A composite active noise reduction method based on inverse filtering frequency equalization
By using an inverse filtering frequency equalization method and an improved RNLMS algorithm, the speaker system is decomposed into a minimum phase and an all-pass system. Combined with a feedforward and feedback structure, the problem of low-frequency noise suppression in active noise-canceling headphones under complex noise environments is solved, achieving smaller steady-state error and better noise reduction effect.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-11-22
- Publication Date
- 2026-03-13
AI Technical Summary
Existing active noise-canceling headphones do not perform well in complex noise environments, especially in suppressing low-frequency noise. Furthermore, the spectral distortion introduced by the speaker system affects the real-time recognition of the secondary channel model.
A composite active noise reduction method based on inverse filtering frequency equalization is adopted. By improving the RNLMS algorithm and decomposing the loudspeaker system into a cascaded minimum phase system and an all-pass system, combined with a feedforward and feedback control structure, the contamination of the reference signal by the secondary sound source is compensated, and a pure time delay system model is established.
It significantly reduces steady-state error, improves the suppression of low-frequency noise, and has a better noise reduction effect than the traditional FXLMS algorithm, especially in the low-frequency noise band.
Smart Images
Figure CN113920974B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of active noise cancellation technology for headphones, and in particular to a composite active noise cancellation technology based on inverse filtering frequency equalization. Background Technology
[0002] With the rapid improvement of modern society's infrastructure, people's lives have become increasingly convenient. However, at the same time, noise generated by various machines has increased in people's living and working environments. Prolonged exposure to harsh noise environments can cause serious physical and mental harm. Therefore, reducing noise levels has become increasingly important, and it is closely related to improving our quality of life. Active noise-canceling headphones can effectively reduce the harmful effects of noise in many working and living environments, making research on active noise-canceling headphones an important topic.
[0003] Currently, noise reduction technologies can be broadly categorized into two types: passive noise cancellation (PNC) and active noise cancellation (ANC). Passive noise cancellation primarily absorbs noise through sound-absorbing materials, and the properties of the materials determine the noise reduction effect. Passive noise cancellation is effective at reducing high-frequency noise, but its applicability is limited due to factors such as material size. Active noise cancellation, on the other hand, generates inverse noise in real time, using the interference of sound waves to reduce noise. Active noise cancellation is effective at suppressing low-frequency noise and has wider applicability. Active noise cancellation technology is widely used in automobiles, home appliances, high-speed rail, and portable wearable devices. A typical example of active noise cancellation in portable wearable devices is active noise-canceling headphones. Active noise-canceling headphones can significantly improve noise levels during work, study, and rest, thereby improving people's quality of life. However, with the rapid development of modern micro digital signal processor chips and integrated circuit technology, active noise cancellation technology based on DSP technology has also faced new challenges; such as active noise cancellation in complex noise environments, and how to efficiently combine the rapidly developing digital signal processor technology with related algorithms. Nevertheless, no matter how active noise cancellation technology develops, effectively improving the amount of active noise cancellation remains the core objective of active noise cancellation headphone technology.
[0004] To improve the inherent spectral distortion problem of loudspeaker systems, this invention proposes a frequency equalization method based on inverse filtering; and applies this method to a feedforward plus feedback composite active noise cancellation headphone system, resulting in a new active noise cancellation control scheme. Summary of the Invention
[0005] Active noise-canceling headphones work by generating anti-phase noise through speakers to cancel out initial noise. However, the addition of a speaker system inevitably introduces spectral distortion, which complicates the secondary channel model and hinders online identification of the primary channel in real-time active noise cancellation. In reality, as sound travels a short distance from one point to another in space, the frequency change is minimal, with phase delay being the primary factor. This invention proposes an erosion-based active noise cancellation method based on inverse filtering frequency equalization. By applying frequency equalization to the active noise cancellation model, the secondary channel model becomes a pure time-delay system. This pure time-delay system only has phase delay, more closely approximating the propagation characteristics of sound between two points.
[0006] The technical solution of the present invention is as follows:
[0007] This invention proposes a composite active noise cancellation method based on inverse filtering frequency equalization. Firstly, based on research into the control structure of active noise-canceling headphones, the FXLMS algorithm, and the classic LMS algorithm, an improved RNLMS algorithm is proposed. This invention analyzes the transmission path of the feedforward plus feedback composite active noise-canceling headphone structure, initially considering the frequency and phase distortion problems caused by the introduction of a speaker. Then, it proposes to decompose the non-minimum-phase system of the speaker and secondary channel model into a cascaded form of a minimum-phase system and an all-pass system, solving the frequency resolution problem of traditional equalization methods. The classic FXLMS algorithm is divided into two LMS system identification models (the difference being that the input reference signal has an additional minimum-phase system component), and the optimal weight solution is obtained through transfer function analysis. Subsequent computer simulations verified the above analysis, showing that the model using inverse filtering equalization has a smaller steady-state error, consistent with the theoretical analysis. Based on the computer simulation-verified inverse filtering equalization model, an active noise reduction model for the reflection channel (the influence of the secondary sound source on the reference microphone) was established. Finally, in a semi-anechoic chamber environment, Gaussian white noise (low-frequency white noise with a bandwidth of 20-1500Hz) was used to model the secondary channel and reflection channel offline. Real-time noise reduction was performed under three conditions: ANC off, ANC on using the classic FXLMS algorithm, and ANC on using the inverse filtering equalization FXLMS algorithm. Experimental results show that the steady-state error of the control system using the inverse filtering equalization scheme is smaller than that of the classic FXLMS algorithm, demonstrating good noise reduction performance.
[0008] The main steps of the method of this invention include:
[0009] Step 1: Compare and analyze the control methods of active noise-canceling headphones, and determine the composite control of feedforward and feedback as the research object;
[0010] Step 2: Based on the research object, and combining the active noise reduction models FXLMS, LMS and NLMS, a signal processing method with a large adaptive dynamic range by controlling the step size through error size is proposed - the variable step size RNLMS algorithm.
[0011] Step 3: Based on the FXLMS algorithm, the pre-modeled secondary channel model is decomposed into a minimum phase system, which is then decomposed into a cascaded minimum phase system and an all-pass system. An inverse system of the minimum phase system is cascaded before the system model to make the system an all-pass system. Similarly, the secondary channel transfer function is processed in the same way to make it a pure delay element. A composite active noise reduction algorithm based on inverse filtering frequency equalization is proposed.
[0012] Step 4: Compare the steady-state errors of the pure delay stage before and after moving it forward in the cascaded secondary channel models of the above algorithm model, and verify the steady-state errors of the two through computer simulation.
[0013] Step 5: Based on the verification results, the influence of the reflection channel is also added to the cascaded model. Real-time noise reduction is performed in a semi-anechoic chamber environment under three conditions: ANC off, ANC on using the classic FXLMS algorithm, and ANC on using the inverse filter equalization FXLMS algorithm. Test results show that the proposed composite active noise reduction model, which includes frequency equalization and secondary sound source interference compensation, is a feedforward plus feedback system. The control scheme is consistent with the computer simulation results and has good noise reduction performance.
[0014] Step 1, determining the research object of active noise cancellation, specifically includes: Active noise-canceling headphone control methods can be classified into feedforward, feedback, and a composite type combining feedforward and feedback, based on their control structure and principle. To fully utilize the advantages of feedforward and feedback control structures and pursue superior noise cancellation performance, the active noise cancellation control structure studied in this invention adopts a composite control scheme combining feedforward and feedback.
[0015] Furthermore, step 2, the proposal and verification of the RXLMS algorithm, specifically includes: to ensure that the NLMS algorithm maintains both a fast convergence speed and a small steady-state error, a novel variable step-size RXLMS algorithm is proposed by introducing a small integer μ(n). The feasibility of the algorithm is verified through theoretical analysis and computer simulation.
[0016] Furthermore, step 3, the non-minimum phase system decomposition, specifically includes: since the loudspeaker system is a non-minimum phase system, the secondary channel model must also be a non-minimum phase system. Any non-minimum phase system H(z) can be decomposed into a minimum phase system H... min (z) and an all-pass system H ap(z) Cascaded form. If a minimum-phase system's inverse is cascaded before this system, then the system becomes an all-pass system. Similarly, the secondary channel transfer function is processed in the same way, turning it into a pure delay element. Finally, the pure delay element generated after cascading the secondary channel models is shifted forward between the two system structures;
[0017] Furthermore, step 4, the computer steady-state error verification for frequency equalization, specifically includes: moving the pure delay stage forward and adding a minimum-phase system stage S to the input reference signal. min (z) The addition of this stage complicates the system, which inevitably slows down the system's identification speed and increases the system's steady-state error. Then, through computer simulation using Gaussian white noise with a mean of 0 and a mean square of 0.1 as the input signal, it is verified that the inverse filtering frequency system without the forward-shifting pure delay stage is significantly smaller than that of the forward-shifting system.
[0018] Further, step 5, system model experimental verification, specifically includes: after theoretical and simulation effect verification tests, running the active noise cancellation algorithm in a semi-anechoic environment using Texas Instruments' TMS320C6748LCDK development board as the platform. Offline modeling was performed on the secondary and reflection channels using Gaussian white noise, and real-time noise reduction results were measured under three conditions: ANC off, ANC on using the classic FXLMS algorithm, and ANC on using the inverse filter equalization FXLMS algorithm. Comparison of steady-state errors shows that the steady-state error of the control system using the inverse filter equalization scheme is smaller than that of the classic FXLMS algorithm (approximately 9dB). The power spectral density indicates that the ANC system based on the FXLMS algorithm has a good noise reduction effect on frequency components with high power proportions, while the frequency equalization scheme enhances the noise reduction effect on frequency components with low power proportions.
[0019] The advantages of the composite active noise reduction method based on inverse filtering frequency equalization described in this invention are as follows:
[0020] 1. This invention proposes a frequency equalization method based on inverse filtering; and applies this method to a feedforward plus feedback composite active noise cancellation headphone system to obtain a new active noise cancellation control scheme.
[0021] 2. In order to improve the noise reduction effect, the active noise reduction control scheme of the present invention also compensates for the pollution of the reference signal caused by the secondary sound source speaker.
[0022] 3. Through computer simulation and DSP-based real-time noise reduction experiments, this invention demonstrates that the novel active noise reduction control system proposed in this invention has a good noise reduction effect. Attached Figure Description
[0023] Figure 1The implementation process of the method of this invention;
[0024] Figure 2 : Feedforward and feedback composite noise reduction control structure;
[0025] Figure 3 Equivalent disassembly diagram of the classic FXLMS system;
[0026] Figure 4 Equivalent decomposition diagram of the inverse filter FXLMS system;
[0027] Figure 5 Simulation results of the classic FXLMS algorithm;
[0028] Figure 6 Simulation results of the inverse filtering FXLMS algorithm;
[0029] Figure 7 Complete ANC control structure diagram;
[0030] Figure 8 Active noise cancellation experiment;
[0031] Figure 9 Experimental results of time-domain waveforms from the classic FXLMS algorithm;
[0032] Figure 10 Experimental results of time-domain waveforms of the inverse filtering equalization FXLMS algorithm;
[0033] Figure 11 Steady-state noise error between the FXLMS algorithm and the inverse filtering algorithm;
[0034] Figure 12 Power spectral density under three test conditions; Detailed Implementation
[0035] The implementation process of the method of this invention is as follows:
[0036] Step 1: Determine the research object of active noise cancellation
[0037] By analyzing the advantages and disadvantages of three conventional active noise cancellation control models for headphones, the control method of composite active noise cancellation headphones was determined as the main research object.
[0038] Step 2: Proposal and Validation of the RXLMS Algorithm
[0039] Based on NLMS, a relative error is introduced in the weight iterative update, which is the ratio of the absolute value of the error to the average amplitude of the reference signal. An RNLMS algorithm is proposed that uses a large step size to accelerate algorithm convergence when the error is large, and a smaller step size to minimize the steady-state error when the steady-state convergence error is small. The algorithm is then analyzed theoretically and verified by computer simulation.
[0040] Step 3: Inverse Transformation of Non-Minimum Phase Systems
[0041] Based on the above research object and the FXLMS algorithm model, corresponding inverse minimum phase systems are added before the secondary and primary sound channels; according to the transformation of the secondary and primary sound channels, they are equivalent to classical and inverse filtered FXLMS systems, and the steady-state errors of the two are theoretically analyzed and compared.
[0042] Step 4: Verification of steady-state error of frequency equalization
[0043] Using Gaussian white noise with a mean of 0 and a mean square of 0.1 as the input signal of the system, the classical LMS algorithm is used for system identification. Simulation results show that the steady-state error of the classical FXLMS system is larger than that of the inverse filtered FXLMS system.
[0044] Step 5: System model experimental verification
[0045] Based on the aforementioned inverse filtering equalization model, and considering the influence of secondary sound sources on the reference microphone, a complete active noise cancellation control model is presented. Real-time active noise cancellation experiments are then conducted in a semi-anechoic chamber based on this control model.
[0046] The effectiveness of the algorithm proposed in this invention will be tested under the established experimental conditions, with reference to the accompanying drawings, to further illustrate this invention.
[0047] The specific implementation steps are as follows:
[0048] Step 1: By comparing the applicability, stability, design difficulty, and noise reduction effect of feedforward, feedback, and combined feedforward and feedback control methods, the active noise reduction control structure to be studied adopts a combined feedforward and feedback control scheme, as follows: Figure 1 .
[0049] Step 2: By analyzing the performance and convergence speed of the FXLMS, LMS, and NLMS algorithms, a variable step-size RNLMS algorithm is proposed. Simulation experiments are conducted using the classic LMS model, and the performance of the algorithm is measured by the ratio of error to input signal power. Theoretical analysis and computer simulation results show that the algorithm achieves the expected results, with a faster convergence speed than NLMS, comparable steady-state error, and is unaffected by changes in signal power.
[0050] Step 3: According to the principle of inverse filtering, any non-minimum phase system H(z) can be decomposed into a minimum phase system Hz. min (z) and an all-pass system H ap (z) Cascaded form H(z) = H min (z)H ap(z). Cascading the inverse of a minimum-phase system before this system will transform it into an all-pass system. Based on this, the secondary and primary channel functions are transformed into pure delay elements. Finally, the pure delay elements generated after cascading the secondary channel models are shifted forward. The two system structures are equivalent to the classic FXLNS system, as shown below. Figure 2 And simplified inverse filter FXLMS system, such as Figure 3 ;
[0051] Step 4: Based on the two system control methods described in Step 3, simulation verification is performed using a computer with Gaussian white noise of mean 0 and mean square 0.1 as the input signal. The steady-state error of the classic FXLMS equivalent system is as follows: Figure 4 And simplified inverse filter FXLMS system, such as Figure 5 The results show that the inverse filtering system without forward shifting of the pure delay stage has a significantly smaller frequency than the forward shifting system.
[0052] Step 5: Based on the inverse filtering frequency equalization control method in Step 4, construct a complete active noise reduction control scheme considering the influence of the reflection channel on the cascaded model, as follows: Figure 6 Subsequently, the experimental environment in a semi-anechoic chamber was as follows: Figure 7 The real-time noise reduction results were measured under three conditions: ANC off, ANC on using the classic FXLMS algorithm, and ANC on using the inverse filter equalization FXLMS algorithm. Figure 8 , 9 By comparing the steady-state errors, it can be seen that the inverse filtering equalization scheme is as follows: Figure 10 The steady-state error obtained by the control system is smaller than that of the classical FXLMS algorithm (approximately 9 dB), as shown by the power spectral density... Figure 11 It can be seen that the ANC system based on the FXLMS algorithm has a good noise reduction effect on frequency components with high power proportion, and the frequency equalization scheme enhances the noise reduction effect on frequency components with low power proportion.
Claims
1. A composite active noise reduction method based on inverse filtering frequency equalization, characterized in that The method comprises the following steps: Step 1: comparative analysis is conducted on active noise reduction earphone control modes, and a compound control of feedforward plus feedback is determined as a research object; Step 2: according to the research object, combining with FXLMS, LMS and NLMS active noise reduction models, a signal processing method of variable step size adaptive dynamic range is proposed, that is, a variable step size RNLMS algorithm is proposed by error size control; Step 3: on the basis of the FXLMS algorithm, a pre-modeled secondary sound channel model is decomposed according to a minimum phase system, and is decomposed into a form of a minimum phase system and a through system in cascade; an inverse system of the minimum phase system is cascaded in front of the secondary sound channel model, so that the secondary sound channel model becomes a through system; similarly, the secondary sound channel transfer function is processed in the same way, so that it becomes a pure delay link, and a compound active noise reduction algorithm based on inverse filtering frequency equalization is proposed; Step 4: the steady-state errors of the pure delay link generated after the secondary sound channel model in cascade in the above-mentioned FXLMS, LMS and NLMS algorithms are compared, and the steady-state errors of the two are verified through computer simulation; Step 5: based on the verification result, the influence of the reflection channel is also added to the model after cascade, so as to compensate for the pollution caused by the reference signal of the secondary sound source loudspeaker, and real-time noise reduction is carried out under three conditions of ANC closed, ANC opened classical FXLMS algorithm and ANC opened inverse filtering equalization FXLMS algorithm in a semi-anechoic chamber environment, and test results show that the feedforward plus feedback compound active noise reduction model containing frequency equalization and secondary sound source interference compensation has a control scheme consistent with the computer simulation effect and has good noise reduction effect.
2. The composite active noise reduction method based on inverse filtering frequency equalization according to claim 1, characterized in that, The step 1 is specifically: by comparing the applicability, stability, design difficulty and noise reduction effect of the feedforward type, feedback type and feedforward plus feedback compound control mode, it is determined that the active noise reduction control structure adopts the compound control scheme of feedforward plus feedback.
3. The composite active noise reduction method based on inverse filtering frequency equalization according to claim 1, characterized in that, The step 2 is specifically: by analyzing the performance and convergence speed of the FXLMS algorithm, LMS algorithm and NLMS algorithm, the RNLMS algorithm with variable step size is proposed, the classical LMS model is used for simulation experiment, the error and input signal power ratio are taken to measure the performance of the algorithm, and theoretical analysis and computer simulation results show that the algorithm achieves the expected result.
4. The composite active noise reduction method based on inverse filtering frequency equalization according to claim 1, characterized in that, The step 3 is specifically: according to the inverse filtering principle, an inverse system of a minimum phase system is cascaded in front of the secondary sound channel model, which becomes a through system, according to which the secondary sound channel and the primary sound channel function become a pure delay link, finally, the pure delay link generated after the secondary sound channel model cascades the inverse system of the minimum system is moved forward, so that the secondary sound channel model system is equivalent to a classical FXLMS system, and a simplified inverse filtering FXLMS system is formed.
5. The composite active noise reduction method based on inverse filtering frequency equalization according to claim 1, characterized in that, The step 4 is specifically: a Gaussian white noise with a mean value of 0 and a mean square of 0.1 is used as the input signal of the system, the classical LMS algorithm is used for system identification, and the simulation results show that the steady-state error of the classical FXLMS system is larger than that of the inverse filtering FXLMS system.
6. The method of claim 1, wherein the method is a composite active noise reduction method based on inverse filtering frequency equalization. The step 5 is specifically: after theoretical and simulation effect verification test, using Texas Instrument TMS320C6748 LCDK development board as a platform to run the active noise reduction algorithm in semi-muffled environment, using Gaussian white noise to model the secondary sound channel and reflection channel offline, and measuring the real-time noise reduction results under the conditions of ANC off, ANC on classic FXLMS algorithm, and ANC on inverse filter equalized FXLMS algorithm respectively.