A method for adaptive active noise cancellation control of a headphone

Abstract

The application discloses an earphone adaptive active noise cancellation control method and relates to the technical field of active noise control. The earphone adaptive active noise cancellation control method continuously acquires environmental noise, residual noise and noise reduction control signals, uses a reliable anchor acoustic path model, extracts abnormal residuals according to the environmental noise and the noise reduction control signals, establishes candidate parameters and generates candidate outputs, constructs a virtual error in combination with the residual noise, drives adaptive updating of the candidate parameters, extracts updating confidence based on the abnormal residuals and the virtual error, sets a reliability integral quantity and updates the reliability integral quantity according to the confidence, and determines the updated candidate control parameters as noise control parameters when the updated reliability integral quantity meets a preset synchronization verification condition, so that active noise cancellation is performed, thereby avoiding adjustment of filter coefficients in the wrong direction due to instantaneous disturbance, and further making the noise reduction parameters relatively stable during daily activities of a wearer.

Description

Technical Field

[0001] This invention relates to the field of active noise control technology, specifically to an adaptive active noise cancellation control method for headphones. Background Technology

[0002] Active noise cancellation uses a speaker to emit sound waves that are out of phase with the ambient noise, creating sound wave interference in the ear canal and thus attenuating external noise. To achieve this effect, headphones are usually equipped with an external reference microphone and an in-ear error microphone, which are used to pick up ambient noise and residual noise, respectively. The control cancellation continuously adjusts the out-of-phase noise signal output to the speaker based on these two signals, so that the residual noise gradually converges to a lower level.

[0003] The core of control cancellation is a set of filter coefficients, which determine the amplitude and phase of the inverted noise signal. The adjustment process of the coefficients usually relies on an adaptive algorithm, which is driven by residual noise and corrects the coefficients frame by frame along the gradient descent direction, so that the noise reduction effect gradually approaches the optimal state under the current acoustic environment.

[0004] The limitations of existing technologies include at least the following problems: In existing technologies, even slight loosening when the wearer speaks, chews, touches the earphone, or walks can cause a momentary change in the acoustic coupling state between the earbud and the ear canal. This introduces short and intense abnormal signal components into the residual noise collected by the error microphone. Existing technologies tend to treat these momentary abnormal signals as changes in environmental noise, which can drive the filter coefficients to adjust in the wrong direction. This causes the parameters that have already converged to deviate from the state that matches the current acoustic path in a short period of time. As a result, the earphone will produce audible impact noise or a brief loss of noise reduction at the moment of disturbance. When the acoustic coupling state recovers, the coefficients that have been deviated need to converge again to return to the normal level. This causes the noise reduction effect to deteriorate intermittently frequently in daily activities. Existing technologies can only pause updates when an anomaly is detected, making it difficult to continuously track changes in real environmental noise. This results in a disconnect between the noise reduction effect and the environment during the protection period. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an adaptive active noise cancellation control method for headphones, which solves the problems of existing technologies being prone to misjudging disturbances and causing parameter shifts, and making it difficult to balance disturbance resistance and continuous tracking of environmental noise changes.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an adaptive active noise cancellation control method for headphones, comprising the following steps: continuously acquiring ambient noise signals, residual noise signals, and active noise cancellation control signals of the headphones; using a pre-established reliable anchored acoustic path model, extracting abnormal residuals based on the ambient noise signals and noise cancellation control signals; establishing candidate control parameters and generating corresponding candidate output signals, and constructing virtual error signals by combining the reliable anchored acoustic path model, residual noise signals, and noise cancellation control signals, and using the virtual error signals to drive the candidate control parameters for adaptive updates; extracting update confidence based on the update process of abnormal residuals, virtual error signals, and candidate control parameters, and setting a confidence integral during the update process of candidate control parameters, and updating according to the update confidence; when the updated confidence integral meets the preset synchronization verification conditions, determining the updated candidate control parameters as noise control parameters to perform active noise cancellation.

[0007] Furthermore, the pre-establishment steps of the credible anchoring path model are as follows: Collect noise excitation response signals, noise response signals, test signals, and error response signals; using the noise excitation response signals as input and the noise response signals as the desired output, employ the normalized minimum mean square error adaptive identification method to estimate the environmental direct path transfer function, which serves as the credible anchoring environmental direct path sub-model; using the test signals as input and the error response signals as the desired output, employ the minimum mean square error adaptive identification method to estimate the secondary path transfer function, which serves as the credible anchoring secondary path sub-model; combine the credible anchoring environmental direct path sub-model and the credible anchoring secondary path sub-model to constitute the credible anchoring acoustic path model.

[0008] Further, the specific steps for extracting abnormal residuals are as follows: The environmental noise signal and noise reduction control signal of the current frame and several consecutive previous frames are input into the trusted anchored environmental direct path sub-model and the trusted anchored secondary path sub-model, respectively, for convolutional filtering to obtain the environmental noise prediction component and the speaker output prediction component; the environmental noise prediction component and the speaker output prediction component are added together to obtain the predicted residual signal, and after stripping with the residual noise signal, the residual signal is obtained; the historical residual signal sequence is acquired, the abnormal threshold is determined, and transient impact discrimination is performed on the residual signal to obtain the abnormal residual.

[0009] Furthermore, the specific steps for constructing the virtual error signal are as follows: subtract the candidate output signal and the noise reduction control signal point by point to obtain the output difference signal; extract the finite impulse response filter coefficient sequence from the reliable anchoring secondary path sub-model and perform linear convolution operation with the output difference signal to obtain the mapping compensation signal; add the residual noise signal and the mapping compensation signal point by point to obtain the virtual error signal.

[0010] Furthermore, the specific steps for adaptively updating candidate control parameters using virtual error signals are as follows: The environmental noise signal is input into the reliable anchoring secondary path sub-model for convolutional filtering to obtain a filtered reference signal; the filtered reference signal is multiplied with the virtual error signal point-by-point to obtain an instantaneous gradient vector, and gradient direction similarity values ​​are extracted; based on the gradient direction similarity values, the update step size is determined, and it is weighted step-by-step with the instantaneous gradient vector to obtain a coefficient adjustment vector; the coefficient adjustment vector is added coefficient-by-coefficient to the coefficients corresponding to the candidate control parameters to obtain the updated candidate control parameters.

[0011] Further, the specific steps for extracting the update confidence are as follows: Based on the update process of the abnormal residual, virtual error signal and candidate control parameters, an update evaluation set is extracted, including the impact disturbance value, gradient direction similarity value, error reduction efficiency value, and parameter change amplitude limit value; the update evaluation set is comprehensively processed to obtain the update score value; and the update score value is subjected to first-order recursive smoothing to obtain the update confidence.

[0012] Further, the specific steps for extracting and updating the evaluation set are as follows: Read the residual noise signal and compare it with the abnormal residual noise signal, respectively calculate the sum of the abnormal energy and the sum of the residual noise energy, and determine the impact disturbance value through energy ratio judgment processing; Based on the virtual error signal, extract the recent average energy and the long-term average energy, and determine the error reduction efficiency value through ratio processing; From the update process of the candidate control parameters, extract the magnitude of the coefficient adjustment vector, and judge it with the preset amplitude threshold, and determine the parameter change amplitude limit value based on the judgment result.

[0013] Furthermore, the specific steps for updating based on the updated confidence level are as follows: compare the updated confidence level with the preset confidence interval; when the updated confidence level is higher than the upper limit of the confidence interval, increase the confidence integral by the preset cumulative increment; when the updated confidence level is within the confidence interval, keep the confidence integral unchanged; when the updated confidence level is lower than the lower limit of the confidence interval, decay the confidence integral by the preset decay function.

[0014] Furthermore, the specific conditions for synchronous verification are as follows: the cumulative amount of the updated confidence integral reaches the preset synchronization threshold; the total virtual error energy of the virtual error signal and the total residual noise energy of the residual noise signal are extracted, and the total virtual error energy is lower than or equal to the total residual noise energy; the short-time energy fluctuation amplitude of the environmental noise signal is extracted and is lower than the preset environmental steady-state threshold; and the impact disturbance value is lower than the preset impact disturbance threshold.

[0015] Furthermore, it also includes a long-term anomaly protection step: when the confidence integral value is continuously lower than the preset locking threshold and the duration exceeds the preset safety time limit, a rollback process is performed; after the rollback is completed, the candidate control parameters are reset, the confidence integral value is reset to the preset initial value, and the adaptive update is restarted; and during the rollback process, the amplitude change of the active noise reduction control signal is extracted and a preset upper limit is applied.

[0016] The present invention has the following beneficial effects: (1) The adaptive active noise cancellation control method of the headphones constructs a reliable anchored acoustic path model to decompose the residual noise into normal acoustic components and abnormal residual components. When the wearer speaks, chews, or adjusts the position of the headphones, the acoustic coupling between the earplug and the ear canal will change instantaneously. This change is manifested as a short and violent impact signal in the residual noise. The dynamic threshold is determined by combining the statistical distribution of the historical residual signal sequence. The residual signal is subjected to transient impact discrimination. The components with amplitude exceeding the threshold and extremely short duration are extracted as abnormal residuals. In this way, the instantaneous fluctuations caused by wearing disturbances are distinguished from the natural changes of environmental noise. The identified abnormal residuals then enter the evaluation stage of updating confidence, suppressing the update amplitude, thereby avoiding the filter coefficients from being adjusted in the wrong direction due to instantaneous disturbances. As a result, the noise reduction parameters remain relatively stable during the wearer's daily activities.

[0017] (2) The adaptive active noise cancellation control method of the headphones uses a virtual error signal to extract the influence of the current effective parameter from the actual residual noise through mapping compensation, and restores the residual noise estimate when the candidate parameter acts alone. All adaptive adjustments of the candidate parameter are based on the virtual error signal. The update result will not directly drive the speaker. Only after long-term verification of the updated confidence assessment and confidence integral, and after meeting multiple synchronous verification conditions, will the candidate parameter replace the current effective parameter. This isolates the error correction caused by disturbance from the actual acoustic output. The noise reduction signal heard by the wearer is always generated by the stable parameters that have been fully verified, thereby avoiding audible impact noise or temporary noise reduction loss caused by parameter misadjustment.

[0018] (3) The adaptive active noise cancellation control method of the headphones extracts an updated evaluation set from multiple dimensions and transforms the evaluation results into an updated confidence score to guide the update authority of the candidate control parameters. The updated evaluation set includes impact disturbance value, gradient direction similarity value, error reduction efficiency value and parameter change range limit value. These indicators evaluate the update quality from the perspectives of wearing status, update direction, error trend and adjustment range, respectively. The updated confidence score is obtained after weighting and first-order recursive smoothing of each indicator in the updated evaluation set to drive the accumulation or decay of the confidence integral, thereby controlling whether the candidate control parameters can ultimately replace the noise reduction control parameters, and thus improving the accuracy of parameter synchronization decision.

[0019] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0020] Figure 1 This is a flowchart of an adaptive active noise cancellation control method for headphones according to the present invention.

[0021] Figure 2 This is a flowchart illustrating the specific steps involved in extracting and updating confidence levels in an adaptive active noise cancellation control method for headphones according to the present invention.

[0022] Figure 3 This is a flowchart illustrating the specific steps of updating based on updated confidence in an adaptive active noise cancellation control method for headphones according to the present invention. Detailed Implementation

[0023] Please see Figure 1 The present invention provides a technical solution: an adaptive active noise cancellation control method for headphones, comprising the following steps: continuously acquiring, frame by frame, the ambient noise signal collected by the external reference microphone of the headphones, the residual noise signal collected by the in-ear error microphone, and the active noise cancellation control signal played by the speaker; wherein, the active noise cancellation control signal acquired in the current frame is calculated by the noise cancellation control parameters that are currently in effect based on the ambient noise signal at that time in the previous frame; Using a pre-established and protected trusted anchored acoustic path model, the abnormal residue of the current frame is extracted based on the environmental noise signal and noise reduction control signal of the current frame and several consecutive previous frames. A copy of the currently effective noise reduction control parameter is made as a candidate control parameter. The candidate control parameter generates a corresponding candidate output signal based on the environmental noise signal. The candidate output signal is independent of the noise reduction control signal. Combined with the reliable anchored acoustic path model, the residual noise signal, and the noise reduction control signal, a virtual error signal is constructed. The virtual error signal of the current frame is equivalent to the residual noise that should have been collected if the speaker was driven by the candidate control parameter alone. The virtual error signal drives the candidate control parameter to perform adaptive updates. Based on the update process of anomaly residuals, virtual error signals, and candidate control parameters across multiple consecutive frames, an update confidence score characterizing the reliability of the current frame update is extracted. During the update process of the candidate control parameters, a confidence integral is set and maintained, with a preset initial value (i.e., the initial value of the confidence integral). Update based on the updated confidence level; When the updated confidence integral meets the preset synchronization verification conditions, the updated candidate control parameters are determined as new noise control parameters. The new noise control parameters are then used to generate active noise cancellation control signals starting from subsequent frames to perform active noise cancellation.

[0024] Specifically, the pre-establishment steps for the trusted anchoring path model are as follows: With the earphone still in the user's ear or on a standard artificial ear, and active noise cancellation disabled, a broadband noise excitation signal covering the target frequency band of active noise cancellation is played from an external sound source. The frequency band of this broadband noise ranges from 20Hz to 2000Hz, and the sound pressure level is typically controlled between 60dB and 75dBSPL. In this example, 70dBSPL is used. The noise excitation response signal received by an external reference microphone is simultaneously acquired. Noise response signal received by the in-ear error microphone ,in This serves as an index for sampling points during the offline identification phase. In a quiet environment, such as when the ambient noise level is below 30 dBSPL, a test signal with a preset frequency band is played through the headphone speaker. This test signal is a logarithmic sweep signal with a sweep range of 20 Hz to 2000 Hz and a sweep duration of 2 seconds. The test signal played through the speaker is recorded simultaneously. Error response signal acquired by in-ear error microphone ; In this offline identification phase, the processing method is to perform the process point by point, using noise to excite the response signal. Input and noise response signals To achieve the desired output, a Normalized Least Mean Square Error (NLMS) adaptive identification algorithm is used to progressively approximate the environmental direct-path transfer function from the external reference microphone to the in-ear error microphone using an FIR filter. This transfer function is then stored as a reliable anchored environmental direct-path sub-model. Let the filter length be The typical value range is from 128 to 512 tap coefficients; in this implementation example, we take... At each sampling point time Perform a coefficient update once, using the following formula: ; in, For the first The estimate of the environmental direct path transfer function at each sampling point is a vector of filter coefficients. For the first The environmental noise excitation response signal vector at each sampling point time. To identify the step size, in this implementation example, we take... , For globally uniform regularization, the value is [value]. , For the first The identification error at each sampling point; After processing all sampling points It has converged, at this time The converged coefficient vector is the estimate of the environmental direct path transfer function, which is stored as a trusted anchored environmental direct path sub-model. With test signal Input and error response signals To achieve the desired output, a minimum mean square error adaptive identification method is employed, which uses an FIR filter to progressively approximate the secondary path transfer function from the loudspeaker to the in-ear error microphone. This approximation is then stored as a reliable anchored secondary path sub-model. Let the filter length be In this implementation example, take At each sampling point time The system performs a coefficient update once, using the following formula: ; in, For the first The estimate of the secondary path transfer function at each sampling point is a vector of filter coefficients. For the first The test signal vector at each sampling point. To identify the step size, in this implementation example, we take... , For the first The identification error at each sampling point.

[0025] After processing all sampling points It has converged, at this time The converged coefficient vector is the estimate of the secondary path transfer function, which is stored as a reliable anchored secondary path sub-model. The trusted anchoring environment direct path sub-model and the trusted anchoring secondary path sub-model together constitute the trusted anchoring acoustic path model. It should be noted that during subsequent normal use, switching to the online processing mode based on frames will only be necessary when updating the confidence level. Impact disturbance value exceeding the preset high threshold and in the current frame Only when the impact disturbance threshold is below the preset threshold can the trusted anchored acoustic path sub-model be updated with a preset minimum step size; otherwise, it remains frozen. During online operation, when updating confidence level The impact disturbance value of the current frame is higher than the preset high confidence threshold (typically ranging from 0.80 to 0.95, and 0.85 in this example). When the shock threshold is below the preset threshold (0.25 in this example), it indicates that the current acoustic environment is stable and the headphones are worn normally, allowing for further disturbance. and Perform restricted updates; Constrained updates are performed separately in a decoupled manner, with a step size much smaller than the initial identification step size. In this implementation example, the constrained update step size is taken. Each frame performs only a single coefficient iteration, that is: In the Frame, update the trusted anchoring environment direct path sub-model First, an error signal stripped of secondary path contributions is constructed. ,in For intra-frame sampling point index, , Frame length: ; With the first The last sample point of the frame The environmental noise signal vector at the location As input, with that point As the desired signal, perform a single normalized minimum mean square error coefficient update: ; Update the trusted anchoring secondary path sub-model First, an error signal stripped of the environmental direct path contribution is constructed. : ; With the first The last sample point of the frame Noise reduction control signal vector at the location As input, with that point As the desired signal, perform a single normalized minimum mean square error coefficient update: ; After the update, continuous verification is required. Frame (in this implementation example) For each frame, the following must be satisfied: the sum of squared prediction errors calculated using the updated anchoring model coefficients for the input signal of that frame is less than the sum of squared prediction errors calculated using the unupdated anchoring model coefficients for the same input signal of that frame. Prediction error is defined as the residual noise signal. Compared with anchored model predictions The difference is: ; The sum of squared prediction errors is the sum of the squares of the differences across the entire frame. ; If any frame during the verification period does not meet the requirements, the update is discarded and the anchoring model coefficients are restored to the state before the update. After performing a single restricted update, immediately pause subsequent restricted update checks and operations until continuous updates are performed. After frame verification is completed, the update result is officially accepted or discarded, the judgment on restricted update conditions is resumed. And when updating the confidence level Below the threshold or shock disturbance value If the threshold is exceeded, immediately freeze all path sub-model update operations, keep the current anchored model unchanged, and prevent disturbance signals from contaminating the path estimation.

[0026] Specifically, the steps for extracting anomaly remnants from the current frame are as follows: The environmental noise signal and noise reduction control signal of the current frame and several consecutive previous frames are input into the trusted anchored environment direct path sub-model and the trusted anchored secondary path sub-model, respectively, for convolution filtering to obtain the environmental noise prediction component and the speaker output prediction component, i.e.: Let the current frame index be Frame length is Each sampling point, in this implementation example Take the current frame and the preceding consecutive frames. frame( The typical value range is 2 to 5 frames; in this implementation example, it is taken as... ) environmental noise signal sequence and noise reduction control signal sequence ; Input the environmental noise signal into the trusted anchored environmental direct path sub-model Perform linear convolution to obtain the predicted components of environmental noise at the in-ear error microphone position: ; in, The first sub-model of the trusted anchored environment direct path Each filter coefficient This is the filter length of the model; Simultaneously, the noise reduction control signal sequence is input into the reliable anchoring secondary path sub-model. Perform linear convolution to obtain the predicted component of the speaker output at the in-ear error microphone position: ; in, For the trusted anchoring of the secondary path sub-model, the first Each filter coefficient The filter length for this model is [length], and both convolution operations truncate the data to be aligned with the current frame. Each sampling point is used as the output; The predicted ambient noise component is added to the predicted loudspeaker output component to obtain the predicted residual signal. This residual signal is then stripped from the predicted residual noise signal to obtain the residual signal. The two prediction components are summed point by point to obtain the prediction residual signal of the current frame. : ; The predicted residual signal represents an instantaneous estimate of the residual noise that should be generated by a known sound source under the current credible anchored acoustic path model; The residual noise signal actually acquired in the current frame With predicted residual signal Subtracting each sample point yields the residual signal. : ; It also acquires historical residual signal sequences, determines anomaly thresholds, and performs transient impact discrimination on the residual signals to obtain the anomalous residuals of the current frame, i.e.: Get from the previous frame Historical residual signal buffer composed of frame residuals ( The typical value range is 50 to 200 frames; in this implementation example, we take... The absolute values ​​of the amplitudes of all sampling points in the historical residual signal are statistically analyzed, and their mean is calculated. and standard deviation This determines the upper limit of the dynamic amplitude threshold. : ; in, This is the threshold coefficient, typically ranging from 3 to 8; in this implementation example, it is set to... Simultaneously, a preset impact duration threshold is set. The typical value range is 10 to 50 sampling points; in this implementation example, we take... One sampling point; The residual signal of the current frame Transient impulse discrimination: For residual signals, if the absolute value of the amplitude exceeds... Extract the continuous sampling point segments. If the duration of the segment is shorter than... If it is a transient impact, the fragment is retained as an abnormal residual of the current frame. ; If the duration exceeds If the change is continuous, it will be considered a continuous change in environmental noise and will not be extracted; otherwise, the absolute value of the amplitude will not exceed [the specified value]. The abnormal residuals corresponding to the sampling points are set to zero.

[0027] Specifically, the steps for constructing the virtual error signal are as follows: The candidate output signal and the noise reduction control signal are subtracted point by point to obtain the output difference signal, i.e.: Let the candidate output signal generated from the environmental noise signal for the current frame's candidate control parameters be... The noise reduction control signal generated by the currently effective noise reduction control parameters is: Both are of length. A time-domain discrete signal sequence; Among them, candidate output signals From environmental noise signals Finite impulse response filter with candidate control parameters The result obtained by filtering is: ; in, The first candidate control parameter Each filter coefficient The filter length is given, and the candidate control parameters and the currently effective noise reduction control parameters use the same filter structure and length, both being finite impulse response filters, differing only in the values ​​of their coefficients. Currently active noise reduction control signal Also caused by environmental noise signals By the currently effective parameters Filtered result, i.e. ; The candidate output signal is subtracted from the noise reduction control signal point by point to obtain the output difference signal. : ; The output difference signal characterizes the difference in output response between the two sets of control parameters under the current ambient noise. The finite impulse response filter coefficient sequence is extracted from the trusted anchoring secondary path sub-model and linearly convolved with the output difference signal to obtain the mapping compensation signal for the current frame, i.e.: Extracting the filter coefficient sequence of the trusted anchored secondary path sub-model from the trusted anchored acoustic path model ; Output difference signal Perform a linear convolution operation with the filter coefficient sequence, for Negative index entries at time The length is obtained as The convolution result is truncated, and the first part aligned with the current frame is extracted. Each sampling point serves as the mapping compensation signal for the current frame. : ; The mapping compensation signal represents the acoustic contribution estimate of the difference between the candidate output signal and the current output signal at the position of the in-ear error microphone after propagation through the secondary path. The residual noise signal is added point-by-point to the mapped compensation signal to obtain the virtual error signal of the current frame, i.e.: The residual noise signal actually acquired in the current frame With the above-mentioned mapping compensation signal The virtual error signal of the current frame is obtained by summing the samples one by one. : ; This virtual error signal is equivalent to the residual noise that the in-ear error microphone should have collected when the speaker was driven solely by the candidate control parameters. Its derivation is as follows: The real residual noise can be expressed as... If changed to candidate output Driven, the ideal residual noise should be ; Using trusted anchoring of secondary path sub-models Approximate Real Secondary Path , that is, .

[0028] The specific steps for adaptively updating candidate control parameters using virtual error signals are as follows: The environmental noise signals of the current frame and several consecutive previous frames are input into the reliable anchoring secondary path sub-model for convolutional filtering to obtain the filtered reference signal for the current frame, i.e.: For filtering reference signal Reserve at least The historical values ​​of points are cached, initially filled with zeros, and pushed in with new values ​​after each frame update. This caches the current frame and consecutive values ​​from previous frames. Environmental noise signal sequence at each sampling point Input trusted anchoring secondary path sub-model Perform linear convolution to obtain the filtered reference signal. : ; The filtered reference signal is multiplied with the virtual error signal of the current frame point by point to obtain the instantaneous gradient vector of the current frame. The gradient direction similarity values ​​of the instantaneous gradient vectors of several consecutive frames are then extracted, i.e.: Assume the candidate control parameters adopt a finite impulse response filter structure, with a filter length of... Consistent with the currently effective parameters, in this implementation example, we take... ; Using the maintained historical filtered reference signal buffer, the filtered reference signal vector for the current frame is constructed as follows: ,in This is the last sampling point of the current frame; The filtered reference signal vector is compared with the virtual error signal of the current frame. The cumulative gradient vector of the current frame is obtained by multiplying and summing the samples one by one. : ; in, for The column vector represents the cumulative gradient direction of each filter coefficient in the current frame throughout the entire frame.

[0029] Maintenance recently frame( The typical value range is 3 to 8 frames; in this implementation example, it is taken as... The cumulative gradient vector history is used to calculate the cumulative gradient vector of the current frame and the cumulative gradient vector of the previous consecutive frames. The mean of the directional dot product of the accumulated gradient vectors of each frame is used as the gradient direction similarity value of the current frame. : ; in, The closer it is to 1, the more consistent the gradient direction; Based on the gradient direction similarity value, the update step size of the current frame is determined, and then weighted stepwise with the instantaneous gradient vector to obtain the coefficient adjustment vector of the current frame, i.e.: according to Dynamically adjust update step size : ; Among them, the basic step size The typical value range is from 0.001 to 0.05; in this implementation example, it is taken as... Then the coefficient adjustment vector is obtained. ; The coefficient adjustment vector is added coefficient by coefficient to the candidate control parameters updated in the previous frame to obtain the candidate control parameters updated in the current frame, i.e.: ; At the same time, the gradient direction similarity value Saved synchronously for subsequent updates to confidence assessment.

[0030] Specifically, such as Figure 2 As shown, the specific steps for extracting the update confidence score, which represents the reliability of the current frame update, are as follows: Based on the update process of anomaly residuals, virtual error signals, and candidate control parameters across multiple consecutive frames, the update evaluation set for the current frame is extracted, including the impact disturbance value, gradient direction similarity value, error descent efficiency value, and parameter change amplitude limit value (the gradient direction similarity value has been calculated and saved during the candidate parameter update process and can be read directly from the update history). The update evaluation set of the current frame is comprehensively processed to obtain the update score value of the current frame, i.e.: The updated score is obtained by weighted summation of the parameters in the updated evaluation set. : ; Wherein, the weight coefficients satisfy In this implementation example, take The higher the impact disturbance value, the greater the disturbance, so its complementary value is used in the positive scoring. The updated score value of the current frame is then subjected to first-order recursive smoothing to obtain the updated confidence score of the current frame, i.e.: ; Among them, the smoothing coefficient The typical value range is from 0.80 to 0.95; in this implementation example, we take... initial value , Update the confidence level for the current frame.

[0031] The specific steps for extracting the updated evaluation set of the current frame are as follows: Residual noise signals from multiple consecutive frames are read and compared with contemporaneous abnormal residual noise. The sum of the abnormal energy and the sum of the residual noise energy are calculated separately, and the impact disturbance value is determined by energy ratio judgment. Take the current frame and the preceding consecutive frames frame( The typical value range is 3 to 10 frames; in this implementation example, it is taken as... ), calculate the total residual energy of the anomaly and total energy of residual noise : ; Abnormal energy percentage , Compare it with the shock disturbance threshold (Using 0.25 in this example) the impact disturbance value is obtained through comparison. : ; Based on the virtual error signal of multiple consecutive frames, the recent average energy and the long-term average energy are extracted, and the error reduction efficiency value is determined by ratio processing. Take the consecutive frames before the current frame Frame (in this implementation example) The virtual error signal. The recent window is the most recent. Frame, the far-term window is further back. Frame, calculate recent average energy and long-term average energy Then the error reduction efficiency value for: ; During the update process of candidate control parameters, the magnitude of the coefficient adjustment vector during the update of candidate control parameters in the current frame is extracted and compared with a preset magnitude threshold. The result of this comparison determines the limit value for the parameter change magnitude. ,Right now: Read the coefficient adjustment vector of the current frame Calculate its Euclidean norm. , and amplitude threshold (Take 0.5 in this implementation example) Comparison: .

[0032] Specifically, such as Figure 3 As shown, the specific steps for updating based on the update confidence level are as follows: The updated confidence level of the current frame is compared with the preset confidence interval, that is: The preset confidence interval is composed of a high confidence threshold. and low confidence threshold In this implementation example, the definition is taken as... , ; When the updated confidence level is higher than the upper limit of the confidence interval, the confidence integral is increased by the preset regular cumulative increment, that is: like Then update the credibility score. : ; Among them, the regular cumulative increment In this implementation example, the value is 0.02, which is the maximum integration value. ; When the updated confidence level is within the confidence interval, the confidence integral remains unchanged in the current frame, that is: like ,but ; When the updated confidence level is lower than the lower limit of the confidence interval, the confidence integral is decayed according to a preset decay function, i.e.: The decay function can be in the form of: ; in, This is the attenuation coefficient, which is 0.70 in this implementation example.

[0033] The specific conditions for synchronous verification are as follows: The updated credibility score has accumulated to a preset synchronization threshold, i.e.: After the integral is updated in the current frame, determine ( ); Furthermore, within the time period from the most recent update of the candidate control parameters until the confidence integral reaches the synchronization threshold, the total virtual error energy of the virtual error signal and the total residual noise energy of the residual noise signal are extracted. During this time period, the total virtual error energy is less than or equal to the total residual noise energy. Let the frame index where the candidate control parameters last started updating be... (That is, the next frame after the last synchronization is completed and the candidate parameters are reset to the currently effective parameters; on the first startup) (Set as frame 1), the frame index where the confidence integral reaches the synchronization threshold is... ; Calculate the total energy of the virtual error signal within this time period. and total energy of residual noise signal : ; Determine if it satisfies Since the two sets of parameters are the same in the initial frame, the virtual error signal energy is equal to the residual noise signal energy. If the candidate parameters are continuously improved in subsequent updates, the total energy must satisfy the relationship of being less than or equal to the current parameter. This condition allows for initial equality and ensures that the overall performance of the candidate parameters is at least not inferior to the current parameters. Within this time period, the short-time energy fluctuation amplitude of the environmental noise signal is extracted and is lower than a preset environmental steady-state threshold, i.e.: Calculate the fluctuation amplitude of the ambient noise signal frame energy within this time period. : ; in , For the first Frame environmental noise energy, if (In this implementation example) This indicates that the environment is stable; And the impact disturbance value of the current frame is lower than the preset impact disturbance threshold, that is: ( ).

[0034] It should be noted that the synchronous verification condition is met when all of the above conditions are met simultaneously. After synchronization is complete, the updated candidate control parameters are determined as the new noise reduction control parameters and assigned... At the same time Save as trusted anchoring control parameters ; and candidate control parameters Reset to the current new noise reduction control parameters This causes the candidate parameters to be adaptively updated again with the next frame as the new starting frame, and the confidence integral is increased. Reset to initial value ; Furthermore, during initialization, the trusted anchoring control parameters are set. Set to factory-pre-stored effective noise control parameters, and the initial effective parameters. same; If no synchronization verification has ever occurred before, this initial value will be used during rollback protection. Each time synchronization verification is successful, [the value will be adjusted accordingly]. Updated to new .

[0035] It also includes long-term anomaly protection steps: When the confidence integral value remains below the preset locking threshold for an extended period exceeding the preset safety time limit, a rollback process is initiated (that is, the currently effective noise reduction control parameters are gradually rolled back frame by frame to the most recently stored trusted anchoring control parameters, which are copies of the noise reduction control parameters saved when the synchronization verification conditions are met and parameter synchronization is completed). like ( The state continues to exceed the safe time limit (Taking 5 seconds in this implementation example), the currently effective parameter will be... Gradually roll back frame by frame to the most recently stored trusted anchoring control parameters ; After entering the rollback process, the adaptive update of the candidate control parameters is paused until the rollback is complete and the candidate parameters are reset. The rollback operation is as follows: ; Among them, the backstep coefficient In this implementation example, the value is 0.05, and it continues until... ; After the rollback is completed, the candidate control parameters are reset with the trusted anchoring control parameters, and the confidence integral is reset to the preset initial value. The adaptive update then restarts. After the rollback is complete, Copy as candidate control parameters and reset Then, the adaptive update of the candidate control parameters is resumed; Furthermore, during the fallback process, the amplitude change of the active noise cancellation control signal output by the speaker is extracted and a preset upper limit is applied to prevent abrupt auditory changes in the output signal during the fallback (and this amplitude limiting protection is only triggered when the inter-frame amplitude change exceeds the preset upper limit, and does not function during normal smooth fallback), that is: Calculate the active noise reduction control signal for the current frame. The maximum amplitude is the same as the previous frame. The ratio of the maximum amplitude to the maximum amplitude; if this ratio exceeds the preset upper limit. (In this implementation example, we take 2.0, corresponding to approximately 3dB of inter-frame gain change), then for the current frame... Multiply the whole by the scaling factor To ensure that its maximum amplitude does not exceed the maximum amplitude of the previous frame. This ensures a smooth transition of the speaker output signal during parameter rollback.

[0036] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0037] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for adaptive active noise cancellation control of headphones, characterized in that, Includes the following steps: Continuously acquire ambient noise signals, residual noise signals, and active noise cancellation control signals from the headphones; Using a pre-established reliable anchored acoustic path model, abnormal residues are extracted based on environmental noise signals and noise reduction control signals; Candidate control parameters are established and corresponding candidate output signals are generated. A virtual error signal is constructed by combining a reliable anchored acoustic path model, residual noise signal, and noise reduction control signal. The virtual error signal is then used to drive the candidate control parameters to be updated adaptively. Based on the update process of abnormal residuals, virtual error signals, and candidate control parameters, the update confidence is extracted, and during the update process of candidate control parameters, a confidence integral is set, and the update is performed according to the update confidence. When the updated confidence integral meets the preset synchronous verification conditions, the updated candidate control parameters are determined as noise control parameters to perform active noise cancellation.

2. The adaptive active noise cancellation control method for headphones according to claim 1, characterized in that, The steps for pre-establishing a trusted anchoring path model are as follows: Acquire noise excitation response signal, noise response signal, test signal, and error response signal; Using the noise excitation response signal as input and the noise response signal as the desired output, the normalized minimum mean square error adaptive identification method is adopted to estimate the environmental direct path transfer function, which serves as a reliable anchoring environmental direct path sub-model. Using the test signal as input and the error response signal as the desired output, the minimum mean square error adaptive identification method is adopted to estimate the secondary path transfer function, which serves as a reliable anchoring secondary path sub-model. The trusted anchored environment direct path sub-model and the trusted anchored secondary path sub-model together constitute the trusted anchored acoustic path model.

3. The adaptive active noise cancellation control method for headphones according to claim 2, characterized in that, The specific steps for extracting anomalous residues are as follows: The environmental noise signal and noise reduction control signal of the current frame and several consecutive frames before it are respectively input into the trusted anchored environment direct path sub-model and the trusted anchored secondary path sub-model for convolution filtering to obtain the environmental noise prediction component and the speaker output prediction component. The predicted environmental noise component is added to the predicted loudspeaker output component to obtain the predicted residual signal, which is then stripped from the residual noise signal to obtain the residual signal. The system acquires historical residual signal sequences, determines anomaly thresholds, and performs transient impact discrimination on residual signals to obtain anomalous residuals.

4. The adaptive active noise cancellation control method for headphones according to claim 2, characterized in that, The specific steps for constructing the virtual error signal are as follows: The candidate output signal and the noise reduction control signal are subtracted point by point to obtain the output difference signal; The finite impulse response filter coefficient sequence is extracted from the trusted anchored secondary path sub-model and linearly convolved with the output difference signal to obtain the mapped compensation signal. The residual noise signal and the mapping compensation signal are added point by point to obtain the virtual error signal.

5. The adaptive active noise cancellation control method for headphones according to claim 4, characterized in that, The specific steps for adaptively updating candidate control parameters using virtual error signals are as follows: The environmental noise signal is input into the trusted anchoring secondary path sub-model for convolution filtering to obtain the filtered reference signal; The filtered reference signal and the virtual error signal are multiplied point by point to obtain the instantaneous gradient vector, and the gradient direction similarity value is extracted. Based on the gradient direction similarity value, the update step size is determined and weighted stepwise with the instantaneous gradient vector to obtain the coefficient adjustment vector; The coefficient adjustment vector is added to the coefficients corresponding to the candidate control parameters one by one to obtain the updated candidate control parameters.

6. The adaptive active noise cancellation control method for headphones according to claim 5, characterized in that, The specific steps for extracting and updating the confidence level are as follows: Based on the update process of abnormal residuals, virtual error signals and candidate control parameters, an updated evaluation set is extracted, including shock disturbance values, gradient direction similarity values, error reduction efficiency values, and parameter change amplitude limits. The updated evaluation set is then processed to obtain the updated score value. The updated score values ​​are then subjected to first-order recursive smoothing to obtain the updated confidence level.

7. The adaptive active noise cancellation control method for headphones according to claim 6, characterized in that, The specific steps for extracting the updated evaluation set are as follows: Read the residual noise signal and compare it with the abnormal residual noise. Calculate the sum of the abnormal energy and the sum of the residual noise energy respectively, and determine the impact disturbance value by processing the energy ratio. Based on the virtual error signal, the recent average energy and the long-term average energy are extracted, and the error reduction efficiency value is determined by ratio processing. During the update process of candidate control parameters, the magnitude of the coefficient adjustment vector is extracted and compared with the preset magnitude threshold to determine the limit value of parameter change magnitude.

8. The adaptive active noise cancellation control method for headphones according to claim 7, characterized in that, The specific steps for updating based on the update confidence level are as follows: The updated confidence level is compared with the preset confidence interval; When the updated confidence level is higher than the upper limit of the confidence interval, the confidence score is increased by the preset cumulative increment. When the updated confidence level is within the confidence interval, the confidence integral remains unchanged; When the updated confidence level is lower than the lower limit of the confidence interval, the confidence integral is decayed and mapped according to the preset decay function.

9. The adaptive active noise cancellation control method for headphones according to claim 7, characterized in that, The specific conditions for synchronous verification are as follows: The updated credibility score has accumulated to the preset synchronization threshold; Extract the total virtual error energy of the virtual error signal and the total residual noise energy of the residual noise signal, wherein the total virtual error energy is lower than or equal to the total residual noise energy; It also extracts the short-time energy fluctuation amplitude of the environmental noise signal, which is below the preset environmental steady-state threshold; Furthermore, the impact disturbance value is lower than the preset impact disturbance threshold.

10. The adaptive active noise cancellation control method for headphones according to claim 1, characterized in that, It also includes long-term anomaly protection steps: When the confidence score remains below the preset lock threshold for an extended period of time, a rollback process is initiated. After the rollback is completed, the candidate control parameters are reset, the confidence integral is reset to the preset initial value, and the adaptive update is restarted. Furthermore, during the fallback process, the amplitude change of the active noise reduction control signal is extracted and a preset upper limit is applied.

Images