A step-by-step virtual sensing noise reduction method

Abstract

The application provides a step-by-step virtual sensing noise reduction method. The method comprises the following steps: step one, temporarily placing a virtual microphone at a target noise reduction position to collect a virtual error signal and model a virtual secondary path; step two, training a noise control filter; step three, placing the virtual microphone in steps one and two at a physical position as a physical microphone to collect a physical error signal and model a physical secondary path; step four, not placing any microphone at the target noise reduction position and placing a physical microphone at the physical position to train an auxiliary filter; and step five, entering a control stage to realize active noise reduction at the target noise reduction position. The step-by-step virtual sensing noise reduction method provided by the application has small hardware resource consumption and good noise reduction effect when the intensity range of primary noise is large, and has high practicability.

Description

Technical Field

[0001] This invention relates to a stepwise virtual sensing noise reduction method. Background Technology

[0002] With the rapid development of science and technology, human society has entered the industrial age. While enjoying the convenience brought by technology, people have gradually discovered some drawbacks, one of which is noise pollution. As noise increasingly affects people's daily lives, the demand for noise reduction is also rising. Traditional noise control is passive noise control (PNC), which achieves noise reduction by having noise waves interact with acoustic materials or structures. However, PNC is difficult to use for low-frequency noise signals and has limitations. The emergence and development of active noise control (ANC) has filled this gap. In recent years, ANC has gradually become one of the main research directions in the field of noise control.

[0003] Active noise cancellation (ANC) technology refers to the technique of collecting ambient noise and generating a signal that is opposite to the noise to cancel it out. When the control source radiates an anti-noise wave with the same amplitude but opposite phase as the noise source's sound wave, the ANC system can reduce noise at the desired location where the error microphone is placed.

[0004] Depending on the type of microphone used, ANC noise reduction systems are structurally divided into feedforward and feedback systems. Feedforward systems offer stable performance and are suitable for handling broadband noise; feedback systems can only reduce single-frequency noise and narrowband noise, and have limitations. Based on the number of loudspeakers used, ANC noise reduction systems can also be divided into single-channel ANC systems and multi-channel ANC systems.

[0005] In some applications, such as wireless Bluetooth headsets, it's impossible to place an error microphone at the target noise reduction location. Virtual sensing noise reduction technology has emerged to address this. Virtual sensing noise reduction technology mainly consists of two stages: a training stage and a control stage. In the training stage, a virtual microphone is temporarily placed at the target noise reduction location to obtain the transmission relationship between the physical and virtual positions. In the control stage, the virtual microphone at the target noise reduction location is removed, and the noise control filter is updated based on the signal collected by the physical microphone and the transmission relationship between the physical and virtual positions obtained in the training stage, eventually reaching a stable state, thus achieving active noise reduction at the target noise reduction location. Currently, two virtual sensing technologies exist: the remote microphone (RM) method and the auxiliary filter (AF) method. For the AF method, the training stage uses one more microphone than the control stage, placing greater demands on the number of interfaces and computational load of the hardware controller. Therefore, this invention aims to simplify the AF algorithm. Summary of the Invention

[0006] The purpose of this invention is to address the drawback of the AF method, which requires an additional microphone during the training phase compared to the control phase, resulting in high hardware resource consumption. Furthermore, to ensure that noise reduction at the target noise reduction location reaches a specified threshold even when the reference signal volume range is large, this invention proposes a step-by-step virtual sensing noise reduction method, comprising the following steps:

[0007] Step 1: Model the virtual secondary path S v A virtual microphone is temporarily placed at the target noise reduction location to collect virtual error signals. According to the LMS algorithm, the secondary sound source emitted by the secondary speaker and the virtual error signals received by the virtual microphone work together to update the virtual secondary path filter. Finally, a stable state is reached, resulting in a virtual secondary path filter.

[0008] Step 2: Train the noise control filter W. According to the FxLMS algorithm, the virtual error signal received by the virtual microphone is the superposition of the reference signal, the output of the noise control filter W, and the primary noise. The reference signal then passes through the virtual secondary path filter. The output of the signal and the virtual error signal received by the virtual microphone work together to update the noise control filter W, eventually reaching a stable state, thus obtaining the noise control filter W.

[0009] Step 3: Model the physical secondary pathway S m The virtual microphone from steps one and two is placed at a physical location as a physical microphone to collect physical error signals. According to the LMS algorithm, the secondary sound source emitted by the secondary speaker and the physical error signals received by the physical microphone work together to update the physical secondary path filter. Finally, a stable state is reached, resulting in a physical secondary path filter.

[0010] Step 4: Train the auxiliary filter H. No microphone is placed at the target noise reduction location, but a physical microphone is placed at the physical location. The noise control filter W trained in Step 2 is used as the fixed-coefficient filter W0. The physical error signal received by the physical microphone is used as the reference signal and passed through the fixed-coefficient filter W0 and the physical secondary path filter. The output of the auxiliary filter is the superposition of the primary noise and the residual signal generated by the auxiliary filter. The residual signal generated by the auxiliary filter is the superposition of the physical error signal received by the physical microphone and the output of the reference signal through the auxiliary filter H. The reference signal and the residual signal generated by the auxiliary filter work together to update the auxiliary filter H, eventually reaching a stable state, and thus obtaining the auxiliary filter H.

[0011] Step 5: Enter the control phase. No microphone is placed at the target noise reduction location, but a physical microphone is placed at the physical location. The physical error signal received by the physical microphone serves as the reference signal and passes through the noise control filter W and the physical secondary path filter. The output of the signal is superimposed with the primary noise. The auxiliary filter H obtained in step four is used as a fixed-coefficient filter H0. The residual signal generated by the auxiliary filter is the superposition of the physical error signal received by the physical microphone and the output of the reference signal through the fixed-coefficient filter H0. The reference signal passes through the physical secondary path filter. The output of the filter and the residual signal generated by the auxiliary filter work together to update the noise control filter W, eventually reaching a stable state and achieving active noise reduction at the target noise reduction location.

[0012] Furthermore, in step two, the process of updating the noise control filter W is as follows:

[0013] Through formula The weight coefficient vector is updated, where w(n) represents the weight coefficient vector before the noise control filter update, w(n+1) represents the updated weight coefficient vector, and x(n) represents the reference signal. Represents a virtual secondary path filter, e v (n) represents the virtual error signal, μ w1 This represents the convergence factor for updating the noise control filter.

[0014] Furthermore, in step four, the process of updating the auxiliary filter H is as follows:

[0015] The formula is: h(n+1)=h(n)-μ h e h(n)x(n) updates the weight coefficient vector, where h(n) represents the weight coefficient vector before the auxiliary filter update, h(n+1) represents the updated weight coefficient vector, and x(n) represents the reference signal, e h (n) represents the residual signal generated by the auxiliary filter, μ h This represents the convergence factor for the auxiliary filter update.

[0016] Furthermore, in step five, the process of updating the noise control filter W is as follows:

[0017] The formula w(n+1)=w(n)-μ is used. w2 e h (n)x s (n) Updates the weight coefficient vector, where w(n) represents the weight coefficient vector before the noise control filter update, w(n+1) represents the updated weight coefficient vector, and x s (n) represents the output of the reference signal through the physical secondary path filter, e h (n) represents the residual signal generated by the auxiliary filter, μ w2 This represents the convergence factor for updating the noise control filter.

[0018] Furthermore, to meet the noise reduction requirements when the primary noise intensity range is large, the convergence step size is set as follows:

[0019]

[0020]

[0021]

[0022] Where, μ w1 μ represents the convergence factor updated by the noise control filter in step two. h μ represents the convergence factor updated by the auxiliary filter in step four. w2 denoted by , represents the convergence factor updated by the noise control filter in step five, and x represents the reference signal.

[0023] In summary, due to the adoption of the above technical solutions, the beneficial effects of the present invention are: it solves the problem that the training phase of the AF method uses one more microphone than the control phase, resulting in high hardware resource consumption, and it also ensures that when the reference signal volume range is large, the noise reduction at the target noise reduction position can reach the specified threshold. Attached Figure Description

[0024] Figure 1 The following are flowcharts of the stepwise virtual sensing noise reduction method of the present invention: a) flowchart of step two; b) flowchart of step four; c) flowchart of step five.

[0025] Figure 2 This is a specific implementation of the secondary path diagram of the present invention when the reference signal volume is medium.

[0026] Figure 3 In a specific implementation, the error signal diagrams for steps two and four are shown in the present invention when the reference signal volume is medium.

[0027] Figure 4 In a specific embodiment, the error signal diagram of step five of the present invention is shown when the reference signal volume is medium.

[0028] Figure 5 In a specific implementation, the present invention provides the noise power spectrum of step five when the reference signal volume is low, medium, and high. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings.

[0030] To address the drawback of the AF method requiring an additional microphone during the training phase compared to the control phase, resulting in high hardware resource consumption, and to ensure that noise reduction at the target noise reduction location reaches a specified threshold when the reference signal volume range is large, this invention proposes a stepwise virtual sensing noise reduction method.

[0031] Appendix Figure 1 The flowchart of the stepwise virtual sensing noise reduction method of the present invention is shown. In this specific embodiment, step one involves temporarily placing a virtual microphone at the target noise reduction location to collect virtual error signals and modeling the virtual secondary path S. v , is represented as:

[0032] y v (n)=s v (n)*y(n)

[0033] Where y(n) represents the secondary sound source emitted by the secondary loudspeaker, y v (n) represents the virtual error signal received by the virtual microphone. The secondary sound source emitted by the secondary speaker and the virtual error signal received by the virtual microphone work together to update the virtual secondary path filter. Finally, a stable state is reached, resulting in a virtual secondary path filter.

[0034] In this specific implementation, step two (see appendix) Figure 1 (a) Training the noise control filter W, according to the FxLMS algorithm, the virtual error signal received by the virtual microphone is the superposition of the reference signal through the output of the noise control filter W and the primary noise, expressed as:

[0035] ev (n)=d v (n)+s v (n)*y(n)

[0036] Where, d v (n) represents primary noise, e v (n) represents the virtual error signal received by the virtual microphone. The reference signal passes through the virtual secondary path filter. The output of the signal and the virtual error signal received by the virtual microphone work together to update the noise control filter W. The update process of the noise control filter W is as follows:

[0037]

[0038] Finally, a stable state is reached, resulting in the noise control filter W. Here, w(n) represents the weight coefficient vector of the noise control filter before the update, w(n+1) represents the updated weight coefficient vector, and x(n) represents the reference signal. Represents a virtual secondary path filter, e v (n) represents the virtual error signal, μ w1 This represents the convergence factor for updating the noise control filter.

[0039] In this specific embodiment, step three involves placing the virtual microphone from steps one and two as a physical microphone at a physical location to collect physical error signals and model the physical secondary path S. m , is represented as:

[0040] y m (n)=s m (n)*y(n)

[0041] Where y(n) represents the secondary sound source emitted by the secondary loudspeaker, y m (n) represents the physical error signal received by the physical microphone. The secondary sound source emitted by the secondary speaker and the physical error signal received by the physical microphone work together to update the physical secondary path filter. Finally, a stable state is reached, resulting in a physical secondary path filter.

[0042] In this specific implementation, step four (see appendix) Figure 1 (b) No microphone is placed at the target noise reduction location, but a physical microphone is placed at the physical location. An auxiliary filter H is trained, and the noise control filter W obtained in step two is used as a fixed-coefficient filter W0. The physical error signal received by the physical microphone is used as a reference signal and passes through the fixed-coefficient filter W0 and the physical secondary path filter. The superposition of the output and the primary noise is expressed as:

[0043]

[0044] Where, d m (n) represents primary noise, e m (n) represents the physical error signal received by the physical microphone. The residual signal generated by the auxiliary filter is the superposition of the physical error signal received by the physical microphone and the output of the reference signal through the auxiliary filter H. The reference signal and the residual signal generated by the auxiliary filter work together to update the auxiliary filter H. The update process of the auxiliary filter H is as follows:

[0045] h(n+1)=h(n)-μ h e h (n)x(n)

[0046] Finally, a stable state is reached, resulting in the auxiliary filter H. Here, h(n) represents the weight coefficient vector before the auxiliary filter update, h(n+1) represents the updated weight coefficient vector, x(n) represents the reference signal, and e... h (n) represents the residual signal generated by the auxiliary filter, μ h This represents the convergence factor for the auxiliary filter update.

[0047] In this specific implementation, step five (see appendix) Figure 1 (c) No microphone is placed at the target noise reduction location, but a physical microphone is placed at the physical location. The system enters the control phase. The physical error signal received by the physical microphone serves as the reference signal and passes through the noise control filter W and the physical secondary path filter. The superposition of the output and the primary noise is expressed as:

[0048]

[0049] Where, d m (n) represents primary noise, e m (n) represents the physical error signal received by the physical microphone. The auxiliary filter H obtained in step four is used as the fixed-coefficient filter H0. The residual signal generated by the auxiliary filter is the superposition of the physical error signal received by the physical microphone and the output of the reference signal through the fixed-coefficient filter H0. The reference signal passes through the physical secondary path filter. The output of the filter and the residual signal generated by the auxiliary filter work together to update the noise control filter W. The process of updating the noise control filter W is as follows:

[0050] w(n+1)=w(n)-μ w2 e h (n)x s (n)

[0051] Finally, a stable state is reached, achieving active noise reduction at the target noise reduction location. Here, w(n) represents the weight coefficient vector before the noise control filter update, w(n+1) represents the updated weight coefficient vector, and x... s (n) represents the output of the reference signal through the physical secondary path filter, e h (n) represents the residual signal generated by the auxiliary filter, μ w2 This represents the convergence factor for updating the noise control filter.

[0052] In terms of computational complexity (where M represents one multiplication operation, A represents one addition operation, L is the filter length, and LM represents L multiplication operations; by default, all filter lengths are the same):

[0053] The computational cost of performing step one once is: (2L+1)M+2LA

[0054] The computational cost of performing step two once is: (4L+1)M+(4L-2)A

[0055] The computational cost of performing step three once is: (2L+1)M+2LA

[0056] The computational cost of performing step four once is: (4L+1)M+(4L-1)A

[0057] The computational cost of performing step five once is: (6L+1)M+(6L-2)A

[0058] The specific implementation method predicts the noise reduction amount at the target noise reduction location by first recording audio and then simulating the acquired recordings using MATLAB R2016b software. The experiment was conducted in a localized semi-enclosed space. The noise source was a recording with low, medium, and high volume levels. A reference microphone was placed behind the localized semi-enclosed space, and a virtual secondary path S... v Physical secondary pathway S m During modeling, the secondary sound source emitted by the secondary speaker is white noise. The left ear canal opening of the local semi-enclosed space is set as the target noise reduction position. The left ear canal opening, the secondary speaker, and the physical microphone are placed on the same horizontal line. The secondary speaker is located on the left side of the local semi-enclosed space, the physical microphone is 1cm away from the secondary speaker, and the left ear canal opening is 8cm away from the physical microphone.

[0059] All recordings obtained in this experiment were 25 seconds long. During simulation, the lengths of both the noise control filter and the auxiliary filter were set to 384. To ensure that noise reduction at the target denoising location reaches the specified threshold when the primary noise intensity range is large, and given that the convergence factor is known to be inversely proportional to the reference signal energy, the convergence step sizes in this invention are set as follows:

[0060]

[0061]

[0062]

[0063] Where, μ w1 μ represents the convergence factor updated by the noise control filter in step two. h μ represents the convergence factor updated by the auxiliary filter in step four. w2 denoted by , represents the convergence factor updated by the noise control filter in step five, and x represents the reference signal.

[0064] First, we analyze the data training and control phases when the reference signal volume is medium. Finally, we discuss the noise reduction effect of this method in local semi-enclosed space applications when the reference signal volume is low, medium, and high. All results below are obtained through MATLAB R2016b simulations:

[0065] Appendix Figure 2 This is the obtained secondary path filter. The system sampling rate is set to 24kHz, and the impulse response lengths of the virtual and physical secondary paths are 16ms.

[0066] Appendix Figure 3 The error signal graphs for steps two and four are shown when the reference signal volume is medium. (Observation appendix) Figure 3 It can be seen that step two reaches a steady state after t = 0.5, and step four reaches a steady state after t = 1s.

[0067] Through append Figure 4 It can be observed that step five reaches convergence when t is very small, meaning that the system reaches a stable state in a very short time.

[0068] Appendix Figure 5 The noise reduction effect is demonstrated when the reference signal volume is low, medium, and high. (See attached...) Figure 5 It can be seen that when the reference signal volume is low, the noise reduction effect at the target noise reduction location reaches about 20dB; when the reference signal volume is medium, the noise reduction effect at the target noise reduction location reaches about 22dB; and when the reference signal volume is high, the noise reduction effect at the target noise reduction location reaches about 22dB. Therefore, this invention can form an active quiet zone at the target noise reduction location when the reference signal is in a relatively large volume range.

[0069] In summary, using a stepwise virtual sensing noise reduction method can reduce the use of one microphone, saving hardware resources. At the same time, when the reference signal is within a large volume range, an active quiet zone can be formed at the target noise reduction location, demonstrating the effectiveness of the method proposed in this invention.

[0070] It should be noted that the above embodiments are not intended to limit the scope of protection of the present invention. Equivalent transformations or substitutions made on the basis of the above technical solutions, as well as several improvements such as extensions based on them, all fall within the scope of protection of the claims of the present invention.

Claims

1. A stepwise virtual sensing noise reduction method, characterized in that, Includes the following steps: Step 1: Model the virtual secondary path S v A virtual microphone is temporarily placed at the target noise reduction location to collect virtual error signals. According to the LMS algorithm, the secondary sound source emitted by the secondary speaker and the virtual error signals received by the virtual microphone work together to update the virtual secondary path filter. Finally, a stable state is achieved, resulting in a virtual secondary path filter. Step 2: Train the noise control filter W. According to the FxLMS algorithm, the virtual error signal received by the virtual microphone is the superposition of the reference signal, the output of the noise control filter W, and the primary noise. The reference signal then passes through the virtual secondary path filter. The output of the signal and the virtual error signal received by the virtual microphone work together to update the noise control filter W, eventually reaching a stable state, thus obtaining the noise control filter W. Step 3: Model the physical secondary pathway S m The virtual microphone from steps one and two is placed at a physical location as a physical microphone to collect physical error signals. According to the LMS algorithm, the secondary sound source emitted by the secondary speaker and the physical error signals received by the physical microphone work together to update the physical secondary path filter. Finally, a stable state is reached, resulting in a physical secondary path filter. Step 4: Train the auxiliary filter H. No microphone is placed at the target noise reduction location, but a physical microphone is placed at the physical location. The noise control filter W trained in Step 2 is used as the fixed-coefficient filter W0. The physical error signal received by the physical microphone is used as the reference signal and passed through the fixed-coefficient filter W0 and the physical secondary path filter. The output of the auxiliary filter is the superposition of the primary noise and the residual signal generated by the auxiliary filter. The residual signal generated by the auxiliary filter is the superposition of the physical error signal received by the physical microphone and the output of the reference signal through the auxiliary filter H. The reference signal and the residual signal generated by the auxiliary filter work together to update the auxiliary filter H, eventually reaching a stable state, and thus obtaining the auxiliary filter H. Step 5: Enter the control phase. No microphone is placed at the target noise reduction location, but a physical microphone is placed at the physical location. The physical error signal received by the physical microphone serves as the reference signal and passes through the noise control filter W and the physical secondary path filter. The output of the signal is superimposed with the primary noise. The auxiliary filter H obtained in step four is used as a fixed-coefficient filter H0. The residual signal generated by the auxiliary filter is the superposition of the physical error signal received by the physical microphone and the output of the reference signal through the fixed-coefficient filter H0. The reference signal passes through the physical secondary path filter. The output of the filter and the residual signal generated by the auxiliary filter work together to update the noise control filter W, eventually reaching a stable state and achieving active noise reduction at the target noise reduction location.

2. The stepwise virtual sensing noise reduction method according to claim 1, characterized in that, In step two, the noise control filter W is updated as follows: Through formula The weight coefficient vector is updated, where w(n) represents the weight coefficient vector before the noise control filter update, w(n+1) represents the updated weight coefficient vector, and x(n) represents the reference signal. Represents a virtual secondary path filter, e v (n) represents the virtual error signal, μ w1 This represents the convergence factor for updating the noise control filter.

3. The stepwise virtual sensing noise reduction method according to claim 1, characterized in that, In step four, the process of updating the auxiliary filter H is as follows: The formula is: h(n+1)=h(n)-μ h e h (n)x(n) updates the weight coefficient vector, where h(n) represents the weight coefficient vector before the auxiliary filter update, h(n+1) represents the updated weight coefficient vector, and x(n) represents the reference signal, e h (n) represents the residual signal generated by the auxiliary filter, μ h This represents the convergence factor for the auxiliary filter update.

4. The stepwise virtual sensing noise reduction method according to claim 1, characterized in that, In step five, the noise control filter W is updated as follows: The formula w(n+1)=w(n)-μ is used. w2 e h (n)x s (n) Updates the weight coefficient vector, where w(n) represents the weight coefficient vector before the noise control filter update, w(n+1) represents the updated weight coefficient vector, and x s (n) represents the output of the reference signal through the physical secondary path filter, e h (n) represents the residual signal generated by the auxiliary filter, μ w2 This represents the convergence factor for updating the noise control filter.

5. A stepwise virtual sensing noise reduction method according to claims 2 to 4, characterized in that, The convergence factor is set as follows: Where, μ w1 μ represents the convergence factor updated by the noise control filter in step two. h μ represents the convergence factor updated by the auxiliary filter in step four. w2 denoted by , represents the convergence factor updated by the noise control filter in step five, and x represents the reference signal.

Images