CN114184999B - Method for processing generated model of cross-coupling small-aperture array - Google Patents
Method for processing generated model of cross-coupling small-aperture array Download PDFInfo
- Publication number
- CN114184999B CN114184999B CN202010966451.6A CN202010966451A CN114184999B CN 114184999 B CN114184999 B CN 114184999B CN 202010966451 A CN202010966451 A CN 202010966451A CN 114184999 B CN114184999 B CN 114184999B
- Authority
- CN
- China
- Prior art keywords
- covariance
- aperture array
- array
- small
- standard
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 49
- 238000012545 processing Methods 0.000 title claims abstract description 33
- 238000006880 cross-coupling reaction Methods 0.000 title claims abstract description 13
- 238000005070 sampling Methods 0.000 claims abstract description 11
- 238000013507 mapping Methods 0.000 claims abstract description 8
- 230000008569 process Effects 0.000 claims abstract description 5
- 238000001228 spectrum Methods 0.000 claims description 13
- 238000012935 Averaging Methods 0.000 claims description 3
- 238000000354 decomposition reaction Methods 0.000 claims description 2
- 239000011159 matrix material Substances 0.000 description 22
- 230000008878 coupling Effects 0.000 description 18
- 238000010168 coupling process Methods 0.000 description 18
- 238000005859 coupling reaction Methods 0.000 description 18
- 238000012549 training Methods 0.000 description 9
- 238000003491 array Methods 0.000 description 8
- 238000004088 simulation Methods 0.000 description 8
- 230000009466 transformation Effects 0.000 description 8
- 230000000694 effects Effects 0.000 description 6
- 238000013135 deep learning Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 238000006243 chemical reaction Methods 0.000 description 3
- 230000004044 response Effects 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 2
- 239000000654 additive Substances 0.000 description 2
- 230000000996 additive effect Effects 0.000 description 2
- 230000015556 catabolic process Effects 0.000 description 2
- 230000001808 coupling effect Effects 0.000 description 2
- 238000006731 degradation reaction Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000003672 processing method Methods 0.000 description 2
- 238000012546 transfer Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 239000011664 nicotinic acid Substances 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 238000002604 ultrasonography Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S3/00—Direction-finders for determining the direction from which infrasonic, sonic, ultrasonic, or electromagnetic waves, or particle emission, not having a directional significance, are being received
- G01S3/02—Direction-finders for determining the direction from which infrasonic, sonic, ultrasonic, or electromagnetic waves, or particle emission, not having a directional significance, are being received using radio waves
- G01S3/14—Systems for determining direction or deviation from predetermined direction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Pure & Applied Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Computational Mathematics (AREA)
- Algebra (AREA)
- Remote Sensing (AREA)
- Radar, Positioning & Navigation (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Measurement Of Velocity Or Position Using Acoustic Or Ultrasonic Waves (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention provides a method for processing a generating model of a cross-coupling small-aperture array, which belongs to the field of array signal processing, and utilizes a depth generating model and a standard array to process signals of unknown array manifold, and is characterized by comprising the following steps: step S1, a computer randomly generates a group of arrival directions, corresponding small-aperture array sample covariance and standard-aperture array sample covariance, and normalizes the sample covariance after subtracting noise covariance; s2, inputting a standard aperture covariance into a depth generation model according to the small aperture array sample covariance, and learning the depth generation model to obtain a mapping relation of probability distribution between the small aperture array sample covariance and the standard aperture array covariance; s3, inputting small-aperture array sample covariance, and sampling to generate a plurality of standard virtual array covariance according to the mapping relation; and S4, estimating the direction of arrival through a subspace algorithm and taking an average value.
Description
Technical Field
The invention belongs to the field of array signal processing, and particularly relates to a signal processing method for improving the performance of a small-aperture array by using a depth generation model.
Background
The direction of arrival (Direction Of Arrival, DOA) estimation of signals is an important content in array signal processing, and has wide application in the fields of radar, mobile communication and the like. Most current arrays rely on inter-element delay to complete DOA estimation, where the estimated performance is proportional to the pitch of the array (i.e., array aperture size). In some applications, however, a smaller aperture is required for the sensor array, which degrades the estimation performance of the array. Furthermore, the coupling between array elements under a small aperture array is not negligible, the manifold of the array no longer meets the preset ideal form, and if calibration is not performed, the performance of the conventional DOA estimation algorithm is rapidly deteriorated, even in error.
Aiming at the problem of performance degradation of a small-aperture array, a bionic-based method is proposed in the prior art, and the method increases the difference between response amplitude and arrival time delay among array elements by designing a group of butterfly filters, so that the method is equivalent to obtaining a large-aperture virtual array in effect. However, the above method only assumes the interactions that exist between adjacent array elements, and for the coupling that exists in small aperture arrays, it is first assumed that the array calibration parameters are known or estimated by other algorithms. However, the calibration coefficient of the array is generally unknown, while the conventional estimation algorithm needs to assume that the mutual coupling phenomenon can be represented by a matrix with a specific structure, and when this assumption is not satisfied, the mutual coupling relationship becomes very complex or even difficult to solve.
For complex and even nonlinear array signal processing problems, the prior art machine learning-based method provides a method for estimating the angle of arrival of a signal by using a multi-layer perceptron or realizing voice enhancement by using deep learning, and improves the recognition rate of a low-angle target according to a covariance transformation network. The above method uses a discriminant model to analyze the array signal, but the discriminant model often cannot obtain an ideal effect in the presence of noise in the input signal.
Therefore, the signal processing methods of the small-aperture array in many documents can only expand the aperture of the small-aperture array or remove the coupling between the array elements, and do not consider the problem that the small-aperture array has the mutual coupling effect or even the array element distance error, which is necessarily needed to be considered in application.
Disclosure of Invention
In order to solve the problems, the invention provides a method for processing a generating model of a cross-coupling small-aperture array, which adopts the following technical scheme:
The invention provides a method for processing a generating model of a cross-coupling small-aperture array, which utilizes a depth generating model and a standard array to process signals of unknown array manifold and is characterized by comprising the following steps: step S1, a computer randomly generates a group of directions of arrival, obtains original small-aperture array sample covariance corresponding to the directions of arrival and original standard-aperture array sample covariance corresponding to the directions of arrival, and normalizes the original small-aperture array sample covariance and the original standard-aperture array sample covariance after subtracting the noise covariance respectively to obtain the small-aperture array sample covariance and the standard-aperture array covariance; s2, inputting a standard aperture array sample covariance into a depth generation model according to the small aperture array sample covariance, and learning the depth generation model to obtain a mapping relation of probability distribution between the small aperture array sample covariance and the standard aperture array sample covariance; s3, inputting small-aperture array sample covariance into a computer, and sampling to generate a plurality of standard virtual array covariance according to a transfer relation; and S4, estimating the arrival direction of the standard virtual array covariance obtained by sampling through a subspace algorithm to obtain a plurality of estimated arrival angles, and averaging the estimated arrival angles to obtain an average arrival angle.
The method for processing the generated model of the cross-coupled small-aperture array can also have the characteristics that in the step S2, the input of the encoder is covariance R yy, and the output of the decoder is covarianceThe encoder and decoder conditions are the small aperture array sample covariance R xx, covariance R yy is the standard aperture array covariance, R yy and/>The covariance R yy includes a real part and an imaginary part, which contain the same arrival information and target energy information.
The method for processing the generated model of the cross-coupling small-aperture array can also have the characteristic that noise is added in the condition and hidden variables in the step S1 and the step S2.
The method for processing the generative model of the cross-coupling small-aperture array provided by the invention can also have the characteristics that in the step S3, a decoder generates covarianceStandard virtual array covariance is/>
The method for processing the generated model of the cross-coupled small-aperture array can also have the characteristic that subspace algorithms comprise, but are not limited to, MUSIC algorithms.
The method for processing the generated model of the cross-coupled small-aperture array can also have the characteristics that in the step S4, the MUSIC algorithm carries out eigenvalue decomposition on the standard virtual array covariance with the size of M multiplied by M to obtain
Wherein U s is the eigenvector corresponding to the K largest eigenvalues, defined as the signal subspace; u n is the eigenvector corresponding to M-K minimum eigenvalues, which is defined as noise subspace, and the spatial spectrum obtained by MUSIC algorithm is
The estimated angle of arrival is the maximum point of the spatial spectrum.
The method for processing the generated model of the cross-coupling small-aperture array can also have the characteristic that the subspace algorithm is an ESPRIT algorithm.
The method for processing the generated model of the cross-coupling small-aperture array can also have the characteristic that the subspace algorithm is a Root-MUSIC algorithm.
The actions and effects of the invention
The method for processing the generated model of the cross-coupled small-aperture array is used for learning the conditional probability between the covariance of the received signal of the small-aperture array and the covariance of the standard array. After training, the covariance of the corresponding standard virtual array with the same arrival information is generated by sampling and taking the signal received by the small-aperture array as a condition, so that the signal processing can be performed by utilizing the array manifold known by the standard array. The invention is different from the prior art that only the aperture of the small aperture array can be singly enlarged or the coupling between the array elements can be removed, but the coupling between the aperture of the enlarged array and the coupling between the array elements can be brought into the same frame, thereby obviously improving the estimation performance of the small aperture array. Meanwhile, the invention learns the conversion relation between the covariance of the small-aperture array and the covariance of the large-aperture standard array by utilizing the depth generation model in a data driving mode, does not need to make assumptions on array forms or array element distance errors, and has practical value. And the invention can process various types of arrays, including but not limited to radar, ultrasonic, microphone array and other sensor array devices, and has wide application value.
Drawings
FIG. 1 is a flow chart of a method for processing a generated model of a cross-coupled small aperture array according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a deep learning network encoding-decoding architecture according to an embodiment of the present invention;
FIG. 3 is a sample generation schematic of a standard array covariance of an embodiment of the invention;
FIG. 4 is a normalized spatial spectrum of angles of arrival of an embodiment of the present invention of [ -25 °, -5 °,30 °,40 ° ];
FIG. 5 is a normalized spatial spectrum of an embodiment of the present invention with an angle of arrival of [5, 15 ].
Detailed Description
The sensor used in the array of the invention comprises an antenna, a hydrophone, a transducer and the like, and the array topology comprises a linear array, a circular array, an area array and the like.
The following describes specific embodiments of the present invention with reference to the drawings and examples.
< Example >
The embodiment provides a method for processing a generated model of a cross-coupled small-aperture array, which is used for improving the performance of the small-aperture array.
FIG. 1 is a flow chart of a method for processing a generated model of a cross-coupled small aperture array according to an embodiment of the present invention.
As shown in fig. 1, the method for processing the generated model of the cross-coupled small-aperture array specifically includes the following steps:
Step S1, input data are obtained. The computer randomly generates a group of directions of arrival, obtains the covariance of the original small-aperture array samples corresponding to the directions of arrival and the covariance of the corresponding original standard-aperture array samples, and normalizes the covariance of the original small-aperture array samples and the covariance of the original standard-aperture array samples after subtracting the noise covariance respectively to obtain the covariance of the small-aperture array samples and the covariance of the standard-aperture array. The computer generates several groups of signals with random directions of arrival, the number of the information sources is random and not more than the number of array elements of the array, and the snapshot numbers of the signals in each group are the same. The data set is divided into a training set, a verification set and a test set according to the proportion of 8:1:1. And then, subtracting the estimated noise covariance matrix from the generated small-aperture array sample covariance, and normalizing the small-aperture array sample covariance as an input of the network.
The generation of standard array sample covariance (equations 1-3) and the generation of small aperture array sample covariance (equations 4-5) are as follows.
A measurement model of a standard Uniform Linear Array (ULA) containing M array elements can be expressed as:
y(t)＝As(t)+ey(t)t＝1,2,...,N (1)
Where y (t) =c M×1 is the output vector of the standard array; s (t) = [ s 1(t),…,sK(t)]T represents K far-field narrowband incident signals; e y (t) represents additive ambient noise, the noise covariance Q y＝σ2 yI;A＝[a(φ1,d),a(φ2,d),…,a(φk, d) of which is the array response matrix, and the steering vector a (phi k) can be expressed as
Wherein λ represents a wavelength; d is the interval of array elements, and the value in the defined standard array is 0.5lambda. In this embodiment, the MUSIC algorithm is used to perform the direction of arrival estimation. The MUSIC algorithm first estimates the sample covariance of the received signal:
Ryy＝E{y(t)yH(t)}＝ARsAH+σ2I (3)
When the small-aperture array researched by the invention is used for testing the far-field narrowband signal which is the same as the formula (1), when the aperture of the array is smaller, the mutual coupling effect is generated between array elements, and the array measurement model is that
x(t)＝CBs(t)+ex t t＝1,2,...,N (4)
Where x (t) =c M×1 is the output vector of the small aperture array; e x (t) represents additive ambient noise, whose noise covariance Qx＝σ2 xI;B＝[a(φ1,ds),a(φ2,ds),…,a(φk,ds)] is the array response matrix, and ds < d; c is a cross-coupling matrix (Mutual Coupling Matrix, MCM), and the cross-coupling and small-array transformation can be practically unified into the same transformation matrix. Thus, the covariance of the output signal of the small aperture array can be expressed as
Wherein C t >0 represents a unified transformation matrix, which can not only represent mutual coupling and conversion relation between the small-aperture array and the standard array, but also represent distance error between array elements; f (·) represents a function that converts the standard array output covariance to a small aperture array output covariance.
Assuming that the inverse of F (-) is formally F -1 (-), this inverse can be used to convert R ww to R yy and estimate the angle of arrival using the manifold information of the standard array. However, since the explicit solution is difficult to obtain in the F -1 (), the present embodiment uses the transformation relationship between the array covariances learned by the generative model to transform the covariances of the small hole array into the covariances of the standard array, which is effectively equivalent to generating a virtual standard array by using the generative model, and performing DOA estimation by using the MUSIC algorithm.
Step S2, learning phase: and inputting the standard aperture array sample covariance into the depth generation type model according to the small aperture array sample covariance, and learning the depth generation type model to obtain the mapping relation of probability distribution between the small aperture array sample covariance and the standard aperture array sample covariance.
Namely, based on the deep learning network coding-decoding structure, the mapping relation of probability distribution between the small aperture array covariance and the standard virtual array covariance is learned.
Fig. 2 is a schematic diagram of a deep learning network encoding-decoding structure according to an embodiment of the present invention.
As shown in fig. 2, in step S2, the conditions of the encoder (Encoder) and the Decoder (Decoder) are the sample covariance R xx of the small aperture array, the input of the encoder is set to be the standard aperture array sample covariance R yy, and the output of the Decoder is set to beR yy and/>Comprising the same arrival information as well as target energy information, such as signals received from both arrays at an angle of incidence of 5, 15, the target. The above condition is a special structure in "CVAE", and in CVAE, the probability distribution may be expressed as P (y|x) = Σp (y|z, x) P (z|y, x) dz. Where x is the condition in the input encoder P (z|y, x) and decoder P (y|z, x); z represents the hidden variable linking the encoder and decoder in the structure, and has to be distributed close to each other during learningTo improve generalization ability and robustness of the model.
The sum of squares of differences between corresponding positions between the matrices is used to describe the difference between the two matrices as a learned objective function. Let input R yy and output/>, when R xx is taken as the conditionAs equal as possible, the conditional probability transition relationships of R xx and R yy can be learned.
Since the covariance of the array is a complex matrix, the present embodiment splits the covariance matrix into real and imaginary parts, so that the covariance of the input network can be considered as a 2-channel image.
In this embodiment, when training the network, the optimizer is Adam, the learning rate is set to a smaller value (between 0.001 and 0.05), and the iteration is performed until the loss function of the network is not significantly reduced. In other embodiments, strategies such as learning rate decline and early fall may also be used to achieve better training results.
The algorithm adds noise in the conditions and hidden variables of the step S2 and the step S3, namely, adds noise in R xx and Z, which is helpful for regularization of the network, so that good virtual array covariance can be generated under the condition that the sample covariance matrix of the small hole array is not ideal enough.
Step S3, sampling: and inputting the small-aperture array sample covariance into a computer, and sampling to generate a plurality of standard virtual array covariance according to the transfer relation.
Once training is completed, the decoder can generate standard array covariance under the condition of taking the small-aperture array covariance as input, and then the obtained standard array covariance is transformed to obtain transformation covariance in the form of a conjugate symmetric array, and the transformation covariance is taken as the required standard array covariance, namely the input of the step S4.
In this embodiment, the encoder inputs normally distributed white gaussian noise, samples N s covariances, and then estimates the angle of arrival using a subspace algorithm.
FIG. 3 is a sample generation schematic of standard array covariance according to an embodiment of the invention.
Wherein, as shown in FIG. 3, after training is completed, the small aperture array covariance R yy is input and the decoder generates the required standard virtual array covarianceThe decoder at this time corresponds to the mapping function F -1 (·).
The covariance matrix should be a conjugate symmetric matrix, but the covariance generated by the network cannot strictly meet the condition, and deviation may occur when using subspace algorithm, so the present embodiment will obtain standard virtual array covarianceThe following transformation is performed
Wherein,Standard virtual array covariance generated for decoder,/>Representing transformed standard virtual array covariance,/>And satisfying the hermitian condition, namely, inputting in the step S4.
Step S4, estimating: sampling the obtained standard virtual array covariance matrix by subspace algorithmDOA estimation is performed and the average of the estimates is taken as the final result.
In this embodiment, the subspace algorithm used is a MUSIC algorithm.
First, the obtained standard virtual array covariance is decomposed by eigenvalue
Wherein U s is the eigenvector corresponding to the K largest eigenvalues, defined as the signal subspace; u n is the eigenvector corresponding to the M-K minimum eigenvalues, defined as the noise subspace. The spatial spectrum obtained by the MUSIC algorithm is
And taking the maximum point of the obtained spatial spectrum as an estimated angle of arrival, and taking an average value to obtain an average angle of arrival estimation result.
Simulation experiment
The present embodiment also provides a simulation experiment, i.e. the effectiveness of the proposed algorithm is verified by using the simulation of the Uniform Linear Array (ULA) in the present experiment.
The main simulation parameters of the simulation experiment are as follows: the array element spacing of the small aperture array is 0.2lambda, and the number of the antenna array elements is 12; the mutual coupling matrix is a toprilz matrix, and the mutual coupling coefficient c 1＝0.6+0.6i,c2 =0.4+0.3i. Each group of signals generates a theoretical covariance under a corresponding standard array, and the grid precision of the MUSIC algorithm is 0.01 degrees. In addition to coupling, the interference of small aperture array positions is also considered in this simulation scenario. Wherein the position d' s of the perturbed small aperture array satisfies
Wherein d s＝[0 0.2λ…2.2λ]T∈R11×1 represents the position of the array element with the ideal small aperture array pitch. At the same time, the array coupling error is disturbed, and meets the requirements
Wherein p, q represents the position of the C' t element; Representing a complex gaussian distribution, i.e. the coupling matrix and the array element positions are unknown in the simulation.
10000 Groups of signals generated in the experiment have random directions of arrival, the number of the information sources is between 1 and 5, the number of the snapshots of each group of signals is 300, and the center frequency of the signals is 1GHz. The signal-to-noise ratio is 5dB and the source power is between 0.8 and 1. The data set is divided into a training set, a verification set and a test set according to the proportion of 8:1:1.
When the network is trained, the optimizer is Adam, the learning rate is 0.005, and the iteration number is 120. After training, three covariances are generated in a sampling mode, the maximum value of the spatial spectrum is respectively calculated to be used as an estimated angle of arrival, and the average angle of arrival is obtained by averaging the estimated angle of arrival to be used as a final result.
Simulation result analysis:
Fig. 4 is a normalized spatial spectrum of the angle of arrival of the embodiment of the present invention [ -25 °, -5 °,30 °,40 ° ], and fig. 5 is a normalized spatial spectrum of the angle of arrival of the embodiment of the present invention [5 °,15 ° ].
Fig. 4 is an estimation of a large aperture standard array (d=0.5λ), an uncalibrated small aperture array (d s =0.2λ), a small aperture array calibrated using a theoretical cross-coupling matrix (d s =0.2λ), and the method of the present invention at different combinations of angles of arrival. All the spatial spectra estimated by the algorithm are normalized to between 0 and 1. As shown in fig. 4, when the angle of arrival of the signal is [ -25 °, -5 °,30 °,40 ° ], the uncorrected small aperture array cannot estimate the correct angle of arrival at all, and has high side lobes. The small aperture arrays with and without calibration cannot estimate the correct angle of arrival due to the factors of array element position errors, small aperture and small resolution degradation, and array element coupling, and the difference between the correct angle of arrival and the actual result is large. The results obtained by the method are not greatly different from those obtained by the large-aperture standard array, which shows that the method has good effects.
Also, as shown in fig. 5, when the angle of arrival of the signal is [5 °, -15 ° ], the estimation performance of the small aperture array, which is not processed and calibrated using the theoretical mutual coupling matrix, is poor, and it is difficult to correctly estimate all the angles of arrival. In contrast, the invention has good effect. This also verifies that the proposed method can significantly improve the performance of small aperture arrays.
Example operation and Effect
The method for processing the generated model of the cross-coupled small-aperture array is used for learning conditional probability between covariance of received signals of the small-aperture array and covariance of a standard array. After training, the covariance of the corresponding standard virtual array with the same arrival information is generated by sampling and taking the signal received by the small-aperture array as a condition, so that the signal processing can be performed by utilizing the array manifold known by the standard array. In the invention, the coupling between the expanded array aperture and the removed array element is brought into the same frame, thus obviously improving the estimation performance of the small aperture array. Meanwhile, the invention learns the conversion relation between the covariance of the small aperture array and the covariance of the large aperture standard array by utilizing the depth generation model in a data driving mode, does not need to make assumptions on array forms or array element distance errors, has practical value, can process arrays in various forms, including but not limited to sensor array equipment such as radar, ultrasound, microphone arrays and the like, and has wide application value.
Furthermore, the algorithm adds noise in both the condition and the hidden variable, and the strategy is favorable for regularization of the network and enhancing the generalization capability of the model, so that better virtual array covariance can be generated under the condition that the sample covariance matrix of the small-hole array is not ideal enough.
Further, the covariance matrix should be a conjugate symmetric matrix, but the covariance generated by the network cannot strictly meet the condition, and a deviation may occur when using the subspace algorithm, so the embodiment transforms the obtained standard virtual array covariance, so that the transformed array covariance meets the hermite condition, and then obtains a result by using the subspace algorithm.
The above examples are only for illustrating the specific embodiments of the present invention, and the present invention is not limited to the description scope of the above examples.
In this embodiment, the subspace algorithm used is a MUSIC algorithm, the result obtained in step S4 is a spatial spectrum, and the maximum value is taken to obtain the estimated angle of arrival. In other embodiments, root-MUSIC, ESPRIT algorithm, etc. may be used, and the form of the obtained result may be changed accordingly.
Claims (8)
1. The method for processing the generation model of the cross-coupling small-aperture array utilizes a depth generation model and a standard aperture array to process signals of unknown array manifold, and is characterized by comprising the following steps:
Step S1, a computer randomly generates a group of directions of arrival, obtains original small-aperture array sample covariance corresponding to the directions of arrival and original standard-aperture array sample covariance corresponding to the directions of arrival, and normalizes the original small-aperture array sample covariance and the original standard-aperture array sample covariance after subtracting noise covariance respectively to obtain small-aperture array sample covariance and standard-aperture array covariance;
S2, inputting the standard aperture array sample covariance into the depth generation model according to the small aperture array sample covariance, and learning the depth generation model to obtain a mapping relation of probability distribution between the small aperture array sample covariance and the standard aperture array sample covariance;
S3, inputting the small-aperture array sample covariance to the computer, and sampling to generate a plurality of standard virtual array covariance according to the mapping relation;
And S4, estimating the arrival direction of the standard virtual array covariance obtained by sampling through a subspace algorithm to obtain a plurality of estimated arrival angles, and averaging the estimated arrival angles to obtain an average arrival angle.
2. The method for generating model processing of the cross-coupled small-aperture array according to claim 1, wherein the method comprises the following steps:
in the step S2, the input of the encoder is covariance R yy, and the output of the decoder is covariance
The conditions of the encoder and the decoder are the small aperture array sample covariance R xx,
The covariance R yy is the standard aperture array sample covariance, R yy andContains the same arrival information and target energy information,
The covariance R yy includes a real part and an imaginary part.
3. The method for generating model processing of the cross-coupled small-aperture array according to claim 2, wherein the method comprises the following steps:
wherein noise is added to the conditions and hidden variables of the step S1 and the step S2.
4. The method for generating model processing of the cross-coupled small-aperture array according to claim 2, wherein the method for generating model processing of the cross-coupled small-aperture array is characterized in that:
wherein in the step S3, the decoder generates covariance
The standard virtual array covariance is
5. The method for generating model processing of the cross-coupled small-aperture array according to claim 1, wherein the method comprises the following steps:
wherein, the subspace algorithm is a MUSIC algorithm.
6. The method for generating model processing of the cross-coupled small aperture array according to claim 5, wherein the method comprises the steps of:
In step S4, the MUSIC algorithm performs eigenvalue decomposition on the standard virtual array covariance with size of mxm
Wherein, U s is the eigenvector corresponding to the K largest eigenvalues, defined as the signal subspace; u n is the eigenvector corresponding to M-K minimum eigenvalues, defined as the noise subspace,
The spatial spectrum obtained by the MUSIC algorithm is
The estimated angle of arrival is the maximum point of the spatial spectrum.
7. The method for generating model processing of the cross-coupled small-aperture array according to claim 1, wherein the method comprises the following steps:
wherein the subspace algorithm is an ESPRIT algorithm.
8. The method for generating model processing of the cross-coupled small-aperture array according to claim 1, wherein the method comprises the following steps:
the subspace algorithm is a Root-MUSIC algorithm.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010966451.6A CN114184999B (en) | 2020-09-15 | 2020-09-15 | Method for processing generated model of cross-coupling small-aperture array |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010966451.6A CN114184999B (en) | 2020-09-15 | 2020-09-15 | Method for processing generated model of cross-coupling small-aperture array |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114184999A CN114184999A (en) | 2022-03-15 |
| CN114184999B true CN114184999B (en) | 2024-05-14 |
Family
ID=80600828
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010966451.6A Active CN114184999B (en) | 2020-09-15 | 2020-09-15 | Method for processing generated model of cross-coupling small-aperture array |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114184999B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115436873B (en) * | 2022-08-09 | 2024-06-18 | 哈尔滨工程大学 | Aperture expansion method, system, computer and storage medium based on array manifold vector mapping |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101131429A (en) * | 2007-09-19 | 2008-02-27 | 浙江大学 | Image restoration method for image quality degraded imaging system with optical aberration and small hole diffraction |
| CN109597047A (en) * | 2018-11-29 | 2019-04-09 | 西安电子科技大学 | Based on the metre wave radar DOA estimation method for having supervision deep neural network |
| CN109829547A (en) * | 2018-12-18 | 2019-05-31 | 中国人民解放军国防科技大学 | Depth learning-based SST (stimulated Raman Scattering) inversion method for one-dimensional synthetic aperture microwave radiometer |
| CN110531313A (en) * | 2019-08-30 | 2019-12-03 | 西安交通大学 | A kind of near-field signals source localization method based on deep neural network regression model |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11340345B2 (en) * | 2015-07-17 | 2022-05-24 | Origin Wireless, Inc. | Method, apparatus, and system for wireless object tracking |
-
2020
- 2020-09-15 CN CN202010966451.6A patent/CN114184999B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101131429A (en) * | 2007-09-19 | 2008-02-27 | 浙江大学 | Image restoration method for image quality degraded imaging system with optical aberration and small hole diffraction |
| CN109597047A (en) * | 2018-11-29 | 2019-04-09 | 西安电子科技大学 | Based on the metre wave radar DOA estimation method for having supervision deep neural network |
| CN109829547A (en) * | 2018-12-18 | 2019-05-31 | 中国人民解放军国防科技大学 | Depth learning-based SST (stimulated Raman Scattering) inversion method for one-dimensional synthetic aperture microwave radiometer |
| CN110531313A (en) * | 2019-08-30 | 2019-12-03 | 西安交通大学 | A kind of near-field signals source localization method based on deep neural network regression model |
Non-Patent Citations (2)
| Title |
|---|
| 基于BP神经网络的烟熏香肠色泽预测研究;陈炎;屠泽慧;聂文;季拓;刘敏;张静;杨潇;蔡克周;陈从贵;姜绍通;;食品研究与开发;20171020(第20期);全文 * |
| 非高斯噪声环境迭代自适应DOA估计的算法;谢晨欣;李旦;张建秋;;复旦学报(自然科学版);20150415(第02期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114184999A (en) | 2022-03-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110927659B (en) | Method and system for estimating arbitrary array manifold DOA (direction of arrival) under cross-coupling condition and cross-coupling calibration | |
| CN110208735B (en) | Sparse Bayesian learning-based coherent signal DOA estimation method | |
| CN109298383B (en) | Mutual-prime array direction-of-arrival estimation method based on variational Bayes inference | |
| CN110045323B (en) | Matrix filling-based co-prime matrix robust adaptive beamforming algorithm | |
| CN109375154B (en) | Coherent signal parameter estimation method based on uniform circular array in impact noise environment | |
| CN111239678B (en) | Two-dimensional DOA estimation method based on L-shaped array | |
| CN109254261B (en) | Coherent signal null deepening method based on uniform circular array EPUMA | |
| Chen et al. | Adaptive beamforming with sensor position errors using covariance matrix construction based on subspace bases transition | |
| CN101150345A (en) | Direction measurement method applicable to phase interference signal source under non stabilized noise background | |
| Gong et al. | Joint DOD and DOA estimation for bistatic multiple‐input multiple‐output radar target discrimination based on improved unitary ESPRIT method | |
| CN110837076A (en) | Tensor decomposition-based vector hydrophone array orientation estimation method | |
| CN110806557A (en) | Monostatic MIMO radar DOA estimation method under impact noise background | |
| CN111352063A (en) | Two-dimensional direction finding estimation method based on polynomial root finding in uniform area array | |
| CN111366893A (en) | Non-circular signal azimuth angle estimation method under unknown mutual coupling condition of uniform circular array | |
| Adhikari | Shift-invariant structure-imposed convolutional neural networks for direction of arrival estimation | |
| CN114184999B (en) | Method for processing generated model of cross-coupling small-aperture array | |
| CN108594165B (en) | Narrow-band signal direction-of-arrival estimation method based on expectation maximization algorithm | |
| Yang et al. | A correlation-aware sparse Bayesian perspective for DOA estimation with off-grid sources | |
| CN115469286B (en) | Super-resolution angle measurement method based on millimeter wave automobile radar minimum redundancy MIMO array | |
| CN115980721A (en) | Array self-correcting method for error-free covariance matrix separation | |
| CN109298381A (en) | A kind of relatively prime battle array coherent signal azimuth estimation method based on variational Bayesian | |
| CN116299150A (en) | Two-dimensional DOA estimation method of dimension-reduction propagation operator in uniform area array | |
| CN113381793B (en) | Coherent information source estimation-oriented non-grid direction-of-arrival estimation method | |
| CN115421098A (en) | Two-dimensional DOA estimation method for nested area array dimension reduction root finding MUSIC | |
| CN111431575B (en) | Incoming wave direction sparse reconstruction method based on conventional beam forming |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |