Active noise reduction method and device based on fxlms algorithm and active noise reduction door

Abstract

The application provides an active noise reduction method and device based on an FxLMS algorithm and an active noise reduction door, and the method comprises the following steps: obtaining noise data to be reduced through a preset microphone array; the noise data to be reduced comprises noise signals collected by each microphone in the microphone array; performing sound source positioning on the noise data to be reduced according to a wave speed formation algorithm to obtain a noise sound source position; obtaining an optimal noise reduction microphone position according to the noise sound source position, enabling a microphone corresponding to the optimal noise reduction microphone position in the microphone array, and generating a resonance inverse rotating wave based on a preset FxLMS algorithm to perform active noise reduction. On the basis of accurately positioning the noise sound source by loading the noise positioning algorithm through the microphone array, the improved FxLMS algorithm based on the Gompertz function adjusts the variable step and adopts the Adam algorithm to optimize the gradient calculation, so that the resonance inverse rotating wave output is controlled, and the flexibility and accuracy of the active noise reduction processing can be effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of active noise reduction technology, in particular to an active noise reduction method and device based on FxLMS algorithm and an active noise reduction door. BACKGROUND

[0002] It is beyond doubt that noise is harmful to human health, and therefore, various countries have set clear sound insulation standards for different buildings such as residences, schools, hospitals, hotels, offices and businesses.

[0003] Installing soundproof doors and windows is an effective way to reduce noise indoors and solve the problem of noise disturbance. However, most existing soundproof doors are based on passive noise reduction principles and are designed using a special sound insulation structure with multiple layers, such as: (1) Steel soundproof door: The inside is filled with damping and sound insulation materials such as sound insulation cotton to block the propagation of sound, and its function is to reduce resonance and isolate noise; (2) Filled core soundproof door: Glass wool or rock wool is filled in the door leaf core, and magnetic rubber strips are used to seal the door leaf joints; (3) Wrapped soundproof door: A layer of artificial leather is wrapped around the ordinary wooden door leaf, and rock wool is filled inside the artificial leather. The long artificial leather strip is nailed tightly with bubble nails, and the gaps around it are sealed tightly with sponge rubber strips; (4) Soundproof fireproof door: Rock wool and other sound-absorbing materials are embedded in the door leaf wooden frame, and the outside is covered with asbestos board, galvanized iron sheet and refractory fiber board. The gaps around it are sealed tightly with magnetic rubber strips. Although the application of existing noise reduction doors can achieve a certain degree of noise reduction effect, they only have good noise reduction effect on medium and high frequency noise, and the control effect on low frequency noise is limited. After traditional passive noise reduction, there are still many complex low frequency noises in the room, which continue to affect human health.

[0004] Although the application of active noise control (ANC) can solve the problem of reducing low frequency wideband noise, the FxLMS (Filtered-x Least Mean Square) algorithm still has the disadvantage of being unable to balance the convergence speed and steady-state error, and the precise positioning of the noise source has not been considered before noise reduction, making it difficult to achieve precise noise reduction effect. SUMMARY

[0005] The purpose of the present application is to provide an active noise reduction method based on FxLMS algorithm, which realizes precise positioning of noise source by configuring a microphone array to load a noise positioning algorithm, and combines an improved FxLMS algorithm that adjusts the variable step size based on the Gompertz function and optimizes the gradient calculation using the Adam algorithm to control the resonance inverse spin wave output. The application solves the application defects of existing active noise reduction control methods, effectively improves the adaptive flexibility of active noise reduction processing, and also improves the accuracy of noise reduction processing, avoids the generation of additional noise, and has high application value.

[0006] In order to achieve the above object, it is necessary to provide an active noise reduction method and an active noise reduction door based on an improved FxLMS algorithm in view of the above technical problems.

[0007] In a first aspect, an embodiment of the present application provides an active noise reduction method based on an FxLMS algorithm, the method comprising the following steps:

[0008] Obtaining noise data to be reduced by a preset microphone array; the noise data to be reduced comprises noise signals collected by each microphone in the microphone array;

[0009] Performing sound source positioning on the noise data to be reduced according to a wave velocity formation algorithm to obtain a noise sound source position;

[0010] Obtaining an optimal noise reduction microphone position according to the noise sound source position, enabling a microphone corresponding to the optimal noise reduction microphone position in the microphone array, and generating a resonant inverse rotational wave based on a preset FxLMS algorithm to perform active noise reduction.

[0011] Further, the microphone array is an elliptical 360-degree microphone array.

[0012] Further, the wave velocity formation algorithm comprises an SRP-PHAT algorithm.

[0013] Further, the step of performing sound source positioning on the noise data to be reduced according to the wave velocity formation algorithm to obtain a noise sound source position comprises:

[0014] Converting each noise signal in the noise data to be reduced into a corresponding noise frequency domain signal through fast Fourier transform;

[0015] Calculating a cross-correlation function spectrum of any two noise frequency domain signals, and performing inverse fast Fourier transform on the cross-correlation function spectrum to obtain a corresponding cross-correlation function;

[0016] Obtaining a total signal correlation value according to all cross-correlation functions, and obtaining a noise sound source position according to the total signal correlation value; the noise sound source position is represented as:

[0017]

[0018] In the formula,

[0019]

[0020] In the formula, represents the noise sound source position; Q represents a point in a preset search direction; q and P(q) represent a sound source point and a corresponding total signal correlation value, respectively. represents the cross-correlation function of the noise frequency domain signals corresponding to microphones l and m; M represents the total number of microphones in the microphone array; represents the n-th input of microphones l and m at the sound source point q.

[0021] Further, the step of obtaining the optimal noise reduction microphone position according to the noise sound source position comprises:

[0022] The distance between each microphone in the microphone array and the noise sound source position is calculated respectively, and the microphone position corresponding to the minimum distance is taken as the optimal noise reduction microphone position.

[0023] Further, the variable step size of the preset FxLMS algorithm is adjusted by the Gompertz function; the variable step size is represented as:

[0024]

[0025] wherein e(k) and μ[e(k)] represent the error signal at the error microphone and the corresponding step size in the preset FxLMS algorithm respectively; α, β and γ are pre-set constants.

[0026] Further, the step of generating the resonant inverse rotational wave based on the preset FxLMS algorithm comprises:

[0027] Based on the Adam optimization algorithm, the reference signal gradient estimated by the FRI filter through the secondary channel is optimized to obtain the optimal compensation signal.

[0028] According to the optimized reference signal, the coefficients of the adaptive noise reduction filter are updated; the adaptive noise reduction filter coefficients are represented as:

[0029]

[0030] wherein w l (k+1) and w l (k) represent the l-th coefficient of the adaptive noise reduction filter at k+1 and k time respectively; represents the optimal compensation signal compensated by the secondary channel at k-1 time; e(k) and μ[e(k)] represent the error signal at the error microphone at k time and the corresponding variable step size respectively.

[0031] Further, the step of obtaining the optimal compensation signal based on the Adam optimization algorithm, optimizing the reference signal gradient estimated by the FRI filter through the secondary channel comprises:

[0032] The estimated reference signal corresponding to the noise data to be reduced is input into the secondary channel FRI filter for filtering processing to obtain the reference signal gradient; the reference signal gradient is represented as:

[0033]

[0034] wherein x(k) represents the reference signal estimated at the kth moment; e(k) represents the error signal at the kth moment in the error microphone; represents the coefficient of the M-order FIR filter at the kth moment;

[0035] y(k-m) represents the resonance inverse wave output by the adaptive noise reduction filter at the k-mth moment;

[0036] According to the reference signal gradient, a first-order momentum and a second-order momentum corresponding to the reference signal gradient are calculated;

[0037] The first-order momentum and the second-order momentum are respectively modified, and the optimal compensation signal is obtained according to the modified first-order momentum and the modified second-order momentum; the optimal compensation signal is represented as:

[0038]

[0039] In the formula,

[0040]

[0041]

[0042] m k = β1·m k-1 +(1-β1)·x'(k),m0=0

[0043] v k = β2·v k-1 +(1-β2)·[x'(k)] 2 ,v0=0

[0044] wherein, represents the optimal compensation signal compensated by the secondary channel at the kth moment; and respectively represent the modified first-order momentum and the modified second-order momentum; m k and v k respectively represent the first-order momentum and the second-order momentum before modification; x'(k) represents the reference signal gradient estimated at the kth moment; α', β1, β2 and ε are all Adam optimization parameters preset in advance.

[0045] In a second aspect, an embodiment of the present application provides an active noise reduction device based on an FxLMS algorithm, the device comprising:

[0046] A noise collection module is configured to acquire noise data to be reduced by a preset microphone array

[0047] The sound source localization module is used to locate the sound source of the noise data to be denoised according to the wave velocity forming algorithm, so as to obtain the location of the noise source;

[0048] The noise reduction processing module is used to obtain the optimal noise reduction microphone position based on the noise source position, and to activate the microphone in the microphone array corresponding to the optimal noise reduction microphone position, and generate a resonant inverse spiral wave based on the preset FxLMS algorithm to perform active noise reduction.

[0049] Thirdly, embodiments of the present invention provide an active noise reduction door, the active noise reduction door including a door body and the aforementioned active noise reduction device; the active noise reduction device is installed on the door body.

[0050] Fourthly, embodiments of the present invention also provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.

[0051] Fifthly, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described method.

[0052] This application provides an active noise reduction method, apparatus, and active noise reduction gate based on the FxLMS algorithm. The method acquires noise data (including noise signals collected by each microphone in a preset microphone array) using a pre-defined microphone array. It then uses a wave velocity formation algorithm to locate the noise source, determines the optimal noise reduction microphone position, and activates the microphone corresponding to that position. Finally, it generates a resonant inverse spiral wave based on a pre-defined FxLMS algorithm for active noise reduction. Compared to existing technologies, this active noise reduction method based on the FxLMS algorithm, while achieving precise noise source localization through a optimized microphone array and noise localization algorithm, combines an improved FxLMS algorithm—using a Gompertz function to adjust the variable step size and the Adam algorithm to optimize gradient calculation—to control the resonant inverse spiral wave output. This effectively enhances the adaptive flexibility of active noise reduction processing while also improving its accuracy and avoiding the generation of additional noise, thus possessing high application value. Attached Figure Description

[0053] Figure 1 This is a schematic diagram of the active noise reduction principle used in this embodiment of the invention;

[0054] Figure 2 This is a schematic diagram of the existing main control noise control system;

[0055] Figure 3is a flowchart of the active noise reduction method based on the FxLMS algorithm in the embodiment of the application;

[0056] Figure 4 is a flowchart of the noise source position obtained according to the wave velocity forming algorithm in the embodiment of the application;

[0057] Figure 5 is an effect diagram of the sound source point sending sound waves in the far field of the sound waves received by the microphones l and m in the embodiment of the application;

[0058] Figure 6 is a principle diagram of the improved FxLMS algorithm in the embodiment of the application;

[0059] Figure 7 is a comparison experiment effect diagram of adjusting the variable step size by the Gompertz function and adjusting the step size by the Sigmoid function in the embodiment of the application;

[0060] Figure 8 is a structural diagram of the active noise reduction device based on the FxLMS algorithm in the embodiment of the application;

[0061] Figure 9 is a schematic diagram of the active noise reduction door of the active noise reduction device shown in the embodiment of the application; Figure 8

[0062] Figure 10 is an internal structure diagram of the computer device in the embodiment of the application. DETAILED DESCRIPTION

[0063] In order to make the purposes, technical solutions and beneficial effects of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. Obviously, the following described embodiments are part of the embodiments of the present application, and are only used to illustrate the present application, but not to limit the scope of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0064] The active noise reduction method based on the FxLMS algorithm provided by the present application belongs to the noise active control ANC (Active Noise Control) technology, which is based on the processing of the original noise sound wave signal shown in the figure, generates a resonant inverse rotating wave (anti-phase sound wave) to realize the noise sound wave cancellation, so as to compensate for the application defects of the passive noise control technology in the low-frequency noise control, solve the noise reduction principle of the low-frequency wideband noise prevention and control problem, and combine the existing application in the noise control field. Figure 1 Figure 2 ​​The FxLMS (Filtered-x Least Mean Square) algorithm shown in the main noise control system introduces a secondary path estimation correction error gradient, which is based on the improvement of the equalization convergence speed and the steady-state error direction. In addition, the method also incorporates the precise positioning of the sound source to assist in selecting the optimal position of the resonant inverse rotational wave output, further improving the noise hedging efficiency. The noise reduction method obtained can effectively improve the active noise reduction effect. It should be noted that, Figure 2 The main noise control system shown mainly includes a reference microphone, an error microphone, a cancellation speaker, and a controller. The controlled object (environmental noise) in the system is a primary sound stream generated by an initial noise source in a primary sound range. The controller will generate a control signal through an adaptive filter based on the noise signal collected by the reference microphone to drive the secondary speaker to generate a secondary sound stream (resonant inverse rotational wave) for controlling the primary noise. The noise cancellation effect is received by the error microphone and fed back to the controller for iterative processing. Through sufficient iteration, the error microphone ultimately receives the minimum sound wave to achieve the expected noise reduction effect.

[0065] The active noise reduction method based on the FxLMS algorithm provided by the present application can be applied to any scene that needs to use active noise reduction to prevent and control noise. In particular, when this method is used in the noise prevention design of a building door, outdoor noise data can be collected by a microphone array, and the sound source position can be accurately positioned by a noise positioning algorithm. After that, the microphone at the appropriate position outputs the resonant inverse rotational wave based on the FxLMS algorithm improved by variable step and gradient optimization to cancel the outdoor noise and achieve precise noise reduction, providing reliable technical support for people's demand for a comfortable living environment. The following embodiments will describe the active noise reduction method based on the FxLMS algorithm in detail.

[0066] In one embodiment, as shown in Figure 3 An active noise reduction method based on the FxLMS algorithm is provided, including the following steps:

[0067] S11, obtain noise data to be de-noised through a preset microphone array; the noise data to be de-noised includes noise signals collected by each microphone in the microphone array; wherein the microphone array can be understood as a device for collecting environmental noise, and in principle, any form of arrangement can be adopted, but considering the requirement of accurate positioning of sound sources involved in the de-noising process, the microphone array is preferably set as an elliptical 360-degree microphone array, which can realize 360-degree full-range collection of noise data in the current environment without dead angle, thereby ensuring the accuracy of the noise data to be de-noised and providing effective data support for subsequent sound source positioning and inverse rotational wave generation.

[0068] S12, position the sound source according to a wave velocity formation algorithm to obtain the position of the noise sound source; wherein the wave velocity formation algorithm can be understood as an algorithm for realizing sound source positioning based on the wave velocity formation principle, including various application algorithms such as Fixed Beamforming, Delay and Sum Beamforming (DSB), Minimum Variance Distortionless Response Beamforming (MVDR), and Linearly Constrained Minimum Variance (LCMV) Beamforming, and the SRP-PHAT (Steered-Response Power Phase Transform) algorithm is preferably adopted in the embodiment, which is not affected by the environment and can realize sound source positioning through multiple acoustic sensors (microphone arrays) in a complex environment, thereby ensuring the accuracy of noise sound source positioning.

[0069] When the SRP-PHAT algorithm is used to realize sound source direction positioning, the cross-correlation function of each pair of microphone array signals needs to be calculated in the frequency domain, and the frequency amplitude information is removed, only the phase information is retained. Specifically, as shown in Figure 4 The step of positioning the sound source according to the wave velocity formation algorithm to obtain the position of the noise sound source includes:

[0070] Each noise signal in the noise data to be de-noised is subjected to fast Fourier transform (FFT) to obtain the corresponding noise frequency domain signal;

[0071] The cross-correlation function spectrum of any two noise frequency domain signals is calculated, and the inverse fast Fourier transform (IFFT) of the cross-correlation function spectrum is performed to obtain the corresponding cross-correlation function; wherein the cross-correlation function is represented as:

[0072]

[0073] In the formula,

[0074]

[0075] Among them, X l (k) and X m (k) represents the noise signal x collected by microphones l and m, respectively. l (n) and x m (n) corresponds to the noise frequency domain signal in the frequency domain k; X represents m (k) conjugate; R lm (k) represents the noise frequency domain signal X. l (k) and X m The cross-correlation function spectrum of (k), where N represents the number of samples of a frame of signal acquired by the microphone array, n = 0, 1, 2, ..., N-1, k = 0, 1, 2, ..., N-1;

[0076] Based on all cross-correlation functions, the total correlation value of the signal is obtained, and based on the total correlation value, the location of the noise source is determined; the location of the noise source is represented as:

[0077]

[0078] In the formula,

[0079]

[0080] in, The noise source location is represented by Q; Q represents the point in the preset search direction; q and P(q) represent the source point and the corresponding total correlation value of the signal, respectively. The cross-correlation function of the noise frequency domain signals corresponding to microphones l and m is given; M is the total number of microphones in the microphone array. This represents the nth input from microphones l and m at the sound source point q;

[0081] Specifically, in practical applications, Represented as:

[0082]

[0083] In the formula,

[0084]

[0085] Among them, f S Indicates the sampling rate; like Figure 5 The figure shows the time difference between the arrival of the sound wave from the sound source point q at the microphones l and m, respectively; d represents the distance between the two microphones; θ represents the angle formed between the direction of sound wave propagation and the line connecting microphones l and m; and c represents the speed of sound.

[0086] S13. Based on the location of the noise source, obtain the optimal noise-canceling microphone position, and activate the microphone in the microphone array corresponding to the optimal noise-canceling microphone position. Generate a resonant inverse vortex wave based on a preset FxLMS algorithm for active noise reduction; wherein, the optimal noise-canceling microphone position can be understood as... Figure 2 The optimal position of the secondary loudspeaker is selected based on the principle that the closer it is to the noise source, the better the noise cancellation effect of the resonant inverse spiral wave emitted; specifically, the step of obtaining the optimal noise-canceling microphone position based on the noise source position includes:

[0087] The distance between each microphone in the microphone array and the noise source is calculated, and the microphone position corresponding to the minimum distance is taken as the optimal noise reduction microphone position. There are many ways to calculate the distance, such as Euclidean distance, Manhattan distance, and Chebyshev distance. The specific distance calculation method in this embodiment can be selected according to the actual application requirements, and no specific restrictions are imposed here.

[0088] In principle, such as Figure 6 The existing improved version of the FxLMS (Filtered-x Least Mean Square) algorithm, as shown, effectively solves the stability problem caused by the existence of secondary channels in the LMS algorithm by introducing secondary path estimation to correct the error gradient. It also has the advantages of simple implementation, low computational load, and stable performance, and can be applied to control the generation of resonant inverse spiral waves in this embodiment. However, considering the drawback of this algorithm in application where convergence speed and steady-state error cannot be simultaneously optimized, this embodiment preferably adopts an optimized variable step size adjustment method and introduces gradient optimization from deep neural networks to further improve the existing FxLMS algorithm, so as to better adapt to the dynamic variability of environmental noise, thereby ensuring the accuracy and reliability of the generated resonant inverse spiral wave signal and effectively improving the noise reduction effect. To facilitate a detailed explanation of the FxLMS (Filtered-x Least Mean Square) algorithm preferred in this embodiment, the following will first be based on... Figure 6 The processing principle of the existing improved version of the FxLMS (Filtered-x Least Mean Square) algorithm is explained:

[0089] Figure 6 The symbols in the table are as follows: d(k) represents primary noise, P(z) represents the primary channel, and S(z) represents the secondary channel. For secondary channel estimation, x(k) is the estimated reference signal, and y(k) is the secondary acoustic wave emitted by the controller. s (k) represents the sound pressure signal generated by the secondary source at the microphone, i.e., the secondary sound wave at the error microphone, y′ s(K) is the estimated secondary sound, e(k) is the output error of the error microphone, and W(z) is an adaptive filter;

[0090] At time k, the output y(k) of the adaptive filter W(z) is:

[0091]

[0092] where W l (k) is the coefficient of the L-order adaptive filter W(z) at time k. Correspondingly, the secondary sound y s (k) is:

[0093]

[0094] where s m is the coefficient of the M-order FIR filter S(z);

[0095] It is known that the error signal e(k) at the error microphone is expressed as:

[0096] e(k) = d(k) - y s (k) (3)

[0097] The filtered reference signal x'(k) at time k is defined as:

[0098]

[0099] where s is the coefficient of the M-order FIR filter ;

[0100] Then, the reference signal x(k) of the estimated primary noise d(k) is synthesized by equation (5) to obtain:

[0101]

[0102] And the coefficient of the adaptive filter W(z) at time k is updated by the FxLMS algorithm to:

[0103]

[0104] where μ is a fixed step or a convergence coefficient.

[0105] Considering that μ in equation (6) is one of the most critical parameters in the FxLMS algorithm, directly determining the algorithm's performance and noise reduction effect, and that the step size in the traditional FxLMS algorithm is preset and fixed, lacking a dynamic adaptation mechanism, it cannot reflect the dynamic variability of environmental noise and cannot adequately guarantee the reliability of adaptive filter coefficient adjustment, this embodiment preferably uses the Gompertz function to effectively adjust the variable step size in equation (6). That is, the variable step size of the preset FxLMS algorithm is adjusted by the Gompertz function; the variable step size is expressed as:

[0106]

[0107] Where e(k) and μ[e(k)] represent the error signal at the error microphone and the corresponding step size in the preset FxLMS algorithm, respectively; α, β and γ are preset constants.

[0108] It should be noted that the Gompertz function used in this embodiment is an sigmoid function. Applying it to active noise reduction by changing the step size μ in the FxLMS algorithm aligns well with the step size adjustment principle of adaptive filtering algorithms: during the initial convergence phase or when unknown system parameters change, the step size should be relatively large to ensure faster convergence and tracking of time-varying systems; conversely, after convergence, regardless of the magnitude of the interference signal at the main input, a small adjustment step size should be maintained to achieve minimal steady-state offset noise. Therefore, this algorithm effectively balances convergence speed and steady-state error, resulting in a faster system response and lower steady-state error.

[0109] Furthermore, to illustrate the application effect of the FxLMS algorithm with variable step size adjusted by the Gompertz function in this embodiment, a comparative experiment was conducted between the classic FxLMS algorithm with step size adjusted by the S-function (Sigmoid function) and the FxLMS algorithm with variable step size adjusted by the Gompertz function: when α = β = 1 and γ = 0.8 in the Gompertz function shown in equation (7), the following results were obtained: Figure 7 The comparison images shown are in the positive real number interval; based on Figure 7 It can be seen that, compared with the Sigmoid function, the Gompertz function, when the system error is large, also obtains a larger step size, just like the Sigmoid function, to ensure the system's rapid convergence. However, when the error is small, the Gompertz function is more cautious than the Sigmoid function, using a smaller step size. It is evident that the Gompertz function can coordinate the relationship between convergence speed and steady-state error, enabling the system to have faster response capability and smaller steady-state error, thereby enhancing the FxLMS algorithm's ability to adapt to noisy environments and improving the generation effect of resonant inverse spiral waves.

[0110] Further, considering that the gradient calculation in the existing FxLMS algorithm does not involve optimization improvement, in principle, the active noise control involves a non-convex function optimization problem as the deep neural network does, and in the application without using gradient optimization, it will inevitably limit the application effect of generating the resonance inverse spin wave by using the FxLMS algorithm. Based on this, in order to further improve the application effect of the FxLMS algorithm with the Gompertz function adjusting the variable step size, the embodiment combines it with the gradient optimization in the deep neural network, optimizes the gradient based on the Adam (Adaptive Moment Estimation) method, and obtains an active noise reduction method integrating a gradient dynamic adaptation mechanism. Specifically, the step of generating the resonance inverse spin wave based on the preset FxLMS algorithm comprises:

[0111] Based on the Adam optimization algorithm, the reference signal gradient estimated by the FRI filter of the secondary channel is optimized to obtain an optimal compensation signal; wherein the Adam optimization algorithm is a momentum-based algorithm, which can fully utilize the historical information of the gradient, combine the advantages of AdaGrad (Adaptive gradient algorithm) and RMSProp (Root Mean Square Propagation) methods, and be suitable for using the same learning rate for each parameter and independently adapting with the learning process. The easy-to-tune characteristics enable it to quickly obtain good results; specifically, the step of optimizing the reference signal gradient estimated by the FRI filter of the secondary channel based on the Adam optimization algorithm to obtain the optimal compensation signal comprises:

[0112] The estimated reference signal corresponding to the noise data to be reduced is input into the secondary channel FRI filter for filtering processing to obtain a reference signal gradient; the reference signal gradient is represented as:

[0113]

[0114] Wherein x(k) represents the estimated reference signal at the kth moment; e(k) represents the error signal at the kth moment at the error microphone; represents the coefficient of the M-order FIR filter at the kth moment; y(k-m) represents the resonance inverse spin wave output by the adaptive noise reduction filter at the k-mth moment;

[0115] According to the reference signal gradient, a first-order momentum and a second-order momentum corresponding thereto are calculated;

[0116] The first-order momentum and the second-order momentum are respectively corrected, and the optimal compensation signal is obtained according to the corrected first-order momentum and the second-order momentum; the optimal compensation signal is represented as:

[0117]

[0118] In the formula,

[0119]

[0120]

[0121] m k = β1·m k-1 + (1-β1)·x'(k), m0=0 (9)

[0122] v k = β2·v k-1 + (1-β2)·[x'(k)] 2 , v0=0 (10)

[0123] wherein, represents the optimal compensation signal compensated by the secondary channel at the kth moment; and respectively represent the first-order momentum and the second-order momentum after correction; m k and v k respectively represent the first-order momentum and the second-order momentum before correction; x'(k) represents the reference signal gradient estimated at the kth moment; α', β1, β2 and ε are all preset Adam optimization parameters.

[0124] It should be noted that the Adam optimization algorithm has the advantages of simple implementation, high efficiency, low memory requirement, parameter update not affected by gradient scaling transformation, suitable for unstable objective functions, and suitable for gradient sparsity or large noise in gradient. In this embodiment, the application of the Adam optimization algorithm in the ANC field to improve the active noise reduction effect is a precedent. Through the way of first-order moment estimation and second-order moment estimation of the gradient, adaptive learning is realized, which can effectively improve the processing efficiency of the FxLMS algorithm. The specific description is as follows:

[0125] (1) First-order exponential smoothing of historical gradient;

[0126] The first-order exponential smoothing of the historical gradient (also called the first-order moment weighted mean), i.e., formula (9) can be understood as a weighted average of the historical gradient, and the farther the distance from the current moment, the smaller the weight. Here, it is similar to the L1 norm of the historical gradient, and this idea comes from the momentum gradient descent algorithm. If the weight parameter is updated, there is a large shock in a certain dimension (the gradient update value alternates between positive and negative), then using the first-order exponential smoothing of the current gradient (canceling the positive and negative of the historical gradient), a smaller gradient value can be obtained. If the variable does not have a large shock in a certain dimension when it is updated, and there is a stable large gradient update in a certain direction, then a smaller value will be obtained after the first-order exponential smoothing. That is, the first-order exponential smoothing of the historical gradient is to use the historical gradient information to correct (eliminate the shock of the variable update) the current gradient, thereby obtaining a stable gradient update value.

[0127] (2) The first-order exponential smoothing of the historical gradient square;

[0128] The first-order exponential smoothing of the historical gradient square (also called the second-order moment weighted mean), i.e., formula (10) can be understood as a weighted average of the historical gradient square, and the farther the distance from the current moment, the smaller the weight. Here, it is similar to the L2 norm of the historical gradient, and this idea comes from the adaptive learning rate gradient descent algorithm. In the iteration process, if a certain dimension is always updated with a very small gradient, it proves that the gradient transformation in this direction is stable, so the learning rate can be increased to update in this dimension with a larger learning rate. A certain dimension is always updated with a very large gradient in the iteration process, which proves that the gradient transformation in this direction is more intense (unstable), so the learning rate can be reduced to update in this dimension with a smaller learning rate. The first-order exponential smoothing of the historical gradient square can also solve the problem of gradient sparsity; frequent updates of the gradient will be assigned a smaller learning rate, while sparse gradients will be assigned a larger learning rate. Through the above mechanism, in the scene of sparse data distribution, the information of sparse gradient can be better utilized, and the standard gradient descent algorithm can be more effectively converged.

[0129] Therefore, the use of the Adam optimization algorithm can update the variable according to the shock of the historical gradient and the true historical gradient after filtering the shock, which has strong adaptability and can effectively improve the noise reduction effect based on ANC.

[0130] According to the optimized reference signal, the coefficients of the adaptive noise reduction filter are updated; the adaptive noise reduction filter coefficients are represented as:

[0131]

[0132] where w l (k+1) and w l(k) respectively represent the 1th coefficient of the adaptive noise reduction filter at k+1 and k time; represents the optimal compensation signal compensated by the secondary path at k-1 time; e(k) and μ[e(k)] respectively represent the error signal at the error microphone at k time and the corresponding variable step size.

[0133] The embodiment of the present application adopts a preset elliptical 360-degree microphone array to obtain noise data to be reduced, which includes noise signals collected by each microphone in the microphone array, performs sound source positioning on the noise data to be reduced according to a wave speed formation algorithm, obtains a noise sound source position, obtains an optimal noise reduction microphone position according to the noise sound source position, and enables a microphone corresponding to the optimal noise reduction microphone position in the microphone array, generates a resonant inverse rotational wave based on an FxLMS algorithm with variable step size and gradient optimization to perform active noise reduction, and effectively improves the adaptive flexibility of active noise reduction processing while improving the accuracy of noise reduction processing, avoids the generation of additional noise, and has high application value.

[0134] It should be noted that although each step in the above flowchart is displayed in sequence according to the arrow, these steps are not necessarily executed in the order indicated by the arrow. Unless otherwise specified herein, the execution of these steps has no strict order limitation, and these steps can be executed in other orders.

[0135] In one embodiment, as shown in Figure 8 An active noise reduction device based on the FxLMS algorithm is provided, and the device comprises:

[0136] A noise collection module 1 is configured to obtain noise data to be reduced through a preset microphone array

[0137] A sound source positioning module 2 is configured to perform sound source positioning on the noise data to be reduced according to a wave speed formation algorithm to obtain a noise sound source position.

[0138] A noise reduction processing module 3 is configured to obtain an optimal noise reduction microphone position according to the noise sound source position, enable a microphone corresponding to the optimal noise reduction microphone position in the microphone array, generate a resonant inverse rotational wave based on a preset FxLMS algorithm, and perform active noise reduction.

[0139] The specific limitation of the active noise reduction device based on the FxLMS algorithm can refer to the limitation of the active noise reduction method based on the FxLMS algorithm in the foregoing, and the corresponding technical effect can also be obtained equally, which will not be repeated here. Each module in the active noise reduction device based on the FxLMS algorithm can be realized by software, hardware, and a combination thereof in whole or in part. The above-mentioned modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to call and execute the operations corresponding to each module by the processor.

[0140] In one embodiment, as shown in Figure 9 An active noise reduction door is provided, which includes a door body and Figure 8 The active noise reduction device is installed on the door body.

[0141] It should be noted that the specific position of each internal module in the active noise reduction device embedded in the building door can be selected according to actual needs. In actual application, after the active noise reduction door opens the active noise reduction device, the noise collection module can obtain the noise data to be reduced in real time through the preset microphone array, and the sound source positioning module can perform sound source positioning on the collected noise data to be reduced according to the wave speed formation algorithm to obtain the noise source position and provide it for the noise reduction processing module. According to the noise source position, the optimal noise reduction microphone position is obtained, and the microphone corresponding to the optimal noise reduction microphone position in the microphone array is enabled. Based on the improved FxLMS algorithm using the Gompertz function to adjust the variable step and using the Adam algorithm to optimize the gradient calculation, the anti-noise wave of the resonance inverse spin is generated to cancel the outdoor noise, active noise reduction is performed, precise noise reduction is achieved, additional noise is avoided, the application demand of people enjoying the comfortable living environment without noise pollution can be effectively met, and high practical value is obtained.

[0142] Figure 10 An internal structure diagram of a computer device in one embodiment is shown, which can be a terminal or a server. As Figure 10As shown, the computer device includes a processor, a memory, a network interface, a display and an input device connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is configured to communicate with an external terminal through a network connection. The computer program is executed by the processor to implement the active noise reduction method based on the FxLMS algorithm. The display of the computer device can be a liquid crystal display or an electronic ink display. The input device of the computer device can be a touch layer overlaid on the display, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0143] Those skilled in the art can understand that, Figure 10 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. A specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have the same component arrangement.

[0144] In one embodiment, a computer device is provided, including a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor executes the computer program to implement the steps of the above method.

[0145] In one embodiment, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by the processor to implement the steps of the above method.

[0146] In summary, the active noise reduction method and the active noise reduction door provided by the embodiment of the application achieve the following technical solutions: a preset 360-degree microphone array in the shape of an ellipse is used to obtain noise data to be reduced, including noise signals collected by each microphone in the microphone array; sound source positioning is performed on the noise data to be reduced according to a wave speed formation algorithm; after the position of a noise sound source is obtained, an optimal noise reduction microphone position is obtained according to the position of the noise sound source, and a microphone corresponding to the optimal noise reduction microphone position in the microphone array is enabled; and the FxLMS algorithm based on variable steps and gradient optimization is used to generate a resonant inverse rotating wave for active noise reduction. Based on the optimal positioning of the noise sound source by using the noise positioning algorithm of the microphone array, the improved FxLMS algorithm based on the adjustment of the variable steps according to the Gompertz function and the optimization of the gradient calculation according to the Adam algorithm is used to control the output of the resonant inverse rotating wave, so that the adaptive flexibility of the active noise reduction process is effectively improved, the accuracy of the noise reduction process is improved, the generation of additional noise is avoided, and the application value is high. In addition, the active noise reduction method is applied to the design of the noise reduction door, so that the influence of outdoor noise on the indoor crowd can be effectively reduced, and reliable technical support is provided for the demand of people to enjoy a comfortable living environment.

[0147] Each of the embodiments in the specification is described in a progressive manner, and the same or similar parts of each embodiment can be referred to each other, and each embodiment mainly describes the difference from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the related parts can be referred to the part of the method embodiment. It should be noted that the technical features of the above embodiments can be combined arbitrarily, and in order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the description.

[0148] The above-described embodiments only express several preferred embodiments of the application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent. It should be noted that for ordinary skilled in the art, without departing from the technical principles of the application, a number of improvements and replacements can be made, and these improvements and replacements should be considered as the protection scope of the application. Therefore, the protection scope of the patent of the application should be subject to the protection scope of the claims.

Claims

1. An active noise reduction method based on the FxLMS algorithm, characterized in that, The method includes the following steps: The noise data to be reduced is acquired through a preset microphone array; the noise data to be reduced includes noise signals collected by each microphone in the microphone array. Based on the wave velocity formation algorithm, the noise source of the noise data to be denoised is located to obtain the location of the noise source. Based on the location of the noise source, the optimal noise-canceling microphone position is obtained, and the microphone in the microphone array corresponding to the optimal noise-canceling microphone position is activated. A resonant inverse spiral wave is generated based on the preset FxLMS algorithm to perform active noise cancellation. The variable step size of the preset FxLMS algorithm is adjusted by the Gompertz function. The step of generating a resonant inverse cyclotron wave based on a preset FxLMS algorithm includes: Based on the Adam optimization algorithm, the gradient of the reference signal estimated by the FRI filter through the secondary channel is optimized to obtain the optimal compensation signal, including: The estimated reference signal corresponding to the noise data to be denoised is input into the secondary channel FRI filter for filtering to obtain the reference signal gradient. Calculate the corresponding first-order momentum and second-order momentum based on the gradient of the reference signal; The first-order momentum and the second-order momentum are corrected respectively, and the optimal compensation signal is obtained based on the corrected first-order momentum and second-order momentum. The coefficients of the adaptive noise reduction filter are updated based on the optimal compensation signal.

2. The active noise reduction method based on the FxLMS algorithm as described in claim 1, characterized in that, The microphone array is an elliptical 360-degree microphone array.

3. The active noise reduction method based on the FxLMS algorithm as described in claim 1, characterized in that, The wave velocity formation algorithm includes the SRP-PHAT algorithm.

4. The active noise reduction method based on the FxLMS algorithm as described in claim 3, characterized in that, The step of locating the noise source in the noise data to be denoised according to the wave velocity formation algorithm to obtain the location of the noise source includes: Each noise signal in the noise data to be denoised is obtained by fast Fourier transform to obtain the corresponding noise frequency domain signal. Calculate the cross-correlation function spectrum of any two noise frequency domain signals, and perform an inverse fast Fourier transform on the cross-correlation function spectrum to obtain the corresponding cross-correlation function; Based on all cross-correlation functions, the total correlation value of the signal is obtained, and based on the total correlation value, the location of the noise source is determined; the location of the noise source is represented as: In the formula, in, Indicates the location of the noise source; This indicates a point in the preset search direction; and These represent the sound source point and the corresponding total correlation value of the signal, respectively. microphone and The cross-correlation function corresponding to the noise frequency domain signal; This represents the total number of microphones in the microphone array; microphone l and m At the source of the sound q The nth input.

5. The active noise reduction method based on the FxLMS algorithm as described in claim 1, characterized in that, The step of obtaining the optimal noise-canceling microphone position based on the noise source position includes: Calculate the distance between each microphone in the microphone array and the location of the noise source, and take the microphone position corresponding to the minimum distance as the optimal noise reduction microphone position.

6. The active noise reduction method based on the FxLMS algorithm as described in claim 1, characterized in that, The variable step size is expressed as: in, and These represent the error signal at the error microphone and the corresponding step size in the preset FxLMS algorithm; , and These are pre-set constants.

7. The active noise reduction method based on the FxLMS algorithm as described in claim 1, characterized in that, The coefficients of the adaptive noise reduction filter are expressed as follows: in, and These represent the adaptive noise reduction filter in... k +1 and k The first moment l One coefficient; Indicates the first k The optimal compensation signal for compensation using the secondary channel at time -1; and They represent the first k The timing error is the error signal at the microphone and the corresponding variable step size.

8. The active noise reduction method based on the FxLMS algorithm as described in claim 1, characterized in that, The gradient of the reference signal is expressed as: in, This represents the reference signal estimated at time k; Indicates the first k The timing error signal at the microphone; This represents the coefficients of the M-order FIR filter at time k; This represents the resonant inverse cyclotron wave output by the adaptive noise reduction filter at time k. The optimal compensation signal is represented as follows: In the formula, in ， Indicates the first k The optimal compensation signal that utilizes the secondary channel for compensation at all times; and These represent the corrected first-order momentum and second-order momentum, respectively; and These represent the first-order momentum and second-order momentum before the correction, respectively; Indicates the first k The gradient of the reference signal obtained from the time-mapping estimation; and All of these are pre-set Adam optimization parameters.

9. An active noise reduction device based on the FxLMS algorithm, characterized in that, The apparatus for applying the active noise reduction method based on the FxLMS algorithm as described in claim 1 includes: The noise acquisition module is used to acquire noise data to be reduced through a preset microphone array. The sound source localization module is used to locate the sound source of the noise data to be denoised according to the wave velocity forming algorithm, so as to obtain the location of the noise source; The noise reduction processing module is used to obtain the optimal noise reduction microphone position based on the noise source position, and to activate the microphone in the microphone array corresponding to the optimal noise reduction microphone position, and generate a resonant inverse spiral wave based on the preset FxLMS algorithm to perform active noise reduction.

10. An active noise cancellation gate, characterized in that, The active noise reduction door includes a door body and an active noise reduction device as described in claim 9; the active noise reduction device is installed on the door body.

Images