Active noise reduction and audio power amplification integrated system

Abstract

The invention relates to the technical field of audio processing, discloses a system named as an integrated system of active noise reduction and audio power amplification, and aims to solve the problem of performance restriction caused by coupling between a power amplifier and a noise reduction module in a highly integrated system. The system is characterized in that in a calibration mode, calibration signals are injected into a D-type power amplifier, coupling noise is collected, and a digital transfer function model from the power amplifier to an active noise reduction microphone is established; in a normal working mode, coupling interference is predicted and eliminated in real time according to a power amplifier driving signal by using the model, and pure environmental noise is obtained for active noise reduction. According to the system, the noise reduction performance is remarkably improved, endogenous interference is eliminated, the robustness is enhanced, and collaborative optimization of noise reduction and amplification efficiency is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of audio processing, and in particular relates to an integrated system of active noise reduction and audio power amplification. Background Art

[0002] The field of audio electronics technology includes multiple links such as signal acquisition, signal processing, power amplification and sound playback. The core of this technology lies in the precise control and high-quality restoration of audio signals. Audio signal processing refers to filtering, enhancing, transforming and other operations on sound signals through electronic circuits or digital algorithms. With the development of consumer electronics, professional audio and communication equipment, audio electronics technology has been widely used in various scenarios such as headphones, speakers, car audio and communication terminals. Technological progress in this field directly determines the user's auditory experience and the market competitiveness of products.

[0003] Active noise reduction (ANC) and audio power amplification are two key technical areas. ANC cancels out ambient noise by generating an anti-noise signal with a phase opposite to the external noise, while audio power amplification boosts low-power audio signals to a level sufficient to drive transducers such as speakers. In modern audio systems, these two functions often need to work together to provide clear, high-quality audio output in noisy environments. The system first uses a noise reduction processing unit to filter out background interference, then feeds the pure audio signal into the power amplifier for amplification, ultimately achieving sound reproduction with a high signal-to-noise ratio.

[0004] Existing technologies mostly adopt a separate architecture, designing the active noise cancellation module and power amplifier module as independent units. This results in a long and complex signal chain, increasing the risk of signal distortion and delay. The electromagnetic interference and heat generated by the power amplifier circuit at high power output can easily couple to the highly sensitive noise reduction front-end circuitry, degrading noise pickup accuracy and noise reduction effectiveness. Independent power management strategies fail to coordinate the real-time processing load of the noise reduction algorithm and the dynamic range of the audio signal, resulting in unnecessary static power consumption and energy waste. The fragmented circuit layout not only consumes valuable hardware space but also increases the number of components and manufacturing costs, making it difficult to meet the trend of miniaturization and lightweighting in consumer electronics. The lack of a unified optimization mechanism for impedance matching and gain calibration between different modules makes it difficult to achieve optimal frequency and transient response for the overall system. These issues are particularly prominent in portable audio devices that require high performance, low power consumption, and a compact design, directly impacting the product's overall performance and user experience. Summary of the Invention

[0005] The purpose of the present invention is to overcome the defects in the prior art of mutual performance constraints caused by electromagnetic coupling and power supply noise coupling between the audio power amplifier module and the active noise reduction module in a highly integrated system, and to propose an integrated system of active noise reduction and audio power amplification.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for integrating active noise reduction and audio power amplification, comprising the following steps: S1. During the system initialization or preset calibration mode, the power amplification of the external audio signal is stopped, and an internal calibration signal generator generates a pseudo-random binary sequence calibration signal having preset spectrum and time domain characteristics; the pseudo-random binary sequence calibration signal is injected into the digital pulse width modulation (PWM) front end of the Class D audio power amplifier inside the system, driving the power stage of the Class D audio power amplifier to perform switching operations at a preset static operating point, thereby generating a deterministic reference coupling noise caused by the power stage switching action in the system's power supply network and physical space.

[0007] S2. Synchronously, the feedforward microphone of the active noise reduction (ANC) module of the system collects the reference coupled noise coupled to the feedforward microphone signal link by the power stage switching action via the system's shared power bus, ground plane, and spatial electromagnetic radiation, and digitizes the collected analog noise signal through a high-precision analog-to-digital converter (ADC) to generate a reference coupled noise digital sequence; the digital signal processing unit of the system receives a digital copy of the pseudo-random binary sequence calibration signal and the reference coupled noise digital sequence.

[0008] S3. The digital signal processing unit executes a system identification algorithm, using the digital replica of the pseudo-random binary sequence calibration signal as the input signal and the reference coupled noise digital sequence as the output signal, to calculate and establish a digital transfer function model that can accurately characterize the complete noise coupling path from the PWM front end of the Class D audio power amplifier to the ADC output of the feedforward microphone; the digital transfer function model is solidified and stored in a non-volatile memory of the system as a mathematical basis for subsequent real-time interference cancellation.

[0009] S4. Under the normal working mode of the system, an audio PWM digital signal stream to be input to the PWM front end of the Class D audio power amplifier is obtained in real time; the real-time audio PWM digital signal stream is used as input, and a convolution operation is performed through the solidified and stored digital transfer function model, so as to generate a predicted coupled interference digital signal sequence in real time and sample point by sample point, wherein the predicted coupled interference digital signal sequence accurately reproduces the noise that will be actually generated and coupled to the front UCE microphone signal link due to the audio PWM signal driving the power stage in terms of time domain waveform and spectral components.

[0010] S5. Perform a sampling point-by-sample subtraction operation in the digital domain on a mixed digital signal collected by the feedforward microphone in normal working mode, where the mixed digital signal includes ambient noise, audio leakage, and actual coupling noise generated by the audio power amplifier, from the predicted coupling interference digital signal sequence to achieve feedforward cancellation of the amplifier coupling noise and generate a purified ambient noise digital sequence from which the amplifier's intrinsic interference has been eliminated; and input the purified ambient noise digital sequence into an adaptive ANC filtering algorithm unit to generate a final reverse noise signal and drive an output transducer to complete active noise reduction.

[0011] As a further solution of the present invention, the pseudo-random binary sequence calibration signal has a specific code length and autocorrelation characteristics, and its code length is set to be greater than four times the system impulse response length of the noise coupling path to ensure that the full-band response of the coupling path can be fully stimulated during the system identification process, and the value of its autocorrelation function at non-zero delay is close to zero to reduce the cross-correlation noise during the identification process.

[0012] As a further embodiment of the present invention, the system identification algorithm is a recursive least squares (RLS) algorithm. The RLS algorithm iteratively updates the coefficient vector of the digital transfer function model to minimize the mean square error between the reference coupled noise digital sequence and the predicted noise sequence calculated using the currently estimated transfer function model. The core iterative update formula is specifically: First, define the input signal vector at time n ,in is the pseudo-random binary sequence calibration signal sample at time n, and L is the order of the digital transfer function model. is the transfer function model coefficient vector estimated at time n-1.

[0013] Secondly, calculate the prior error ,in is the digital sequence sample of the reference coupled noise at time n.

[0014] Again, calculate the gain vector in is the inverse covariance matrix at time n-1, and λ is the forgetting factor, which ranges from 0.98 to 1.0 and is used to adjust the algorithm's ability to track the time-varying characteristics of the system.

[0015] Then, update the coefficient vector .

[0016] Finally, update the inverse covariance matrix Through the above iterative process, until The change in is less than a preset convergence threshold, and the final These are the coefficients of the digital transfer function model.

[0017] As a further solution of the present invention, the digital transfer function model is a finite impulse response (FIR) filter model, whose coefficients are the coefficient vector identified by the RLS algorithm. In step S4, the convolution operation is specifically implemented as follows: in is the predicted coupled interference digital signal sample generated at time n, for The real-time audio PWM digital signal sample at the time instant, L is the order of the FIR filter, and the order is set to be sufficient to cover the main impulse response delay of the coupling path.

[0018] As a further embodiment of the present invention, the method further comprises step S6: S6. After performing feedforward cancellation in step S5, the digital signal processing unit maintains a parallel, low-update-rate residual correction loop. This loop extracts the purified ambient noise digital sequence after the feedforward cancellation and performs a short-time Fourier transform (STFT) on it to analyze whether there are residual spectral components corresponding to the switching frequency of the Class D audio power amplifier and its harmonic frequencies. If the energy of the residual spectral components exceeds a preset dynamic correction threshold, a slow adaptive algorithm is activated to fine-tune the coefficients of the fixed-stored digital transfer function model to compensate for time-varying coupling path characteristics caused by factors such as system operating temperature and power supply voltage fluctuations.

[0019] As a further embodiment of the present invention, the slow adaptive algorithm in step S6 is a Normalized Least Mean Square (NLMS) algorithm. This algorithm uses the real-time audio PWM digital signal stream as a reference signal and the residual component extracted from the purified ambient noise digital sequence as an error signal to fine-tune the coefficients h_c of the digital transfer function model. The update formula is: in is the updated coefficient vector, is a very small, fixed step size factor to ensure the stability of the update process, is the extracted residual signal, is the real-time audio PWM signal vector, is a regularization constant to prevent the denominator from being zero.

[0020] As a further solution of the present invention, the method further includes step S7: S7: The digital signal processing unit executes a joint optimization strategy. The strategy establishes a joint cost function ,in is the final system error signal after ANC filtering, r(n) is the residual coupling noise energy in the residual correction loop, is the power consumption evaluation value of the class-D audio power amplifier. and The digital signal processing unit dynamically adjusts the switching frequency of the Class D audio power amplifier according to the real-time changing audio content. and the core parameters of the adaptive ANC filter to minimize the joint cost function while meeting specific audio playback requirements .

[0021] An integrated system of active noise reduction and audio power amplification, for performing the above method, comprising: The calibration signal generation and injection module integrates a pseudo-random binary sequence (PRBS) generator. The PRBS generator is configured to generate a calibration signal sequence with deterministic statistical characteristics in the system calibration mode and provide the sequence as a digital input to the pulse width modulation (PWM) logic unit of the Class D audio power amplifier.

[0022] A noise coupling path identification module includes a high-precision analog-to-digital converter (ADC) connected to the output of an active noise cancellation (ANC) feedforward microphone and connected to a digital signal processing (DSP) core; the DSP core is configured to receive a copy of the calibration signal sequence generated by the PRBS generator and a reference coupled noise sequence digitized by the ADC, and execute a recursive least squares (RLS) system identification algorithm to calculate and generate a digital transfer function model that accurately characterizes the noise coupling characteristics between the amplifier and the microphone link based on these two input sequences.

[0023] The predicted interference generation module is configured, in the normal operating mode of the system, to intercept the audio PWM digital signal stream from the front end of the PWM logic unit of the Class D audio power amplifier in real time, and input this signal stream into a finite impulse response (FIR) filter defined by the digital transfer function model generated by the noise coupling path identification module. Through real-time convolution operation, a predicted coupling interference digital signal sequence is generated that is highly consistent with the actual coupling noise in the time domain and frequency domain.

[0024] A feedforward digital cancellation module is provided between the ADC output of the ANC feedforward microphone and the input of the ANC core filtering algorithm. The module comprises a digital subtractor; one input of the subtractor receives a mixed digital signal containing ambient noise and actual coupled noise from the ADC, and the other input receives a sequence of predicted coupled interference digital signals from the predicted interference generation module. By performing sample-by-sample subtraction, the module outputs a purified ambient noise signal from which the amplifier's intrinsic interference has been filtered out, to the ANC core filtering algorithm.

[0025] A residual monitoring and model correction module is connected in parallel to the output of the pre-UCE digital cancellation module. This module includes a bandpass filter bank and a slow adaptive algorithm unit. The bandpass filter bank is configured to detect residual energy related to the amplifier switching frequency in the purified ambient noise signal. When the energy exceeds the limit, the slow adaptive algorithm unit is activated. This unit adopts the normalized least mean square (NLMS) algorithm and uses the real-time audio PWM signal as a reference to fine-tune the coefficients of the digital transfer function model to adapt to the slow drift of system parameters.

[0026] Compared with the prior art, the advantages and positive effects of the present invention are: By establishing a precise mathematical model of endogenous interference at the system level, this method transforms the uncertain coupling noise, difficult to suppress in traditional designs, into a predictable, deterministic signal that can be precisely canceled. Through an offline system identification process, this method captures the complete and complex coupling path from the power amplifier to the ANC sensor, including all channels through the power supply, ground, and spatial radiation, generating a highly accurate transfer function model. In real-time operation, this model is used to proactively predict and digitally cancel impending coupling noise, eliminating the source of the power amplifier's interference with ANC performance. This allows the ANC system to operate in an extremely low-noise environment, significantly improving the depth and width of noise reduction. Furthermore, the proposed residual error correction loop dynamically compensates for model mismatches caused by temperature drift, aging, and other factors, ensuring long-term robustness. Ultimately, by jointly optimizing noise reduction performance, interference suppression, and power consumption, this method achieves a synergistic optimization of active noise reduction performance and audio amplification efficiency within the constraints of a compact physical space and shared power supply, overcoming the inherent contradiction between the two in traditional integrated solutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 A schematic diagram of the overall process of an integrated method for active noise reduction and audio power amplification provided by an embodiment of the present invention; Figure 2 A schematic diagram of the functional modules of an integrated system for active noise reduction and audio power amplification provided by an embodiment of the present invention; Figure 3 for Figure 1 Detailed flowchart of the noise coupling path identification and modeling steps in the method shown; Figure 4 for Figure 1 A detailed flowchart of the real-time interference cancellation and active noise reduction processing steps in the method shown; Figure 5 The figure is a schematic diagram of a signal interaction process in a normal working mode according to an embodiment of the present invention. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be noted that in the description of the present invention, unless otherwise clearly specified or limited, the terms "connect", "fix" and the like should be understood in a broad sense.

[0029] In a specific embodiment, referring to Figures 1 to 5A method and system for integrating active noise reduction (ANC) with audio power amplification is proposed. The core of this method is to establish an accurate digital model to characterize the coupling path from the audio power amplifier switching noise to the ANC microphone signal chain through an offline system identification process. This model is then used to perform proactive, real-time digital cancellation in normal operating mode.

[0030] Reference Figure 1 and Figure 3 The initial stage of this method is system calibration and noise coupling path identification modeling. The specific steps are as follows: Step S1: In system initialization or a preset calibration mode, power amplification of an external audio signal is stopped, and an internal calibration signal generator generates a pseudo-random binary sequence calibration signal having preset spectrum and time domain characteristics; the pseudo-random binary sequence calibration signal is injected into a digital pulse width modulation front end of a Class D audio power amplifier within the system, driving the power stage of the Class D audio power amplifier to perform switching operations at a preset static operating point, thereby generating a deterministic baseline coupled noise caused by the power stage switching action in the system's power supply network and physical space.

[0031] Specifically, step S1 can be further decomposed into several sub-steps.

[0032] In step S101, the system enters calibration mode. This mode can be triggered automatically when the system is first powered on, when a specific calibration command is received from a host processor or user interface, or when the system detects a significant change in a key operating parameter (such as ambient temperature or power supply voltage) exceeding a preset threshold. Upon entering this mode, the system controller asserts a control signal, which acts on a digital switch or multiplexer in the audio signal path, disconnecting the external audio source (such as a Bluetooth audio stream or line input) from the Class-D audio power amplifier PWM front end. This ensures that no external audio signal drives the power stage during calibration.

[0033] Step S102: Start an internal calibration signal generator to generate a pseudo-random binary sequence (PRBS) calibration signal. The calibration signal generator is a hardware logic module based on a linear feedback shift register (LFSR) or a software module running in a digital signal processing unit. In one specific embodiment, the PRBS generator is configured to generate a maximum length sequence (m-sequence).

[0034] Furthermore, to ensure the accuracy and robustness of system identification, the PRBS parameters are precisely set. The parameter sources and specific values ​​are as follows: 1. Generating polynomial: Choose a 16th-order primitive polynomial, for example This choice ensures that the generated sequence has The code length is much longer than the impulse response length estimated by the noise coupling path.

[0035] 2. Impulse response length estimation: Through preliminary physical testing and simulation analysis of the system prototype, it is estimated that the impulse response duration of the main energy of the complete coupling path from the power amplifier switch node to the feedforward microphone ADC output is about 12 milliseconds. Corresponding to the system's 48kHz sampling rate, this duration is equivalent to 12ms*48samples / ms=576 sampling points. According to system identification theory, in order to avoid the time aliasing effect during the identification process, the PRBS code length N should be at least the impulse response length Four times, that is, N>4* Here, 65535>4*576=2304, and this condition is fully satisfied.

[0036] 3. Sequence clock frequency: The update clock of the PRBS generator is synchronized with the system main sampling clock and is set to =48kHz, ensuring the generated calibration signal sequence It is strictly aligned with the subsequent ADC sampling sequence in the time domain.

[0037] 4. Sequence Amplitude: The generated logic "0" and "1" are mapped to 16-bit signed integers with specific digital amplitudes. Specifically, a logic "1" is mapped to +16384, and a logic "0" is mapped to -16384. This bipolar representation eliminates the DC component of the signal, avoiding the introduction of unnecessary DC offset into the power stage during calibration.

[0038] Step S103: calibrate the generated PRBS signal sequence The PWM is injected into the digital pulse-width modulation (PWM) front end of the Class-D audio power amplifier. The injection point is located at the data input of the PWM modulator. The PWM modulator receives the bipolar PRBS sequence and linearly converts it into a duty cycle instruction for the PWM waveform. In a specific embodiment, the PWM modulator operates in natural sampling mode. When the input is +16384, the duty cycle of the output PWM waveform is set to 52.5%; when the input is -16384, the duty cycle is set to 47.5%. The center duty cycle is 50%. This small duty cycle perturbation (±2.5%) is sufficient to stimulate the power stage to generate measurable switching noise. At the same time, because the average duty cycle is close to 50%, the average voltage from the power amplifier output to the speaker load is close to zero during calibration, that is, operating at a preset static operating point, avoiding external audible noise.

[0039] Step S104: driving the power stage and generating a reference coupled noise. The high frequency pulse sequence (eg, carrier frequency) generated by the PWM modulator and carrying the PRBS information =400kHz) is sent to the gate driver of the power stage to control a full-bridge or half-bridge power MOSFET for high-speed switching. Due to the rapid turn-on and turn-off of the MOSFET, significant current transients (di / dt) will be generated on the shared power bus VDD_AMP of the system. These transients form voltage ripples through the finite impedance of the power network. At the same time, the switching current flows through the shared PCB ground plane, causing ground potential fluctuations, namely ground bounce. In addition, the switching node (SwitchingNode) and output filter inductor of the power stage will radiate high-frequency electromagnetic fields into the surrounding space. These three physical phenomena - power supply noise coupling, ground plane noise coupling and spatial electromagnetic radiation coupling - together constitute a complete noise coupling path from the power amplifier to other sensitive circuits in the system, and ultimately superimpose on the signal chain of the ANC feedforward microphone to form a deterministic and PRBS signal. Strictly correlated reference coupled noise.

[0040] In step S2, synchronously, the feedforward microphone of the active noise reduction (ANC) module of the system collects the reference coupled noise coupled to the feedforward microphone signal link by the power stage switching action via the system's shared power bus, ground plane, and spatial electromagnetic radiation, and digitizes the collected analog noise signal through a high-precision analog-to-digital converter (ADC) to generate a reference coupled noise digital sequence; the digital signal processing unit of the system receives a digital copy of the pseudo-random binary sequence calibration signal and the reference coupled noise digital sequence.

[0041] Specifically, step S2 can be further decomposed into the following sub-steps.

[0042] Step S201, synchronously collect the reference coupling noise. The core sensing element of the ANC system, the feedforward microphone, is activated and working. The microphone and its front-end analog preamplifier circuit are physically close to the power amplifier and share part of the power supply and ground network, so they will inevitably pick up the reference coupling noise described in step S104. The collection process is strictly synchronized with the PRBS generation in step S1, which means that the ADC sampling clock and the PRBS generation clock are co-origin or phase-locked, ensuring and subsequently generated There is a definite causal and temporal relationship between them.

[0043] Step S202: Analog signal processing and digitization. The weak analog noise signal picked up by the microphone is first amplified by a low-noise preamplifier. The gain (e.g., +20dB) is set high enough to bring the noise signal level within the optimal input range of the ADC, but low enough to saturate it. The signal then passes through an anti-aliasing filter. This filter is a third-order Butterworth low-pass filter with a -3dB cutoff frequency set to 21.6kHz, or 0.45 times the sampling frequency. This filter removes high-frequency components above the Nyquist frequency and prevents spectral aliasing. The processed analog signal is then fed into a high-precision analog-to-digital converter (ADC).

[0044] To ensure accurate quantification of weak coupled noise, the performance parameters of the ADC are strictly specified: 1. Type: Σ-Δ (Sigma-Delta) ADC, which excels in high-resolution applications due to its inherent oversampling and noise shaping characteristics.

[0045] 2. Resolution: 24 bits. This resolution provides a theoretical dynamic range of 144dB, which is sufficient to capture noise signals with a wide dynamic range.

[0046] 3. Sampling rate fs_adc: 48kHz, consistent with the PRBS generation rate.

[0047] 4. Effective number of bits (ENOB): not less than 17 bits, corresponding to a signal-to-noise ratio (SNR) better than 104dB.

[0048] The output of the ADC is a 24-bit signed integer sequence, namely the reference coupled noise digital sequence y_cal(n).

[0049] Step S203: Data stream aggregation. The digital signal processing unit (DSP) core receives data from two sources in parallel. One is the calibration signal sequence obtained from the PRBS generator. The first is a digital copy of the reference coupled noise digital sequence obtained from a high-precision ADC. The two sequences are written in real time into a designated RAM area within the DSP, typically two independent circular buffers, each with a size of at least the PRBS code length N, or 65535 samples, providing a complete data set for subsequent system identification algorithms.

[0050] In step S3, the digital signal processing unit executes a system identification algorithm, using the digital replica of the pseudo-random binary sequence calibration signal as an input signal and the reference coupled noise digital sequence as an output signal, to calculate and establish a digital transfer function model that can accurately characterize the complete noise coupling path from the PWM front end of the Class-D audio power amplifier to the ADC output of the feedforward microphone; the digital transfer function model is permanently stored in a non-volatile memory of the system and serves as a mathematical basis for subsequent real-time interference cancellation.

[0051] Reference Figure 3 ,The modeling process is specifically implemented as follows.

[0052] Step S301, select the system identification algorithm. In this embodiment, the Recursive Least Squares (RLS) algorithm is used. The reason for choosing the RLS algorithm is that it has a faster convergence speed than other algorithms (such as LMS) and can obtain high-precision model estimation in a shorter time, which is crucial for a one-time offline calibration process. The goal of this algorithm is to iteratively adjust the coefficient vector h_c of a finite impulse response (FIR) filter model so that the input signal is The output after the filter The actual measured noise signal The mean square error between them is the smallest.

[0053] Step S302: Initialize the RLS algorithm. Before starting the iterative calculation, the various parameters of the algorithm must be initialized.

[0054] 1. Model order L: This parameter defines the length of the identified FIR filter, that is, the dimension of the coefficient vector h_c. Its value must be greater than or equal to the impulse response length of the actual coupling path Based on the estimated value in step S102 =576, and considering computational efficiency (choosing a power of 2), the model order L is set to 1024. This setting ensures that the model has enough capacity to fully capture the dynamic characteristics of the coupling path, including its main delay and decay oscillations.

[0055] 2. Forgetting factor λ: This parameter controls the algorithm's weighting of historical data. During the calibration phase, the coupling paths are assumed to be time-invariant, so all historical data are given equal weight. The forgetting factor λ is set very close to 1.0, specifically λ = 0.9998. This value ensures effective averaging over the entire dataset while allowing the algorithm to converge smoothly from inaccurate initial states.

[0056] 3. Initial coefficient vector h_c(0): Without any prior information, the initial coefficient vector of the model is set to a zero vector with a length of L=1024, that is, =[0,0,...,0]^T.

[0057] 4. Initial inverse covariance matrix P(0): This matrix reflects the confidence level of the initial coefficient vector h_c(0). Initialize it as , where I is the LxL identity matrix, is a large positive number. In this embodiment, is set to 100. The larger A value of indicates that the confidence in the initial coefficients is very low, which will cause the algorithm to use a larger correction step size in the initial stage, thereby accelerating convergence.

[0058] Step S303: Perform RLS iterative calculation. After receiving at least L sampling points, the RLS algorithm begins to iteratively update each sampling point. For each new sampling point moment n (n>=L), perform the following calculation: 1. Construct input signal vector: This is a column vector containing the current and past L-1 calibration signal samples.

[0059] 2. Calculating the Prior Error : This error indicates that the current model For actual noise prediction bias.

[0060] 3. Calculate the gain vector k(n): The gain vector k(n) determines the weight of the coefficient vector modified by the current error e(n).

[0061] 4. Update coefficient vector : This is the core update step of the RLS algorithm, which adjusts the coefficient vector in the direction of reducing the error.

[0062] 5. Update the inverse covariance matrix P(n): This step prepares for the next iteration.

[0063] As a specific calculation example, assume that at time n=1024: Vector constructed The value is -5231 (24-bit quantization value), is the coefficient vector obtained from the previous iteration.

[0064] First calculate the predicted value = * , assuming the result is -4800.

[0065] Then the prior error e(1024)=-5231-(-4800)=-431.

[0066] Then, using P(1023) and The gain vector k (1024) is calculated.

[0067] Then update the coefficient vector: = +k(1024)*(-431).

[0068] Finally, the inverse covariance matrix P (1024) is updated.

[0069] This iterative process continues until the convergence condition is met.

[0070] Step S304: Convergence judgment and model solidification. The iterative process continues until the change in the coefficient vector converges. The convergence criterion is defined as the moving average of the change in the coefficient vector's two-norm for M consecutive sampling points (e.g., M=1000) is less than a preset convergence threshold. .Right now ton is less than . The value of is set to 1e-12, a trade-off based on system simulations between ensuring sufficient stability of the model coefficients and avoiding excessively long calibration times. The iterative process terminates when the convergence criteria are met, and the final coefficient vector h_c is considered to be an accurate FIR approximation of the noise coupling path transfer function H(z).

[0071] Step S305: Model storage. The final coefficient vector The coefficients (an array of 1024 elements) are read from the DSP's RAM and written to a nonvolatile memory on the system board, such as EEPROM or Flash memory. Each coefficient is stored as a 32-bit single-precision floating-point number, occupying a total of 1024 * 4 = 4096 bytes of memory. This stored model will be used later in the system's normal operating mode.

[0072] As an alternative, the system identification algorithm can also use the Normalized Least Mean Square (NLMS) algorithm. The coefficient update formula of the NLMS algorithm is: in, is the step size factor, a constant (e.g., 0.05) that needs to be set empirically or experimentally; δ is a small positive constant used to prevent the denominator from being zero. Compared to the RLS algorithm, the NLMS algorithm significantly reduces computational complexity (from O(L^2) to O(L)), requiring less DSP computing resources. However, its convergence rate is typically much slower than RLS, and longer calibration time may be required to achieve equivalent model accuracy. In applications that are not sensitive to calibration time but are computationally cost-sensitive, NLMS can serve as an effective alternative.

[0073] As an alternative, a frequency domain-based identification method can be used. This method first acquires and stores the complete PRBS input sequence. and the corresponding coupled noise output sequence , each sequence length is N=65535. Then, fast Fourier transform (FFT) is performed on the two sequences respectively to obtain their frequency domain representation and The frequency domain response of the transfer function It can be calculated by point-by-point complex division: = / Since the power spectrum of the m-sequence is theoretically flat, is a close constant, making the division numerically stable. Perform an inverse fast Fourier transform (IFFT) to get the impulse response in the time domain This method is very accurate when the signal-to-noise ratio is high and is conceptually simple. However, it is a batch method that requires storing the entire dataset, which places a high demand on RAM. Its performance is also sensitive to the synchronization and periodicity of the data acquisition windows.

[0074] In addition, an exception handling mechanism is designed in the calibration process. For example, if the RLS algorithm processes the entire PRBS sequence (65535 samples) and the change in its coefficient vector still does not meet the convergence threshold, , the system will determine that the calibration has failed. In this case, the system will log a specific error code and can choose to load a pre-existing universal (or "golden") transfer function model suitable for similar hardware as a fallback, or directly disable the interference cancellation function and operate only in standard ANC mode. At the same time, the system will issue a calibration failure alert to the upper-layer application or user.

[0075] After the system has been successfully calibrated and the transfer function model has been solidified After that, the system will exit the calibration mode and enter the normal operating mode. In the normal operating mode, the system performs real-time interference prediction, cancellation and active noise reduction functions. Figure 1 、 Figure 4 and Figure 5 The specific steps of this stage are as follows: Step S4, in the normal working mode of the system, obtaining in real time an audio PWM digital signal stream to be input into the PWM front end of the Class D audio power amplifier; using the real-time audio PWM digital signal stream as input, performing a convolution operation through the solidified stored digital transfer function model, thereby generating a predicted coupled interference digital signal sequence in real time and sample point by sample point, wherein the predicted coupled interference digital signal sequence accurately reproduces, in terms of time domain waveform and spectral components, the noise that will actually be generated by the audio PWM signal driving the power stage and coupled to the feedforward microphone signal chain.

[0076] Specifically, the execution process of step S4 includes the following sub-steps.

[0077] Step S401: Model loading and system mode switching. When switching from calibration mode to normal operating mode, the system controller first reads the transfer function model coefficient vector solidified in step S305 from a non-volatile memory (such as EEPROM). , and loads it into the high-speed static random access memory (SRAM) inside the digital signal processing unit (DSP) for low-latency real-time access. At the same time, the system controller reconfigures the digital switches or multiplexers on the audio path to connect the external audio source (such as the I2S digital audio stream from the Bluetooth SoC) to the signal processing chain of the Class-D audio power amplifier.

[0078] Step S402: Real-time capture of the audio PWM digital signal stream. The original audio signal (e.g., PCM format) provided by the external audio source first undergoes a series of standard audio processing, including volume control, equalizer (EQ) adjustment, dynamic range compression (DRC), etc. The processed audio signal is sent to the digital pulse width modulator (PWM) of the Class D amplifier. At the digital input end of the PWM modulator, the signal stream is copied in real time and sample by sample point, and sent to the calculation unit responsible for interference prediction. It is a 16-bit or 24-bit digital sequence whose amplitude directly corresponds to the duty cycle of the PWM waveform to be generated. The interception operation must be completed before the signal enters the PWM modulator to ensure timely prediction.

[0079] Step S403: Perform real-time convolution operation to generate a predicted coupled interference signal. This step is the core of interference cancellation. The digital signal processing unit implements a finite impulse response (FIR) filter, the coefficients of which are the transfer function model coefficient vector loaded in step S401. The intercepted real-time audio PWM digital signal stream As the input of the FIR filter. At each sampling time n, the DSP performs a convolution operation, and the calculation formula is: in, is the predicted coupled interference digital signal sample generated at time n; is the kth element of the model coefficient vector; It is at the moment The audio PWM signal samples are obtained; L is the order of the FIR filter, which is 1024.

[0080] To achieve efficient real-time calculation, the convolution operation is performed in the DSP through an optimized multiply-accumulate (MAC) instruction set. are stored in a circular buffer of length L = 1024. In each sampling cycle, the latest input sample are written to the buffer, overwriting the oldest sample At the same time, the MAC unit of the DSP completes 1024 multiplications and 1024 additions in a single instruction cycle or a few cycles. For example, in a DSP with an operating frequency of 200MHz, the time required to complete a 1024-order FIR filter calculation is much less than a sampling cycle (1 / 48kHz≈20.8 microseconds), thus ensuring the prediction signal All calculations are performed using 32-bit single-precision floating-point format to ensure sufficient numerical accuracy and avoid quantization errors during the convolution process that may reduce the prediction accuracy.

[0081] Step S404: output the predicted signal. The calculated predicted coupling interference digital signal sequence is output to the feedforward digital cancellation module. It should be emphasized that due to is based on the signal that will drive the power stage The calculation is almost instantaneous compared to the physical delay of the signal through the power stage and coupling path to the microphone, so This constitutes an accurate forward-looking prediction of the actual coupled noise that will occur.

[0082] Step S5: performing a sampling point-by-sample subtraction operation in the digital domain on a mixed digital signal collected by the feedforward microphone in a normal working mode, the mixed digital signal including ambient noise, audio leakage, and actual coupling noise generated by the audio power amplifier, from the predicted coupling interference digital signal sequence to achieve feedforward cancellation of the amplifier coupling noise, thereby generating a purified ambient noise digital sequence from which the amplifier's intrinsic interference has been eliminated; and inputting the purified ambient noise digital sequence into an adaptive ANC filtering algorithm unit to generate a final reverse noise signal and drive an output transducer to complete active noise reduction.

[0083] Reference Figure 4 and Figure 5 , the specific implementation of step S5 is as follows.

[0084] Step S501: Collect mixed signals in normal working mode. In normal working mode, the feedforward microphone continues to work and collects the acoustic signals around it. At this time, the signal picked up by the microphone is a complex mixture, whose main components include: 1. External environmental noise: This is the target signal that the active noise reduction system aims to eliminate.

[0085] 2. Audio Leakage: The audio signal played by the system speaker leaks back to the feedforward microphone through the acoustic path.

[0086] 3. Actual coupling noise d(n): The noise generated by the Class D power amplifier when amplifying the audio signal The intrinsic interference generated when the microphone is operating is coupled into the microphone signal chain through paths such as power supply, ground and space radiation.

[0087] This mixed analog signal is also subjected to the low-noise preamplification, anti-aliasing filtering and high-precision ADC digitization described in step S202 to generate a mixed digital signal sequence y_mix(n).

[0088] Step S502: Perform feedforward digital cancellation. A digital subtractor is deployed between the output of the ADC and the input of the main ANC filter algorithm. The two inputs of the subtractor receive the mixed digital signals from the ADC and the input of the main ANC filter algorithm. and the predicted coupled interference digital signal from step S403 At each sampling time n, a subtraction operation is performed: .

[0089] Since the model h_c established in step S3 is very accurate, the predicted signal The waveform and phase are highly consistent with the actual coupling noise d(n). Therefore, the subtraction operation can effectively remove the actual coupling noise d(n) from the mixed signal y_mix(n). Ideally, ≈d(n), then the signal after subtraction is Will primarily contain only ambient noise and audio leakage components.

[0090] Step S503: Input the purified ambient noise signal into the adaptive ANC filtering algorithm. The signal is used as the feedforward input signal of the main ANC processing loop. In a typical feedback or hybrid ANC system, the signal is input to an adaptive ANC filter (e.g., a filter based on the Filtered-xLeastMeanSquare, FxLMS algorithm). The high-frequency amplifier-coupled noise that is strongly correlated with the audio signal no longer exists, and the ANC adaptive algorithm no longer needs to expend its adaptive capacity to model and eliminate this interference. This brings significant performance advantages: 1. Improved noise reduction depth: The ANC algorithm can focus more on modeling external environmental noise, thereby achieving lower residual error, that is, deeper noise reduction effect.

[0091] 2. Expanded noise reduction bandwidth: Eliminates the interference of high-frequency coupled noise, allowing the ANC system to remain stable within a wider frequency band and effectively suppress higher-frequency noise.

[0092] 3. Improved stability: It avoids the nonlinear positive feedback that may be formed between the amplifier noise and the ANC loop, reduces the risk of system howling or instability, and allows the use of higher ANC loop gain.

[0093] As a further solution of the present invention, in order to cope with the slow drift of system parameters due to changes in the working environment (such as temperature, voltage) or aging of components, the solidified transfer function model To address the mismatch problem, the method also includes a parallel, low-update-rate residual correction loop.

[0094] In step S6, after performing feedforward cancellation in step S5, the digital signal processing unit maintains a parallel, low-update-rate residual correction loop. This loop extracts the purified ambient noise digital sequence after the feedforward cancellation and performs a short-time Fourier transform (STFT) on it to analyze whether residual spectral components corresponding to the switching frequency of the Class D audio power amplifier and its harmonics are present. If the energy of the residual spectral components exceeds a preset dynamic correction threshold, a slow adaptive algorithm is activated to fine-tune the coefficients of the stored digital transfer function model to compensate for time-varying coupling path characteristics caused by factors such as system operating temperature and power supply voltage fluctuations.

[0095] Specifically, the implementation details of step S6 are as follows.

[0096] Step S601, residual signal monitoring. The loop is based on the output signal of step S502. As input. Theoretically, if the model Perfect match, There should be no deterministic frequency content associated with the switching behavior of the amplifier. Here it is regarded as the residual signal r(n). In order to detect the weak residual coupling noise, an efficient frequency domain analysis method is adopted. In a specific embodiment, a set of parallel Goertzel algorithm filters are used. The nominal switching frequency of the class D amplifier is set to =400kHz, which will appear at a lower frequency due to aliasing at a 48kHz sampling rate. More importantly, its modulation sideband noise will fall into the audio frequency band. The focus of monitoring is the specific frequency points related to the PWM signal spectrum characteristics, such as the PWM carrier frequency. and its main harmonic components in the baseband. Assume that K key frequency points are monitored { , ,..., }. DSP for each frequency point Configure a Goertzel filter, each processing sampling points (e.g. =512), the energy at that frequency point is calculated .

[0097] Step S602, trigger condition determination. The calculated total residual energy With a dynamic correction threshold The threshold is compared. It is not a fixed value, but is related to the overall power of the current audio signal. Related, ,in is a basic noise threshold, and γ is a proportional coefficient. This dynamic threshold setting can avoid being falsely triggered by small fluctuations in the audio leakage spectrum when playing loud audio. Only when this condition persists for more than a preset time window (e.g., 100 milliseconds) is it determined that the model has significantly mismatched and the model fine-tuning procedure is initiated.

[0098] Step S603, execute the slow adaptive algorithm to fine-tune the model. Once triggered, a Normalized Least Mean Square (NLMS) algorithm is activated. This algorithm uses the real-time audio PWM signal stream As the reference input signal, the monitored residual signal r(n) (i.e. ) as the error signal, the transfer function model coefficients currently used Fine-tune. The coefficient update formula is: in, is the new coefficient vector after fine-tuning; is the current coefficient vector; is composed of the current and past L-1 The vector of samples; δ is a small regularization constant (e.g. 1e-9) to prevent the denominator from being zero.

[0099] The most critical parameter is the step factor . This value is set to a very small, fixed constant, such as = 5e-7. The extremely small step size ensures an extremely slow update process, with an adaptation time constant on the order of seconds or even minutes. This slowness is crucial because it only responds to persistent, static model mismatches caused by system parameter drift, without inadvertently adapting to rapidly changing ambient noise or the audio signal itself, thus ensuring the stability and performance of the main ANC loop.

[0100] Step S604: Update the solidified model. The new coefficient vector after fine-tuning by the NLMS algorithm is The model is updated by writing data back to non-volatile memory at a very low frequency (e.g., once a minute, or when the cumulative coefficient changes exceed a certain threshold). This enables the system to self-learn and adapt, maintaining optimal interference cancellation performance over the long term.

[0101] As another further solution of the present invention, in order to achieve a global optimal balance between noise reduction performance, interference suppression level and system power consumption, the method also introduces a joint optimization strategy.

[0102] Step S7: The digital signal processing unit executes a joint optimization strategy. The strategy establishes a joint cost function in is the final system error signal after ANC filtering (usually collected by a dedicated error microphone), r(n) is the residual coupled noise energy in the residual correction loop, is the power consumption evaluation value of the Class D audio power amplifier. α and β are preset weight coefficients used to balance the noise reduction depth, interference suppression level and system power consumption. The digital signal processing unit dynamically adjusts the switching frequency of the Class D audio power amplifier according to the real-time changing audio content. and the core parameters of the adaptive ANC filter to minimize the joint cost function J while meeting specific audio playback requirements.

[0103] Specifically, the implementation mechanism of step S7 is as follows.

[0104] Step S701: Define and calculate the joint cost function. The DSP calculates the three components of the cost function in real time: one, :The digital signal output by the error microphone ADC Obtained by calculating the short-term mean square value. This represents the residual noise level that the user can ultimately perceive and is a core indicator of ANC performance.

[0105] two, : Directly reuse the residual coupling noise energy calculated in step S601 This item represents the completeness of interference cancellation.

[0106] three, : Power consumption estimate. This value is not measured directly but estimated using a power consumption model. The model expresses the power consumption as a function of the switching frequency based on the amplifier's design parameters. and the output audio signal RMS value Function: .in Represents the switching loss coefficient related to the switching frequency, and R_loss represents the conduction loss coefficient related to the output current. DSP real-time calculation , and according to the current setting estimated .

[0107] Step S702: Dynamically adjust the weight coefficients α and β. These two weight coefficients are not fixed, but can be adaptively adjusted according to the user scenario or audio content. For example, in a "high-fidelity music" mode, the system will set β to a lower value, giving priority to sound quality and noise reduction performance, allowing higher power consumption. In a "long battery life" mode, β will be set to a higher value, with minimizing power consumption as the main goal. When making a voice call, the weight of α will be increased to minimize any endogenous interference that may affect the clarity of the call.

[0108] Step S703: Execute optimization search. The joint optimization controller within the DSP periodically (e.g., once per second) attempts to make small perturbations to the controllable parameters and observes the changing trend of the joint cost function J to implement a gradient descent or hill climbing search strategy. The controllable parameters mainly include: 1. Class D amplifier switching frequency :The system is designed to support dynamic adjustment of the switching frequency within a preset range (e.g. 350kHz to 600kHz). will increase , but it may push the switching noise spectrum to higher frequencies, reducing its interference with the ANC band, thus possibly reducing and .

[0109] 2. ANC filter step size : This is the update step size of the main ANC adaptive filter. Can speed up the tracking of environmental noise and may reduce , but too high may cause system instability or amplify residual noise.

[0110] A specific optimization scenario is: when the system detects that the energy of the audio signal being played is very low (for example, when the music is paused), the optimization controller will try to reduce To reduce Although the reduction This may cause the coupled noise characteristics to change, but due to the slow adaptive loop in S6 To make compensation, At the same time, since there is no audio playback, The requirements of are relatively loose. Finally, the joint cost function J reaches a new minimum value at a lower power consumption point, realizing intelligent power management.

[0111] Reference Figure 2 To implement the above method, the present invention also provides an integrated system for active noise reduction and audio power amplification. The system includes a series of highly coordinated functional modules, as follows: The calibration signal generation and injection module integrates a pseudo-random binary sequence (PRBS) generator implemented using a linear feedback shift register (LFSR) with a 16th-order primitive polynomial. In system calibration mode, the PRBS generator is triggered by the system controller to generate a bipolar digital calibration signal sequence with a code length of 65535 and a clock frequency of 48kHz. This sequence is then provided as a digital input to the front-end data path of the pulse-width modulation (PWM) logic unit of the Class-D audio power amplifier.

[0112] The noise coupling path identification module, at its core, is a high-performance digital signal processing (DSP) unit. This module includes a 24-bit Σ-Δ analog-to-digital converter (ADC) connected to the output of the active noise cancellation (ANC) feedforward microphone and is connected to the DSP core. The DSP core is configured to receive, in calibration mode, a copy of the calibration signal sequence from the PRBS generator and a reference coupled noise sequence digitized by the ADC in parallel. The DSP runs a recursive least squares (RLS) system identification algorithm that takes the PRBS sequence as input and the coupled noise sequence as output. Through iterative calculations, it ultimately generates a finite impulse response (FIR) filter coefficient vector consisting of 1024 32-bit floating-point numbers that accurately characterizes the noise coupling characteristics of the amplifier-to-microphone link and stores it in non-volatile memory.

[0113] The predicted interference generation module runs within the DSP core during the system's normal operating mode. It is configured to intercept, in real time, the audio PWM digital signal stream, which has undergone sound effects processing, from the front end of the Class D audio power amplifier's PWM logic unit. This module feeds this signal stream into a FIR filter. The coefficients of this FIR filter are dynamically loaded using the coefficient vector generated by the noise coupling path identification module during calibration. By performing a 1024-order multiplication-accumulation operation per sampling period, the module generates, in real time, a predicted coupling interference digital signal sequence that is highly consistent with the actual coupling noise to be generated in both the time and frequency domains.

[0114] The feedforward digital cancellation module, also implemented in the DSP, is logically located between the feedforward microphone ADC output and the input of the ANC core filtering algorithm. The core of this module is a digital subtractor. One input of the subtractor receives a mixed digital signal from the ADC containing both ambient noise and actual coupled noise, while the other input receives a sequence of predicted coupled interference digital signals from the predicted interference generation module. By performing a sample-by-sample digital subtraction of the two inputs, the module outputs a purified ambient noise signal, essentially free of amplifier-derived interference, to the ANC core filtering algorithm, providing it with a clean working input.

[0115] The residual monitoring and model correction module, connected in parallel to the output of the feedforward digital cancellation module, runs as a background process in the DSP. This module comprises a set of parallel Goertzel filter banks and a slow adaptive algorithm unit. The Goertzel filter bank is configured to continuously monitor the residual energy in the purified ambient noise signal at a specific frequency relative to the amplifier switching frequency. When this energy exceeds a dynamic threshold, the slow adaptive algorithm unit is activated. This unit uses the Normalized Least Mean Square (NLMS) algorithm with a very small step size factor and uses the real-time audio PWM signal as a reference to fine-tune the coefficients of the currently used digital transfer function model. The updated coefficients are then written back to non-volatile memory to accommodate slow drift in system parameters.

[0116] The joint optimization control module, a top-level policy control unit implemented in the DSP, performs a gradient descent-like optimization algorithm by periodically perturbing and observing key parameters of the Class D amplifier switching frequency and ANC filter, based on a joint cost function that incorporates the final noise reduction residual, interference cancellation residual, and amplifier power consumption estimate. This module dynamically adjusts the optimization objectives based on different user scenarios and audio content to achieve a synergistically optimal balance between noise reduction performance, interference suppression, and power consumption while ensuring a consistent user experience.

[0117] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A method for integrating active noise reduction and audio power amplification, characterized in that: The following steps are involved: S1. In a system calibration mode, an internal calibration signal generator generates a pseudo-random binary sequence calibration signal, and injects the pseudo-random binary sequence calibration signal into a digital pulse width modulation front end of a Class D audio power amplifier to drive the power stage of the amplifier to generate a reference coupled noise; S2. Synchronously collect the reference coupled noise coupled by the power stage via a shared power supply, a ground plane, and spatial radiation to an active noise reduction feedforward microphone signal link, digitize the noise, generate a reference coupled noise digital sequence, and simultaneously obtain a digital copy of the pseudo-random binary sequence calibration signal; S3. Execute a system identification algorithm, using the digital replica of the pseudo-random binary sequence calibration signal as input and the reference coupled noise digital sequence as output, to calculate and establish a digital transfer function model representing the noise coupling path from the digital pulse width modulation front end to the digitized output of the feedforward microphone; S4. In a normal operating mode of the system, the audio pulse width modulation digital signal stream obtained in real time is input into the digital transfer function model for convolution operation, and a predicted coupling interference digital signal sequence is generated in real time.

2. The method for integrating active noise reduction and audio power amplification according to claim 1, characterized in that: The pseudo-random binary sequence calibration signal has a code length that is four times greater than the impulse response length of the noise coupling path. The reference coupled noise digital sequence is a 24-bit digital sequence generated by a Σ-Δ analog-to-digital converter. The digital transfer function model is a finite impulse response filter model. The predicted coupled interference digital signal sequence replicates the actual coupled noise that will be generated by the audio pulse width modulated digital signal stream in terms of time domain waveform and spectral components.

3. The method for integrating active noise reduction and audio power amplification according to claim 1, wherein: The steps for generating the reference coupling noise are specifically as follows: S111, after the system enters the calibration mode, disconnecting the external audio source from the digital pulse width modulation front end of the Class D audio power amplifier, and starting a calibration signal generator based on a linear feedback shift register; S112, the calibration signal generator generates a pseudo-random binary sequence calibration signal having a preset generating polynomial and code length, and maps the logic level of the sequence into a bipolar digital amplitude; S113. Injecting the bipolar pseudo-random binary sequence calibration signal into the digital pulse width modulation front end to modulate the duty cycle of the output PWM waveform, driving the power stage to perform switching operations at a preset static operating point, thereby generating the baseline coupling noise in the shared power supply network and the physical space.

4. The method for integrating active noise reduction and audio power amplification according to claim 3, wherein: The steps for obtaining the reference coupling noise digital sequence are specifically as follows: S211, synchronously with the generation of the pseudo-random binary sequence calibration signal, picking up the analog signal of the reference coupled noise through the feedforward microphone and its front-end analog preamplifier circuit; S212, sending the analog signal through an anti-aliasing filter to a high-precision Σ-Δ analog-to-digital converter for digitization to generate the reference coupled noise digital sequence; S213 , sending the reference coupled noise digital sequence and the digital copy of the pseudo-random binary sequence calibration signal to a designated storage area of ​​a digital signal processing unit in parallel.

5. The method for integrating active noise reduction and audio power amplification according to claim 4, characterized in that: The steps for establishing the digital transfer function model are specifically as follows: S311, selecting a recursive least squares algorithm as the system identification algorithm and initializing it, including setting the model order, forgetting factor, initial coefficient vector and initial inverse covariance matrix; S312, performing iterative calculations of the recursive least squares algorithm in a sampling point-by-sample manner, updating the coefficient vector and inverse covariance matrix of the model by calculating the prior error and the gain vector, so as to minimize the mean square error between the reference coupled noise digital sequence and the model prediction output; S313. When the change in the second norm of the model coefficient vector is less than a preset convergence threshold, the iteration is terminated, and the finally obtained coefficient vector is used as the implementation of the digital transfer function model and solidified and stored in a non-volatile memory.

6. The method for integrating active noise reduction and audio power amplification according to claim 5, characterized in that: The steps of generating the predicted coupled interference digital signal sequence are specifically as follows: S411, constructing a finite impulse response filter in the digital signal processing unit based on the coefficient vector solidified in the non-volatile memory; S412, in a normal system working mode, intercepting in real time the audio pulse width modulation digital signal stream to be input to the digital pulse width modulation front end; S413: Input the real-time audio pulse width modulation digital signal stream into the finite impulse response filter sample by sample point, perform a convolution operation, and output the predicted coupled interference digital signal sequence.

7. The method for integrating active noise reduction and audio power amplification according to claim 1, wherein: The method further comprises step S5: S5. Perform a sampling point-by-sample subtraction operation in the digital domain on a mixed digital signal containing ambient noise and actual coupled noise, collected by the feedforward microphone in a normal working mode, and the predicted coupled interference digital signal sequence to generate a purified ambient noise digital sequence from which the intrinsic interference of the amplifier has been eliminated.

8. The method for integrating active noise reduction and audio power amplification according to claim 7, wherein: The method further comprises step S6: S6. Maintain a parallel residual correction loop to perform spectral analysis on the purified ambient noise digital sequence to detect whether there are residual spectral components related to the switching frequency of the Class D audio power amplifier and its harmonics. If the energy of the residual spectral components exceeds a dynamic correction threshold, initiate a slow adaptive algorithm to fine-tune the coefficients of the digital transfer function model.

9. The method for integrating active noise reduction and audio power amplification according to claim 8, wherein: The model coefficient fine-tuning in step S6 is specifically as follows: S611, performing short-time Fourier transform on the purified environmental noise digital sequence, analyzing and extracting residual spectral components corresponding to the amplifier switching frequency and its harmonic frequencies as residual signals; S612: Compare the energy of the residual signal with the dynamic correction threshold, and activate the slow adaptive algorithm if the energy exceeds the limit; S613. Execute a normalized least mean square algorithm, using the real-time audio pulse width modulation digital signal stream as a reference signal and the residual signal as an error signal, to iteratively update the coefficients of the digital transfer function model to compensate for the time-varying nature of the coupling path.

10. An integrated system of active noise reduction and audio power amplification, characterized in that: The system is used to implement the integrated method of active noise reduction and audio power amplification according to any one of claims 1 to 9, and the system includes: a calibration signal generation and injection module, for generating a pseudo-random binary sequence calibration signal in a system calibration mode, and injecting the signal into a digital pulse width modulation front end of a class D audio power amplifier to stimulate the generation of a reference coupled noise; a noise coupling path identification module, configured to synchronously acquire and digitize the reference coupling noise, receive a digital copy of the calibration signal, execute a recursive least squares system identification algorithm, and calculate and generate a digital transfer function model characterizing the noise coupling path; A predicted interference generation module is used to perform a convolution operation on the real-time audio pulse width modulated digital signal stream through the digital transfer function model in a normal system working mode to generate a predicted coupled interference digital signal sequence; A feedforward digital cancellation module is used to subtract the predicted coupled interference digital signal sequence from the mixed digital signal collected by a feedforward microphone, sampling point by sampling point, and output a purified environmental noise digital sequence; The residual monitoring and model correction module is used to analyze the residual coupling noise energy in the purified environmental noise digital sequence. When the energy exceeds the limit, a normalized least mean square algorithm is activated to fine-tune the coefficients of the digital transfer function model using the real-time audio pulse width modulation signal as a reference.

Images