality performance through systematic modeling, optimization, and iterative process. This method is not only applicable to earphones, but also can be extended to other audio devices that require noise reduction, with broad application prospects.

[0120] Example 1: One of the best embodiments of the present application adopts the above-mentioned earphone hybrid active-passive noise reduction collaborative optimization method. We selected a common in-ear earphone on the market as the optimization object. First, we constructed a passive acoustic unit database containing 100 different structural parameters. Then, a transfer function prediction model was established using a 5-layer neural network (including 1 input layer, 3 hidden layers, and 1 output layer). In the multi-objective optimization stage, we adopted a two-stage optimization structure, with the particle number of the first stage optimization set to 50 and the iteration number set to 100; the particle number of the second stage optimization set to 30 and the iteration number set to 50. Finally, in the collaborative optimization stage, we set the weights of noise reduction effect, sound quality performance, and comfort to 0.4, 0.4, and 0.2, respectively.

[0121] Comparative Example 1: In order to verify the superiority of the method of the present application, we selected a traditional earphone noise reduction optimization method as a comparison. This method only considers the optimization of active noise reduction algorithm, without combining passive noise reduction technology, nor adopting multi-objective optimization strategy. It uses a fixed structure acoustic unit to achieve noise reduction by adjusting the parameters of the feedback controller.

[0122] In order to comprehensively evaluate the performance of the two methods, we designed the following detection indicators:

[0123] 1. Noise reduction effect: The noise attenuation in different frequency bands is measured using a standard sound level meter, with units in dB.

[0124] 2. Frequency response linearity: The frequency response curve in the range of 20Hz-20kHz is measured using an audio analyzer, and the deviation from the ideal straight line is calculated.

[0125] 3. Clarity of sound: PESQ (Perceptual Evaluation of Speech Quality) score, ranging from 1 to 4.5.

[0126] 4. Comfort of wearing: Subjective score from 20 volunteers, ranging from 1 to 10.

[0127] 5. Battery life: Measured in hours under standard usage conditions.

[0128] The following is a comparison table of test results for Example 1 and Comparative Example 1:

[0129] From the above test results, it can be seen that the method of the present application is significantly better than the traditional method in various aspects. Specifically:

[0130] 1. Noise reduction effect: The noise reduction effect of the method of the present application in the low frequency band (100-500Hz) and the medium frequency band (500-2000Hz) is increased by 7dB and 7dB respectively compared with the traditional method. This means that in noisy environments such as airplane cabins or busy offices, users can enjoy a quieter listening experience. This significant improvement is mainly due to our method combining passive noise reduction and active noise reduction technology, and achieving the synergistic effect of the two through multi-objective optimization.

[0131] 2. Frequency response linearity: The deviation of the frequency response curve is controlled within ±2.5dB, which is significantly improved compared with the ±4.5dB of the traditional method. This indicates that our method can achieve efficient noise reduction while better preserving the original characteristics of the audio. This balance is achieved through our multi-objective optimization strategy, which considers both noise reduction effect and sound quality performance.

[0132] 3. Clarity of sound: PESQ score increased from 3.7 to 4.2, close to the theoretical maximum value of 4.5. This means that users can hear clearer and more natural sound when using our earphones. This improvement is mainly due to our method considering multiple audio quality indicators such as frequency response linearity and clarity in the optimization process.

[0133] 4. Comfort of wearing: Subjective score increased from 7.2 to 8.5, indicating that our method not only improves audio performance but also improves user wearing experience. This is because we introduced the comfort of human ear as an indicator in the optimization process, and through synergistic optimization with other performance indicators, a good balance has been achieved.

[0134] 5. Battery life: Despite the better noise reduction and sound quality performance achieved by our method, the battery life has been improved from 25 hours to 30 hours. This shows that our optimization method not only improves performance, but also improves energy efficiency. This may be because our method reduces the dependence on active noise reduction by optimizing the passive noise reduction structure, thereby reducing power consumption.

[0135] Overall, these test results fully demonstrate the superiority of the method of the present invention. By combining passive noise reduction and active noise reduction techniques and using a multi-objective collaborative optimization strategy, we have achieved a comprehensive improvement in noise reduction effect, sound quality performance, and user experience. This method not only has technical innovation, but also brings actual use experience improvement to users, so it has broad application prospects.

[0136] It should be noted that: the above only describes the preferred embodiments of the present invention, and is not intended to limit the present invention. Any modification, equivalent replacement, improvement, etc. within the principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A collaborative optimization method for hybrid active-passive noise reduction of headphones, characterized by: The method comprises the following steps: first, constructing a passive acoustic unit database, specifically constructing different forms of passive acoustic units by adjusting the shape and size of the material, then using modal analysis and finite element simulation to extract the transfer function characteristics of these units, with the structural parameters of the unit as input and the transfer function characteristics as output; second, establishing a transfer function neural network model of the passive acoustic unit based on the above database; then, constructing a multi-objective optimization problem for the passive acoustic unit structure, and optimizing the design of a passive acoustic unit with acoustic reciprocity and comprehensive consideration of the noise reduction effect within the noise frequency band and the sound quality performance outside the noise frequency band for the target headphone structure; then, constructing a hybrid active-passive noise reduction system, combining a feedforward active noise reduction algorithm with the passive acoustic unit, and introducing a multi-input and multi-output matrix, a feedforward filter, a secondary sound source, and a weight matrix to construct a linear matrix model; Finally, based on the linear matrix model constructed above, the targets within and outside the noise frequency band in the hybrid noise reduction system are determined to complete the multi-objective collaborative optimization, wherein the collaborative optimization adopts the following objective function: J(ω)=α·TL(ω)+β·L(ω)+γ·C(ω)+δ·SWR(ω)+∈·EP(ω) Among them, J(ω) is the overall objective function, ω is the frequency; TL(ω) is the transmission loss function, which represents the noise reduction effect within the noise frequency band; L(ω) is the linearity function, which represents the sound quality performance outside the noise frequency band; C(ω) is the clarity function; SWR(ω) is the standing wave ratio function; EP(ω) is the ear pressure function; α, β, γ, δ, ∈ are the weight coefficients of each objective, and they satisfy α+β+γ+δ+∈=1.

2. The collaborative optimization method for hybrid active-passive noise reduction of headphones according to claim 1, characterized in that: In the step of constructing the passive acoustic unit database, the method of adjusting the material to construct the passive acoustic unit includes: removing pore structures with different volume fractions in the passive acoustic unit while maintaining the mass unchanged, thereby realizing units of different shapes; or adding pore structures with different volume fractions in the passive acoustic unit while maintaining the same mass increase, thereby realizing units of different sizes.

3. The collaborative optimization method for hybrid active-passive noise reduction of headphones according to claim 1, characterized in that: In the step of constructing the passive acoustic unit database, the transfer function modeling method of the passive acoustic unit includes the following sub-steps: first, extracting the modal vibration shape, natural frequency and transfer function of the passive acoustic unit based on the modal analysis results; second, establishing the frequency response of the unit based on the above transfer function; Then, a modal identification model of the unit transfer function is established, with the unit's natural frequency and modal vibration shape as input and the transfer function as output, and a nonlinear model of the two is established; finally, the model is tested based on the model training results. If the test requirements are not met, the natural frequency and modal vibration shape are re-extracted. If the test requirements are met, the transfer function output of the unit is obtained, wherein the modal identification model adopts the following function: Where H(ω) is the transfer function, ω is the frequency, and A i is the participation factor of the i-th mode, is the i-th order mode shape, ω i is the i-th order natural frequency, ζ i is the i-th order modal damping ratio, and j is an imaginary unit.

4. The collaborative optimization method for hybrid active-passive noise reduction of headphones according to claim 1, characterized in that: In the step of establishing the transfer function neural network model of the passive acoustic unit, the neural network structure used includes: Input layer: consists of an input node, which inputs the extracted unit parameters into the network; Output layer: consists of an output node, representing the structural transfer function; Hidden layer: consists of multiple hidden nodes and uses activation function to perform nonlinear fitting; Among them, the connection weights from the input layer to the hidden layer, and from the hidden layer to the output layer are random variables, with the connection weights from the input layer to the hidden layer, and from the hidden layer to the output layer as parameters. The mathematical expression of the neural network model is: y=f(∑w i ·x i +b) Where y is the output value, that is, the predicted transfer function; f is the activation function, using the ReLU function f(x) = max(0, x); w i is the connection weight; x i is the input parameter; b is the bias term.

5. The collaborative optimization method for hybrid active-passive noise reduction of headphones according to claim 4, characterized in that: In the step of establishing the transfer function neural network model of the passive acoustic unit, the method for adjusting the weights of the neural network parameters includes the following sub-steps: first, initializing the initial weights and thresholds of the neural network, randomly generating weights and thresholds from the input layer to the hidden layer, and from the hidden layer to the output layer, and setting the maximum number of iterations and the minimum error threshold; Secondly, the unit parameter data is fed into the neural network, and the error between the network output data and the target data is used as the objective function E i ; Then, judge E i If the size of E i is less than or equal to the minimum error threshold, then proceed to the next step. If E i If the error is greater than the minimum error threshold, the next step is performed; then, the weights of the connection structures of each layer in the network are adjusted through the back propagation algorithm; then, the number of iterations is judged. If the number of iterations is greater than or equal to the maximum number of iterations, the next step is performed; otherwise, the parameter data input step is repeated; finally, the weights and thresholds of each layer of the neural network are obtained, wherein the back propagation algorithm adopts the gradient descent method, and the weight update formula is: Among them, w(t) is the weight of the tth iteration, η is the learning rate, E is the error function, is the partial derivative of the error function with respect to the weight.

6. The collaborative optimization method for hybrid active-passive noise reduction of headphones according to claim 1, characterized in that: In the step of constructing the multi-objective optimization problem of the passive acoustic unit structure, a two-level multi-objective optimization structure is adopted: The first-level multi-objective optimization is to select the structural parameters of the passive acoustic unit. The input is the parameters of the shape and size of the passive acoustic unit, and the output is the component transmission loss T in the noise frequency band. n (ω) and linearity outside the noise band (Linearity(ω), where ω represents the frequency; The second-level multi-objective optimization is to select the number of units of the target earphone. Its input is the output result of the first-level optimization, and the output parameter is the in-ear sound pressure amplitude P. maxI , the difference between the sound pressure amplitude in the ear and the mean sound pressure amplitude in the ear ΔP maxI , the difference in sound pressure amplitude between the inside and outside of the ear ΔP inT , human ear comfort P E ; The optimization parameter is the noise reduction frequency band impedance value Z(ω) where, Both levels of optimization use the particle swarm algorithm, and the speed and position update formula of the algorithm is: v(t+1)=w·v(t)+c1·r1·(p best -x(t))+c2·r2·(g best -x(t)) x(t+1)=x(t)+v(t+1) Among them, v is the particle velocity, x is the particle position, w is the inertia weight, c1 and c2 are acceleration constants, r1 and r2 are random numbers between 0 and 1, and p best is the optimal position of the individual, g best is the global optimal position.

7. The collaborative optimization method for hybrid active-passive noise reduction of headphones according to claim 1, characterized in that: In the step of constructing the hybrid active-passive noise reduction system, the method for constructing a linear matrix model includes: first, establishing a system block diagram, wherein the feedforward secondary sound source is an audio unit, the input is the desired signal of the active secondary front-end audio amplifier, and the output is the echo signal played by the feedforward secondary speaker; the primary sound source is the secondary sound source audio signal inside the earphone, the input is the secondary error signal picked up by the earphone microphone, and the output is the feedforward secondary signal and the primary secondary signal superimposed by the secondary speaker output; the secondary sound source output is the target signal of the target space, the input is the superimposed signal output by the primary secondary speaker, and the output is the superimposed signal of the active and passive noise suppression in the target space; Then, according to the system block diagram, a linear matrix model is established to obtain the relationship between the active secondary front-end audio signal, the desired signal in the ear, the secondary speaker error signal, the superimposed signal and the target space signal. The linear matrix model is expressed as: Y=HX+N, where Y is the output signal vector, H is the system transfer function matrix, X is the input signal vector, and N is the noise vector.

8. The collaborative optimization method for hybrid active-passive noise reduction of headphones according to claim 1, characterized in that: The step of completing the multi-objective collaborative optimization includes the following sub-steps: First, the noise reduction target of the hybrid noise reduction system within the noise frequency band is defined, including the total number of units, the sound pressure amplitude in the ear, and the output amplitude of the secondary sound source; Secondly, define the noise reduction targets of the hybrid noise reduction system outside the noise frequency band, including linearity, clarity, standing wave ratio and ear pressure; Then, according to the linear matrix model, the initial value of each objective function is determined; Next, the linearity (ω) is determined using the secondary loudspeaker error signal, the superimposed signal, and the target spatial signal as inputs. Then, determine whether the linearity meets the requirements. If so, proceed to the next step. Otherwise, use linearity (ω) as the optimization target and return to the step of building a hybrid active and passive noise reduction system to update the weights. Then, the three noise reduction targets except linearity (ω) are used as optimization targets to judge the solution situation. If it is within the error threshold range, the optimization is completed. Otherwise, the noise reduction frequency band impedance value Z (ω) is used as a variable to perform weight correction. Finally, return to the step of establishing the transfer function neural network model of the passive acoustic unit to determine whether the active secondary front-end audio signal and the expected signal in the ear meet the requirements. If so, the optimization is completed, otherwise return to the previous step, where the weight correction adopts an adaptive algorithm, and the weight update formula is: w(n+1)=w(n)+μ·e(n)·x(n) Where w(n) is the weight of the nth iteration, μ is the step size factor, e(n) is the error signal, and x(n) is the input signal.

9. The collaborative optimization method for hybrid active-passive noise reduction of headphones according to claim 8, characterized in that: The method for determining linearity (ω) includes: First, according to the linear matrix model, the active secondary front-end audio signal is input to obtain the in-ear sound pressure L e ; Secondly, input the desired signal in the ear and obtain the sound pressure T in the ear e ; Then, according to the following formula, the linearity of the sound pressure in the ear is obtained: Among them L e (ω), T e (ω) is a dimensionless quantity with the unit being dB.

10. The collaborative optimization method for hybrid active-passive noise reduction of headphones according to claim 8, characterized in that: In the step of taking the three noise reduction targets other than linearity (ω) as the optimization targets, the linear matrix model is linearized within the noise frequency band to obtain the following formula: Y(ω)=A1(ω)X(ω)+B1(ω)Z(ω)+C1(ω) Among them, A1(ω), B1(ω), C1(ω) are transformation matrices, and Z(ω) represents the frequency domain impedance of the secondary speaker. Define the optimization weights, substitute the above transformation matrices into the optimization weights, and obtain the three noise reduction targets through the following formulas: Among them, ε I , ε ωT and ε nT They represent the error terms of the total number of units in the noise frequency band, the sound pressure amplitude in the ear, and the difference between the sound pressure amplitude inside and outside the ear, respectively. The superscript H represents the Hermitian transpose, f is the objective function, Z(ω) is the variable to be optimized, and W ω 、W ωT 、W ωp Represent the optimization weights of the three objectives respectively.

Images