Signal processing device, signal processing method and program

Abstract

To achieve active noise control suitable for continuous spaces.SOLUTION: A signal processor of the present disclosure includes: one or more error microphones (11), one or more speakers (12), an adaptive filter (14) that generates a drive signal (y) for the speakers, and a control unit that updates filter coefficients of the adaptive filter (14) based on an interpolation filter matrix (A(i)) in the time domain to minimize sound pressure (L) of an entire target region (Ω) determined by an error signal (e) acquired with the error microphone (11).SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to a signal processing device, a signal processing method, and a signal processing program for active noise control. [Background technology]

[0002] Active noise control (ANC) is a well-known technique that uses a secondary sound source (speaker) to suppress sound pressure at a target position. In many cases, a microphone is placed at or near the target position, and the observed sound pressure is fed back to update an adaptive filter, which then sequentially determines the speaker drive signal.

[0003] When applying ANC to spatial control, a method known as multipoint pressure control (MPC) is known, which extends one-dimensional adaptive filter theory and suppresses sound pressure at multiple control points placed in a target area (for example, Non-Patent Document 1). In the multipoint control method, the positions of multiple error microphones placed in the target area become the multiple control points for suppressing sound pressure. [Prior art documents] [Non-patent literature]

[0004] [Non-Patent Document 1] SJElliott, IMStothers, and PANelson, "A multiple error LMS algorithm and its application to the active control of sound and vibration," IEEE Transactions on Acoustics, Speech, and Signal Processing, vol.35, no.10, pp.1423-1434, 1987 Summary of the Invention [Problem to be solved by the invention]

[0005] In the multipoint control method, the squared l2 norm of the error signal acquired by the error microphone placed in the target region is used as the objective function, and the filter coefficients of the adaptive filter are updated by solving an optimization problem to minimize this. In other words, the multipoint control method is based on an optimization problem that minimizes the power at only a finite number of control points. For this reason, while the multipoint control method is effective in suppressing sound pressure at the placement positions of each error microphone placed in the target region (i.e., each control point), there is a risk that sound pressure may not be sufficiently suppressed throughout the entire target region.

[0006] Therefore, an object of the present invention is to provide a signal processing device, a signal processing method, and a program that realize active noise control suitable for a continuous space. [Means for solving the problem]

[0007] A signal processing device according to one aspect of the present invention includes one or more error microphones, one or more speakers, an adaptive filter that generates drive signals for the speakers, and a control unit that updates the filter coefficients of the adaptive filter based on a time-domain interpolation filter matrix so as to minimize the sound pressure across an entire target region determined based on error signals acquired by the error microphones.

[0008] According to this aspect, the filter coefficients of the adaptive filter are updated based on the time-domain interpolation filter matrix so as to minimize the sound pressure in the entire target region, thereby suppressing the sound pressure in the entire target region, not just the placement position of the error microphone.

[0009] In the above aspect, the control unit may calculate the time-domain interpolation filter matrix based on the relative positional relationship of the error microphones. According to this aspect, the sound pressure distribution in the target region can be estimated based on the relative positional relationship of the error microphones, thereby improving the accuracy of sound pressure suppression in the entire target region.

[0010] In the above aspect, the adaptive filter may further include one or more error microphones, and the adaptive filter may generate the drive signal based on a reference signal detected by the reference microphone and the filter coefficients. According to this aspect, the feedforward ANC can suppress the sound pressure in the entire target area.

[0011] In the above aspect, the adaptive filter may generate the drive signal based on the filter coefficient and a reference signal derived artificially based on the error signal and the drive signal. According to this aspect, in a feedback ANC, sound pressure in the entire target area can be suppressed.

[0012] In the above aspect, the control unit may update the filter coefficients based on the interpolation filter matrix in the time domain, the error signal in the time domain, and the reference signal in the time domain.

[0013] In the above aspect, the control unit may update the filter coefficients based on a gradient generated in the time domain based on the time-domain interpolation filter matrix, a transfer function of a secondary path, the time-domain error signal, and the time-domain reference signal.

[0014] In the above aspect, the control unit may update the filter coefficients for each block based on the time-domain interpolation filter matrix, a reference signal block including a plurality of the reference signals in the time domain, and an error signal block including a plurality of the error signals in the time domain.

[0015] In the above aspect, the control unit may generate a frequency domain block using a frequency domain filter and the reference signal block in the frequency domain obtained by Fourier transforming the reference signal block, the frequency domain filter being configured by combining the interpolation filter matrix in the time domain and a transfer function of a secondary path, or by Fourier transforming each of the interpolation filter matrix and the transfer function, the frequency domain block being generated using the frequency domain block and the reference signal block in the frequency domain obtained by Fourier transforming the error signal block, and may generate a gradient block in the time domain by performing an inverse Fourier transform on the frequency domain block, and update the filter coefficients based on the gradient block.

[0016] A signal processing method according to one aspect of the present invention comprises the steps of acquiring an error signal from one or more error microphones, updating filter coefficients of an adaptive filter based on a time-domain interpolation filter matrix so as to minimize sound pressure across an entire target region determined based on the error signal, and generating drive signals for one or more speakers using the adaptive filter.

[0017] A program according to one aspect of the present invention causes a computer to execute the steps of acquiring an error signal from one or more error microphones, updating filter coefficients of an adaptive filter based on a time-domain interpolation filter matrix so as to minimize sound pressure across an entire target region determined based on the error signal, and generating drive signals for one or more speakers using the adaptive filter. [Effects of the Invention]

[0018] According to the present invention, active noise control suitable for continuous spaces can be realized. [Brief explanation of the drawings]

[0019] [Figure 1] FIG. 2 is a diagram illustrating an example of an arrangement in a feedforward ANC according to the first embodiment. [Figure 2]1 is a diagram illustrating an example of the configuration of a signal processing device of a feedforward type ANC according to a first embodiment. [Figure 3] FIG. 2 is a diagram illustrating an example of a high-speed block KI-FxLMS algorithm according to the first embodiment. [Figure 4] FIG. 1 is a diagram illustrating an example of the physical configuration of a signal processing device for feedforward ANC according to a first embodiment. [Figure 5] 5A and 5B are diagrams showing an example of a comparison in the time domain of sound pressure between the signal processing device according to the first embodiment and ANC according to a conventional method. [Figure 6] FIG. 4 is a diagram showing an example of the amount of sound pressure suppression by the signal processing device according to the first embodiment and ANC according to a conventional method. [Figure 7] 5 is a flowchart showing an example of the operation of the feedforward ANC according to the first embodiment. [Figure 8] FIG. 10 is a diagram illustrating an example of an arrangement in feedback ANC according to a second embodiment. [Figure 9] FIG. 10 is a diagram illustrating an example of the configuration of a feedback ANC signal processing device according to a second embodiment. [Figure 10] 10 is a flowchart showing an example of the operation of feedback ANC according to the second embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0020] An embodiment of the present invention will be described with reference to the accompanying drawings. In each drawing, components with the same reference numerals have the same or similar configurations. In this embodiment, spatial active noise control (spatial ANC) is active noise control (ANC) that has a predetermined space (for example, a two-dimensional space or a three-dimensional space) as a target region (also called a control region).

[0021] In the ANC according to this embodiment, a control signal y (hereinafter referred to as a "drive signal" for the speaker 12) for controlling a secondary sound source (for example, a speaker 12 described later) is generated using an adaptive filter to suppress a sound pressure signal (hereinafter referred to as a "control target signal") d that reaches a control point (for example, the arrangement position of an error microphone 11 described later) via a transmission path from a primary sound source (for example, a noise source N described later). The control target signal d is suppressed at the control point by a sound pressure signal v (hereinafter referred to as a "cancellation signal") that reaches the control point via a transmission path from the secondary sound source based on the drive signal y. Furthermore, an error signal e is obtained based on the cancellation signal v and the control target signal d, and the filter coefficients of the adaptive filter are updated based on the error signal e.

[0022] The filter coefficients of the adaptive filter are adaptively updated using a given algorithm. Hereinafter, the given algorithm will be described as an example of a kernel interpolation-based filtered-X least mean square (KI-FxLMS) or fast block KI-FxLMS, ​​but is not limited to these. By updating the filter coefficients, the control signal y and the cancellation signal generated using the adaptive filter with the filter coefficients are updated, allowing ANC to be performed appropriately.

[0023] The filter coefficients of the adaptive filter are updated using a time-domain interpolation filter matrix A(k) (or an algorithm based on the interpolation filter matrix A(k)). The interpolation filter matrix A(k) is a coefficient matrix used in a kernel interpolation filter (described later) and is also called a weighting matrix filter. The interpolation filter matrix A(k) may be updated based on the relative positional relationship of the error microphones 11. Here, the relative positional relationship of the error microphones 11 may be the relative relationship between the positions of the multiple error microphones 11, such as the arrangement of the multiple error microphones 11 or the relative positions of the multiple error microphones 11. This allows the sound pressure distribution of the target region Ω to be estimated, so that the sound pressure of not only the control points but also the entire target region Ω can be suppressed.

[0024] In this embodiment, for example, the weighting matrix A(ω) in the frequency domain is derived based on the relative positional relationship of the error microphone 11, and the interpolation filter matrix A(k) in the time domain is derived by performing an inverse Fourier transform of the weighting matrix A(ω), but this is not limiting. The interpolation filter matrix A(k) in the time domain may be derived without deriving the weighting matrix A(ω) in the frequency domain.

[0025] Furthermore, the ANC according to this embodiment can be applied to a feedforward type or a feedback type. Below, an example of a feedforward type ANC (first embodiment) and an example of a feedback type ANC (second embodiment) will be specifically described.

[0026] (First embodiment) Fig. 1 is a diagram showing an example of the arrangement of a feedforward ANC according to the first embodiment. As shown in Fig. 1, the feedforward ANC may use a plurality of error microphones 11, a plurality of loudspeakers 12, and a plurality of reference microphones 13. Note that the number and arrangement of the error microphones 11, loudspeakers 12, and reference microphones 13 are not limited to those shown in Fig. 1. The number of each of the error microphones 11, loudspeakers 12, and reference microphones 13 may be one or more.

[0027] 1, for example, error microphones 11 are arranged in a substantially circular shape to surround a target region Ω, which is a region to be subjected to sound pressure suppression. Speakers 12 are also arranged in a substantially circular shape to surround error microphone 11. Reference microphone 13 is also arranged so as to be able to detect reference signal x representing noise source N.

[0028] In FIG. 1, a control-target signal d from a noise source N reaches the error microphone 11 via a transmission path from the noise source N to the error microphone 11 (hereinafter referred to as the "primary path"). A reference microphone 13 detects a reference signal x. A speaker 12 outputs a sound pressure signal based on a drive signal y generated from the reference signal x using an adaptive filter. The sound pressure signal from the speaker 12 reaches the error microphone 11 via a transmission path from the speaker 12 to the error microphone 11 (hereinafter referred to as the "secondary path") and is observed as a cancellation signal v at the error microphone 11. At the error microphone 11, the control-target signal d is canceled by the cancellation signal v (i.e., the noise is canceled). Note that cancellation may be interchangeably referred to as suppression or reduction, etc.

[0029] Fig. 2 is a diagram showing an example of the configuration of a signal processing device for feedforward ANC according to the first embodiment. As shown in Fig. 2, a signal processing device 10 may include an error microphone 11, a speaker 12, a reference microphone 13, an adaptive filter 14, a filter coefficient update unit 15, and an interpolation filter matrix calculation unit 16.

[0030] Although not shown, the signal processing device 10 may be configured without including at least one of the error microphone 11, the speaker 12, the reference microphone 13, and the adaptive filter 14. Furthermore, there may be one or more error microphones 11, speakers 12, and reference microphones 13, and they may be arranged as described in FIG.

[0031] The reference microphone 13 shown in FIG. 2 detects a reference signal x representing a noise source N and outputs the generated reference signal x to the adaptive filter 14. Specifically, the reference microphone 13 may generate a reference signal x(n) at time n based on an input signal (e.g., frequency) from the noise source N and output the generated reference signal x(n) to the adaptive filter 14. The reference signal x(n) is a signal in the time domain (e.g., discrete time domain). The reference signal x(n) is also referred to as a reference microphone signal, an observed signal at the reference microphone 13, etc. Note that the time n is a predetermined time unit identified by an index n, and may also be referred to as a sample time, a time instant, a period of time, etc.

[0032] The adaptive filter 14 generates a drive signal y(n) for the speaker 12 from a reference signal x(n) detected by the reference microphone 13, using the filter coefficient W(i) updated by the filter coefficient update unit 15. For example, if the adaptive filter 14 is a finite impulse response (FIR) type adaptive digital filter with a filter length I, the drive signal y(n) at time n may be expressed by the following equation (1). Note that the drive signal y(n) is a signal in the time domain (for example, the discrete time domain). The drive signal y(n) is also called a speaker drive signal, a secondary sound source signal, a control signal, or an output signal of the adaptive filter.

number

[0033] The speaker 12 outputs a sound pressure signal based on the drive signal y(n) input from the adaptive filter 14. The sound pressure signal passes through a secondary path and is detected by the error microphone 11 as a cancellation signal v.

[0034] The error microphone 11 observes an error signal e based on the control target signal d and the cancellation signal v. Specifically, the error microphone 11 observes, as the error signal e, a signal obtained by superimposing the control target signal d arriving from the noise source N via the primary path and the cancellation signal v arriving from the speaker 12 via the secondary path. For example, the error microphone 11 may generate an error signal e(n) at time n based on the control target signal d(n) and the cancellation signal v(n) at time n (for example, by the following equation (2)). The error signal e is a signal in the time domain (for example, the discrete time domain). The error signal e is also called an error microphone signal, an observation signal at the error microphone 11, etc. Formula (2) e(n)=d(n)+v(n)

[0035] The cancellation signal v(n) can be expressed as in the following equation (3) if the transfer characteristics of the secondary path (hereinafter referred to as "secondary path characteristics") are approximated to an FIR filter with a filter length J. The secondary path characteristics can also be rephrased as an impulse response from the speaker 12 to the error microphone 11.

number

number

[0036] The filter coefficient update unit 15 updates the filter coefficient W(i) used in the adaptive filter 14. Specifically, the filter coefficient update unit 15 may update the filter coefficient W(i) by solving an optimization problem that minimizes a cost function L, where the cost function L is the square integral value of the sound pressure in the entire target region Ω estimated based on the pressure distribution u(r,n) in the target region Ω at time n. The cost function L is expressed, for example, by Equation (5). The square integral value of the sound pressure in the entire target region Ω may also be referred to as the power or acoustic energy of the sound field in the target region Ω. The filter coefficient update unit 15 functions as a control unit that controls the update of the filter coefficient W(i) based on a time-domain interpolation filter matrix A(k) so as to minimize the sound pressure in the entire target region Ω determined based on the error signal e(n) at time n.

number

[0037] As shown in Figure 1, when error microphones 11 are placed at discrete positions r within or near a target region Ω, the pressure distribution u(r,n) in the target region Ω at time n can be estimated by applying a sound field kernel interpolation method from the error signal e(n) acquired by the error microphone 11 at position r at time n. Here, the sound field kernel interpolation method is an interpolation method based on kernel ridge regression, which is constrained to obey the Helmholtz equation when estimating a continuous sound pressure distribution in the target region Ω from multiple dispersedly placed error microphones 11. A kernel interpolation filter is a filter that receives the error signal e(n) as input and outputs a sound pressure value at any position r within the target region Ω based on the kernel interpolation method.

[0038] The filter coefficient update unit 15 may estimate the pressure distribution u(r,n) in the target region Ω from the error signal e(n) acquired by the error microphone 11 at the position r using a time-domain kernel interpolation filter z(r,i).3 where r is a position vector or position in three-dimensional Euclidean space.

[0039] The filter coefficient update unit 15 may use KI-FxLMS or fast block KI-FxLMS as a time-domain adaptive filter algorithm to update the filter coefficients W(i) based on the time-domain interpolation filter matrix A(k) so as to minimize the sound pressure (i.e., the above-mentioned cost function L) over the entire target region Ω. The KI-FxLMS algorithm and the fast block KI-FxLMS algorithm will be described below.

[0040] <KI-FxLMSアルゴリズム> The time-domain kernel interpolation filter z(r,i) used to estimate the pressure distribution u(r,n) in the target region Ω from the error signal e(n) acquired by the error microphone 11 at the position r at time n may be derived based on the frequency-domain kernel interpolation filter z(r,ω). For example, the time-domain kernel interpolation filter z(r,i) may be obtained by performing an inverse Fourier transform (e.g., a discrete-time Fourier transform) on the frequency-domain interpolation filter z(r,ω), as shown in the following equation (6):

number

number

number

number

number

[0041] The filter coefficient update unit 15 may use the time-domain kernel interpolation filter z(r, i) as described above to estimate the pressure distribution u(r, n) of the target region Ω from the error signal e(n) using, for example, the following equation (12):

number

[0042] Based on the pressure distribution u(r,n) estimated by equation (12), the cost function L of equation (5) may be rewritten as equation (13) below.

number

number

[0043] The filter coefficient update unit 15 may derive a gradient Δ(i) of the filter coefficient W(i) based on the coefficient matrix Γ(i,j) of the time-domain kernel interpolation filter z(r,i). The filter coefficient update unit 15 may derive the gradient Δ(i) using, for example, the following equation (15):

number

[0044] Next, we apply a variable transformation such that μ = ν + k. If the reference signal x(n) and the error signal e(n) at time n are assumed to be stationary within a local time window, then the cross correlation E[e(n)x T (n i )] depends on the time difference i. Therefore, the filter coefficient update unit 15 may derive the gradient Δ(i) based on the interpolation filter matrix A(k) in the time domain. The filter coefficient update unit 15 may derive the gradient Δ(i) using, for example, the following equation (16):

number

[0045] The time-domain interpolation filter matrix A(k) must be implemented as a causal FIR filter. The interpolation filter matrix A(k) may be approximated by truncating it to a symmetric FIR filter with a filter length of 2K+1, and then made causal by adding a delay of K samples. In this case, the approximated time-domain interpolation filter matrix Â(i) may be expressed as the following equation (17):

number

[0046] Furthermore, the filter coefficient update unit 15 may derive the gradient Δ(i) as shown in the following equation (18), for example, by adding a delay of K samples, which is the same as that of the interpolation filter matrix A^(k), to the error signal e(n).

number

[0047] The filter coefficient update unit 15 updates the FIR filter G T The filter coefficients W(i) of the adaptive filter 14 may be updated based on a single filter H previously combined with an FIR filter A^ relating to the interpolation filter matrix A. The single filter H(i) may be, for example, shown in the following equation (19).

number

[0048] The filter coefficient update unit 15 may update the filter coefficient W(i) of the adaptive filter 14 based on the single filter H(i), for example, as shown in the following equation (20). Note that in the following equation (20), the expected value calculation E[e(nK)x T (nij-k)] is the instantaneous value e(nK)x T It can be replaced by (nij―k).

number

number

[0049] In the above description, the filter coefficient update unit 15 updates the filter coefficient W(i) using a single filter H obtained by combining in advance an FIR filter related to the secondary path characteristics G and a time-domain kernel interpolation filter Â, as shown in, for example, equations (19) and (20). However, the present invention is not limited to this. The filter coefficient update unit 15 may update the filter coefficient W(i) using an FIR filter related to the secondary path characteristics G and an FIR filter related to the time-domain interpolation filter matrix A, instead of the single filter H. For example, the filter coefficient update unit 15 may update the filter coefficient W(i) using, for example, the following equation (22) or the following equation (23) based on the gradient Δ(i) shown in the above equation (18).

number

[0050] The FIR filter related to the secondary path characteristic G may be rephrased as the weighting matrix of the FIR filter, the transfer function G(j), etc. Furthermore, the FIR filter related to the time-domain interpolation filter matrix A(k) may be rephrased as the weighting matrix of the FIR filter, the time-domain kernel interpolation filter, the filter coefficients of the kernel interpolation filter, the weighting matrix filter, the interpolation filter, etc. Furthermore, the single filter H may be rephrased as the filter coefficients of the single filter, etc.

[0051] The KI-FxLMS algorithm described above allows for the application of an adaptive filtering algorithm in the time domain. Implementing a time domain algorithm from a frequency domain algorithm is not easy for broadband noise. Therefore, the KI-FxLMS algorithm can be suitably used as an adaptive filtering algorithm for ANC.

[0052] <High-speed block KI-FxLMS algorithm> The KI-FxLMS algorithm described above may require high computational costs when multiple channels and long filters are required. In such cases, computational costs can be reduced by making the adaptive filter 14 constant within a finite block, employing block-based calculations, and implementing the algorithm using a fast Fourier transform (FFT). This method is also used in standard FxLMS and is also called fast block FxLMS. The filter coefficient update unit 15 may update the filter coefficients W using a fast block KI-FxLMS algorithm, which is a block implementation of the KI-FxLMS algorithm.

[0053] FIG. 3 is a diagram showing an example of the fast block KI-FxLMS algorithm according to this embodiment. Each linear convolution and correlation operation may be performed as an overlap-save algorithm using FFT. In FIG. 3, the size of a block is B, the block index is b, and each block of a signal is represented by square brackets. For example, a reference signal block x[b] is the bth reference signal block and is composed of B reference signals x(n) (where n∈[bB, (b+1)B-1]). The length of the control filter I is also B. Furthermore, F 2B indicates the FFT for a sequence of two blocks of block length B. -1 2B indicates the inverse FFT for a sequence of two blocks of block length B.

[0054] In FIG. 3, two reference signal blocks x[b-1] and x[b] are generated by performing serial-parallel conversion on 2B reference signals x(n). The two reference signal blocks x[b-1] and x[b] in the time domain are subjected to FFT (F 2B ) to convert it to the frequency domain, multiply it by the frequency domain filter H(ω), and then perform the inverse FFT (F -1 2B Here, the frequency domain filter H(ω) is obtained by, for example, performing FFT (F 2B ) and converted from the time domain to the frequency domain. Two reference signal blocks X[b-1] and x[b] are generated by using a filter H(ω) to take into account the secondary path characteristic G and to which kernel interpolation is applied (hereinafter referred to as "interpolated filtered reference signal blocks") F [b-1] and X F [b] is generated.

[0055] Inverse FFT(F -1 2B ) is generated by one of two interpolated filtered reference signal blocks in the time domain (e.g., X F [b-1]) is discarded, and the other interpolated filtered reference signal block (e.g., X F [b]) is converted into 2B interpolated filtered reference signals X F (n) is converted into two interpolation filtered reference signal blocks X F [b-1] and X F [b] is generated. Two interpolated filtered reference signal blocks X F [b-1] and X F [b] is FFT(F 2B ) is transformed into the frequency domain.

[0056] On the other hand, for B error signals e(n) (where n∈[bB,(b+1)B-1]), z -K A delay of K samples is added to the error signal e(n) to generate B error signals e(nK). By adding B zeros to the front of the B error signals e(n-K) and performing serial-to-parallel conversion, a block consisting of B zeros (hereinafter referred to as the "zero block") and an error signal block e[b] consisting of B error signals e(nK) are generated, and these two blocks are subjected to the FFT (F 2B ) is transformed into the frequency domain.

[0057] The zero block and error signal block e[b] transformed into the frequency domain are multiplied by the two interpolated filtered reference signal blocks transformed into the frequency domain, and the inverse FFT (F -1 2B ) and then parallel-to-serial conversion is performed to generate the gradient Δ. The gradient Δ may be multiplied by a block of step parameters η[b], and the filter coefficients W of the adaptive filter 14 may be updated block by block based on the multiplication result η[b]Δ(0:B).

[0058] In FIG. 3, the filter coefficient update unit 15 updates the filter coefficient W in units of blocks based on the time domain filter H(i) (the frequency domain filter H(ω) based on the time domain filter H(i)) which is previously combined with an FIR filter related to the secondary path characteristic G and a time domain kernel interpolation filter A^, but this is not limited to this. The filter coefficient update unit 15 may also update the filter coefficient W in units of blocks using the time domain secondary path filter G and the time domain kernel interpolation filter A separately. In this case, the time domain secondary path filter G and the time domain kernel interpolation filter A may each be transformed into the frequency domain by inverse FFT and multiplied by a reference signal block in the frequency domain.

[0059] The interpolation filter matrix calculation unit 16 calculates the time domain interpolation filter matrix A(k) used to update the filter coefficients W(i) in the filter coefficient update unit 15. As described above, the time domain interpolation filter matrix A(k) is used in the KI-FxLMS algorithm or the high-speed block KI-FxLMS algorithm. The interpolation filter matrix calculation unit 16 may be configured as part of the control unit.

[0060] Specifically, the interpolation filter matrix calculation unit 16 calculates the time-domain interpolation filter matrix A(k) based on the relative positional relationship of the error microphone 11. For example, as shown in the following equations (24) and (25), the interpolation filter matrix calculation unit 16 may calculate the frequency-domain weighting matrix A(ω) based on the relative positional relationship of the error microphone 11, and then perform an inverse Fourier transform on the frequency-domain weighting matrix A(ω) to calculate the time-domain interpolation filter matrix A(k). Note that calculation may simply be rephrased as derivation.

number

[0061] Note that the above is merely an example, and the interpolation filter matrix calculation unit 16 may derive the time-domain interpolation filter matrix A(k) based on the relative positional relationship of the error microphone 11, without deriving the weighting matrix A(ω) in the frequency domain. Furthermore, the interpolation filter matrix calculation unit 16 may calculate the time-domain interpolation filter matrix A(k) based on at least one of the wavenumber k controlled in the target region Ω, the size of the target region Ω (for example, the radius if the target region Ω is circular), and prior information regarding the noise source N (for example, the power distribution in each direction of the noise source N, the initial noise direction, etc.), in addition to the relative positional relationship of the error microphone 11.

[0062] Furthermore, the interpolation filter matrix calculation unit 16 may derive the frequency domain weighting matrix A(ω) based on a kernel interpolation filter z(r,ω) based on the relative positional relationship of the error microphone 11 (and at least one of the wave number k controlled in the target region Ω, the size of the target region Ω, and prior information related to the noise source N). The interpolation filter matrix calculation unit 16 may perform an inverse FFT on the frequency domain weighting matrix A(ω) to derive the time domain interpolation filter matrix A(k).

[0063] FIG. 4 is a diagram showing an example of the physical configuration of a signal processing device for feedforward ANC according to the first embodiment. The signal processing device 10 includes a CPU (Central Processing Unit) 10a corresponding to a calculation unit, a RAM (Random Access Memory) 10b corresponding to a storage unit, a ROM (Read Only Memory) 10c corresponding to a storage unit, a communication unit 10d, an input unit 10e, and a display unit 10f. These components are connected via a bus so as to be able to transmit and receive data to and from each other. In this example, the signal processing device 10 is configured by a single computer, but the signal processing device 10 may also be realized by combining multiple computers. The configuration shown in FIG. 2 is an example, and the signal processing device 10 may include other components or may not include some of these components.

[0064] The CPU 10a is a control unit that controls the execution of programs stored in the RAM 10b or the ROM 10c and performs data calculations and processing. The CPU 10a is a calculation unit or control unit that executes a program that controls the update of the filter coefficient W(i) so as to minimize a cost function L based on the square integral value of the sound pressure over the entire target region Ω. The CPU 10a receives various data from the input unit 10e and the communication unit 10d, and displays the calculation results of the data on the display unit 10f or stores them in the RAM 10b.

[0065] The RAM 10b is a rewritable storage unit and may be configured, for example, with a semiconductor memory element. The RAM 10b may store the program executed by the CPU 10a, the interpolation filter matrix A, the filter coefficients H, etc. Note that these are merely examples, and the RAM 10b may store data other than these, or may not store some of these.

[0066] The ROM 10c is a storage unit from which data can be read, and may be configured, for example, with a semiconductor memory element. The ROM 10c may store, for example, a signal processing program or data that is not rewritten.

[0067] The communication unit 10d is an interface that connects the signal processing device 10 to other devices, and may be connected to a communication network such as the Internet.

[0068] The input unit 10e receives data input from a user, and may include, for example, a keyboard and a touch panel.

[0069] The display unit 10f visually displays the results of calculations performed by the CPU 10a, and may be configured with, for example, an LCD (Liquid Crystal Display).

[0070] The signal processing program may be provided by being stored in a computer-readable storage medium such as RAM 10b or ROM 10c, or may be provided via a communication network connected by communication unit 10d. In signal processing device 10, CPU 10a executes the signal processing program to realize the various operations described with reference to FIG. 1. Note that these physical configurations are merely examples and do not necessarily have to be independent configurations. For example, signal processing device 10 may include an LSI (Large-Scale Integration) in which CPU 10a is integrated with RAM 10b and ROM 10c.

[0071] Next, the signal processing device 10 according to the first embodiment will be compared with ANC according to a conventional method with reference to Figures 5 and 6. In Figures 5 and 6, the signal processing device 10 updates the filter coefficients W using the KI-FxLMS algorithm or the fast block KI-FxLMS algorithm. On the other hand, the signal processing device according to the conventional method updates the filter coefficients W using the FxLMS algorithm or the fast block FxLMS algorithm.

[0072] The target area Ω is a cube of 0.6m x 0.6m x 0.1m, with its center at the origin. The numbers of error microphones 11 and speakers 12 are M = 48 and L = 32, respectively. 24 error microphones 11 are regularly arranged along the 0.6m x 0.6m x 0.1m square target area Ω. Meanwhile, 24 error microphones 11 are regularly arranged on two planes at z = 0.5m and -0.5m along the 0.6m x 0.6m square, with half of them shifted outward by 0.03m. 32 speakers 12 are arranged along a 2.0m x 2.0m square. Seventy-two evaluation points were set within the target area Ω, with 36 of them regularly spaced at 0.1 m intervals within a 0.6 m x 0.6 m square. Another 36 points were regularly spaced at 0.1 m intervals at heights of z = 0.025 m and -0.025 m. Furthermore, it was assumed that the control target signal d was obtained directly from noise source N, and noise source N and speaker 12 were conventional closed-type loudspeakers. Error microphone 11 was omnidirectional. Impulse responses between noise source N and speaker 12 and between error microphone 11 and the evaluation points were measured one at a time using a swept sine wave signal. The room dimensions were approximately 7.0 m x 6.4 m x 2.7 m, and the reverberation time T60 was approximately 0.38 seconds.

[0073] The sampling frequency was set to 4000 Hz. The truncation length of the kernel interpolation filter was set to K=77, and the truncation length of the adaptive filter 14 was set to I=2048. The block size B of the high-speed block KI-FxLMS or high-speed block FxLMS was set to 2048. The normalization parameter λ in equation (7) was set to 10 -3, the parameter β used to update the step size parameter in equation (21) is also 10 -3 For the fast block KI-FxLMS, ​​the constant parameter η0 is set to [10 1 ,10 3 ] and divided equally into 30 values ​​on a logarithmic scale. For KI-FxLMS, ​​the constant parameter η0 is [10 1 ,10 3 ] was determined from the same range, but divided by B = 2048. A similar procedure was followed for FxLMS and Fast FxLMS, ​​with [10- 2, ,10 1 ] was carried out within the scope of

[0074] Next, for performance measurement, the amount of power reduction in the target region Ω in the time domain is expressed by the following equation (26).

number

[0075] In addition, the following two types of noise were considered as noise sources N for evaluation. White Gaussian noise (noise) filtered with a bandpass filter with a passband of 50 to 900 Hz High Horse (music) by Secret Mountains taken from the dataset

[0076] In Fig. 5(a)(b), the power reduction amount P red is shown on the time axis. In Fig. 6, the power reduction amount P red Here, the power reduction amount P redThe time interval for calculating was set to 60 seconds, and 240,000 samples were calculated after 60 seconds of adaptation. As for noise, the results showed that KI-FxLMS and fast block KI-FxLMS achieved greater noise reduction effects than FxLMS and fast block FxLMS. This is because the kernel interpolation-based method takes local noise into account. KI-FxLMS had a slightly higher noise reduction effect than fast block KI-FxLMS, ​​but the noise reduction effects of FxLMS and fast block FxLMS were almost the same. Since music is non-stationary, the power reduction amount P red Although changes over time, the kernel interpolation-based KI-FxLMS and fast block KI-FxLMS still outperformed the MPC-based FxLMS and fast block FxLMS in terms of noise reduction effect for music, as shown in Figure 6. Because fast adaptation is required for non-stationary noise, the difference in noise reduction effect between KI-FxLMS and fast block KI-FxLMS was larger for music than for noise.

[0077] Fig. 7 is a flowchart showing an example of the operation of the feedforward ANC according to the first embodiment. As shown in Fig. 7, the error microphone 11 acquires an error signal e(n) based on a control target signal d(n) transmitted from a noise source N via a primary path and a cancellation signal v(n) transmitted from a speaker 12 via a secondary path (step S101).

[0078] The filter coefficient update unit 15 updates the filter coefficient W(i) of the adaptive filter 14 based on the time domain weight sequence A(i) so as to minimize the sound pressure of the entire target region Ω determined based on the error signal e(n) at the error microphone 11 (step S102). Specifically, the filter coefficient update unit 15 may update the filter coefficient W(i) of the adaptive filter 14 using the above-mentioned KI-FxLMS algorithm or the high-speed block KI-FxLMS algorithm.

[0079] The adaptive filter 14 generates a drive signal y(n) for the speaker 12 from the reference signal x(n) at the reference microphone 13 using the filter coefficient W(i) updated in step S102 (step S103). The speaker 12 outputs a sound pressure signal based on the drive signal y(n) (step S104). The sound pressure signal passes through a secondary path and is observed as a cancellation signal v(n) at the error microphone 11.

[0080] The signal processing device 10 determines whether or not to end the processing (step S105), and if the processing is not to end, returns to step S101.

[0081] As described above, according to the first embodiment, the filter coefficients w(i) of the adaptive filter 14 are updated based on the time-domain interpolation filter matrix A(k) so as to minimize the sound pressure in the entire target region Ω. Therefore, in the feedforward ANC, it is possible to suppress not only the placement position of the error microphone 11 but also the sound pressure in the entire target region Ω.

[0082] Furthermore, the gradient Δ(i) used to update the filter coefficients W(i) is derived in the time domain based on the time-domain interpolation filter matrix A(k). Therefore, when the gradient Δ is derived in the frequency domain based on the frequency-domain weighting coefficients A(ω), the gradient Δ can be reduced due to the conversion from the frequency domain to the time domain.

[0083] (Second embodiment) Fig. 8 is a diagram showing an example of the arrangement of a feedback ANC according to the second embodiment. As shown in Fig. 8, the feedback ANC includes multiple error microphones 11 and multiple speakers 12, but differs from the feedforward ANC according to the first embodiment in that it does not include multiple reference microphones 13. Note that the number and arrangement of the error microphones 11 and speakers 12 are not limited to those shown in Fig. 8. The number of error microphones 11 and the number of speakers 12 need only be one or more. Below, the second embodiment will be described, focusing on the differences from the first embodiment.

[0084] In FIG. 8, a control target signal d from a noise source N reaches the error microphone 11 via the primary path. In a feedback ANC, the reference microphone 13 of FIG. 1 is not provided, so a pseudo reference signal (hereinafter referred to as a "pseudo reference signal") x^ is generated based on a drive signal y of the speaker 12 and an error signal e acquired by the error microphone 11. The speaker 12 outputs a sound pressure signal based on a drive signal y generated from the pseudo reference signal x^ using an adaptive filter. The sound pressure signal from the speaker 12 reaches the error microphone 11 via the secondary path, and a cancellation signal v from the speaker 12 reaches the error microphone 11, where it is observed as a cancellation signal v. At the error microphone 11, the control target signal d is canceled by the cancellation signal v (i.e., the noise is canceled).

[0085] Fig. 9 is a diagram showing an example of the configuration of a feedback ANC signal processing device according to the second embodiment. As shown in Fig. 9, a signal processing device 20 includes an error microphone 11, a speaker 12, an adaptive filter 14, a filter coefficient update unit 15, and an interpolation filter matrix calculation unit 16, but does not include a reference microphone 13. The differences from Fig. 2 will be mainly described in Fig. 9.

[0086] Although not shown, the signal processing device 20 may be configured without including at least one of the error microphone 11, the speaker 12, and the adaptive filter 14. Furthermore, there may be one or more error microphones 11, one or more speaker 12, and one or more reference microphones 13, and they may be arranged as described in Fig. 8. Furthermore, the signal processing device 20 may, of course, include a generating unit that generates a pseudo reference signal x̂.

[0087] 9, the signal processing device 20 does not include the reference microphone 13, and therefore the pseudo reference signal x^ may be generated based on the error signal e acquired by the error microphone 11 and the drive signal y of the speaker 12. Specifically, the pseudo reference signal x^(n) at time n may be generated based on the drive signal y(n) generated by the adaptive filter 14, the transfer function G(j) of the secondary path, and the error signal e(n) acquired by the error microphone 11. The pseudo reference signal x^(n) may be generated based on, for example, the following equation (27):

number

[0088] The adaptive filter 14 generates a drive signal y(n) for the speaker 12 from the pseudo reference signal x^(n) using the filter coefficient W updated by the filter coefficient update unit 15. As described above, the signal processing device 20 differs from the signal processing device 10 in that the signal processing device 20 uses the error signal e(n) and the pseudo reference signal x^(n) instead of the reference signal x(n) detected by the reference microphone 13. The signal processing device 20 can apply the adaptive filter 14 and the filter coefficient update unit 15 of the signal processing device 10 described in the first embodiment by replacing the reference signal x(n) with the pseudo reference signal x^(n). Note that the interpolation filter matrix calculation unit 16 of the signal processing device 20 may be the same as the interpolation filter matrix calculation unit 16 of the signal processing device 10. Furthermore, the physical configuration of the signal processing device 20 can appropriately use the physical configuration of the signal processing device 10 shown in FIG. 4.

[0089] 10 is a flowchart showing an example of the operation of feedback ANC according to the second embodiment. Steps S201, S204, and S205 in Fig. 10 are the same as steps S101, S104, and S205 in Fig. 7.

[0090] The filter coefficient update unit 15 updates the filter coefficients W(i) of the adaptive filter 14 based on the time-domain interpolation filter matrix A(k) so as to minimize the sound pressure over the entire target region Ω determined based on the error signal e(n) acquired by the error microphone 11 (step S202). Specifically, the filter coefficient update unit 15 may update the filter coefficients W of the adaptive filter 14 using the above-mentioned KI-FxLMS algorithm or the fast block KI-FxLMS algorithm. In the above-mentioned KI-FxLMS algorithm or the fast block KI-FxLMS algorithm, a pseudo reference signal x^(n) is used instead of the reference signal x(n).

[0091] The adaptive filter 14 generates a drive signal y(n) for the speaker 12 from the pseudo reference signal x^(n) using the adaptive filter 14 with the filter coefficient W(i) updated in step S202 (step S203).

[0092] As described above, according to the first embodiment, the filter coefficients w(i) of the adaptive filter 14 are updated based on the time-domain interpolation filter matrix A(k) so as to minimize the sound pressure over the entire target region Ω, and therefore, in a feedback ANC, it is possible to suppress the sound pressure over the entire target region Ω, not just the placement position of the error microphone 11. Furthermore, because the gradient Δ(i) used to update the filter coefficients W(i) is derived in the time domain based on the time-domain interpolation filter matrix A(k), it is possible to prevent delays due to conversion from the frequency domain to the time domain, compared to when the gradient Δ is derived in the frequency domain based on the frequency-domain weighting coefficients A(ω).

[0093] (others) In the first and second embodiments, the multiple error microphones 11 and the multiple speakers 12 are arranged in a substantially circular shape, but this is not limited to this and they may be arranged in any shape, such as a linear shape, a rectangular shape, a triangle, a square, or a rectangle. Similarly, in the first embodiment, the reference microphone 13 is arranged in a substantially triangular shape, but this is not limited to this and they may be arranged in any shape. Furthermore, the error microphone 11, the speaker 12, and the reference microphone 13 may be arranged in three dimensions, not just two dimensions. In a feedforward ANC, the error microphone 11, the speaker 12, and the reference microphone 13 may be arranged in this order near the target region Ω. In a feedback ANC, the error microphone 11, the speaker 12, and the reference microphone 13 may be arranged in this order near the target region Ω. Furthermore, the target region Ω may have any shape, such as a circular region or an elliptical region.

[0094] Furthermore, in the first and second embodiments, KI-FxLMS and fast block FxLMS are exemplified as KI-based algorithms used to update the filter coefficients W(i) of the adaptive filter 14, but the KI-based algorithm is not limited to these. For example, as long as it is KI-based, it is not limited to FxLMS or fast block FxLMS, ​​and other algorithms such as LMS, normalized LMS (NLMS), and recursive least-squares may also be used. As such, in this embodiment, any algorithm based on a time-domain interpolation filter matrix A(k) for kernel interpolation can be used.

[0095] The above-described embodiments are intended to facilitate understanding of the present invention and are not intended to limit the present invention. The elements of the embodiments, as well as their arrangement, materials, conditions, shapes, sizes, etc., are not limited to those illustrated and can be modified as appropriate. Furthermore, configurations shown in different embodiments can be partially substituted or combined with each other. [Explanation of symbols]

[0096] 10...signal processing device, 10a...CPU, 10d...communication unit, 10e...input unit, 10f...display unit, 11...error microphone, 12...speaker, 13...reference microphone, 14...adaptive filter, 15...filter coefficient update unit, 16...interpolation filter matrix calculation unit, 20...signal processing device, N...noise source

Claims

1. one or more error microphones; one or more speakers; an adaptive filter that generates a drive signal for the speaker; a control unit that updates the filter coefficients of the adaptive filter based on a time-domain interpolation filter matrix so as to minimize the sound pressure in the entire target region determined based on the error signal acquired by the error microphone; one or more reference microphones; the adaptive filter generates the drive signal based on a reference signal detected by the reference microphone and the filter coefficients; the control unit updates the filter coefficients for each block based on the time-domain interpolation filter matrix, a reference signal block including a plurality of the time-domain reference signals, and an error signal block including a plurality of the time-domain error signals; the control unit generates a frequency domain block using a frequency domain filter and the reference signal block in the frequency domain obtained by Fourier transforming the reference signal block, the frequency domain filter being configured by combining the time domain interpolation filter matrix and a transfer function of a secondary path, or by Fourier transforming each of the interpolation filter matrix and the transfer function; generating the time-domain gradient block by performing an inverse Fourier transform on a frequency-domain block generated using the frequency-domain block and the frequency-domain error signal block obtained by Fourier transforming the error signal block; updating the filter coefficients based on the gradient block; Signal processing device.

2. the control unit calculates the time domain interpolation filter matrix based on a relative positional relationship of the error microphones. The signal processing device according to claim 1 .

3. the adaptive filter generates the drive signal based on a reference signal derived artificially based on the error signal and the drive signal, and the filter coefficient.

3. The signal processing device according to claim 1 or 2.

4. the control unit updates the filter coefficients based on the interpolation filter matrix in the time domain, the error signal in the time domain, and the reference signal in the time domain.

4. The signal processing device according to claim 1 or 3.

5. the control unit updates the filter coefficients based on a gradient generated in the time domain based on the time domain interpolation filter matrix, a secondary path transfer function, the time domain error signal, and the time domain reference signal. The signal processing device according to claim 4 .

6. acquiring an error signal at one or more error microphones; updating filter coefficients of an adaptive filter based on a time-domain interpolation filter matrix so as to minimize a sound pressure across a target region determined based on the error signal; generating drive signals for one or more speakers using the adaptive filter; detecting a reference signal at one or more reference microphones; In the step of generating the drive signal, the adaptive filter generates the drive signal based on a reference signal detected by the reference microphone and the filter coefficients; In the step of updating the filter coefficients, the filter coefficients are updated for each block based on the time-domain interpolation filter matrix, a reference signal block including a plurality of the reference signals in the time domain, and an error signal block including a plurality of the error signals in the time domain; In the step of updating the filter coefficients, a frequency domain block is generated using a frequency domain filter and the reference signal block in the frequency domain obtained by Fourier transforming the reference signal block, and the frequency domain filter is configured by combining the time domain interpolation filter matrix and a transfer function of a secondary path, or by Fourier transforming each of the interpolation filter matrix and the transfer function; generating the time-domain gradient block by performing an inverse Fourier transform on a frequency-domain block generated using the frequency-domain block and the frequency-domain error signal block obtained by Fourier transforming the error signal block; updating the filter coefficients based on the gradient block; Signal processing methods.

7. acquiring an error signal at one or more error microphones; updating filter coefficients of an adaptive filter based on a time-domain interpolation filter matrix so as to minimize a sound pressure across a target region determined based on the error signal; generating drive signals for one or more speakers using the adaptive filter; detecting a reference signal at one or more reference microphones; In the step of generating the drive signal, the adaptive filter generates the drive signal based on a reference signal detected by the reference microphone and the filter coefficients; In the step of updating the filter coefficients, the filter coefficients are updated for each block based on the time-domain interpolation filter matrix, a reference signal block including a plurality of the reference signals in the time domain, and an error signal block including a plurality of the error signals in the time domain; In the step of updating the filter coefficients, a frequency domain block is generated using a frequency domain filter and the reference signal block in the frequency domain obtained by Fourier transforming the reference signal block, and the frequency domain filter is configured by combining the time domain interpolation filter matrix and a transfer function of a secondary path, or by Fourier transforming each of the interpolation filter matrix and the transfer function; generating the time-domain gradient block by performing an inverse Fourier transform on a frequency-domain block generated using the frequency-domain block and the frequency-domain error signal block obtained by Fourier transforming the error signal block; updating the filter coefficients based on the gradient block; program.

Images