Frequency-invariant microphone array beam forming method
By selecting the angular frequency ω0 in the microphone array, setting the narrowband beam, calculating the array element weight and delay, converting the frequency spectrum and reducing the side lobe level, the problems of large amount of frequency invariant beamforming algorithms and widening in the existing technology are solved. , the effect of maintaining beam consistency and reducing side lobe level at low frequencies is achieved.
Patent Information
- Application Number
- CN202411014380.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-26
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-07-26
AI Technical Summary
The existing frequency-invariant beamforming algorithm has a large amount of computation and the beam will widen when the frequency is low, resulting in poor results in voice signal processing.
By selecting the angular frequency ω0, setting the narrowband beam, calculating the weight of the array element input signal, converting the frequency spectrum, calculating the delay, obtaining a new spectrum, calculating the gain, and using IFFT to convert the output signal to the time domain, combining with a nonlinear function to reduce the side lobe level, achieving constant frequency microphone array beamforming.
When the frequency is extremely low, the beam parameters are kept consistent, the calculation is simple, the calculation amount is low, the side lobe level is reduced, and the phase continuity of the speech signal is maintained.
Smart Images

Figure CN118972747A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of signal processing, and in particular to a frequency-invariant microphone array beamforming method. Background Art
[0002] Beamforming is a spatial filtering technology that is used to receive signals in a specific spatial direction while attenuating signals from other directions. Depending on the signal bandwidth, beamforming can be divided into two categories: narrowband beamforming and wideband beamforming. Compared with narrowband beamforming, when the received signal is a wideband signal, the phase difference of the received signal of each array element is related to the frequency. Therefore, wideband beamforming requires additional structures to control the frequency-related beam shape, such as sub-band processing, filter addition, etc.
[0003] The beam width of conventional beamforming algorithms usually becomes wider at lower frequencies, which makes the array's directivity worse at low frequencies. This has great limitations when processing some low-frequency signals. For example, in speech signal processing, the beam of conventional microphone arrays has poor directivity at low frequencies, and low frequencies are more important to hearing than high frequencies, which limits the effect of conventional microphone array beamforming algorithms. Frequency-invariant beamforming algorithms are often used to solve the problem that the beam width increases as the frequency decreases, and keep the beam width as constant as possible in the required frequency band, including algorithms based on multidimensional inverse Fourier transform, least squares, convex optimization, and frequency focusing. A broadband robust adaptive beamforming algorithm based on microphone arrays (Lin Jingran, Peng Qicong, Shao Huaizong, et al. Broadband robust adaptive beamforming algorithm based on microphone arrays [J]. Journal of Communications, 2006, (12): 132-138.), Abstract: The algorithm combines frequency focusing technology and diagonal loading technology. On this basis, the diagonal loading factor is determined by optimizing the worst-case beam performance, and the frequency focusing technology is used to achieve the frequency invariance of the array, and the approximate analytical expression of the optimal loading factor is obtained. However, this method requires matrix decomposition, which is computationally intensive. Existing methods based on multidimensional inverse Fourier transform, least squares, and convex optimization (Wei Liu, Stephan Weiss. Wideband Beamforming Concepts And Techniques [M]. 2010.) all require iterative approximation, involving operations such as matrix inversion and matrix decomposition, which are computationally intensive and require a trade-off between sidelobe suppression and frequency invariance performance, resulting in poor low-frequency performance.
[0004] It can be seen that the existing frequency-invariant beamforming algorithm still has some shortcomings: 1. The existing frequency-invariant beamforming algorithm generally has a large amount of computation, involving matrix approximation, matrix decomposition and other operations; 2. When the frequency is very low, there will also be a problem of beam widening, so the effect is not ideal in speech signal processing based on microphone arrays. Summary of the invention
[0005] In view of the problems that the existing frequency-invariant beamforming algorithm has a large amount of computation and still becomes wider at very low frequencies, the present invention provides a frequency-invariant microphone array beamforming method. The method can convert broadband beamforming into a narrow beam design of a specific frequency. The method has the advantages of small computational complexity and applicability to any non-zero frequency. It can not only be used for speech processing, but also can be extended to the fields of underwater acoustic processing, communication signal processing, etc.
[0006] The present invention is achieved by at least one of the following technical solutions.
[0007] A frequency-invariant microphone array beamforming method comprises the following steps:
[0008] Step 1, select the angular frequency ω0 as the reference frequency, set a narrowband beam suitable for the angular frequency ω0 according to the specified incoming wave direction, and obtain the weight of the input signal of each array element;
[0009] Step 2: Frame the input signals of each array element of the microphone and convert them into the frequency domain to obtain the spectrum X corresponding to each array element. l,m (ω), where l represents the lth frame, m represents the mth array element, and ω is the angular frequency;
[0010] Step 3: For any angular frequency ω<πf s The signal is taken as the reference element, and the delay τ of each element relative to the reference element is calculated. m (ω), where f s is the sampling frequency;
[0011] Step 4: Convert the spectrum to the reference frequency ω0 according to the delay to obtain a new spectrum X′ l,m (ω0);
[0012] Step 5: Based on the weights calculated in step 1, use the new spectrum X′ obtained in step 4 l,m (ω0) Calculate the output spectrum of the array at the lth frame when the frequency is ω0
[0013] Step 6: Calculate the ratio of the above output amplitude to the reference array element input amplitude as the gain of the array when the frequency is ω;
[0014] Step 7: Multiply the gain by the spectrum X of the 0th element in the lth frame. l,0(ω), and obtain the output spectrum of the lth frame of the array at frequency ω
[0015] Step 8: Use IFFT to convert the output signal from the frequency domain to the time domain, and perform weighted superposition with the previous frame to obtain the output signal of the array.
[0016] Further, setting a narrowband beam suitable for the angular frequency ω0 according to the specified incoming wave direction includes the following steps:
[0017] Step 1.1, sample the angle [-90°, 90°], and the number of sampling points is N;
[0018] Step 1.2: For each angle θ i , calculate the direction vector in
[0019]
[0020] where d m (ω0,θ i ) is the vector d(ω0) at an angle θ i The mth component above, i=0~N-1 is the angle number, m=0~M-1 is the array element number, M is the number of array elements, a is the spacing between microphone array elements, c is the sound speed, and j is the complex unit;
[0021] Step 1.3, given the target direction vector p(ω0) = [p(ω0,θ0), p(ω0,θ1), …, p(ω0,θ N )] T , where p(ω0,θ N ) is the vector p(ω0) at an angle θ N The component on , the superscript T indicates transpose;
[0022] Step 1.4, calculate the covariance matrix B of the direction vector d(ω0) and the cross-correlation vector A between the direction vector d(ω0) and the target direction vector P(ω0):
[0023] B=d(ω0)d T (ω0)
[0024] A=d(ω0)p(ω0)
[0025] where d T (ω0) is the transpose of d(ω0);
[0026] Step 1.5, calculate the optimal weight vector w:
[0027] w=B -1 A
[0028] where \(w = [w_0, w_1, \ldots, w M-1 \) is the weight vector of the array elements, and each component \(w_0, w_1, \ldots, w M-1 \) corresponds to the weight of each microphone array element, and \(M\) is the number of array elements.
[0029] Furthermore, in step 3, calculating the delay of the \(m\)th array element relative to the reference array element includes the following steps:
[0030] Step 3.1: Let \(m = 1\), and the delay \(\tau_0(\omega)=0\) of the reference array element;
[0031] Step 3.2: Calculate the complex correlation \(c l,m-1 (\omega)\) of \(X l,m (\omega)\) and \(X m-1,m (\omega)\):
[0032]
[0033] where \(X l,m (\omega)\) is the spectrum of the \(m\)th array element in the \(l\)th frame, is the conjugate of \(X l,m (\omega)\);
[0034] Step 3.3: Calculate the delay \(\tau m (\omega)\) of the \(m\)th array element relative to the reference array element by the method of accumulation:
[0035]
[0036] where \(\arg(c k-1,k (\omega))\) represents the principal value of the phase of the complex correlation \(c k-1,k (\omega)\), and \(\tau k-1 (\omega)\) is the delay of the \((k - 1)\)th array element, \(k\in(1\sim m)\);
[0037] Step 3.4: Let \(m = m + 1\). If \(m\lt M\), then go to step 3.2; otherwise, end. Here, \(M\) is the number of array elements.
[0038] Furthermore, the spectrum \(X' l,m (\omega_0)\) is obtained by the following formula:
[0039]
[0040] where \(\tau m (\omega)\) is the delay of the \(m\)th array element relative to the reference array element, and \(j\) is the imaginary unit.
[0041] Furthermore, the output spectrum of the \(l\)th frame of the array is:
[0042]
[0043] Among them, X′ l,m (ω0) is the frequency spectrum, ω0 is the angular frequency, w m is the weight of the input signal of the array element, and M is the number of array elements.
[0044] Furthermore, in step 6, the gain g l (ω) is:
[0045]
[0046] Where ω0 is the angular frequency, is the output spectrum of the array frame l, X l,0 (ω) is the spectrum of the 1st frame of the 0th array element.
[0047] Furthermore, in step 6, a nonlinear function is used to calculate the gain to further reduce the sidelobe level g of the beam. l (ω):
[0048]
[0049] Where ω is the angular frequency and f(x) is a nonlinear function.
[0050] Furthermore, the nonlinear function f(x) includes a step function and a sigmoid function.
[0051] Further, in step 6, according to the given direction pointing width θ B The sigmoid function is used to calculate the gain:
[0052] Step 6.1, calculate the gain g(ω0,θ) of the array for different incoming wave directions according to the weight w in step 1:
[0053] g(ω0,θ)=w T d(ω0,θ)
[0054] Where θ is the angle, and d(ω0,θ) is the direction vector of the array when the frequency is ω0;
[0055] Step 6.2, calculate the normalized gain g′(ω0,θ):
[0056]
[0057] where g max (ω0) and g min (ω0) are the maximum and minimum values of g(ω0,θ) respectively;
[0058] Step 6.3: Calculate the threshold g c (ω0):
[0059] g c(ω0)=g′(ω0,θ0±θ B / 2)
[0060] Where θ0 is the central angle of the specified direction;
[0061] Step 6.4: Calculate the linear gain g of the array at frequency ω l (ω):
[0062]
[0063] Step 6.5: Calculate the nonlinear gain g′ of the array at frequency ω using the following formula: l (ω):
[0064]
[0065] where g′ l (ω) is the nonlinear gain of the array, α is an empirical parameter, g max (ω0) and g min (ω0) are the maximum and minimum values of g(ω0,θ), respectively. c (ω0) is the threshold.
[0066] Further, step 8 includes the following steps:
[0067] First, the output spectrum of the array's first frame Use IFFT to convert to time domain signal x l [n]:
[0068]
[0069] Where n is the discrete time sequence number;
[0070] The output signal of the array is calculated using the following formula:
[0071]
[0072] in is the output signal of the array, x l-1 [n] and x l [n] are the time domain signals after IFFT of the l-1th and lth frames respectively, L is the frame length, v is the frame shift, and n is the discrete time sequence number.
[0073] Compared with the prior art, the present invention has the following advantages:
[0074] 1. A frequency-invariant microphone array beamforming method of the present invention can keep beam parameters consistent with a reference frequency even when the frequency is extremely low.
[0075] 2. The present invention is simple to calculate and has a low amount of computation.
[0076] 3. The method of the present invention can further reduce the sidelobe level through nonlinear gain.
[0077] 4. The present invention utilizes the characteristic that the human ear is insensitive to phase and adopts a method of adjusting the amplitude of the reference microphone to well maintain the continuity of the phase of the voice signal. BRIEF DESCRIPTION OF THE DRAWINGS
[0078] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention.
[0079] Figure 1 A flow chart of a frequency-invariant microphone array beamforming method provided by an embodiment;
[0080] Figure 2 is the beam pattern of Example 2;
[0081] Figure 3 This is the beam pattern of Example 3. DETAILED DESCRIPTION
[0082] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0083] Example 1
[0084] A frequency-invariant microphone array beamforming method of this embodiment includes the following steps:
[0085] Step 1: Select the angular frequency ω0, set a narrowband beam suitable for the angular frequency ω0 according to the specified wave direction, and obtain the weight w of each array element input signal m ;
[0086] Step 2: Frame the input signal of each array element and convert it to the frequency domain to obtain the spectrum X corresponding to each array element. l,m (ω), where m=0~M-1 is the array element number, M is the number of array elements, l is the frame number, and ω is the angular frequency;
[0087] Step 3: For any angular frequency ω<πf s The signal is taken as the reference element, and the delay τ of each element relative to the reference element is calculated. m(ω), where f s is the sampling frequency;
[0088] Step 4: X l,m (ω) is converted to the reference frequency according to the delay to obtain the spectrum X′ l,m (ω0);
[0089] Step 5: Based on the weights calculated in step 1, use X′ l,m (ω0) Calculate the output spectrum of the array at the lth frame when the frequency is ω0
[0090] Step 6: Calculate the ratio of the above output amplitude to the reference array element input amplitude as the gain of the array when the frequency is ω;
[0091] Step 7: Multiply the gain by the spectrum x of the 0th element in the lth frame. l,0 (ω), and obtain the output spectrum of the lth frame of the array at frequency ω
[0092] Step 8: Use IFFT to convert the output signal from the frequency domain to the time domain, and perform weighted superposition with the previous frame to obtain the output signal of the array.
[0093] Example 2
[0094] In this embodiment, the microphone array is a linear array of 4 elements, and the sampling frequency of the speech signal is f s =8kHz, the microphone array element spacing a=4cm, and the reference array element is the leftmost array element. The flowchart and beam direction diagram of the embodiment are as follows Figure 1 , Figure 2 As shown, the following steps are included:
[0095] Step 1: Select an angular frequency ω0, design a narrowband beam suitable for ω0 according to the specified wave direction, and obtain the weight w of the input signal of each array element m .
[0096] In the above embodiment, ω0=πc / a is selected, the incoming wave direction is specified as [-20°, 20°], and then the following steps are used to design a narrowband beam suitable for ω0:
[0097] Step 1.1: Sample the angle [-90°, 90°] at N points. In the above embodiment, N=181.
[0098] Step 1.2: For each angle θ i , calculate the direction vector in
[0099]
[0100] dm (ω0,θ i ) is the vector d(ω0) at an angle θ i For the mth component, i=0~N-1 is the angle number, m=0~M-1 is the array element number, M is the number of array elements, a is the spacing between microphone array elements, c is the sound speed, and j is a complex unit.
[0101] Step 1.3, given the target direction vector p(ω0) = [p(ω0,θ0), p(ω0,θ1), …, p(ω0,θ N )] T , where p(ω0,θ N ) is the vector p(ω0) at an angle θ N The superscript T indicates the transpose.
[0102] In the above embodiment, it is specified that p(ω0,θ i )=1, other angles p(ω0,θ i )=0.
[0103] Step 1.4. Calculate the covariance matrix B of d(ω0) and the cross-correlation vector A between d(ω0) and p(ω0)
[0104] B=d(ω0)d T (ω0)
[0105] A=d(ω0)p(ω0)
[0106] where d T (ω0) is the transpose of d(ω0).
[0107] Step 1.5: Calculate the optimal weight vector
[0108] w=B -1 A
[0109] where w=[w0,w1,…,w M-1 ] is the weight vector of the array element, and each component w0,w1,…,w M-1 They correspond to the weights of each array element of the microphone, and M is the number of array elements.
[0110] Step 2: Frame the input signal of each array element and convert it into the frequency domain to obtain the spectrum X of each frame corresponding to each array element. l,m (ω), where m=0~M-1 is the array element number, M is the number of array elements, l is the frame number, and ω is the angular frequency.
[0111] In the above embodiment, the input signal is divided into frames with a frame length of 16 ms and a frame shift of 10 ms, and FFT is performed to obtain the spectrum X of each frame. l,m (ω).
[0112] Step 3. For a signal with any frequency ω < πf s take the 0th element as the reference element, and calculate the delay τ m (ω) of each element relative to the reference element.
[0113] Step 3.1. Let m = 1, and the delay τ0(ω) of the reference element is 0.
[0114] Step 3.2. Calculate the complex correlation of X l,m-1 (ω) and X l,m (ω)
[0115]
[0116] where X l,m (ω) is the spectrum of the mth element in the lth frame, is the conjugate of X l,m (ω).
[0117] Step 3.3. Calculate the delay of the mth element relative to the reference element
[0118]
[0119] where arg(c k-1,k (ω)) represents the principal value of the phase of the complex correlation c k-1,k (ω), τ k-1 (ω) is the delay of the (k - 1)th element, and k ranges from 1 to m.
[0120] Step 3.4. m = m + 1. If m < M, go to Step 3.2; otherwise, end. M is the number of elements.
[0121] Step 4. Use the following formula to convert X l,m (ω) to the reference frequency according to the delay, and obtain the spectrum X′ l,m (ω0).
[0122]
[0123] Step 5. According to the weights calculated in Step 1, use X′ l,m (ω0) to calculate the output of the array at the lth frame when the frequency is ω0
[0124]
[0125] Step 6. Calculate the ratio of the above output amplitude to the input amplitude of the reference element as the gain g l (ω) of the array at the frequency ω:
[0126]
[0127] Step 7: Multiply the gain by the spectrum X of the 0th element in the lth frame. l,0 (ω), and get the output of the array when the frequency is ω:
[0128]
[0129] Step 8: Use IFFT to convert the output signal from the frequency domain to the time domain, and perform weighted superposition with the previous frame to obtain the output signal of the array.
[0130] In the above embodiment, first Use IFFT to convert to time domain signal:
[0131]
[0132] The output signal of the array is then calculated using the following formula:
[0133]
[0134] in is the output signal of the array, x l-1 [n] and x l [n] are the time domain signals after IFFT of the l-1th and lth frames respectively, L is the frame length, v is the frame shift, and n is the discrete time sequence number.
[0135] Example 3
[0136] Based on the above embodiment, a frequency-invariant microphone array beamforming method of this embodiment includes the following steps:
[0137] Step 1, select the angular frequency ω0 as the reference frequency, set a narrowband beam suitable for the angular frequency ω0 according to the specified incoming wave direction, and obtain the weight of the input signal of each array element;
[0138] Step 2: Frame the input signals of each array element of the microphone and convert them into the frequency domain to obtain the spectrum X corresponding to each array element. l,m (ω), where l represents the lth frame, m represents the mth array element, and ω is the angular frequency;
[0139] Step 3: For any angular frequency ω<πf s The signal is taken as the reference element, and the delay τ of each element relative to the reference element is calculated. m (ω), where f s is the sampling frequency;
[0140] Step 4: Convert the spectrum to the reference frequency ω0 according to the delay to obtain a new spectrum X′ l,m (ω0);
[0141] Step 5: Based on the weights calculated in step 1, use the new spectrum X′ obtained in step 4 l,m (ω0) Calculate the output spectrum of the array at the lth frame when the frequency is ω0
[0142] Step 6: Calculate the ratio of the above output amplitude to the reference array element input amplitude as the gain of the array when the frequency is ω;
[0143] Step 7: Multiply the gain by the spectrum X of the 0th element in the lth frame. l,0 (ω), and obtain the output spectrum of the first frame of the array at frequency ω
[0144] Step 8: Use IFFT to convert the output signal from the frequency domain to the time domain, and perform weighted superposition with the previous frame to obtain the output signal of the array.
[0145] In step 6, a nonlinear function is used to further reduce the sidelobe level of the beam:
[0146]
[0147] Where f(x) is a nonlinear function, including step function and S-type function.
[0148] In the above embodiment, f(x) uses a sigmoid function, and the gain is calculated using the following formula:
[0149]
[0150] Its beam pattern is as follows Figure 3 shown.
[0151] As a preferred embodiment, in step 6, according to the given direction pointing width θ B The sigmoid function is used to calculate the gain:
[0152] Step 6.1: Calculate the gain of the array for different wave directions according to the weights in step 1.
[0153] g(ω0,θ)=w T d(ω0,θ)
[0154] Step 6.2: Calculate the normalized gain
[0155]
[0156] where g max (ω0) and g min (ω0) are the maximum and minimum values of g(ω0,θ) respectively.
[0157] Step 6.3: Calculate the threshold
[0158] g c (ω0)=g′(ω0,θ0±θ B / 2)
[0159] Where θ0 is the center angle of the specified direction.
[0160] Step 6.4: Calculate the linear gain of the array at frequency ω
[0161]
[0162] Step 6.5: Calculate the nonlinear gain of the array at frequency ω using the following formula:
[0163]
[0164] where g′ l (ω) is the nonlinear gain of the array.
[0165] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well.
Claims
1. A frequency-invariant microphone array beamforming method, characterized in that: It includes the following steps: Step 1: Select the angular frequency ω0 as the reference frequency, set a narrowband beam applicable to the angular frequency ω0 according to the specified incoming wave direction, and obtain the weights of the input signals of each array element; Step 2: Frame the input signals of each array element of the microphone and convert them into the frequency domain to obtain the spectrum X corresponding to each array element. l,m (ω), where l represents the lth frame, m represents the mth array element, and ω is the angular frequency; Step 3: For any angular frequency ω<πfs s The signal is taken as the reference element, and the delay τ of each element relative to the reference element is calculated. m (ω), where f s is the sampling frequency; Step 4: Convert the spectrum to the reference frequency ω0 according to the delay to obtain a new spectrum X′ l,m (ω0); Step 5: Based on the weights calculated in step 1, use the new spectrum X′ obtained in step 4 l,m (ω0) Calculate the output spectrum of the array at the lth frame when the frequency is ω0 Step 6: Calculate the ratio of the above output amplitude to the input amplitude of the reference array element as the gain of the array at the frequency ω; Step 7: Multiply the gain by the spectrum X of the 0th element in the lth frame. l,0 (ω), and obtain the output spectrum of the first frame of the array at frequency ω Step 8: Use IFFT to convert the output signal from the frequency domain to the time domain, and perform weighted superposition with the previous frame to obtain the output signal of the array.
2. The frequency-invariant microphone array beamforming method according to claim 1, characterized in that: Setting a narrowband beam applicable to the angular frequency ω0 according to the specified incoming wave direction includes the following steps: Step 1.1: Sample the angle range [-90°, 90°], and the number of sampling points is N; Step 1.2: For each angle θ i , calculate the direction vector in where d m (ω0,θ i ) is the vector d(ω0) at an angle θ i The mth component above, i=0~N-1 is the angle number, m=0~M-1 is the array element number, M is the number of array elements, a is the spacing between microphone array elements, c is the sound speed, and j is the complex unit; Step 1.3, given the target direction vector p(ω0) = [p(ω0,θ0), p(ω0,θ1), …, p(ω0,θ N )] T , where p(ω0,θ N ) is the vector p(ω0) at an angle θ N The component on , the superscript T indicates transpose; Step 1.4: Calculate the covariance matrix B of the direction vector d(ω0) and the cross-correlation vector A between the direction vector d(ω0) and the target direction vector p(ω0): B=d(ω0)d T (ω0) A = d(ω0)p(ω0) where d T (ω0) is the transpose of d(ω0); Step 1.5: Calculate the optimal weight vector w; w=B -1 A where w=[w0,w1,…,w M-1 ] is the weight vector of the array element, and each component w0,w1,…,w M-1 They correspond to the weights of each array element of the microphone, and M is the number of array elements.
3. The frequency-invariant microphone array beamforming method according to claim 1, characterized in that: In Step 3, calculating the delay of the m-th array element relative to the reference array element includes the following steps: Step 3.1: Let m = 1, and the delay τ0(ω) of the reference array element is 0; Step 3.2, calculate X l,m-1 (ω) and X l,m The complex correlation c of (ω) m-1,m (ω): Where X l,m (ω) is the spectrum of the mth element in the lth frame, For X l,m (ω) conjugation; Step 3.3: Calculate the delay τ of the mth array element relative to the reference array element by using the accumulation method. m (ω): where arg(c k-1,k (ω)) represents the complex correlation c k-1,k The phase principal value of (ω), τ k-1 (ω) is the delay of the k-1th array element, k∈(1~m); Step 3.4: m = m + 1. If m < M, go to Step 3.2, otherwise end. M is the number of array elements.
4. The frequency-invariant microphone array beamforming method according to claim 1, characterized in that: The spectrum X′ is obtained by the following formula: l,m (ω0); Among them, τ m (ω) is the delay of the mth array element relative to the reference array element, and j is a complex unit.
5. The frequency-invariant microphone array beamforming method according to claim 1, characterized in that: The output spectrum of the array's first frame for: Among them, X′ l,m (ω0) is the frequency spectrum, ω0 is the angular frequency, w m is the weight of the input signal of the array element, and M is the number of array elements.
6. The frequency-invariant microphone array beamforming method according to claim 1, characterized in that: In step 6, the gain g l (ω) is: Where ω0 is the angular frequency, is the output spectrum of the array frame l, X l,0 (ω) is the spectrum of the 1st frame of the 0th array element.
7. The frequency-invariant microphone array beamforming method according to claim 6, characterized in that: In step 6, a nonlinear function is used to calculate the gain to further reduce the sidelobe level f of the beam. l (ω): Where ω is the angular frequency and f(x) is a non-linear function.
8. The frequency-invariant microphone array beamforming method according to claim 7, characterized in that: The non-linear function f(x) includes a step function and an S-shaped function.
9. The frequency-invariant microphone array beamforming method according to claim 1, characterized in that: In step 6, the width θ is pointed according to the given direction B The sigmoid function is used to calculate the gain: Step 6.1: Calculate the gain g(ω0, θ) of the array for different incoming wave directions according to the weight w in Step 1: g(ω0,θ)=w T d(ω0,θ) Where θ is the angle and d(ω0, θ) is the direction vector of the array at the frequency ω0; Step 6.2: Calculate the normalized gain g′(ω0, θ); where g max (ω0) and g min (ω0) are the maximum and minimum values of g(ω0,θ) respectively; Step 6.3: Calculate the threshold g c (ω0): g c (ω0)=g′(ω0,θ0±θ B / 2) Where θ0 is the central angle of the specified direction; Step 6.4: Calculate the linear gain g of the array at frequency ω l (ω): Step 6.5: Calculate the nonlinear gain g′ of the array at frequency ω using the following formula: l (ω): where g′ l (ω) is the nonlinear gain of the array, α is an empirical parameter, g max (ω0) and g min (ω0) are the maximum and minimum values of g(ω0,θ), respectively. c (ω0) is the threshold.
10. The frequency-invariant microphone array beamforming method according to claim 1, characterized in that: Step 8 includes the following steps: First, the output spectrum of the array's first frame Use IFFT to convert to time domain signal x l [n]: Where n is the discrete time sequence number; The output signal of the array is calculated using the following formula: in is the output signal of the array, x l-1 [n] and x l [n] are the time domain signals after IFFT of the l-1th and lth frames respectively, L is the frame length, v is the frame shift, and n is the discrete time sequence number.
Citation Information
Patent Citations
Design method and device of frequency-invariant beam forming device
CN107577872A
Microphone array voice enhancing method based on statistic model
CN107993670A
Optimal modal beamformer for sensor arrays
GB0906269D0
Optimal modal beamformer for sensor arrays
US20120093344A1
Method and apparatus for generating steering matrices for beamforming
US8345797B1
