Robust differential beam forming method based on Kronecker product

Through the Cronek product, the uniform linear array is decomposed into virtual sub-arrays. The differential and conventional beamforming methods are used to solve the problem of difficult trade-off between space gain and white noise gain in the traditional method, and the efficient beamforming of small-size arrays is achieved. Enhanced robustness and flexibility.

CN118944728AInactive Publication Date: 2024-11-12QINGDAO GUOSHU INFORMATION TECH CO LTD +1

Patent Information

Application Number
CN202411433010.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-15
Publication Date
2024-11-12
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

Traditional beamforming methods are difficult to achieve high space gain and resolution on small-sized underwater unmanned submarine platforms, and it is difficult to flexibly design beam patterns and control white noise gain, especially in low-frequency or high-order differential algorithms.

Method used

Through the Cronek product, the uniform linear array is decomposed into virtual small-aperture sub-arrays and virtual large-aperture sub-arrays. The differential beamforming method and conventional beamforming method are used to design the corresponding beamformer to improve overall performance and reduce operational complexity. Spend.

Benefits of technology

The trade-off between improving the spatial gain and white noise gain of the array under finite aperture is achieved, enhancing the robustness and flexibility of the array and reducing the computational complexity.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118944728A_ABST
    Figure CN118944728A_ABST
Patent Text Reader

Abstract

The invention discloses a robust differential beam forming method based on a Kronecker product, and belongs to the technical field of signal processing. The method comprises the following steps: firstly, decomposing a uniform linear array into two virtual sub-arrays by using a Kronecker product, namely a virtual small-aperture sub-array and a virtual large-aperture sub-array; for the virtual small-aperture sub-array, a differential beam forming method is used to realize the overall target gain requirement of the array; and aiming at the virtual large-aperture subarray, the white noise gain is improved by utilizing a conventional beam forming method. And a corresponding beam former is flexibly designed on the basis of the virtual small-aperture sub-array and the virtual large-aperture sub-array. According to the method, the uniform linear array is decomposed into the two virtual sub-arrays by using the Kronecker product, so that a corresponding beam former can be flexibly designed on the basis of each sub-array, the overall performance of the array is improved, and the operation complexity is reduced.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of signal processing, and in particular relates to a robust differential beamforming method based on Kronecker product. Background Art

[0002] In the early 1990s, with the development of microelectronics technology, intelligent AI technology, virtual reality technology, etc., underwater unmanned underwater vehicle (UUV) platforms emerged in large numbers. With their small size, low cost, and flexible and diverse payloads, they have become an important combat force in the future ocean battlefield. However, the development of sonar systems carried by small underwater platforms such as UUVs is relatively lagging. Due to the limited size of the UUV platform, it is difficult to support a large-aperture array, resulting in limited spatial gain and resolution obtained by traditional beamforming methods under the limitation of the array aperture. Therefore, how to use signal processing methods to obtain high spatial gain of the sonar array under limited aperture size and realize the detection of long-range targets is a key problem that needs to be solved urgently. Based on this, small-size, high-gain signal processing methods have become the focus of people's attention.

[0003] The traditional differential beamforming method uses the sound pressure difference between two adjacent sensors to achieve beamforming. When the sensor spacing is much smaller than half the wavelength of the minimum wavelength of the signal, the differential of the sound pressure field is approximated by performing a finite-order spatial difference to obtain the sound pressure gradient information or the higher-order sound pressure gradient of the sound field. Therefore, the maximum directional gain can be achieved with a given number of sensors. However, this method lacks flexibility in designing the required beam pattern, and more importantly, it is difficult to control the white noise amplification, especially in low-frequency or high-order differential algorithms, which makes it difficult to apply in practice.

[0004] The prior art also proposes a differential beamforming method based on zero-point constraints in the short-time Fourier transform domain. This method transforms the cascade structure used in the traditional differential beamforming method into a linear structure according to the zero-point position relationship of the ideal beam pattern and the expected constraints. The beam pattern can be designed very flexibly, and more sensors (more than the order of the differential beamforming algorithm plus 1) can be used to improve the white noise gain. That is, based on the zero-point constraint method, a method for maximizing the white noise gain is proposed. Additional redundancy is obtained by increasing the number of sensors to improve the white noise gain. However, as the number of sensors increases, although the low-frequency white noise gain is significantly improved, it also brings about a decrease in the directional gain in the mid- and low-frequency bands. Therefore, the flexible design of the differential beamforming method, which enables it to maximize the white noise gain under the condition of meeting the target gain, has become a key technology for the differential beamforming method to be applied in practice. Summary of the invention

[0005] Based on the above technical problems, the present invention proposes a robust differential beamforming method based on the Kronecker product. The method decomposes the uniform linear array into two virtual sub-arrays by utilizing the Kronecker product, so that the corresponding beamformer can be flexibly designed based on each sub-array, thereby improving the overall performance of the array and reducing the computational complexity.

[0006] The technical solution adopted by the present invention is: A robust differential beamforming method based on Kronecker product comprises the following steps: (1) Using the Kronecker product, the uniform linear array is decomposed into two virtual sub-arrays, namely a virtual small-aperture sub-array and a virtual large-aperture sub-array. (2) For the virtual small aperture subarray, the differential beamforming method is used to achieve the overall target gain requirement of the array; (3) For the virtual large-aperture subarray, the conventional beamforming method is used to improve the white noise gain and complete the robust differential beamforming method.

[0007] Furthermore, a corresponding beamformer is designed based on the virtual small-aperture subarray and the virtual large-aperture subarray.

[0008] The above step (1) includes the following steps: The steering vector of the uniform linear array composed of M sensors is decomposed into the steering vector of two virtual sub-arrays through Kronecker integral, thus reconstructing into two virtual sub-arrays; the number of sensors in the first virtual sub-array is , the number of sensors in the second virtual subarray is , ; The uniform linear array composed of M sensors and the first virtual sub-array have an array element spacing of , and the element spacing of the second virtual subarray is .

[0009] Further, the steps of decomposing the steering vector of the uniform linear array by Kronecker product are as follows: The steering vector of the uniform linear array is , where the steering vector of the uniform linear array can be expressed as the Kronecker product of the steering vectors of the two virtual sub-arrays: (8) (9) (10) According to formula (8), the global filter can be rewritten as the Kronecker product form of the filters corresponding to the two virtual sub-arrays: (11) On this basis, two virtual sub-array beamformers are designed: , , and finally the length is Linear filter.

[0010] Furthermore, in the case where the array sensor cannot be decomposed into the integer product form of two virtual sub-arrays, the number and layout of the first virtual sub-array are as follows: (12) in, , , are all non-negative integers; when When is 0, the steering vector of the first virtual subarray is as shown in formula (9); and when When it is not 0, the steering vector of the first virtual subarray is: (13) in, The length is p+1, which is the first virtual subarray The number is p+1; in this case, .

[0011] Furthermore, the compressed sensing matrix is ​​introduced to solve the problem of data dimension mismatch; (14) in, is a random Gaussian matrix composed of independent and incoherent Gaussian random variables.

[0012] Furthermore, the uniform linear array beam pattern is expressed as the product of two sub-array beam patterns, which allows for more flexible and convenient design of beam patterns or beamformers to meet performance requirements: (15) The white noise gain is simplified into the product of the white noise gains of two virtual sub-arrays: (16) Substitute the beam pattern product relationship into formula (7): (7) Obtain the directivity gain evaluation index: (17).

[0013] The above step (2) includes the following steps: According to the passive sonar equation, the target directivity gain is obtained: ,in is the number of sensors; according to the target directivity gain, the design order N of the differential beamformer in the ideal case is obtained ( ); The directional gain of the sensor array is related to the sonar range, ocean environment, target sound source level and receiving depth. The relationship is expressed by the following passive sonar equation: (18) Among them, SL is the sound source level, unit is dB; TL is the propagation loss, which consists of two parts: geometric expansion loss and dielectric absorption loss. Among them, geometric expansion loss refers to the geometric effect when the sound signal is regularly weakened when expanding outward from the sound source, and dielectric absorption loss includes the effects of absorption, scattering and sound energy leakage out of the sound channel, so there is an empirical formula: (19) Where: K is a parameter used to describe the type of geometric expansion. When K=2, it is spherical expansion, and K=1 is cylindrical expansion. In the far field distance, it is assumed that the sound wave mainly propagates spherically, and K=2 is taken; r is the propagation distance, in meters; f is the propagation frequency, in kHz; NL is the ambient noise level; DI is the directivity gain, which represents the parameter of the sensor or array's ability to suppress interference from non-target directions. If a non-directional sensor is used, the receiving directivity index DI is ignored; DT is the detection threshold. The target can be detected when according to Relationship, to obtain the design order N of the differential beamformer under ideal conditions; to design an ideal differential beam pattern according to the obtained order; Design of ideal differential beam pattern at order N: (20) in, is a real coefficient. Different beam patterns of the N-order differential beamformer are determined according to different coefficient values. When the main lobe of the beam pattern points to the target direction, , the beam pattern is equal to 1; therefore, the coefficient satisfy: (twenty one) For the design of the virtual small-aperture differential beamformer, under the given minimum directivity gain and maximum white noise gain criteria, the final beam pattern is made to approach the ideal beam pattern; based on this, the corresponding beamformer is equivalent to solving the following problem: (twenty two) in, is the differential beamformer of the virtual small aperture subarray, is the actual beam pattern formed, and The ideal beam pattern is obtained by using the least squares method to make the error norm between the actual beam pattern and the ideal beam pattern less than a specific value. , and then maximize the white noise gain under the distortion-free constraint and given target gain.

[0014] Furthermore, for formula (22), the alternating direction multiplier method or the CVX toolbox of Matlab is used to solve it.

[0015] The above step (3) includes the following steps: The conventional beamforming method realizes beamforming by compensating the phase difference caused by the distance between array elements when the signal is received by each array element; therefore, the weight vector of the conventional beamformer is set as: (twenty three) The size of its weight is related to the direction of the expected signal, so the array has different gains for signals from different directions in space, thus forming spatial filtering; The array output signal is: (twenty four) Substituting formula (23) into formula (6): W [ h ( w ) ] = | h H ( w ) d ( w , θ s ) | 2 h H ( w ) h ( w ) (6) In the formula, is the direction of the target desired signal, and the superscript H is the conjugate transpose operator; Find: (25).

[0016] The beneficial technical effects of the present invention are: The present invention uses the Kronecker product to decompose and reconstruct the steering vector of the uniform linear array into two virtual sub-arrays, so that the two virtual sub-arrays are used to achieve the desired performance indicators respectively, thereby improving the overall performance. In view of the high gain robustness requirement, a differential beamforming method is designed for the virtual small aperture sub-array to achieve the desired target gain. A conventional beamforming method is designed for the virtual large aperture sub-array to improve the white noise gain. In this way, a good compromise of the two target performances is achieved. In addition, the present invention has the advantages of flexible and convenient design and greatly reduced computational complexity. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 Schematic diagram of the process of the robust differential beamforming method based on Kronecker product of the present invention; Figure 2 Schematic diagram of decomposing a uniform linear array into two virtual sub-arrays; Figure 3is the structural block diagram of the beamformer; Figure 4 It is a schematic diagram of decomposing a uniform linear array into two virtual sub-arrays in general; Figure 5 A flowchart of a virtual small aperture subarray according to the present invention; Figure 6 DF diagram of the robust differential beamforming method under virtual small aperture subarray; Figure 7 It is the WNG diagram of the robust differential beamforming method under the virtual small aperture subarray; Figure 8 It is the DF diagram of the conventional beamforming method under the virtual large aperture subarray; Fig. 9 It is the WNG diagram of the conventional beamforming method under the virtual large aperture subarray; Fig.10 DF diagram of the method proposed in the present invention (kro-r) and the comparative method (kro-p); Fig.11 WNG diagrams of the method proposed in the present invention (kro-r) and the comparative method (kro-p). DETAILED DESCRIPTION

[0018] The Kronecker product-based beamforming method has attracted people's attention because of its flexibility and convenience in designing beam patterns. However, how to better combine the virtual subarray after the Kronecker product decomposition with the beamforming method to achieve the goal of meeting the target gain and improve the robustness of the overall array method, so as to better apply it in practice, still needs to be solved.

[0019] Based on this, the present invention proposes a robust differential beamforming method based on Kronecker product. The method first uses Kronecker product to decompose the uniform linear array into two virtual sub-arrays, namely virtual small aperture sub-array and virtual large aperture sub-array; for the virtual small aperture sub-array, the differential beamforming method is used to achieve the overall target gain requirement of the array; for the virtual large aperture sub-array, the conventional beamforming method is used to improve the white noise gain. And the corresponding beamformer is flexibly designed based on the virtual small aperture sub-array and the virtual large aperture sub-array, so as to improve the overall performance of the array and reduce the computational complexity.

[0020] In the present invention, since the number of virtual small aperture subarrays is obtained according to the target gain, the number of sensors in the global array may not be decomposed into two integer product forms. Therefore, the Kronecker integral decomposition is considered in general, but this often leads to data dimension mismatch problems. Therefore, the compressed sensing matrix is ​​introduced to transform the final beamformer, so that it can be better applied to the beamforming method.

[0021] In order to meet the need to improve the white noise gain of the array under the target gain, the corresponding design is carried out according to the characteristics of each subarray to meet the overall demand. For virtual small-aperture subarrays, the differential beamforming method can be used to achieve the overall target gain requirement of the array. Compared with the direct use of the zero-point constrained differential beamforming method, this method uses the least squares criterion to approximate the ideal beam pattern after the target gain and distortion-free constraints are given, and maximizes the white noise gain, thereby improving the robustness of the array. Compared with the zero-point constrained differential beamforming method, this method is more flexible, avoids strict spatial response constraints, and increases the degree of freedom. For virtual large-aperture subarrays, the array element spacing no longer meets the requirements of the differential method, so the conventional beamforming method can be used to increase the white noise gain again, which is better applied in practice. This method solves the problem that the directivity gain and white noise gain cannot be well compromised.

[0022] In order to achieve the above object, a flow chart of a robust differential beamforming method based on Kronecker product is designed in the present invention, as shown in FIG. Figure 1 shown.

[0023] Before introducing the specific method, we first briefly describe the beamforming method: In the far-field signal model, the sound source signal In an anechoic underwater channel propagation, impacting at a distance of A uniform linear array of M omnidirectional sensors ( The uniform linear array model is as follows: Figure 2 The Global ULA is shown in Figure 1.

[0024] Then the steering vector of the array is expressed as: (1) Where superscript T is the transposition factor; is the angular frequency, is the frequency; Represents the imaginary part of a complex number; c is the propagation speed of the signal in the medium (usually c=1500m / s in water and 340m / s in air).

[0025] Then the frequency domain output signal of these M sensors can be expressed as: (2) represents the additive noise at the M hydrophones. Beamforming is accomplished by applying a filter of length M From the signal to be processed Get the expected signal ,like Figure 3 As shown, that is: (3) in, represents the conjugation operator, the superscript H stands for the conjugate transpose operator. When designing a beamformer, constraints are usually required to avoid or reduce distortion of the desired signal, namely: (4) Next, we will introduce the three main evaluation indicators for measuring the beamformer, namely beam pattern, white noise gain (White Noise Gain, WNG) and directivity factor (Directivity Factor, DF).

[0026] (1) The beam pattern reflects the beamformer's response to a specific angle. The sensitivity of the starting plane wave, the corresponding beam pattern can be defined as: B [ h ( w ) , θ ] = h H ( w ) d ( w , ϑ ) (5) (2) White noise gain is used to evaluate the sensitivity of the beamformer to the sensor’s own defects (e.g., noise from its own hardware components). The white noise gain evaluation index is: W [ h ( w ) ] = | h H ( w ) d ( w , θ s ) | 2 h H ( w ) h ( w ) (6) in, is the direction of the target desired signal, and the superscript H is the conjugate transpose operator.

[0027] (3) Directivity gain is used to indicate the ability of the beamformer to suppress spatial noise from directions other than the observation direction. It is defined as the ratio between the directivity pattern in the target direction and the average directivity pattern in the entire space. The directivity gain evaluation indicators are: (7) The following is a Kronecker product-based robust differential beamforming method, or the specific design of a Kronecker product-based robust differential beamformer, which is divided into the following three steps.

[0028] (1) The Kronecker integral decomposition is used to decompose the global uniform linear array into two virtual sub-arrays.

[0029] First, The steering vector of the uniform linear array composed of sensors is decomposed into the steering vector of two virtual sub-arrays through Kronecker integral, thus reconstructing into two sub-arrays. The beamformer is redesigned based on each sub-array to meet the final required indicators. Among them, the number of sensors in the first sub-array is , the number of sensors in the second subarray is , The uniform linear array composed of M sensors and the array element spacing of the first sub-array are both , and the element spacing of the second subarray is The sensor uniform linear array layout and virtual sub-array layout are as follows Figure 2 shown.

[0030] The steps involved in decomposing the steering vector using Kronecker are: The steering vector for the array is: , where the global steering vector can be expressed as the Kronecker product of the steering vectors of the two sub-arrays: (8) (9) (10) According to the steering vector, that is, formula (8), the global filter can be rewritten as the Kronecker form of the filter corresponding to the two virtual sub-arrays: (11) On this basis, the beamformers of two virtual sub-arrays are designed as follows: , The beamformers of the two sub-arrays are designed and optimized separately according to the target requirements. Finally, the length is Linear filter.

[0031] The above method requires that the number of sensors in the global array is the integer product of the two sub-arrays, but due to The specific value of is determined based on the directivity gain value obtained later. Often, the array sensor cannot be well decomposed into the integer product form of two sub-arrays. In this case, the number and layout of the first virtual sub-array are as follows (such as Figure 4 shown): (12) in, , , are all non-negative integers. When is 0, the steering vector of the first virtual subarray is as shown in formula (9). When it is not 0, the steering vector of the first virtual subarray is: (13) in, The length is p+1, which is the first virtual subarray The number is p+1. In this case, , then the global array steering vector obtained by the Kronecker expression in mathematical expression is no longer exactly the same as the steering vector obtained by the global array based on the physical structure. This is because under the effect of q, some sensors will share the same position on the generated array, so the overall steering vector obtained will be repeated compared to the steering vector obtained from the original physical array. But in fact, the two are equivalent.

[0032] pass Figure 3 The beamformer block diagram shows that beamforming requires weighted summation of each physical array element to achieve effective reception. In order to make the length obtained in this case The linear filter can finally be directly applied to the physical array, so the compressed sensing matrix can be introduced into it to solve the problem of data dimension mismatch.

[0033] (14) in, is a random Gaussian matrix, which is composed of independent and incoherent Gaussian random variables. It can extract as much information as possible from the original vector.

[0034] According to the Kronecker expression of the steering vector and the global filter, the three performance evaluation indicators of the beamformer are re-expressed.

[0035] Beam pattern: The global array beam pattern can be expressed as the product of two sub-array beam patterns, which means that the beam pattern or beamformer can be designed more flexibly and conveniently to meet performance requirements: (15) This further shows that smaller virtual beamformers can be designed and optimized for each virtual array, which means that the matrix to be inverted is smaller, thus increasing the robustness of the beamformer and reducing the size of the observation data.

[0036] Correspondingly, the white noise gain can also be simplified into the product of the white noise gains of two virtual sub-arrays.

[0037] (16) Since the directional gain measures the ability to suppress isotropic noise fields, the directional gains of the two sub-arrays cannot be simply multiplied. However, the product relationship of the beam pattern can be brought into equation (7), which facilitates the directional gain calculation based on the Kronecker beamformer and obtains the directional gain evaluation index at this time: (17) Obviously, the Kronecker product beamforming method can be used to design any type of beamformer. Therefore, according to the layout characteristics of the virtual subarray and the target performance requirements, the corresponding filter is designed and optimized.

[0038] (2) Compared with other traditional beamforming methods, the differential beamforming method has higher directivity gain under small aperture array. Figure 2 or Figure 4 In the ULA2 in the figure, the spacing between two adjacent array elements is consistent with the spacing between array elements of the global uniform linear array, which is much smaller than half a wavelength. Selecting a virtual small aperture subarray and using the differential beamforming method can better meet the directivity gain requirements of the array. At the same time, redundancy is used as much as possible to increase the white noise gain, thereby improving the robustness of the overall method. Through optimized design, combined with the characteristics of the differential beamforming method and the virtual small aperture subarray, higher directivity gain can be achieved under the small aperture array, and the performance and stability of the system can be improved.

[0039] First, according to the passive sonar equation, the target directivity gain can be obtained when the actual detection distance, environmental noise, and target sound source level are known. For super-directivity methods such as traditional differential beamforming, in the ideal case where the array element spacing is much smaller than half a wavelength, the directivity gain ,in is the number of sensors. Therefore, according to the target gain, the design order N of the differential beamformer in the ideal case can be obtained ( ). On this basis, the corresponding target beam pattern is designed.

[0040] Specific implementation steps: The directional gain of the sensor array is closely related to the sonar range, the specific ocean environment, the target sound source level and the receiving depth. The relationship between the systems can be expressed by the following passive sonar equation: (18) Among them, SL is the sound source level, unit is dB; TL is the propagation loss, which consists of two parts: geometric expansion loss and dielectric absorption loss. Among them, geometric expansion loss refers to the geometric effect when the sound signal is regularly weakened when expanding outward from the sound source, and dielectric absorption loss includes the effects of absorption, scattering and sound energy leakage out of the sound channel, so there is an empirical formula: (19) Where: K is a parameter used to describe the type of geometric expansion. When K=2, it is spherical expansion, and K=1 is cylindrical expansion. In the far field distance, it is assumed that the sound wave mainly propagates spherically, and K=2 is taken; r is the propagation distance, in m; f is the propagation frequency, in kHz. If the propagation distance is small or the frequency is small, the second term can be ignored.

[0041] NL is the ambient noise level. DI is the directional gain, which represents the parameter of the sensor or array's ability to suppress interference from non-target directions. If a non-directional sensor is used, the receiving directivity index DI can be ignored. DT is the detection threshold. It is customary to refer to the signal-to-noise ratio at the processor input required for the device to just complete its intended function as the detection threshold, that is, how many decibels higher the signal level is than the noise level. Usually when The target can be detected.

[0042] according to The relationship can be used to obtain the design order N of the differential beamformer under ideal conditions ( ). Design the ideal differential beam pattern according to the obtained order.

[0043] Design of ideal differential beam pattern at order N: (20) in, is a real coefficient, and different beam patterns of the N-order differential beamformer can be determined according to different coefficient values. When the beam pattern points to the target direction, for example , the beam pattern should be equal to 1. Therefore, the coefficient Should meet: (twenty one) Generally, the purpose of designing a differential beamformer is to find an optimal filter under certain criteria so that the beam pattern it forms is the same as or as close as possible to a given ideal pattern. Therefore, for the design of a virtual small-aperture differential beamformer, the final beam pattern can be made to approach the ideal beam pattern under the given minimum directivity gain and maximum white noise gain criteria. Based on this, the corresponding beamformer can be equivalent to solving the following problem: (twenty two) in, is the differential beamformer of the virtual small subarray, is the actual beam pattern formed, and The ideal beam pattern is obtained by using the least squares criterion to make the error norm between the actual beam pattern and the ideal beam pattern less than a specific value. , It is determined according to actual needs and empirical values. Then, under the condition of distortion-free constraint and given target gain, the white noise gain is maximized to improve the robustness of the method. The formula can be solved by using the alternating direction multiplier method or the CVX toolbox of Matlab.

[0044] Through the above method, the least squares criterion is used to optimize and solve the differential beamformer, which is more flexible in dealing with equality constraints than the zero-point constraint method. This flexibility can avoid the situation where many constraints cannot be satisfied at the same time due to excessive equality constraints. The least squares criterion aims to minimize the error vector norm between the output and the ideal target, so that the output beam is as close to the ideal target beam as possible without requiring strict identity. Therefore, compared with the zero-point constraint method, this method can better adapt to complex scenarios in practical applications and improve the performance and stability of the differential beamformer.

[0045] (3) For the virtual large aperture subarray, that is Figure 2 or Figure 4 ULA1 in the figure. It can be seen that the array element spacing between two adjacent array elements is , at this time, the array element spacing may no longer meet the requirement of being much smaller than half a wavelength. In this case, the differential beamforming method may no longer be applicable. Therefore, when using a small aperture subarray to meet the target directivity gain, for the virtual large aperture subarray, a conventional beamforming method is used to improve the overall white noise gain of the array by taking advantage of the weak noise correlation.

[0046] The conventional beamforming method is to set the weight vector To compensate for the phase difference of the received signal of each array element, realize the in-phase accumulation of the signal, and improve the signal-to-noise ratio, which is conducive to extracting the signal under low signal-to-noise ratio conditions and realizing target signal detection. Because this method has the characteristics of high robustness, it is widely used in real life. Therefore, the conventional beamforming method is selected to make up for the problem of insufficient low-frequency white noise gain caused by the differential method, so as to improve the robustness of the overall method.

[0047] The specific implementation steps of the conventional beamforming method are as follows: The conventional beamforming method realizes beamforming by compensating the phase difference caused by the distance between array elements when the signal is received by each array element. Therefore, the weight vector of the conventional beamformer is set as: (twenty three) It can be seen that the size of the weight is related to the direction of the expected signal, so the array has different gains for signals from different directions in space, thus forming spatial domain filtering.

[0048] The array output signal is: (twenty four) Substituting formula (23) into the white noise gain formula (6) yields: (25) According to formula (25), it can be further proved that the white noise gain of the conventional beamforming method is only related to the number of array elements, but has nothing to do with factors such as frequency. Therefore, in the low-frequency band of the signal, under the action of the Kronecker product, the virtual small subarray uses the robust differential beamforming method to improve the white noise gain by using the remaining redundancy while meeting the target gain, and at the same time uses the least squares criterion to approximate the ideal beam pattern to avoid directly adding the zero-point constrained differential beamforming method to the constraint conditions. The virtual large-aperture subarray uses the high robustness of the conventional beamforming method to improve the white noise gain of the array. Therefore, under the premise of meeting the target gain, the white noise gain of the array is also greatly improved, which can better ensure the detection of the target signal. At the same time, due to the Kronecker decomposition, the scale of the observation data is greatly reduced, which can be better used for real-time data processing and target observation.

[0049] The present invention is a robust differential beamforming method based on Kronecker product. The Kronecker product is used to decompose and reconstruct the steering vector of the uniform linear array into two virtual sub-arrays, so that the two virtual sub-arrays are used to achieve the desired performance indicators respectively, thereby improving the overall performance. In view of the high gain robustness requirement, a differential beamforming method is designed for the virtual small aperture sub-array to achieve the desired target gain. For the virtual large aperture sub-array, a conventional beamforming method is designed to improve the white noise gain. In this way, a good compromise of the two target performances is achieved. Moreover, the design is flexible and convenient, and the computational complexity is greatly reduced.

[0050] Finally, the effectiveness of the method of the present invention is further verified through simulation.

[0051] Simulation 1: Consider a uniform linear array consisting of 16 elements, with an element spacing of The frequency range of interest is 200Hz-4000Hz. Assuming the target directivity gain is 5dB, 7dB, and 8dB, the maximum differential method is the second order according to the virtual small aperture subarray design method. Based on this, the linear array can be divided into two sub-linear arrays. , , then the virtual small aperture ( ) are as follows: Figure 6 , Figure 7 As shown, it is proved that the robust differential beamforming method proposed under the virtual small aperture subarray can be flexibly designed according to the target directivity gain.

[0052] Simulation 2: For a virtual large aperture subarray Using the conventional beamforming method, the virtual sub-array element spacing is The corresponding DF and WNG curves are as follows: Figure 8 , Fig. 9As shown in the figure, it can be seen that the WNG of the conventional beamforming method has nothing to do with factors such as frequency. Get the maximum value .

[0053] Simulation 3: Under the same physical formation and other conditions as simulation 1, the target directivity gain is as low as 6 dB. The robust differential beamforming method based on Kronecker proposed in the present invention is compared with the differential method based on the spatial differential operator of Kronecker product and the zero point constraint method. The corresponding DF and WNG curve comparison diagrams are shown as follows: Fig.10 , Fig.11 As shown, it can be seen that the DF of the two methods are both above 6 dB, among which the DF of the Kronecker-based robust differential beamforming method (kro-r) proposed in the present invention keeps an upward trend with the increase of frequency, and the WNG of this method is much higher than that of the compared method at low frequency, which once again proves that the method of the present invention can improve the robustness of the method while ensuring the target gain.

Claims

1. A robust differential beamforming method based on Kronecker product, characterized in that The following steps are involved: (1) Using the Kronecker product, the uniform linear array is decomposed into two virtual sub-arrays, namely a virtual small-aperture sub-array and a virtual large-aperture sub-array. (2) For the virtual small aperture subarray, the differential beamforming method is used to achieve the overall target gain requirement of the array; (3) For the virtual large-aperture subarray, the conventional beamforming method is used to improve the white noise gain and complete the robust differential beamforming method.

2. The robust differential beamforming method based on Kronecker product according to claim 1, characterized in that: The corresponding beamformer is designed based on the virtual small-aperture subarray and the virtual large-aperture subarray.

3. The robust differential beamforming method based on Kronecker product according to claim 1, characterized in that: Step (1) includes the following steps: The steering vector of the uniform linear array composed of M sensors is decomposed into the steering vector of two virtual sub-arrays through Kronecker integral, thus reconstructing into two virtual sub-arrays; the number of sensors in the first virtual sub-array is , the number of sensors in the second virtual subarray is , ; The uniform linear array composed of M sensors and the first virtual sub-array have an array element spacing of , and the element spacing of the second virtual subarray is .

4. The robust differential beamforming method based on Kronecker product according to claim 3, characterized in that: The steps to solve the steering vector of a uniform linear array using the Kronecker product are as follows: The steering vector of the uniform linear array is , where the steering vector of the uniform linear array can be expressed as the Kronecker product of the steering vectors of the two virtual sub-arrays: (8) (9) (10) According to formula (8), the global filter can be rewritten as the Kronecker product form of the filters corresponding to the two virtual sub-arrays: (11) On this basis, two virtual sub-array beamformers are designed: , , and finally the length is Linear filter.

5. The robust differential beamforming method based on Kronecker product according to claim 4, characterized in that: In the case where the array sensor cannot be decomposed into the integer product form of two virtual sub-arrays, the number and layout of the first virtual sub-array are as follows: (12) in, , , are all non-negative integers; when When is 0, the steering vector of the first virtual subarray is as shown in formula (9); and when When it is not 0, the steering vector of the first virtual subarray is: (13) in, The length is p+1, which is the first virtual subarray The number is p+1; in this case, .

6. The robust differential beamforming method based on Kronecker product according to claim 5, characterized in that: Introducing compressed sensing matrix to solve the problem of data dimension mismatch; (14) in, is a random Gaussian matrix composed of independent and incoherent Gaussian random variables.

7. The robust differential beamforming method based on Kronecker product according to claim 6, characterized in that: The uniform linear array beam pattern is represented as the product of two sub-array beam patterns, which allows for more flexible and convenient design of beam patterns or beamformers to meet performance requirements: (15) The white noise gain is simplified into the product of the white noise gains of two virtual sub-arrays: (16) Substitute the beam pattern product relationship into formula (7): (7) Obtain the directivity gain evaluation index: (17)。 8. The robust differential beamforming method based on Kronecker product according to claim 1, characterized in that: Step (2) includes the following steps: According to the passive sonar equation, the target directivity gain is obtained: ,in is the number of sensors; according to the target directivity gain, the design order N of the differential beamformer in the ideal case is obtained ( ); The directional gain of the sensor array is related to the sonar range, ocean environment, target sound source level and receiving depth. The relationship is expressed by the following passive sonar equation: (18) Among them, SL is the sound source level, unit is dB; TL is the propagation loss, which consists of two parts: geometric expansion loss and dielectric absorption loss. Among them, geometric expansion loss refers to the geometric effect when the sound signal is regularly weakened when expanding outward from the sound source, and dielectric absorption loss includes the effects of absorption, scattering and sound energy leakage out of the sound channel, so there is an empirical formula: (19) Where: K is a parameter used to describe the type of geometric expansion. When K=2, it is spherical expansion, and K=1 is cylindrical expansion. In the far field distance, it is assumed that the sound wave mainly propagates spherically, and K=2 is taken; r is the propagation distance, in meters; f is the propagation frequency, in kHz; NL is the ambient noise level; DI is the directivity gain, which represents the parameter of the sensor or array's ability to suppress interference from non-target directions. If a non-directional sensor is used, the receiving directivity index DI is ignored; DT is the detection threshold. The target can be detected when according to Relationship, to obtain the design order N of the differential beamformer under ideal conditions; to design an ideal differential beam pattern according to the obtained order; Design of ideal differential beam pattern at order N: (20) in, is a real coefficient. Different beam patterns of the N-order differential beamformer are determined according to different coefficient values. When the main lobe of the beam pattern points to the target direction, , the beam pattern is equal to 1; therefore, the coefficient satisfy: (21) For the design of the virtual small-aperture differential beamformer, under the given minimum directivity gain and maximum white noise gain criteria, the final beam pattern is made to approach the ideal beam pattern; based on this, the corresponding beamformer is equivalent to solving the following problem: (22) in, is the differential beamformer of the virtual small aperture subarray, is the actual beam pattern formed, and The ideal beam pattern is obtained by using the least squares method to make the error norm between the actual beam pattern and the ideal beam pattern less than a specific value. , and then maximize the white noise gain under the distortion-free constraint and given target gain.

9. The robust differential beamforming method based on Kronecker product according to claim 8, characterized in that: Formula (22) is solved using the alternating direction multiplication method or the CVX toolbox of Matlab.

10. The robust differential beamforming method based on Kronecker product according to claim 1, characterized in that: Step (3) includes the following steps: The conventional beamforming method realizes beamforming by compensating the phase difference caused by the distance between array elements when the signal is received by each array element; therefore, the weight vector of the conventional beamformer is set as: (23) The size of its weight is related to the direction of the expected signal, so the array has different gains for signals from different directions in space, thus forming spatial filtering; The array output signal is: (24) Substituting formula (23) into formula (6): (6) In the formula, is the direction of the target desired signal, and the superscript H is the conjugate transpose operator; Find: (25)。

Citation Information

Patent Citations

  • Direction-adjustable differential beam former design method based on vector hydrophone

    CN117471523A

  • Multi-granular feedback reporting and feedback processing for precoding in telecommunications

    US20110080969A1

Cited By

  • Underwater target accurate orientation method for vector hydrophone array signal synthesis

    CN119291606A