Short-time-delay active noise control method and system

Abstract

The invention discloses a short-time-delay active noise control method and system. The method comprises: acquiring, by a reference sensor, a reference signal at a noise source; the reference signal is represented by a discrete time sequence; inputting the reference signals of a preset number of continuous time steps into a long-short term memory (LSTM) neural network model, and predicting a predicted reference signal corresponding to the next time step through the LSTM neural network model; inputting the prediction reference signal into an adaptive filter to obtain a first offset signal output by the adaptive filter, and updating a weight coefficient vector of the adaptive filter; and outputting the first offset signal through the loudspeaker, so that the second offset signal and the target noise are superposed and offset, and active noise control is realized. By predicting the future sample of the reference signal and performing active noise reduction processing based on the prediction result, the delay of the system can be shortened to a certain extent, and the noise reduction performance of the ANC system is more effectively improved.

Description

Technical Field

[0001] This application relates to the technical field of signal processing, and particularly to a short-delay active noise control method and system. Background Art

[0002] Active Noise Control (ANC) technology collects a reference signal through a reference sensor. After being processed by an ANC controller, a reverse sound wave is emitted through a loudspeaker and superimposed with the target signal to cancel it out, which has a significant noise reduction effect on low-frequency noise and is widely used in fields such as transportation, household appliances, and industrial production.

[0003] However, in the existing ANC systems, there is a time delay between the input and output of the signal, which will cause additional phase shift of the signal, resulting in a reduction in the noise reduction bandwidth and affecting the noise reduction effect of the ANC system. Summary of the Invention

[0004] In view of this, this application provides a short-delay active noise control method and system, which can reduce the time delay of ANC and improve the noise reduction performance of ANC.

[0005] To solve the above problems, the technical solutions provided in this application are as follows:

[0006] The first aspect of this application provides a short-delay active noise control method. The noise emitted by a noise source is transmitted through a primary path to a region to be noise-reduced to form a target noise. The method includes:

[0007] Collecting a reference signal at the noise source by a reference sensor; the reference signal is represented by a discrete time series;

[0008] Inputting the reference signals of a continuous preset number of time steps into a Long Short-Term Memory (LSTM) neural network model, and predicting the predicted reference signal corresponding to the next time step through the LSTM neural network model;

[0009] Inputting the predicted reference signal into an adaptive filter, obtaining a first cancellation signal output by the adaptive filter, and updating the weight coefficient vector of the adaptive filter; outputting the first cancellation signal through a loudspeaker;

[0010] Superimposing and canceling the second cancellation signal with the target noise to achieve active noise control; where the second cancellation signal is the signal formed in the region to be noise-reduced after the first cancellation signal is transmitted through a secondary path.

[0011] Preferably, the LSTM neural network model includes LSTM neural network units; the LSTM neural network units specifically include an input gate, a forget gate, and an output gate;

[0012] Input the reference signal of a continuous preset number of time steps into the long short-term memory (LSTM) neural network model, and predict the predicted reference signal corresponding to the next time step through the LSTM neural network model. Specifically, it includes:

[0013] Use the reference signals of a continuous preset number of discrete time series as input signals; the input gate filters the input signals and obtains the cell state;

[0014] The forget gate discards the non-critical information in the historical information and updates the cell state;

[0015] The output gate determines the output part of the updated cell state;

[0016] Determine the output information through the output part and the updated cell state;

[0017] The fully connected layer obtains the predicted reference signal according to the output information and the output weight matrix.

[0018] Preferably, input the predicted reference signal into an adaptive filter to obtain the first cancellation signal output by the adaptive filter. Specifically, it includes:

[0019] Construct the reference signal vector of the adaptive filter according to the predicted reference signal and the length of the adaptive filter;

[0020] Obtain the first cancellation signal according to the reference signal vector and the weight coefficient vector.

[0021] Preferably, update the weight coefficient vector of the adaptive filter. Specifically, it includes:

[0022] Collect the error signal remaining after the target noise and the second cancellation signal are superimposed and canceled;

[0023] Update the weight coefficient vector according to the error signal, the weight update step size, the length of the adaptive filter, and the filtered reference signal.

[0024] Preferably, the adaptive filter specifically includes a finite impulse response (FIR) filter and a weight update module.

[0025] In the second aspect of the present application, a short-delay active noise control system is provided. The noise emitted by the noise source is transmitted to the area to be noise-reduced through the primary path to form the target noise. The system includes: a reference sensor, an ANC controller, and a speaker;

[0026] The reference sensor is used to collect the reference signal at the noise source; the reference signal is represented by a discrete time series;

[0027] An ANC controller is used to input reference signals of a continuous preset number of time steps into a long short-term memory (LSTM) neural network model, and predict the predicted reference signal corresponding to the next time step through the LSTM neural network model; according to the predicted reference signal, obtain the first cancellation signal output by the adaptive filter, and update the weight coefficient vector of the adaptive filter.

[0028] A loudspeaker is used to output the first cancellation signal, so that the second cancellation signal is superimposed and cancelled with the target noise to achieve active noise control; wherein, the second cancellation signal is a signal formed in the area to be noise-reduced after the first cancellation signal is transmitted through the secondary path.

[0029] Preferably, the LSTM neural network model includes LSTM neural network units; the LSTM neural network units specifically include an input gate, a forget gate, and an output gate.

[0030] The ANC controller is specifically used to use the reference signals of a continuous preset number of time steps as input signals; filter the input signals through the input gate and obtain the cell state.

[0031] Discard the non-critical information in the historical information through the forget gate and update the cell state.

[0032] Determine the output part of the updated cell state through the output gate.

[0033] Determine the output information through the output part and the updated cell state.

[0034] Obtain the predicted reference signal through the fully connected layer according to the output information and the output weight matrix.

[0035] Preferably, the ANC controller is specifically used to construct a reference signal vector of the adaptive filter according to the predicted reference signal and the length of the adaptive filter; obtain the first cancellation signal according to the reference signal vector and the weight coefficient vector.

[0036] Preferably, the system further includes an error sensor; the error sensor is specifically used to collect the error signal remaining after the target noise and the second cancellation signal are superimposed and cancelled; the ANC controller is used to update the weight coefficient vector according to the error signal, the weight update step size, the length of the adaptive filter, and the filtering reference signal.

[0037] Preferably, the adaptive filter specifically includes a finite impulse response (FIR) filter and a weight update module.

[0038] Thus, the present application has the following beneficial effects:

[0039] The short-delay active noise control method provided by the embodiments of the present application is as follows: The noise emitted by the noise source is transmitted through the primary path to the area to be noise-reduced to form the target noise, and a reference signal is collected by the reference sensor at the noise source; the reference signal is represented by a discrete time series; the reference signals of a continuous preset number of time steps are input into the long short-term memory (LSTM) neural network model, and the LSTM neural network model predicts the predicted reference signal corresponding to the next time step; the predicted reference signal is input into the adaptive filter to obtain the first cancellation signal output by the adaptive filter, and the weight coefficient vector of the adaptive filter is updated; the first cancellation signal is output through the speaker, so that the second cancellation signal is superimposed and cancelled with the reference signal to achieve active noise control; wherein, the second cancellation signal is the signal formed in the area to be noise-reduced after the first cancellation signal is transmitted through the secondary path. The embodiments of the present application predict the future samples of the reference signal, obtain the cancellation signal through the adaptive filter based on the predicted reference signal and update the filter weight coefficient vector. By predicting the future samples of the reference signal and performing active noise reduction processing based on the prediction results, the system delay can be shortened to a certain extent, and the noise reduction performance of the ANC system can be improved more effectively. Description of the Drawings

[0040] Figure 1 It is a flowchart of a short-delay active noise control method provided by the embodiments of the present application;

[0041] Figure 2 It is a processing flowchart of an LSTM neural network model provided by the embodiments of the present application;

[0042] Figure 3 It is a processing flowchart of an adaptive filter provided by the embodiments of the present application;

[0043] Figure 4 It is a noise data diagram of a duct fan provided by the embodiments of the present application;

[0044] Figure 5 It is an amplitude-frequency response curve diagram of a secondary path provided by the embodiments of the present application;

[0045] Figure 6 It is a phase-frequency response curve diagram of a secondary path provided by the embodiments of the present application;

[0046] Figure 7 It is an effect diagram of a predicted reference signal provided by the embodiments of the present application;

[0047] Figure 8 It is a schematic diagram of a short-delay active noise control system provided by the embodiments of the present application;

[0048] Figure 9 It is a module control schematic diagram of a short-delay active noise control system provided by the embodiments of the present application. Detailed implementation manners

[0049] In an ANC system, there is a time delay between the input and output of a signal; specifically, it includes the response time delay of a speaker, the sampling time delay of a reference sensor, and the control time delay of an ANC controller. They will cause additional phase shifts of the signal, reduce the noise reduction bandwidth that the ANC system can achieve, and thus affect the overall noise reduction effect.

[0050] The time delay of the reference sensor is relatively small, and the impact is relatively small, which is generally ignored. The time delay of the speaker is determined by its inherent factors such as size and structure, and it is usually difficult to change; however, shortening the control time delay of the ANC controller is achievable.

[0051] The time delay of the ANC controller mainly includes the hardware time delays of an analog-to-digital converter / digital-to-analog converter, an anti-aliasing filter / reconstruction filter, and a power amplifier module, as well as the software time delay of the operation of the ANC algorithm. The time delay of the hardware part is basically unchanged, while the time delay of the software part is determined by the response frequency of a Digital Signal Processors (DSP) chip.

[0052] In recent years, the response frequency of the DSP chip has been significantly improved. However, higher speed will bring the negative impact of high power consumption. Therefore, the system delay reduced by increasing the response frequency of the DSP chip is also limited. To maximize the noise reduction performance of the system, this application provides a short-time-delay active noise control method, which can further shorten the delay through an algorithm.

[0053] The embodiments of this application will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.

[0054] See Figure 1 , which is a flowchart of a short-time-delay active noise control method provided by an embodiment of this application.

[0055] For the short-time-delay active noise control method provided by the embodiment of this application, the noise emitted by a noise source is transmitted to a region to be noise-reduced through a primary path to form a target noise.

[0056] This method includes:

[0057] S101: Collect a reference signal at the noise source by a reference sensor.

[0058] The collected reference signal is represented by a discrete time series.

[0059] The reference signal at time n is expressed as follows:

[0060] x(n) = [x 1 (n)x 2 (n)…x D(n)] T

[0061] where n is a discrete-time exponent, n = 0, 1, 2, 3, …, x D (n) represents a reference signal with D-dimensional sampling data at time n. The D-dimensional sampling data refers to D types or dimensions of reference data with high coherence to the target noise, such as vibration signals, acoustic signals, etc.

[0062] S102: Input the reference signals of a continuous preset number of time steps into a long short-term memory (LSTM) neural network model, and predict the predicted reference signal corresponding to the next time step through the LSTM neural network model.

[0063] Its input time series data is represented as follows:

[0064] X p (n)=[x(n - K + 1) x(n - K + 2) … x(n)]

[0065] where X P (n) represents the model input with K time steps at time n. When n ≥ K - 1, the LSTM neural network model uses the input time series data X p (n) to predict the reference signal value at the next moment, that is, to predict the sampling data of the next time step through the sampling data of the previous K time steps.

[0066] S103: Input the predicted reference signal into an adaptive filter, obtain the first cancellation signal output by the adaptive filter, and update the weight coefficient vector of the adaptive filter.

[0067] Preferably, the adaptive filter includes a finite impulse response (FIR) filter and a weight update module. The weight update module includes an adaptive algorithm, and the FIR filter has the advantage of high stability. Of course, the type of the adaptive filter is not specifically limited in this application. In other embodiments, an infinite impulse response (IIR) filter can also be selected.

[0068] S104: Output the first cancellation signal through a speaker, so that the second cancellation signal is superimposed and cancelled with the target noise, and active noise control is completed.

[0069] where the second cancellation signal is the signal formed in the area to be noise-reduced after the first cancellation signal is transmitted through the secondary path; the target noise is the noise signal when the noise emitted by the noise source is transmitted to the area to be noise-reduced through the primary path.

[0070] The short-delay active noise control method provided by the embodiment of the present application is as follows: the reference signal emitted by the noise source is transmitted through the primary path to the area to be noise-reduced to form a target noise, and the reference signal is collected by the reference sensor; the reference signal is represented by a discrete time series; the reference signals of continuously preset number of time steps are input into the long short-term memory (LSTM) neural network model, and the predicted reference signal corresponding to the next time step is predicted through the LSTM neural network model; the predicted reference signal is input into the adaptive filter to obtain the first cancellation signal output by the adaptive filter, and the weight coefficient vector of the adaptive filter is updated; the first cancellation signal is output through the speaker, so that the second cancellation signal is superimposed and cancelled with the reference signal to achieve active noise control; wherein, the second cancellation signal is the signal formed in the area to be noise-reduced after the first cancellation signal is transmitted through the secondary path. By predicting the future samples of the reference signal and performing active noise reduction processing based on the prediction result, the embodiment of the present application can shorten the system delay to a certain extent and more effectively improve the noise reduction performance of the ANC system.

[0071] The following will specifically describe the processing process of the LSTM neural network model for the reference signal with reference to the accompanying drawings.

[0072] See Figure 2 , which is the processing flow chart of an LSTM neural network model provided by the embodiment of the present application.

[0073] Among them, the LSTM neural network model consists of an input layer with D nodes, a hidden layer with K LSTM neural network units, and a fully connected layer. The LSTM neural network unit includes an input gate, a forget gate, and an output gate.

[0074] S201: Take the reference signals of continuously preset number of time steps as the input signals; the input gate filters the input signals and obtains the cell state.

[0075] Assume that x t represents the input of the LSTM neural network unit in the hidden layer at time t, t = 1,..., K, x1 = x(n - K + 1), x2 = x(n - K + 2),..., x K = x(n).

[0076] For the convenience of introduction, the parameters involved in the operations of each layer of the LSTM neural network model are listed below first. For the LSTM neural network model, W f , W c , W i and W o represent the recurrent layer weight matrices, U f , U c , U i and U o represent the input layer weight matrices, b f , bc , b i and b o denotes the bias vector; h t denotes the output vector of the LSTM neural network unit, i.e., the hidden layer state.

[0077] The input gate filters the new information of the input signal, and the process is as follows:

[0078] i t = σ(U i x t + W i h t-1 + b i )

[0079] In the formula, σ is the sigmoid activation function.

[0080] Obtain the cell state

[0081]

[0082] In the formula, tanh is the hyperbolic tangent function.

[0083] S202: The forget gate discards the non-critical information in the historical information and updates the cell state.

[0084] The process carried out by the forget gate is as follows:

[0085] f t = σ(U f x t + W f h t-1 + b f )

[0086] The old cell state C t-1 is updated to the new cell state C t The process is as follows:

[0087]

[0088] In the formula, ⊙ represents the element-wise multiplication operation of matrices.

[0089] S203: The output gate determines the output part of the updated cell state.

[0090] O t = σ(U o x t + W o h t-1 + b o )

[0091] S204: Determine the output information based on the output part and the updated cell state.

[0092] Partially select the current cell state C by the tanh function t and determine the output information in combination with the output gate:

[0093] h t = O t ⊙tanh(C t )

[0094] S205: The fully connected layer obtains the prediction reference signal according to the output information and the output weight matrix.

[0095]

[0096] In the formula, W out is the output weight matrix, and h t＝K is the hidden state at t = K, which is the predicted value of the reference signal at the (n + 1)-th moment by the LSTM neural network model.

[0097] The following specifically describes the processing process of the adaptive filter with reference to the accompanying drawings.

[0098] See Figure 3 , which is a processing flow chart of an adaptive filter provided by an embodiment of the present application.

[0099] S301: Construct the reference signal vector of the adaptive filter according to the prediction reference signal and the length of the adaptive filter.

[0100] The prediction reference signal adopted by the adaptive filter module is expressed as follows:

[0101]

[0102] Among them, x q (n) means that when the collected reference signal is not sufficient to predict the reference signal at the next moment, the directly collected reference signal is input into the adaptive filter.

[0103] The number of adaptive filters is consistent with the dimension D of the reference signal. Among them, the reference signal vector of the q-th filter is:

[0104]

[0105] Among them, L is the length of the filter.

[0106] S302: Obtain the first cancellation signal according to the reference signal vector and the weight coefficient vector.

[0107] Calculate the first cancellation signal:

[0108]

[0109] Wherein, is the weight coefficient vector of the q-th filter.

[0110] S303: Collect the error signal remaining after the target noise and the second cancellation signal are superimposed and cancelled.

[0111] The result of the error signal is as follows:

[0112] e(n) = d(n) - y s (n)

[0113] Wherein, d(n) is the target noise, and y s (n) is the second cancellation signal.

[0114] S304: Update the weight coefficient vector according to the error signal, the weight update step size, the length of the adaptive filter, and the filter reference signal.

[0115] As a possible implementation manner, step S304 can update the weight coefficient vector of the adaptive filter based on the FXLMS algorithm.

[0116] The update result of the weight coefficient vector of the adaptive filter is as follows:

[0117]

[0118] Wherein, μ is the weight update step size, e(n) is the error signal, is the filter reference signal vector; is the filter reference signal, which can be calculated by the following formula:

[0119]

[0120] Wherein, are the coefficients of the M-1 order FIR filter, which are obtained by modeling the secondary path.

[0121] Continuously repeat the above process to effectively control the target noise.

[0122] In order to more intuitively reflect the technical effect brought by the present application, this embodiment performs active control on the noise of the duct fan.

[0123] See Figure 4 , this figure is the duct fan noise data diagram provided by the embodiment of the present application. The abscissa represents time, and the ordinate represents the amplitude of the collected noise data.

[0124] The LSTM neural network model is trained using the collected noise of the duct fan.

[0125] Specifically, an LSTM neural network model with an input feature dimension D of 1, a hidden layer number of 1, a hidden state dimension of 4, an LSTM neural network unit number of 8, and a fully connected layer output dimension of 1 is adopted; an FIR filter with a filter length L of 300 is used; and the weight update step size μ is 0.001.

[0126] The sampling frequency of ANC control is set to 2000 Hz; an FIR filter with an order (M - 1) of 127 is used to model the secondary path. Refer to Figure 5 , which is the amplitude - frequency response curve graph of a secondary path provided by an embodiment of the present application; refer to Figure 6 , which is the phase - frequency response curve graph of a secondary path provided by an embodiment of the present application. Figure 5 and Figure 6 reflect the characteristic curves after the FIR filter models the secondary path.

[0127] Refer to Figure 7 , which is the effect graph of a predicted reference signal provided by an embodiment of the present application.

[0128] In the figure, the dashed line represents the predicted reference signal, and the solid line represents the actually collected reference signal. It can be seen that the predicted reference signal is generally close to the sample value of the actually collected reference signal. According to the display of the actually measured sound level meter, for the active noise reduction of the above - mentioned duct fan noise, the total noise reduction amount of the traditional active noise reduction method is 1.8 dB, while the short - delay active noise control method provided by the embodiment of the present application can achieve a total noise reduction amount of 2.8 dB, having more prominent noise reduction performance.

[0129] Based on the short - delay active noise control method provided in the above embodiments, the embodiment of the present application also provides a short - delay active noise control system, which will be introduced in detail below with reference to the accompanying drawings.

[0130] Refer to Figure 8 , which is the schematic diagram of a short - delay active noise control system provided by an embodiment of the present application.

[0131] The short - delay active noise control system provided by the embodiment of the present application includes: a reference sensor 810, an ANC controller 820, and a speaker 830. The ANC controller can be specifically implemented using a DSP chip.

[0132] Among them, the noise emitted by the noise source is transmitted through the primary path to the area to be noise - reduced to form the target noise.

[0133] The reference sensor 810 is used to collect the reference signal at the noise source.

[0134] The collected reference signal is represented by a discrete-time sequence.

[0135] The ANC controller 820 is configured to input the reference signals of a continuous preset number of time steps into the long short-term memory (LSTM) neural network model, predict the predicted reference signal corresponding to the next time step through the LSTM neural network model; obtain the first cancellation signal output by the adaptive filter according to the predicted reference signal, and update the weight coefficient vector of the adaptive filter.

[0136] Preferably, the adaptive filter may include a finite impulse response (FIR) filter and a weight update module.

[0137] The speaker 830 is configured to output the first cancellation signal, so that the second cancellation signal and the target noise are superimposed and cancelled to achieve active noise control.

[0138] Wherein, the second cancellation signal is the signal formed in the area to be noise-reduced after the first cancellation signal is transmitted through the secondary path; the target noise is the noise signal when the noise emitted by the noise source is transmitted to the area to be noise-reduced through the primary path.

[0139] This embodiment further includes an error sensor 840; the error sensor 840 is placed in the area to be noise-reduced, and is configured to monitor the result after noise reduction, that is, monitor the residual error signal after the second cancellation signal and the target noise are superimposed and cancelled, so as to feedback to the ANC controller for updating the weight coefficient vector; the error sensor 840 may be specifically implemented by an error microphone.

[0140] The short-delay active noise control system provided by the embodiment of the present application includes: a reference sensor, an ANC controller, and a speaker. The noise emitted by the noise source is transmitted to the area to be noise-reduced through the primary path to form the target noise, and the reference sensor collects the reference signal at the noise source; the reference signal is represented by a discrete-time sequence; the reference signals of a continuous preset number of time steps are input into the long short-term memory (LSTM) neural network model of the ANC controller, and the predicted reference signal corresponding to the next time step is predicted through the LSTM neural network model; the cancellation signal output by the adaptive filter is obtained according to the predicted reference signal, and the weight coefficient vector of the adaptive filter is updated; the speaker outputs the first cancellation signal, so that the second cancellation signal and the target noise are superimposed and cancelled to achieve active noise control; wherein, the second cancellation signal is the signal formed in the area to be noise-reduced after the first cancellation signal is transmitted through the secondary path. The embodiment of the present application predicts the future samples of the reference signal, obtains the cancellation signal through the adaptive filter based on the predicted reference signal and updates the filter weight coefficient vector. By predicting the future samples of the reference signal and performing active noise reduction processing based on the prediction result, the delay of the system can be shortened to a certain extent, and the noise reduction performance of the ANC system can be more effectively improved.

[0141] In some embodiments, the active noise control system further includes devices such as a power amplifier; the power amplifier amplifies the output signal of the adaptive filter and drives the speaker to emit sound. The power amplifier can be integrated on the ANC controller or can be an independent device from the ANC controller, and the present application does not make specific limitations on this.

[0142] In some embodiments, the LSTM neural network model consists of an input layer with D nodes, a hidden layer with K LSTM neural network units, and a fully connected layer. The LSTM neural network unit includes an input gate, a forget gate, and an output gate. Accordingly, the ANC controller is specifically configured to:

[0143] Use the reference signal of a continuous preset number of time steps as the input signal; filter the input signal through the input gate and obtain the cell state;

[0144] Discard the non-critical information in the historical information through the forget gate and update the cell state;

[0145] Determine the output part of the updated cell state through the output gate;

[0146] Determine the output information through the output part and the updated cell state;

[0147] Obtain the predicted reference signal through the fully connected layer according to the output information and the output weight matrix.

[0148] In some embodiments, the ANC controller is specifically configured to construct a reference signal vector of the adaptive filter according to the predicted reference signal and the length of the adaptive filter; obtain the first cancellation signal according to the reference signal vector and the weight coefficient vector.

[0149] In some embodiments, the error sensor is specifically configured to obtain an error signal according to the desired signal and the secondary cancellation signal after passing through the secondary path; the ANC controller is specifically configured to update the weight coefficient vector according to the error signal, the weight update step size, and the length of the adaptive filter.

[0150] Corresponding to the above process, see Figure 9 , this figure is a schematic diagram of the module control of a short-delay active noise control system provided by an embodiment of the present application.

[0151] Specifically, the ANC controller includes a reference signal prediction module, a filtering module, a weight update module, and an error synthesis module; the reference signal prediction module includes an LSTM neural network model, and the filter module and the weight update module constitute an adaptive filter.

[0152] From Figure 9It can be seen that after the reference signal is processed by the reference signal prediction module and the filtering module, and the first cancellation signal is output, the second cancellation signal is formed through the transmission of the secondary path. The error synthesis module obtains the error signal after the second cancellation signal and the actual target noise are superimposed and cancelled; the error signal returns to the weight update module, and the weight coefficient of the adaptive filter is updated together with the data output by the reference signal prediction module; the updated weight coefficient is used for the subsequent filtering process of the filtering module.

[0153] Specifically, P(z) is the transfer function of the primary path between the reference sensor and the area to be noise-reduced; S(z) is the transfer function of the secondary path between the loudspeaker and the area to be noise-reduced; is the transfer function of the secondary path estimated through modeling. d(n) is the target noise signal, y(n) is the output signal of the filtering module, and y s (n) is the second cancellation signal of the output signal passing through the secondary path, and e(n) is the residual noise signal after the target noise and the second cancellation signal are superimposed in the area to be noise-reduced, also known as the error signal, which is used to feedback to the weight update module for weight update; X f (n) is the filtered reference signal input to the weight update module after the output signal of the reference signal prediction module passes through . It should be noted that the various embodiments in this specification are described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The same or similar parts among the various embodiments can be referred to each other. For the systems or devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0154] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A short-delay active noise control method, characterized in that, Noise emitted by a noise source is transmitted through a primary path to a region to be noise-reduced to form target noise. The method includes: Collecting a reference signal at the noise source by a reference sensor; the reference signal is represented by a discrete time series; Inputting reference signals of a continuous preset number of time steps into a long short-term memory (LSTM) neural network model, and predicting a predicted reference signal corresponding to the next time step through the LSTM neural network model; Inputting the predicted reference signal into an adaptive filter to obtain a first cancellation signal output by the adaptive filter, and updating a weight coefficient vector of the adaptive filter; outputting the first cancellation signal through a speaker; Superposing and canceling the second cancellation signal with the target noise to achieve the active noise control; wherein, the second cancellation signal is a signal formed in the region to be noise-reduced after the first cancellation signal is transmitted through a secondary path.

2. The method according to claim 1, wherein The LSTM neural network model includes LSTM neural network units; the LSTM neural network units specifically include an input gate, a forget gate, and an output gate; The step of inputting reference signals of a continuous preset number of time steps into a long short-term memory (LSTM) neural network model and predicting a predicted reference signal corresponding to the next time step through the LSTM neural network model specifically includes: Taking reference signals of a continuous preset number of discrete time series as input signals; the input gate filters the input signals and obtains a cell state; The forget gate discards non-critical information in historical information and updates the cell state; The output gate determines an output part of the updated cell state; Determining output information through the output part and the updated cell state; A fully connected layer obtains a predicted reference signal according to the output information and an output weight matrix.

3. The method according to claim 1, wherein The step of inputting the predicted reference signal into an adaptive filter to obtain a first cancellation signal output by the adaptive filter specifically includes: Constructing a reference signal vector of the adaptive filter according to the predicted reference signal and the length of the adaptive filter; Obtaining a first cancellation signal according to the reference signal vector and the weight coefficient vector.

4. The method according to claim 3, wherein The step of updating the weight coefficient vector of the adaptive filter specifically includes: Collecting an error signal remaining after the target noise and the second cancellation signal are superposed and canceled; Updating the weight coefficient vector according to the error signal, a weight update step size, the length of the adaptive filter, and a filtered reference signal.

5. The method according to any one of claims 1 to 4, characterized in that, The adaptive filter specifically includes a finite impulse response (FIR) filter and a weight update module.

6. A short-delay active noise control system, characterized in that, Noise emitted by a noise source is transmitted through a primary path to a region to be noise-reduced to form target noise. The system includes: a reference sensor, an ANC controller, and a speaker; The reference sensor is configured to collect a reference signal at the noise source; the reference signal is represented by a discrete time series; The ANC controller is configured to input reference signals of a continuous preset number of time steps into a long short-term memory (LSTM) neural network model, and predict a predicted reference signal corresponding to the next time step through the LSTM neural network model; obtain a first cancellation signal output by the adaptive filter according to the predicted reference signal, and update the weight coefficient vector of the adaptive filter; The loudspeaker is configured to output the first cancellation signal, so that the second cancellation signal is superimposed and cancelled with the target noise, thereby realizing the active noise control; wherein, the second cancellation signal is a signal formed in the area to be noise-reduced after the first cancellation signal is transmitted through the secondary path.

7. The system according to claim 6, characterized in that, The LSTM neural network model includes LSTM neural network units; the LSTM neural network units specifically include an input gate, a forget gate, and an output gate; The ANC controller is specifically configured to use reference signals of a continuous preset number of time steps as input signals; filter the input signals through the input gate and obtain a cell state; Discard non-critical information in the historical information through the forget gate and update the cell state; Determine the output part of the updated cell state through the output gate; Determine output information through the output part and the updated cell state; Obtain a predicted reference signal through a fully connected layer according to the output information and the output weight matrix.

8. The system according to claim 6, wherein The ANC controller is specifically configured to construct a reference signal vector of the adaptive filter according to the predicted reference signal and the length of the adaptive filter; obtain a first cancellation signal according to the reference signal vector and the weight coefficient vector.

9. The system according to claim 8, wherein The system further includes an error sensor; the error sensor is specifically configured to collect an error signal remaining after the target noise and the second cancellation signal are superimposed and cancelled; the ANC controller is configured to update the weight coefficient vector according to the error signal, the weight update step size, the length of the adaptive filter, and the filtered reference signal.

10. The system according to any one of claims 6-9, characterized in that, The adaptive filter specifically includes a finite impulse response (FIR) filter and a weight update module.

Images