A narrowband active noise control method and system thereof

Abstract

The application discloses a narrow-band active noise control method and a system thereof. The method and the system thereof perform frequency estimation and tracking of a noise signal based on frequency estimation of DFT coefficients, and control the frequency estimation of the DFT coefficients in combination with error gradient monitoring. The frequency estimation of the DFT coefficients is started when the error signal is monitored to increase due to frequency offset, and the frequency estimation of the DFT coefficients is stopped when the error signal change is lower than a threshold. The application mainly proposes a DFT frequency estimation method combined with error gradient monitoring for the NANC system to reduce the calculation burden of the NANC system, and meanwhile, the estimation accuracy of the system can be ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of noise control, in particular to a narrow-band active noise control method and system, especially suitable for controlling narrow-band active noise. BACKGROUND

[0002] Active noise cancellation (ANC) is an important application in current audio technology, widely used in headphones, vehicles, industrial environments, and aviation scenarios. Unlike passive noise reduction (which isolates noise through physical structure), active noise reduction uses an electronic system to emit a sound signal with opposite phase to the noise, thereby achieving noise reduction effect. With the advancement of electronic components and audio processing algorithms, ANC technology continues to develop and innovate, becoming an important guarantee for high-quality audio experience.

[0003] The basic working principle of active noise reduction is based on the interference principle, that is, using the mutual interference phenomenon of sound waves to eliminate noise. Specifically, sound waves have phase characteristics, when two sound waves of the same frequency but opposite phase meet, they will cancel each other out, thereby reducing or completely eliminating sound.

[0004] The core of active noise reduction lies in the signal processing algorithm, especially the ability to generate signals with opposite phase to noise. Common algorithms include the following:

[0005] 1. Wiener filtering algorithm

[0006] The Wiener filtering algorithm is the most classic and simplest algorithm in active noise reduction, its core idea is to minimize the mean square error, offline calculate the optimal filter coefficients of the controller, and then import the coefficients into the embedded system for real-time convolution operation. Because it consumes less computing resources and is sensitive to channel response changes, this algorithm is commonly used in one-dimensional acoustic scenes where channel response is simple and changes little, such as active noise reduction headphones with strict power consumption requirements. However, since this algorithm cannot adjust the filter coefficients in real time according to the external environment, its performance is poor for three-dimensional active noise reduction scenarios, such as active noise reduction headrests.

[0007] 2. Adaptive filtering algorithm (Adaptive Filtering) Adaptive filtering is a commonly used algorithm in active noise reduction, which can dynamically adjust the parameters of the filter according to the changes in the input signal, thereby generating signals opposite to the noise. Typical adaptive filtering algorithms include LMS (Least Mean Squares) algorithm and RLS (Recursive Least Squares) algorithm.

[0008] LMS Algorithm: The LMS algorithm adjusts the filter coefficients step by step to minimize the output error. It has lower computational complexity and is suitable for most active noise reduction systems, but performs slightly worse in high-frequency noise processing.

[0009] RLS Algorithm: The RLS algorithm has faster convergence speed than LMS, but higher computational complexity, suitable for scenarios with higher noise response requirements.

[0010] 3. Deep Learning and ANC

[0011] In recent years, deep learning has been introduced into active noise reduction technology, learning the pattern of environmental noise through a neural network model, predicting future noise changes, and generating corresponding anti-phase signals. This algorithm can handle more complex noise environments, such as the mixture of speech and background noise. The advantage of deep learning ANC algorithm is strong adaptive ability, especially suitable for variable noise environments.

[0012] In summary, active noise control (ANC) is a very effective low-frequency noise control technology, which has better control effect on low-frequency noise, especially low-frequency line spectrum noise, compared with traditional passive noise control. In addition, ANC system has the advantages of light weight, low cost, small size, etc. According to the different types of noise spectrum to be controlled, ANC system is divided into broadband ANC (BANC) and narrowband ANC (NANC). Among them, NANC system mainly aims at line spectrum noise, compared with BANC system, NANC system does not need to collect reference signal by reference sensor, can avoid the influence of secondary sound feedback, and has lower computational complexity. However, NANC system needs to accurately obtain the frequency information of line spectrum noise signal, in order to accurately generate reference signal, otherwise it will greatly affect the noise reduction performance of the system. Therefore, NANC system often needs to have a frequency estimation module to estimate and track the frequency of noise signal.

[0013] In recent years, many scholars have studied the frequency estimation module in the NANC system, which can be mainly divided into the method based on adaptive notch filter and the method based on DFT coefficient. Among them, the method based on adaptive notch filter has smaller calculation amount, but has poor tracking performance when facing sudden change of noise frequency, and has lower frequency estimation accuracy and more parameters to be set artificially. The method based on DFT coefficient only needs to set the data block length for each estimation, and can flexibly change the estimation accuracy and tracking speed of the system by adjusting the data block length. However, in order to obtain higher estimation accuracy, more data need to be buffered and higher-point DFT coefficient calculation is needed, which will bring great storage and calculation burden to the embedded system. SUMMARY

[0014] The present application aims at the problem that the calculation complexity of the frequency estimation module in the NANC system is restricted by the frequency estimation accuracy, and provides a narrowband active noise control method and system. The present application proposes a NANC method combining error gradient monitoring and DFT coefficient frequency estimation. The method monitors the error signal gradient, and only when the error signal increases due to frequency offset is monitored, the frequency estimation algorithm is started. When the error signal change is lower than the threshold, the frequency estimation algorithm is stopped, without real-time frequency estimation to waste calculation resources. The method can effectively reduce the calculation burden of the NANC system caused by the frequency estimation algorithm, and is beneficial to the implementation of the multi-channel NANC system on the embedded platform with limited calculation resources.

[0015] To achieve the above-mentioned application purposes, the present application provides a narrowband active noise control method. The method performs frequency estimation and tracking of the noise signal based on the DFT coefficient frequency estimation, and combines error gradient monitoring to control the DFT coefficient frequency estimation. The error signal gradient is monitored, and when the error signal increases due to frequency offset is monitored, the DFT coefficient frequency estimation is started. When the error signal change is lower than the threshold, the DFT coefficient frequency estimation is stopped, without real-time frequency estimation to waste calculation resources. The method can effectively reduce the calculation burden of the NANC system caused by the frequency estimation algorithm, and is beneficial to the implementation of the multi-channel NANC system on the embedded platform with limited calculation resources.

[0016] As an improvement of the above technical solution, the step of adjusting the DFT coefficient frequency estimation by the error gradient monitoring includes:

[0017] (1) setting specific parameters for error gradient calculation: calculating the data length of error gradient buffer error gradient threshold Frequency estimation frame length Frequency change threshold and filter iteration step size ;

[0018] (2) The estimated noise signal is obtained by using the internal model estimation method as the reference signal. It also buffers the reference signal and the error signal collected by the error microphone;

[0019] (3) When the buffered reference signal reaches the data length required for error gradient calculation When using the formula Calculate the current error gradient ;

[0020] (4) Execution judgment: when ,cache After sampling points, a DFT-based frequency estimation algorithm is performed; when When the residual error changes very little, the system tends to be stable, and there is no need to perform the frequency estimation algorithm anymore. Therefore, the frequency estimator is turned off.

[0021] As a further improvement to the above technical solution, the data length required for the error gradient calculation is... The value is determined based on the background noise signal-to-noise ratio (SNR). A lower value is set for high SNR scenarios to reduce storage burden; a higher value is set for low SNR scenarios. The aforementioned error gradient threshold... The threshold needs to be set based on the background noise energy. Before active noise control begins, a noise signal for a certain period of time is collected to calculate its background noise energy as a threshold; the frequency estimation frame length... According to the sampling rate and noise frequency component spacing To set it up, it needs to meet the following requirements. The frequency change threshold mentioned above The filter iteration step size is set according to the frequency interval and the bandwidth of the low-pass filter; All are set to 0.01.

[0022] To achieve the aforementioned other objective, the present invention also provides a narrowband active noise control system. This system performs frequency estimation and tracking of noise signals based on a frequency estimation module of DFT coefficients. The system further includes an error gradient monitoring module for controlling the frequency estimation module of DFT coefficients. By monitoring the error signal gradient, when an increase in the error signal due to frequency offset is detected, the frequency estimation module of DFT coefficients is activated. When the change in the error signal is lower than a threshold, the frequency estimation module of DFT coefficients is stopped.

[0023] As an improvement of the above technical solution, the frequency estimation module is based on DFT coefficient interpolation algorithm, and for a periodic signal containing only one frequency The estimated frequency can be expressed as:

[0024] ;

[0025] Wherein, is an integer frequency component, is a residual fractional frequency component.

[0026] As another improvement of the above technical solution, the narrow-band active noise control system further comprises a spectrum shifter; the frequency obtained by the frequency estimation module is used to generate a spectrum shift signal The spectrum shifter performs spectrum shift on the reference signal, and obtains the final reference signal through a low-pass filter.

[0027] The present application has the advantages that: the present application combines the DFT frequency estimation method with error gradient monitoring to reduce the calculation burden of the NANC system, while ensuring the estimation accuracy of the system, providing an accurate reference signal for the NANC system, thereby avoiding the influence of frequency offset problem on the noise reduction performance of the NANC system. At the same time, the low calculation complexity also ensures that the algorithm of the present application can be realized on an embedded platform with limited computing resources. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 An algorithm implementation block diagram of a NANC method combining error gradient monitoring and DFT frequency estimation according to the present application is shown;

[0029] Figure 2 A comparison diagram of frequency estimation results with error monitoring and frequency estimation results without error monitoring in the simulation experiment of the present application is shown, and the noise source signal used is a line spectrum signal with a frequency component of 300Hz;

[0030] Figure 3 A comparison diagram of frequency estimation results with error monitoring and frequency estimation results without error monitoring in the simulation experiment of the present application is shown, and the noise source signal used is a line spectrum signal with a frequency component of 600Hz;

[0031] Figure 4 A comparison diagram of frequency estimation results with error monitoring and frequency estimation results without error monitoring in the simulation experiment of the present application is shown, and the noise source signal used is a line spectrum signal with a frequency component of 900Hz;

[0032] Figure 5 The mean square error convergence result of the residual noise in the simulation experiment of the present application is shown;

[0033] ​Figure 6 The figure shows the power spectrum density of residual noise at error microphone before and after noise reduction in simulation experiment of the present application.

[0034] Figure 7 The figure shows the time domain variation of error gradient monitoring value used in simulation experiment of the present application. DETAILED DESCRIPTION

[0035] The technical solution of the present application will be described in further detail in combination with the following figures and simulation experiment of the present application.

[0036] The present application provides a DFT frequency estimation method combined with error gradient monitoring to reduce the calculation burden of NANC system, while ensuring the estimation accuracy of the system, and the specific steps are as follows:

[0037] (1) Set the specific parameters used in the method: data length required for error gradient calculation , error gradient threshold , frequency estimation frame length , frequency variation threshold , and filter iteration step size ;

[0038] (2) Obtain the estimated noise signal as the reference signal by using the internal model estimation method , and buffer the reference signal and the error signal collected by the error microphone;

[0039] (3) When the buffered reference signal reaches the data length required for error gradient calculation , calculate the current error gradient by using the formula ;

[0040] (4) Perform judgment: when , buffer sampling points and then perform DFT-based frequency estimation algorithm; when , it is proved that the residual error changes very little at this time, and the system tends to be stable, so the frequency estimator does not need to be performed again, and thus the frequency estimator is closed;

[0041] The specific DFT-based frequency estimation algorithm is as follows:

[0042] For a periodic signal containing only one frequency , the estimated frequency can be represented as:

[0043] ;

[0044] Wherein, is an integer frequency component, is a residual fractional frequency component.

[0045] The frequency estimation algorithm based on DFT first performs... Perform FFT to obtain its spectrum The frequency index at which the maximum amplitude is obtained is the integer frequency component. :

[0046] ;

[0047] ;

[0048] in, The imaginary unit is represented, and then two iterations are performed to obtain the precise residual fractional frequency components. ,Right now

[0049] ;

[0050] ;

[0051] in,

[0052] ;

[0053] ;

[0054] in, , ; and These are the indexes of the periodic signals at their largest values. nearby and The DFT coefficients; for ease of expression, the above estimation process for a single frequency is simplified as follows: The above single-frequency estimation process can be extended to the estimation of multi-frequency signals as follows: Assume... Represents the first The FFT signal block, i.e. the th FFT signal block The frames contain a total of [number] frequencies in the multi-frequency signal to be estimated. Each frequency component. First, the frequency component with the current largest amplitude. Estimated by the HAQSE algorithm:

[0055] ;

[0056] in, It contains A multi-frequency signal with several frequency components. The estimated frequencies can be obtained by projecting this signal onto the estimated subspace. amplitude :

[0057]

[0058] Thus, we can get the estimated frequency The time-domain sinusoidal signal is:

[0059]

[0060] Finally, we subtract the sinusoidal signal from the original multi-frequency signal , and get a multi-frequency signal vector containing frequency components, which is expressed as:

[0061]

[0062] Repeat the above process until all frequencies are estimated, and we can get the frequency estimation results obtained in the current frame .

[0063] (5) Calculate the change of the 1-norm of the estimated frequency obtained in the current frame and the estimated frequency obtained in the last frame, if , do not replace the shift frequency used in the next spectrum shifter, if , replace and recalculate the Fourier transform coefficients of the secondary path, the calculation formula is as follows:

[0064] ;

[0065] ;

[0066] (6) According to the frequency obtained by the frequency estimation module, generate a spectrum shift signal , and perform spectrum shift and low-pass filtering on the reference signal to obtain the final reference signal, the specific process is as follows:

[0067] First, multiply the reference signal by the complex exponential signal to obtain the complex signal ;

[0068] ;

[0069] Among them, is the shift frequency of the th frequency component. The spectrum of this complex signal can be expressed as:

[0070] ;​

[0071] where, is the amplitude of the th frequency component in the frequency domain, is the unit impulse response function, and only is the term close to 0 Hz; therefore, it is passed through a low-pass filter to obtain the spectrum of the filtered signal is:

[0072] ;

[0073] where, is the amplitude of the filtered signal, in order to facilitate the rapid adjustment of the low-pass filter in hardware and reduce the system calculation complexity, the low-pass filter is a second-order IIR filter with adjustable bandwidth, as follows:

[0074]

[0075] where, the parameter can adjust the bandwidth of the filter, and is usually set to a decimal close to . Then, the inverse frequency shift is performed on , and the time domain is multiplied by the complex exponential signal , so that its spectrum is:

[0076] ;

[0077] Its time domain expression is:

[0078] ;

[0079] where, is a complex exponential signal containing only the th frequency component. Therefore, we can take the real part and the imaginary part as the reference signal and of the th frequency component, that is:

[0080] ;

[0081] ;

[0082] (7) After obtaining the reference signal, the output of the filter at this time is calculated:

[0083] Then, the update of the filtered reference signal is performed, and the formula is as follows:

[0084] ; ;

[0085] Next, update the filter of the NANC system using the following formula:

[0086] ;

[0087] ;

[0088] (8) Return to step (2) until the entire noise control process is completed.

[0089] like Figure 1 As shown, Figure 1 The simulation experiment was conducted in MATLAB. The noise source signal used was a line spectrum signal containing three frequency components: 300Hz, 600Hz, and 900Hz. The amplitudes of the three line spectra were randomly set. The secondary path model was generated using the built-in MATLAB function `fir1`, with a cutoff frequency set to 0.4 and an order of 50. The entire simulation experiment lasted 10 seconds, with a sampling rate of 8000Hz. At 5 seconds, all frequency components were increased by a 5% frequency shift, becoming 315Hz, 630Hz, and 945Hz respectively. The required data length for error gradient calculation was determined. The value should typically be greater than 16, depending on the background noise signal-to-noise ratio (SNR). A lower value can be set for high SNR scenarios to reduce storage burden, but a larger value is needed for low SNR scenarios. In this simulation, it is set to 32. (Error gradient threshold) The threshold needs to be set based on the background noise energy. A threshold can be calculated by pre-collecting signals for a period of time before active control begins; in this simulation, it is set to 0.001. Frequency estimation frame length. It needs to be based on the sampling rate and noise frequency component spacing To set it up, it needs to meet the following requirements. In this simulation, the frequency change threshold is set to 512. The value needs to be set according to the frequency spacing and the bandwidth of the low-pass filter, typically between 5 and 40; in this simulation, it is set to 15. The filter iteration step size... The step size is set to 0.01 for both algorithms to ensure that they have the same convergence speed and steady-state error, thus avoiding interference with the results due to the step size selection.

[0090] like Figure 2The comparative diagram of the frequency estimation results with error monitoring and the frequency estimation results without error monitoring in the simulation experiment of the application is shown in Figure 2, and the noise source signal used is a line spectrum signal with a frequency component of 300 Hz. It can be seen that the frequency estimator with error monitoring only performs frequency estimation once when the primary noise has no frequency change, while the frequency estimator without error monitoring performs frequency estimation once every 512 sampling points, but there is no obvious difference in the estimation accuracy of the two. When the frequency suddenly changes at 5s, the noise source frequency changes from 300 Hz to 330 Hz, and the frequency estimator with error monitoring and the frequency estimator without error monitoring can quickly track the change of the noise frequency. This shows that when the noise frequency is constant, it is not necessary to perform frequency estimation every 512 sampling points, and frequency estimation is only needed when the frequency changes. The frequency estimator with error monitoring of the application effectively determines whether frequency estimation is needed at this time, thereby avoiding the calculation burden caused by excessive frequency estimation.

[0091] As shown in Figure 2, the comparative diagram of the frequency estimation results with error monitoring and the frequency estimation results without error monitoring in the simulation experiment of the application is shown, and the noise source signal used is a line spectrum signal with a frequency component of 300 Hz. It can be seen that the frequency estimator with error monitoring only performs frequency estimation once when the primary noise has no frequency change, while the frequency estimator without error monitoring performs frequency estimation once every 512 sampling points, but there is no obvious difference in the estimation accuracy of the two. When the frequency suddenly changes at 5s, the noise source frequency changes from 300 Hz to 330 Hz, and the frequency estimator with error monitoring and the frequency estimator without error monitoring can quickly track the change of the noise frequency. This shows that when the noise frequency is constant, it is not necessary to perform frequency estimation every 512 sampling points, and frequency estimation is only needed when the frequency changes. The frequency estimator with error monitoring of the application effectively determines whether frequency estimation is needed at this time, thereby avoiding the calculation burden caused by excessive frequency estimation. Figure 3 As shown in Figure 2, the comparative diagram of the frequency estimation results with error monitoring and the frequency estimation results without error monitoring in the simulation experiment of the application is shown, and the noise source signal used is a line spectrum signal with a frequency component of 300 Hz. It can be seen that the frequency estimator with error monitoring only performs frequency estimation once when the primary noise has no frequency change, while the frequency estimator without error monitoring performs frequency estimation once every 512 sampling points, but there is no obvious difference in the estimation accuracy of the two. When the frequency suddenly changes at 5s, the noise source frequency changes from 300 Hz to 330 Hz, and the frequency estimator with error monitoring and the frequency estimator without error monitoring can quickly track the change of the noise frequency. This shows that when the noise frequency is constant, it is not necessary to perform frequency estimation every 512 sampling points, and frequency estimation is only needed when the frequency changes. The frequency estimator with error monitoring of the application effectively determines whether frequency estimation is needed at this time, thereby avoiding the calculation burden caused by excessive frequency estimation.

[0092] As shown in Figure 2, the comparative diagram of the frequency estimation results with error monitoring and the frequency estimation results without error monitoring in the simulation experiment of the application is shown, and the noise source signal used is a line spectrum signal with a frequency component of 300 Hz. It can be seen that the frequency estimator with error monitoring only performs frequency estimation once when the primary noise has no frequency change, while the frequency estimator without error monitoring performs frequency estimation once every 512 sampling points, but there is no obvious difference in the estimation accuracy of the two. When the frequency suddenly changes at 5s, the noise source frequency changes from 300 Hz to 330 Hz, and the frequency estimator with error monitoring and the frequency estimator without error monitoring can quickly track the change of the noise frequency. This shows that when the noise frequency is constant, it is not necessary to perform frequency estimation every 512 sampling points, and frequency estimation is only needed when the frequency changes. The frequency estimator with error monitoring of the application effectively determines whether frequency estimation is needed at this time, thereby avoiding the calculation burden caused by excessive frequency estimation. Figure 4 Therefore, as shown in Figure 2, the comparative diagram of the frequency estimation results with error monitoring and the frequency estimation results without error monitoring in the simulation experiment of the application is shown, and the noise source signal used is a line spectrum signal with a frequency component of 300 Hz. It can be seen that the frequency estimator with error monitoring only performs frequency estimation once when the primary noise has no frequency change, while the frequency estimator without error monitoring performs frequency estimation once every 512 sampling points, but there is no obvious difference in the estimation accuracy of the two. When the frequency suddenly changes at 5s, the noise source frequency changes from 300 Hz to 330 Hz, and the frequency estimator with error monitoring and the frequency estimator without error monitoring can quickly track the change of the noise frequency. This shows that when the noise frequency is constant, it is not necessary to perform frequency estimation every 512 sampling points, and frequency estimation is only needed when the frequency changes. The frequency estimator with error monitoring of the application effectively determines whether frequency estimation is needed at this time, thereby avoiding the calculation burden caused by excessive frequency estimation.

[0093] Figures 2-4 ​The frequency estimation results of the three frequency components shown are presented with and without error gradient estimation. It can be seen that without error monitoring, although the actual frequency remains unchanged (e.g., consistently 300 Hz, 600 Hz, and 900 Hz for the first 5 seconds, and 330 Hz, 660 Hz, and 990 Hz for the next 5 seconds), the frequency estimator continuously operates, fluctuating around the actual frequency, but within a range of less than 0.5 Hz. With error monitoring, however, the frequency estimator stops operating when the actual frequency remains unchanged, performing only one frequency estimation. It can be seen that the frequency estimator without error monitoring does not improve estimation accuracy with an increase in the number of frequency estimations. At 5 seconds, when the noise frequency undergoes a sudden change, the frequency estimator with error monitoring in this invention can track the frequency change as quickly as the frequency estimator without error monitoring, promptly activating the frequency estimator for frequency estimation. This is because when the frequency changes abruptly, the frequency of the reference signal deviates from the actual noise frequency, significantly affecting the noise reduction performance of the NANC system. Therefore, the residual error signal at the error point suddenly increases. This invention addresses this by monitoring the error gradient. To determine the state of the system, such as Figure 7 As shown, the error gradient value increases significantly beyond the set threshold at 5 seconds, thus activating the frequency estimator.

[0094] like Figure 5 As shown, the mean square error convergence result of residual noise in the simulation experiment of this invention is shown. It can be seen from the mean square error convergence result of residual noise that the algorithm of this application can achieve the same convergence speed as the traditional NANC system with high computational complexity, which verifies that the proposed algorithm does not reduce the convergence performance of the system even if the computational complexity is reduced.

[0095] like Figure 6 As shown, the power spectral density of the residual noise at the error microphone before and after noise reduction in the simulation experiment of this invention is represented. It can be seen from the power spectral density results of the residual noise that the algorithm of this application can achieve the same noise reduction as the traditional NANC system with high computational complexity, which verifies that the proposed algorithm does not reduce the noise reduction performance of the system while reducing computational complexity.

[0096] Therefore, from the mean square error convergence result of the residual noise, it can be seen that when the noise frequency mutates at 5s, the application can obtain the same convergence speed as the traditional NANC system with high computational complexity, and both can converge to a steady-state background noise of-40dB within 0.5s, verifying that the algorithm of the application does not reduce the convergence performance of the system even if the computational complexity is reduced; from the power spectral density result of the residual noise, it can be seen that the algorithm of the application can obtain the same noise reduction as the traditional NANC system with high computational complexity, and the three line spectra of 300Hz, 600Hz and 900Hz before the frequency mutation and the three line spectra of 330Hz, 660Hz and 990Hz after the frequency mutation can all achieve a noise reduction performance of more than 50dB, verifying that the algorithm of the application does not reduce the noise reduction performance of the system while reducing the computational complexity. Therefore, through Figure 4 and Figure 5 It is verified that the error monitoring subsystem does not affect the convergence performance and noise reduction performance of the NANC system. In this simulation experiment, the number of secondary sources and the number of error microphones are both 1, the number of multiplications with error monitoring and without error monitoring at a sampling point when the real frequency does not change is 92 and 138 respectively, and in the whole 10 seconds, 312 frequency estimations are performed without error monitoring, and only 8 frequency estimations are performed with error monitoring, it can be seen that whether the computational amount used at a single sampling point or the computational amount required in the whole process, the NANC system with error monitoring is lower.

[0097] As shown in Figure 7 , the time domain variation result of the error gradient monitoring value used in the simulation experiment of the application is shown. It can be seen that the error gradient monitoring value will only change significantly at 5s, that is, when the noise frequency mutates; and at the remaining time, the noise is in a steady state, the frequency does not change, and the error gradient monitoring value also does not change. Through this result, it can be found that the error gradient monitoring value has high sensitivity to the change of the noise frequency, verifying the feasibility of determining the change of the noise frequency through the error gradient monitoring value.

[0098] Through this simulation experiment, it is further verified that the NANC system combined with the error gradient monitoring proposed in the application has the same convergence speed and noise reduction performance as the original system, but has lower computational complexity, and when the number of secondary sources and error points increases, the improvement is more obvious.

[0099] Finally, it should be explained that the above embodiments are only used to illustrate the technical solutions of the application and not to limit. Although the technical solutions of the application are described in detail with reference to the embodiments, those skilled in the art should understand that the modification or equivalent replacement of the technical solutions of the application does not deviate from the spirit and scope of the technical solutions of the application, and they should be covered in the scope of the claims of the application.

Claims

1. A narrowband active noise control method, characterized in that, This method is based on frequency estimation of DFT coefficients to estimate and track the frequency of noise signals, and combines error gradient monitoring to control the frequency estimation of DFT coefficients. By monitoring the error signal gradient, when the error signal is detected to increase due to frequency offset, the frequency estimation of DFT coefficients is started. When the error signal change is lower than the threshold, the frequency estimation of DFT coefficients is stopped. The step of adjusting the frequency estimate of the DFT coefficients by error gradient monitoring includes: (1) Set the specific parameters for error gradient calculation: data length Error gradient threshold Frequency estimation frame length Frequency change threshold and filter iteration step size (2) The estimated noise signal is obtained by using the internal model estimation method as the reference signal. It also buffers the reference signal and the error signal collected by the error microphone; (3) When the buffered reference signal reaches the data length required for error gradient calculation When using the formula Calculate the current error gradient ; (4) Execution judgment: when ,cache After sampling points, a DFT-based frequency estimation algorithm is performed; when When this happens, the frequency estimator is turned off; The required data length for the error gradient calculation The setting value is determined based on the background noise signal-to-noise ratio. When the signal-to-noise ratio is high, the value should be set lower to reduce the storage burden; when the signal-to-noise ratio is low, the value should be set higher. The error gradient threshold The threshold needs to be set based on the background noise energy. Before the active noise control starts, a noise signal for a period of time is collected in advance to calculate its background noise energy as the threshold. The frequency estimation frame length According to the sampling rate and noise frequency component spacing To set it up, it needs to meet the following requirements. ; The frequency change threshold It is set according to the frequency spacing and the bandwidth of the low-pass filter.

2. The narrowband active noise control method according to claim 1, characterized in that, The filter iteration step size All are set to 0.

01.

3. A narrowband active noise control system, characterized in that, The system uses a frequency estimation module based on DFT coefficients to estimate and track the frequency of noise signals. The system also includes an error gradient monitoring module for controlling the frequency estimation module of DFT coefficients. By monitoring the error signal gradient, when an increase in the error signal due to frequency offset is detected, the frequency estimation module of DFT coefficients is started. When the change in the error signal is lower than a threshold, the frequency estimation module of DFT coefficients is stopped. The error gradient monitoring module executes the method as described in any one of claims 1 to 2.

4. The narrowband active noise control system according to claim 3, characterized in that, The frequency estimation module is based on the DFT coefficient interpolation algorithm, for frequencies containing only one frequency. periodic signal Its estimated frequency can be expressed as: ； in, For integer frequency components, The residual fractional frequency component.

5. The narrowband active noise control system according to claim 4, characterized in that, The narrowband active noise control system also includes a spectrum shifter; the frequency obtained by the frequency estimation module is used to generate a spectrum shift signal. The spectrum shifter performs spectrum shifting on the reference signal and then obtains the final reference signal through a low-pass filter.

Images