JP2024000187A - Learning device, arrival angle estimation device, arrival angle estimation system, learning model generation method, arrival angle estimation method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To use a learning model for highly accurately estimating an arrival angle of a radio wave in an actual environment.

SOLUTION: A learning device includes: signal generation means for generating a transmission signal vector so as to allow signal intensity of a transmission signal corresponding to a designated arrival angle to be stronger than a transmission signal corresponding to the other arrival angle; correlation matrix generation means for generating a correlation matrix based on the transmission signal vector and a mode vector of an actual environment array antenna which is used in an actual environment; input data generation means for generating input data from an element of the correlation matrix; and learning means for generating a learning model by using the input data as an input so as to execute learning processing with the use of teacher data where data indicating the arrival angle corresponding to the input data is defined as a correct label.

SELECTED DRAWING: Figure 13

COPYRIGHT: (C)2024,JPO&INPIT

Description

The present invention relates to a learning device, an arrival angle estimating device, an arrival angle estimation system, a learning model generation method, an arrival angle estimation method, and a program.

As a method for estimating the arrival angle of radio waves using an array antenna, the MUSIC (Multiple Signal Classification) method based on a subspace method using eigenvalue decomposition of a correlation matrix of a received signal is widely known. In addition, a method as disclosed in Patent Document 1 has been proposed, and in recent years, an angle of arrival estimation method using deep learning (hereinafter referred to as a learning angle of arrival estimation method) has been proposed. (For example, see Non-Patent Documents 1 and 2).

Japanese Patent Application Publication No. 2005-291916

Y. Kase, et al., “Fundamental Trial on DOA Estimation with Deep Learning”, IEICE Trans. Commun., Vol.E103-B, Oct. 2020. W. Zhu, et al., “Broadband Direction of Arrival Estimation Based on Convolutional Neural Network”, IEICE Trans. Commun., Vol.E103-B, Mar. 2020.

The learning-based angle of arrival estimation method has the advantage of being able to estimate with simple calculations when using a trained learning model, and that under certain conditions, it has a higher estimation speed than non-learning-based angle of arrival estimation methods such as the MUSIC method. It has advantages such as superior accuracy in some cases. On the other hand, when using the learning angle of arrival estimation method, even if the received signal obtained by receiving radio waves has a high signal-to-noise ratio (SNR), it is not included in the training data. This has the disadvantage that the estimation accuracy for angles of arrival that are not calculated is low.

In order to generate a learning model with high estimation accuracy for various angles of arrival, training data with various angles of arrival as correct labels is required. In order to obtain such teacher data, received signals for each of a plurality of angles of arrival at fine angular intervals are required. However, it takes a very long time to actually measure received signals for each of a plurality of angles of arrival at small angular intervals.

On the other hand, in the techniques disclosed in Non-Patent Documents 1 and 2, for example, an ideal array antenna is assumed, and training data is generated using a mode vector of a theoretical value corresponding to the ideal array antenna. A learning model is generated using the generated training data. By using mode vectors, it is possible to obtain a correlation matrix for each of multiple angles of arrival with fine angular intervals without actually measuring them, so training data that uses each of the multiple angles of arrival with fine angular intervals as the correct label can be obtained. By performing learning processing using the training data obtained in this way, the characteristics of the array antenna, for example, the characteristics of the antenna that affects the estimation of the angle of arrival among the multiple antennas forming the array antenna, are reflected. It becomes possible to generate a learning model based on the

However, there are differences in the distance between the transmission source and each antenna element forming the array antenna and subtle differences in the characteristics of the antenna elements between the array antenna used in real environments and the ideal array antenna. be. Therefore, there is a difference between the mode vector of the array antenna used in the actual environment and the mode vector of the theoretical value. Because of this difference, even if an attempt is made to use a learning model generated from teacher data generated based on a theoretical value mode vector in a real environment, the accuracy of estimating the angle of arrival will decrease. On the other hand, there is a problem of using a learning model that estimates the arrival angle of radio waves with high accuracy in a real environment.

Therefore, an object of the present invention is to provide a learning device, an angle of arrival estimation device, an angle of arrival estimation system, a learning model generation method, an arrival angle estimation method, and a program that solve the above-mentioned problems.

According to the first aspect of the present invention, the learning device generates a transmission signal vector such that the signal strength of a transmission signal corresponding to a specified arrival angle is stronger than transmission signals corresponding to other arrival angles. a generation means; a correlation matrix generation means for generating a correlation matrix based on the transmitted signal vector and a mode vector of a real environment array antenna used in a real environment; and an input for generating input data from elements of the correlation matrix. a data generation means, and a learning means that generates a learning model by performing a learning process using teacher data in which the input data is input and data indicating the arrival angle corresponding to the input data is used as a correct answer label. Be prepared.

According to a second aspect of the present invention, the arrival angle estimating device includes an array antenna, a receiving means that generates a received signal vector from radio waves arriving at the array antenna, and a correlation matrix that generates a correlation matrix based on the received signal vector. a correlation matrix generating means for generating input data from the elements of the correlation matrix; and a learning model generated to be an evaluation function for evaluating the probability of existence of a radio wave for each angle of arrival at the array antenna. an inference means for obtaining output data by giving the input data to a learning model generated according to a learning process based on training data generated using the mode vector of the array antenna; Angle of arrival calculation means for calculating an estimated angle of arrival.

According to a third aspect of the present invention, an angle of arrival estimation system includes the learning device described above and the angle of arrival estimation device described above, and the real environment array antenna in the learning device is configured to The learning model that is the array antenna of the angle of arrival estimation device and that is generated by the learning means of the learning device is the learning model of the inference means of the angle of arrival estimation device.

According to the fourth aspect of the present invention, a learning model generation method generates a transmission signal vector such that the signal strength of a transmission signal corresponding to a specified arrival angle is stronger than transmission signals corresponding to other arrival angles. and generate a correlation matrix based on the generated transmission signal vector and a mode vector of a real environment array antenna used in a real environment, and generate input data from the elements of the generated correlation matrix. A learning model is generated by performing a learning process using teacher data in which the input data is input and data indicating the angle of arrival corresponding to the input data is used as a correct label.

According to the fifth aspect of the present invention, the arrival angle estimation method includes: generating a received signal vector from radio waves arriving at an array antenna; generating a correlation matrix based on the generated received signal vector; A learning model that generates input data from elements of a matrix and is generated using a mode vector of the array antenna to become an evaluation function for evaluating the probability of existence of a radio wave for each angle of arrival at the array antenna. The input data is given to a learning model generated according to a learning process based on the teacher data obtained, and output data is obtained, and an estimated arrival angle of a radio wave is calculated from the obtained output data.

According to the sixth aspect of the present invention, the program causes the computer to generate a transmission signal vector such that the signal strength of the transmission signal corresponding to the specified arrival angle is stronger than the transmission signal corresponding to other arrival angles. a step of generating a correlation matrix based on the generated transmission signal vector and a mode vector of a real environment array antenna used in a real environment; a step of generating input data from the elements of the generated correlation matrix; A program for executing a procedure of generating a learning model by performing a learning process using training data in which the generated input data is input and data indicating the arrival angle corresponding to the input data is used as a correct answer label. It is.

According to the seventh aspect of the present invention, the program instructs a computer to perform a procedure for generating a received signal vector from radio waves arriving at an array antenna, a procedure for generating a correlation matrix based on the generated received signal vector, and a procedure for generating a correlation matrix based on the generated received signal vector. A procedure for generating input data from the elements of the correlation matrix, and a learning model generated to be an evaluation function for evaluating the probability of existence of a radio wave for each angle of arrival at the array antenna, using the mode vector of the array antenna. A step of giving the input data to a learning model generated according to a learning process based on the training data generated by the method to obtain output data, and a step of calculating an estimated arrival angle of a radio wave from the obtained output data. This is a program for

According to the present invention, it is possible to use a learning model that estimates the arrival angle of radio waves with high accuracy in a real environment.

FIG. 1 is a block diagram showing the configuration of an angle of arrival estimation device according to a first embodiment. FIG. 2 is a diagram showing an example of the configuration of an array antenna according to the first embodiment, and a positional relationship between the array antenna and radio waves arriving at the array antenna. 1 is a block diagram showing an example of a hardware configuration of an angle of arrival estimation device according to a first embodiment; FIG. It is a flow chart which shows the flow of processing performed by the arrival angle estimation device concerning a 1st embodiment. FIG. 2 is a diagram illustrating a flowchart of a subroutine of teacher data generation processing performed in processing performed by the angle of arrival estimation device according to the first embodiment. FIG. 3 is a diagram showing a flowchart of a subroutine of a learning process performed in the process performed by the angle of arrival estimation device according to the first embodiment. FIG. 3 is a diagram showing a flowchart of a subroutine of inference processing performed in the processing performed by the angle of arrival estimating device according to the first embodiment. FIG. 2 is a block diagram showing the configuration of an angle of arrival estimation device according to a second embodiment. FIG. 7 is a block diagram showing the configuration of an angle of arrival estimation system and the internal configuration of an operating device according to a third embodiment. FIG. 7 is a block diagram showing the internal configuration of an arrival angle estimation device and a learning device according to a third embodiment. FIG. 7 is a block diagram showing an example of the hardware configuration of an arrival angle estimation device, a learning device, and an operating device according to a third embodiment. 12 is a flowchart showing the flow of processing performed by the angle of arrival estimation system according to the third embodiment. FIG. 2 is a block diagram showing the configuration of a learning procedure according to an embodiment of the present invention. It is a flowchart showing the flow of processing performed by the learning device according to one embodiment of the present invention. FIG. 1 is a block diagram showing the configuration of an angle of arrival estimation device according to an embodiment of the present invention. It is a flow chart which shows the flow of processing performed by an arrival angle estimation device concerning one embodiment of the present invention.

(First embodiment)
FIG. 1 is a block diagram showing the configuration of an angle of arrival estimation device 1 according to the first embodiment. The angle of arrival estimation device 1 includes a plurality of antenna elements 10-1 to 10-K, a receiving section 11, a mode vector storage section 12, a correlation matrix generation section 13, an input data generation section 14, a signal generation section 15, and a teacher data storage section. 16, a learning section 17, an inference section 18, and an angle of arrival calculation section 19. Here, K is an integer of 2 or more.

The plurality of antenna elements 10-1 to 10-K form an array antenna 10. The receiving unit 11 acquires K electrical digital reception signals from radio waves of predetermined wavelengths among the radio waves reaching the antenna elements 10-1 to 10-K. Here, the radio waves are, for example, radio waves for wireless communication. The receiving unit 11 generates a received signal vector including the acquired K received signals as elements. Note that the received signal is represented by a complex number. The mode vector storage unit 12 stores mode vector data generated in advance. Here, the mode vector is a vector indicating a direction, also called a steering vector.

Assume that the array antenna 10 is, for example, a circularly arrayed array antenna with K elements shown in FIG. 2 at equal intervals. The angle measured clockwise from the antenna element 10-1 is defined as the arrival angle of the radio wave. Here, it is assumed that a plane wave radio wave 90 arrives at an angle of arrival θ _i . The dashed-dotted line segments 91 and 92 are perpendicular to the traveling direction of the plane wave radio wave 90, and the line segment 91 corresponds to the k-th antenna element 10-k (here, k is 1 to The line segment 92 is a line segment that intersects the center position 95 of the array antenna 10. In this case, the distance l between the line segment 91 and the line segment 92 is expressed by the following equation (1).

In the above equation (1), α _k is the line segment connecting the center position 95 and the position of the antenna element 10-1, and the line segment connecting the center position 95 and the position of the antenna element 10-k. This is the angle that is formed. r is the radius of a circle that is the shape of the array antenna 10. The plane wave radio waves 90 on the line segment 91 have the same phase, and the plane wave radio waves 90 on the line segment 92 have the same phase. Therefore, when the center position 95 is used as a reference, the phase difference φ _k (θ _i ) at the position of the antenna element 10-k of the plane wave radio wave 90 arriving at the arrival angle θ _i is calculated by the following equation ( 2).

In the above equation (2), λ is the wavelength of the plane wave radio wave 90, and matches the wavelength predetermined in the receiving section 11. In this case, the element a _k,i of the mode vector of the array antenna 10 that can be theoretically derived is expressed as the following equation (3).

For example, when the number of arrival angles θ _i is N, that is, when i is an integer from 1 to N, K antenna elements 10-1 to 10- The mode vector of the array antenna formed by K will be expressed as a matrix having K×N elements. In this way, mode vectors can be calculated theoretically. However, in reality, it is not possible to construct an array antenna 10 that completely matches the theoretical mode vector. Therefore, there is a difference between the actual mode vector of the array antenna 10 and the theoretical mode vector. In order to prevent this difference from occurring, mode vector data to be stored in the mode vector storage unit 12 is generated using the following method.

When designing the array antenna 10 to be used in a real environment, it is common to use an electric field simulator or the like to perform a simulation evaluation of frequency bands that allow reception and angles of arrival that allow reception. During this simulation evaluation, for example, the antenna element 10-1 is used as a reference antenna element, and the amplitudes and phase differences of the other antenna elements 10-2 to 10-K with respect to the reference antenna element 10-1 are calculated. , the mode vector of the array antenna 10 used in the actual environment is calculated based on the calculated amplitude and phase difference. Furthermore, the mode vector of the array antenna 10 used in a real environment may be calculated based on measurement data obtained through actual measurements using an anechoic chamber. Data indicating the mode vector of the array antenna 10 used in the actual environment calculated in this way is written in advance into the mode vector storage unit 12 and stored. However, here, the wavelength of the wireless signal received by the array antenna 10 used in the actual environment is determined in advance to be one, and the mode vector storage unit 12 stores data of the mode vector corresponding to the one predetermined wavelength. It shall be. Here, the wavelength of the wireless signal received by the array antenna 10 used in a real environment is a wavelength predetermined in the receiving section 11.

Returning to FIG. 1, the signal generation unit 15 generates a transmission signal vector such that the signal strength of the transmission signal corresponding to the specified arrival angle is stronger than the transmission signal corresponding to other unspecified arrival angles. Here, the signal strength is, for example, the amplitude of a transmitted signal. Note that the transmission signal is represented by a complex number. For example, assume that the mode vector data stored in the mode vector storage unit 12 indicates θ ₁ to θ _N as the angle of arrival. In this case, when the angle of arrival specified by the signal generation unit 15 is θ _i , the amplitude of one transmission signal corresponding to the angle of arrival θ _i is set to “1”, and other angles of arrival θ ₁ to θ i are not specified. By setting the amplitudes of N-1 transmission signals corresponding to θ _i-1 , θ _i+1 to θ _N to "0", a transmission signal vector including N transmission signals as elements is generated. The number of arrival angles specified by the signal generation unit 15 may be plural. For example, if the angular range of the arrival angle is 360° and the resolution is “1°”, then N=360. In this case, if the number of arrival angles specified by the signal generation section 15 is one, the signal generation section 15 will generate 360 different transmission signal vectors. When the number of arrival angles specified by the signal generation unit 15 is two, the signal generation unit 15 generates ₃₆₀ C ₂ =64620 transmission signal vectors. Note that the range of the number of arrival angles specified by the signal generation unit 15 is determined as appropriate depending on the desired estimation accuracy and the like.

More specifically, the signal generation unit 15 specifies an angle of arrival based on a plurality of angles of arrival included in signal generation parameters given from the outside and the maximum number of angles of arrival to be specified, and specifies the angle of arrival of the specified angle of arrival. A transmit signal vector is generated for each combination. Here, the plurality of arrival angles included in the signal generation parameters are the arrival angles θ ₁ to θ _N indicated in the mode vector data stored in the mode vector storage unit 12. Further, N is an integer of 2 or more. Further, when the maximum number of angles of arrival to be specified is, for example, "3", the signal generation unit 15 generates a combination of angles of arrival that specifies one angle of arrival, a combination that specifies two angles of arrival, and 3. A transmit signal vector will be generated for each combination of arrival angles that specifies two arrival angles. The signal generation unit 15 generates identification information that can uniquely identify each of the transmitted signal vectors for each combination of generated angles of arrival, and uses the generated identification information to generate a transmitted signal vector and the transmitted signal vector. specified arrival angle data, which is data indicating the combination of arrival angles specified when

When the correlation matrix generation unit 13 receives the transmission signal vector generated by the signal generation unit 15, it generates a correlation matrix _Rxx from the transmission signal vector and the mode vector data stored in the mode vector storage unit 12. . Here, when the transmission signal vector is represented by a vector s(t), the correlation matrix generation unit 13 calculates the matrix S using the following equation (4).

In the above equation (4), the operator E[·] is an operator that means an expected value, and specifically, an operator that calculates an ensemble average. Further, the symbol H is a symbol indicating Hermitian transposition. Here, if the mode vector data stored in the mode vector storage unit 12 is represented by a matrix A, the correlation matrix generation unit 13 generates a correlation using the following equation (5) based on the matrix A and the matrix S. Calculate matrix _Rxx .

In the above equation (5), the matrix I is a K×K unit matrix, σ ² is the noise variance, and σ ² I in the second term on the right side indicates the noise component.

On the other hand, when the correlation matrix generation unit 13 captures the received signal vector generated by the reception unit 11, it generates the correlation matrix _Rxx based on the captured received signal vector. Here, when the received vector is represented by a vector x(t), the correlation matrix generation unit 13 calculates the correlation matrix _Rxx using the following equation (6).

The correlation matrix _Rxx generated by the correlation matrix generation unit 13 is a K×K matrix whether it is calculated based on the transmission signal vector and the mode vector or when it is calculated based on the reception signal vector.

The input data generation unit 14 generates input data from the correlation matrix _Rxx generated by the correlation matrix generation unit 13. The input data generation unit 14 generates a vector y shown in the following equation (7) as input data from the components of the lower triangular matrix of the correlation matrix _Rxx .

In the above equation (7), "u _p,q " is a variable indicating each element of the lower triangular matrix of the correlation matrix R _xx , and "p" appended to u is the variable of the lower triangular matrix of the correlation matrix R _xx . This is a variable that indicates the position of a row in the lower triangular matrix, and "q" is a variable that indicates the position of a column in the lower triangular matrix of the correlation matrix _Rxx . Therefore, “u _1,1 , u _2,2 , . . . , u _K,K ” becomes a diagonal element of the correlation matrix R _xx . Since the correlation matrix _Rxx is a Hermitian matrix, the diagonal elements are real numbers. The off-diagonal elements of the correlation matrix _Rxx are complex numbers, and the operator shown in the following equation (8) in equation (7) is an operator that extracts the numerical value of the real component of the complex number, and is expressed in the following equation (9). The operator shown is an operator that extracts the numerical value of the imaginary component of a complex number. That is, the input data generated by the input data generation unit 14 includes the numerical values of the diagonal elements included in the lower triangular matrix of the correlation matrix _Rxx , the numerical values of the real part of the off-diagonal elements included in the lower triangular matrix, It becomes a vector containing the imaginary part of the value as an element.

The teacher data storage unit 16 stores, for each piece of identification information, specified arrival angle data corresponding to the identification information and input data corresponding to the identification information. Hereinafter, data in which one set of identification information, designated arrival angle data corresponding to the identification information, and input data corresponding to the identification information are referred to as teacher data, and data stored in the teacher data storage unit 16. The entire set is called the teacher dataset.

The learning section 17 includes a learning processing section 20, a function approximator 21, and a learning model data storage section 23. The function approximator 21 is, for example, a multilayer neural network including an input layer, an intermediate layer including a plurality of layers, and an output layer. In the input layer, there are neurons whose number matches the number of elements included in the vector y that is input data. Each of the plurality of neurons in the input layer is associated with a different element included in the vector y. For example, according to the arrangement of vector y shown in equation (7), the first neuron in the input layer is associated with the first element "u _1,1 ", and the second neuron in the input layer is , is associated with the second element "u _2,2 ", and is similarly associated up to the last element of vector y. Each of the plurality of neurons in the input layer captures the numerical values of the elements of the vector y associated with it.

The output layer has a number of neurons depending on the range of angles of arrival to be estimated. For example, if the range to be estimated is 360° and the resolution is 1°, 360 neurons will be present in the output layer, and each of the 360 neurons will be 0°, 1°, 2°,... , 359°.

The number of layers in the intermediate layer is two or more, and the number of layers and the number of neurons included in each layer are determined in advance so as to have an appropriate scale depending on the number of data included in the teacher data.

The learning model data storage unit 23 stores weights and bias values applied to each of the intermediate layer and output layer neurons included in the function approximator 21. Hereinafter, the weight and bias data will be referred to as learning model data. The learning model data storage unit 23 stores in advance an initial value of the learning model data in an initial state.

The learning processing unit 20 applies the learning model data stored in the learning model data storage unit 23 to the function approximator 21 . That is, the learning processing unit 20 applies each of the weights and biases, which are learning model data stored in the learning model data storage unit 23, to the corresponding neurons of the function approximator 21. The function approximator 21 to which the learning model data is applied becomes the learning model 22. The learning processing unit 20 reads the teacher data one by one from the teacher data storage unit 16, and applies each numerical value included in the input data included in the read teacher data to the neuron of the input layer of the learning model 22 that corresponds to each numerical value included in the input data. give. The learning processing unit 20 generates a correct label from the designated arrival angle data included in the read teacher data according to the following equation (10).

The meaning of the above equation (10) will be explained. As described above, the number of neurons in the output layer of the function approximator 21 corresponds to the angular range of the angle of arrival to be estimated. Therefore, for example, if the range to be estimated is 360° and the resolution is 1°, 360 neurons will exist in the output layer. In this case, the learning processing unit 20 generates a vector z, which is data having 360 elements. Here, each of the elements of the vector z is associated with each neuron of the output layer of the function approximator 21 on a one-to-one basis. If the angle of arrival angle included in the specified arrival angle data included in the read teacher data is, for example, only "60°", the learning processing unit 20 selects a neuron of the output layer corresponding to "60°". The numerical values of the elements of the vector z are set under the condition that the numerical values of the elements of the vector z associated with are set to "1" and the numerical values of the other elements are set to "0". Similarly, if the specified angle of arrival data included in the teacher data that has been read includes two angles of arrival angle, for example, "60°" and "90°", the learning processing unit 20 performs " The numerical value of the element of the vector z associated with the output layer neuron corresponding to "60°" and the output layer neuron corresponding to "90°" is set to "1", and the numerical value of the other elements is set to "0". ” Set the numerical values of the elements of vector z. The learning processing unit 20 sets the vector z, which has a numerical value set under the condition, as the correct label.

Note that if the angular resolution expressed in the output layer is different from the angular resolution of the arrival angle included in the designated arrival angle data, the values of the elements of the vector z shall be determined as follows. . For example, assume that the resolution on the output layer side is "1 degree" and the resolution on the specified arrival angle data side is "0.1 degree." In this case, for example, the values of the elements of the vector z are determined after rounding off the specified arrival angle data to the nearest whole number. That is, if the specified arrival angle data includes "60.4°", the value of the element of the vector z associated with the neuron of the output layer corresponding to the rounded "60°" is set to "1". . On the other hand, if the specified arrival angle data includes "60.5°", the value of the element of the vector z associated with the neuron of the output layer corresponding to the rounded "61°" is set to "1". ”.

The learning processing unit 20 takes in output values output from each of the output layers of the learning model 22 by providing input data. Hereinafter, a vector whose elements are a plurality of output values will be referred to as output data. The learning processing unit 20 outputs the output data, the correct label corresponding to the input data when the output data was obtained, a predetermined loss function, and the learning model data stored in the learning model data storage unit 23. Based on this, new learning model data is calculated using, for example, an error backpropagation method so that the output data approaches the correct label. The learning processing unit 20 rewrites the learning model data stored in the learning model data storage unit 23 with the calculated new learning model data, and applies the calculated new learning model data to the function approximator 21. As a result, the learning model 22 will be updated.

The inference section 18 includes an inference processing section 30 and a function approximator 31. The function approximator 31 has the same configuration as the function approximator 21. The inference processing unit 30 applies the learned learning model data stored in the learning model data storage unit 23 to the function approximator 31 when the learning process performed by the learning processing unit 20 is completed. The function approximator 31 to which trained learning model data is applied becomes a trained learning model 32. The inference processing unit 30 obtains output data output from the trained learning model 32 by the input data generation unit 14 providing input data to the trained learning model 32 . The output data output by the trained learning model 32 is such that the value at the angle where the arriving radio wave exists is close to "1", and the value at the angle where there is no arriving radio wave is close to "0". become. Therefore, it can be said that the trained learning model 32 performs calculations so as to produce an evaluation function for evaluating the existence probability of radio waves for each angle of arrival, that is, a result similar to the evaluation function of the MUSIC method. In this case, if the output data is expressed in a graph with the vertical axis representing the numerical values of the elements included in the output data and the horizontal axis representing the angle of arrival, the output data indicates the spectrum related to the angle of arrival of the incoming radio waves. It can be considered as data.

As described above, the arrival angle calculation unit 19 regards the output data acquired by the inference processing unit 30 as data indicating the spectrum related to the arrival angle of the incoming radio wave, and performs a peak search on the spectrum to calculate the radio wave. Calculate the arrival angle of The angle of arrival calculated by the angle of arrival calculation unit 19 becomes the angle of arrival estimated by the angle of arrival estimating device 1. Hereinafter, the arrival angle estimated by the arrival angle estimating device 1 will be referred to as "estimated arrival angle."

FIG. 3 is a diagram showing an example of the hardware configuration of the angle of arrival estimation device 1. As shown in FIG. The angle of arrival estimation device 1 includes, for example, a CPU (Central Processing Unit) 201, a RAM (Random Access Memory) 202, a ROM (Read Only Memory) 203, an auxiliary storage device 204, a wireless communication module 205, and an input/output interface 206. It's a computer. The CPU 201, RAM 202, ROM 203, auxiliary storage device 204, wireless communication module 205, and input/output interface 206 are interconnected by a bus. Here, the auxiliary storage device 204 is, for example, an HDD (Hard Disk Drive) or an SDD (Solid State Drive). The wireless communication module 205 corresponds to the antenna elements 10-1 to 10-K and the hardware portion of the receiving section 11. A software portion that generates a received signal vector from a received signal of the receiving unit 11 by executing an application program stored in advance in the ROM 203 or the auxiliary storage device 204 by the CPU 201, the correlation matrix generating unit 13, and input data. The functional units include a generation unit 14, a signal generation unit 15, a learning processing unit 20, function approximators 21 and 31, an inference processing unit 30, and an angle of arrival calculation unit 19. Storage areas corresponding to each of the vector storage section 12, teacher data storage section 16, and learning model data storage section 23 are allocated. Input devices such as a keyboard and mouse, and output devices such as a liquid crystal display are connected to the input/output interface 206. The correlation matrix generation section 13, input data generation section 14, signal generation section 15, learning processing section 20, and inference processing section 30 take in data given from an input device via the input/output interface 206. The input data generation section 14, the learning processing section 20, and the angle of arrival calculation section 19 output data to the output device via the input/output interface 206.

(Processing in the first embodiment)
The processing performed by the arrival angle estimating device 1 will be described with reference to FIGS. 4 to 7. FIG. 4 is a flowchart showing the entire process performed by the arrival angle estimation device 1. Before the process shown in FIG. 4 is started, data of a mode vector generated in advance, which is a mode vector of the array antenna 10 used in a real environment, is written into the mode vector storage unit 12. Further, the initial value of the learning model data is written into the learning model data storage unit 23.

The operator of the angle of arrival estimating device 1 performs an operation to provide the angle of arrival estimating device 1 with a processing type signal indicating the type of “teacher data generation” using an input device connected to the angle of arrival estimating device 1 . The input device operated by the operator performs a process of indicating the type of "teacher data generation" to the correlation matrix generation unit 13, input data generation unit 14, learning processing unit 20, and inference processing unit 30 of the angle of arrival estimation device 1. Output type signal. Each of the correlation matrix generation unit 13, input data generation unit 14, learning processing unit 20, and inference processing unit 30 takes in a processing type signal indicating the type of “teacher data generation” output by the input device (step S1).

Each of the correlation matrix generation unit 13, input data generation unit 14, learning processing unit 20, and inference processing unit 30 determines the type of the captured processing type signal (step S2). When the learning processing unit 20 and the inference processing unit 30 determine that the type of the captured processing type signal is “teacher data generation”, they do not perform any processing. On the other hand, when the correlation matrix generation unit 13 and the input data generation unit 14 determine that the type of the captured processing type signal is “teacher data generation” (step S2, teacher data generation), the correlation matrix generation unit 13 and the input data generation unit 14 The subroutine of the teacher data generation process shown in (step S3) is started.

(Teacher data generation process in the first embodiment)
The correlation matrix generation unit 13 reads mode vector data from the mode vector storage unit 12, and switches the import source to the signal generation unit 15 (step Sa1-1). The input data generation unit 14 switches the output destination of the generated input data to the teacher data storage unit 16 (step Sa1-2). Note that the processing in step Sa1-1 and the processing in step Sa1-2 are processing performed after the correlation matrix generation unit 13 and the input data generation unit 14 perform the determination processing in step S2, and are performed in parallel. It will be held on.

The operator performs an operation to apply predetermined signal generation parameters to the angle of arrival estimation apparatus 1 using an input device connected to the angle of arrival estimation apparatus 1 . As described above, the signal generation parameters include N arrival angles representing each of the N arrival angles θ ₁ to θ _N shown in the mode vector data stored in the mode vector storage unit 12. , the maximum number of angles of arrival to specify. The input device operated by the operator outputs the signal generation parameter to the signal generation unit 15 of the angle of arrival estimation device 1 . The signal generation unit 15 takes in the signal generation parameters output by the input device (step Sa2).

The signal generation unit 15 specifies the angle of arrival within the range of the maximum number of specified arrival angles included in the signal generation parameter from the N angles of arrival included in the imported signal generation parameter. . The signal generation unit 15 generates specified arrival angle data indicating a specified combination of arrival angles. The signal generation unit 15 generates a transmission signal vector corresponding to the specified combination of arrival angles. The signal generation unit 15 generates identification information and adds the generated identification information to the designated arrival angle data and the generated transmission signal vector. The signal generation unit 15 writes the designated arrival angle data to which identification information has been added into the teacher data storage unit 16 and stores it. The signal generation unit 15 outputs the transmission signal vector to which identification information has been added to the correlation matrix generation unit 13 (step Sa3).

The correlation matrix generation unit 13 captures the transmission signal vector to which identification information is outputted from the signal generation unit 15, and calculates the matrix S using equation (4) based on the captured transmission signal vector. The correlation matrix generation unit 13 generates a correlation matrix _Rxx based on the calculated matrix S and the mode vector data read from the mode vector storage unit 12, that is, the matrix A, using equation (5). The correlation matrix generation unit 13 adds the identification information that was added to the captured transmission signal vector to the generated correlation matrix _Rxx . The correlation matrix generation unit 13 outputs the correlation matrix _Rxx to which identification information has been added to the input data generation unit 14 (step Sa4).

The input data generation unit 14 captures the correlation matrix R _xx to which identification information is output, which is output from the correlation matrix generation unit 13, and generates input data from the captured correlation matrix R _xx using equation (7) (step Sa5). The input data generation unit 14 detects a record of identification information that matches the identification information given to the correlation matrix _Rxx from the teacher data storage unit 16. The record detected by the input data generation unit 14 includes identification information and designated arrival angle data. Therefore, the input data generation unit 14 writes and stores the generated input data in the teacher data storage unit 16 so as to add the generated input data to the detected record. As a result, one piece of teacher data is completed (step Sa6).

When the process of step Sa3 is completed, the signal generation unit 15 specifies a pattern of combinations of arrival angles that have not yet been specified within the maximum number of specified arrival angles included in the imported signal generation parameters. Then, the process of step Sa3 is performed again. After the processing in step Sa3, the processing in steps Sa4 to Sa6 is performed again (loop La1s to La1e). For example, if the signal generation parameters include 360 arrival angles from θ ₁ to θ ₃₆₀ and the maximum number of arrival angles to be specified is “2”, the processing of loops La1s to La1e is as follows: ₃₆₀ C ₂ = This will be repeated 64,620 times.

When the signal generation unit 15 specifies all the combinations of arrival angles according to the signal generation parameters, the signal generation unit 15 ends the process. If the correlation matrix generation unit 13 does not output the correlation matrix _Rxx for a predetermined period of time, the input data generation unit 14 outputs a teacher data generation process completion notification signal to the output device (step Sa7). This completes the subroutine of the teacher data generation process, and returns to the process of the flowchart in FIG.

If the output device is, for example, a liquid crystal display, upon receiving the teacher data generation process completion notification signal output by the input data generation unit 14, the output device displays a message indicating the completion of the teacher data generation process on the screen of the liquid crystal display. do. When the operator of the angle-of-arrival estimation device 1 confirms the message indicating completion of the training data generation process, the operator of the angle-of-arrival estimation device 1 transmits a processing type signal indicating the type of “learning” to the input device connected to the angle-of-arrival estimation device 1. Perform the operation given to 1. The input device operated by the operator sends a processing type signal indicating the type of "learning" to the correlation matrix generation unit 13, input data generation unit 14, learning processing unit 20, and inference processing unit 30 of the angle of arrival estimation device 1. Output. Each of the correlation matrix generation unit 13, input data generation unit 14, learning processing unit 20, and inference processing unit 30 takes in a processing type signal indicating the type of “learning” output by the input device (step S1).

Each of the correlation matrix generation unit 13, input data generation unit 14, learning processing unit 20, and inference processing unit 30 determines the type of the captured processing type signal (step S2). When the correlation matrix generation unit 13, the input data generation unit 14, and the inference processing unit 30 determine that the type of the captured processing type signal is “learning,” they do not perform any processing. On the other hand, if the learning processing unit 20 determines that the type of the captured processing type signal is "learning" (step S2, learning), it starts the learning processing subroutine shown in FIG. 6 (step S4). .

(Learning process in the first embodiment)
The learning processing unit 20 reads the initial value learning model data stored in the learning model data storage unit 23 and applies the read learning model data to the function approximator 21 . As a result, the learning model 22 is generated (step Sb1). The learning processing section 20 reads any one of the teacher data from the teacher data storage section 16. For example, if the identification information included in the teacher data is indicated by consecutive positive integer numbers, the teacher data containing the smallest number as the identification information is read out (step Sb2). The learning processing unit 20 provides input data included in the teacher data to the input layer of the learning model 22. The learning processing unit 20 obtains output data output from the output layer of the learning model 22 by providing input data (step Sb3).

The learning processing unit 20 generates a correct label from the specified arrival angle data included in the teacher data according to the condition of equation (10). The learning processing unit 20 generates new learning model data based on the acquired output data, the generated correct label, a predetermined loss function, and the learning model data stored in the learning model data storage unit 23. Calculate (step Sb4). The learning processing unit 20 rewrites the learning model data stored in the learning model data storage unit 23 with the calculated new learning model data, and applies the calculated new learning model data to the function approximator 21. As a result, the learning model 22 is updated (step Sb5).

The learning processing unit 20 reads teacher data that is not subject to learning processing from the teacher data storage unit 16. As described above, when the identification information included in the teacher data is a number, the teacher data including the identification information with the next highest number than the identification information included in the teacher data targeted for the previous learning process is read. The learning processing unit 20 performs the processing of steps Sb2 to Sb5 again based on the read teacher data. The learning processing unit 20 performs the processing of steps Sb2 to Sb5 (loop Lb1s to Lb1e) until there is no teacher data that is not subject to learning processing in the teacher data storage unit 16. When the processing of loops Lb1s to Lb1e is completed, the learning processing unit 20 outputs a learning processing completion notification signal to the output device (step Sb6). This completes the subroutine of the learning process and returns to the process of the flowchart in FIG.

When the output device receives the learning processing completion notification signal output from the learning processing section 20, it displays a message indicating the completion of the learning processing on the screen. When the operator of the angle of arrival estimation device 1 confirms the message indicating the completion of the learning process, the operator of the angle of arrival estimation device 1 sends a processing type signal indicating the type of “inference” to the angle of arrival estimation device 1 using the input device connected to the angle of arrival estimation device 1. Perform the giving operation. The input device operated by the operator sends a processing type signal indicating the type of "inference" to the correlation matrix generation unit 13, input data generation unit 14, learning processing unit 20, and inference processing unit 30 of the angle of arrival estimation device 1. Output. Each of the correlation matrix generation unit 13, input data generation unit 14, learning processing unit 20, and inference processing unit 30 takes in a processing type signal indicating the type of “inference” output by the input device (step S1).

Each of the correlation matrix generation unit 13, input data generation unit 14, learning processing unit 20, and inference processing unit 30 determines the type of the captured processing type signal (step S2). If the learning processing unit 20 determines that the type of the captured processing type signal is "inference", it does not perform any processing. When the correlation matrix generation unit 13, the input data generation unit 14, and the inference processing unit 30 determine that the type of the captured processing type signal is “inference” (step S2, inference), the inference shown in FIG. 7 is performed. A processing subroutine is started (step S5).

(Inference processing in the first embodiment)
The correlation matrix generation unit 13 switches the import source to the reception unit 11 (step Sc1-1). The input data generation unit 14 switches the output destination of the generated input data to the inference processing unit 30 (step Sc1-2). The inference processing unit 30 reads learned learning model data from the learning model data storage unit 23 and applies the read learned learning model data to the function approximator 31 . As a result, a trained learning model 32 is generated (step Sc1-3). Note that the processing in step Sc1-1, the processing in step Sc1-2, and the processing in step Sc1-3 are such that the correlation matrix generation section 13, the input data generation section 14, and the inference processing section 30 perform the processing in step S2. This process is performed after the determination process of , and is performed in parallel.

The receiving unit 11 acquires K electric digital reception signals from radio waves of predetermined wavelengths among the radio waves reaching the antenna elements 10-1 to 10-K, and uses the acquired K reception signals as elements. Generate a received signal vector containing: The receiving unit 11 outputs the generated received signal vector to the correlation matrix generating unit 13 (step Sc2). The correlation matrix generation unit 13 captures the received signal vector output by the receiving unit 11, and generates a correlation matrix _Rxx based on the captured received signal vector using equation (6). The correlation matrix generation unit 13 outputs the generated correlation matrix _Rxx to the input data generation unit 14 (step Sc3).

The input data generation unit 14 takes in the correlation matrix _Rxx output from the correlation matrix generation unit 13, and generates input data from the taken in correlation matrix _Rxx according to equation (7). The input data generation unit 14 outputs the generated input data to the inference processing unit 30 (step Sc4). The inference processing unit 30 captures the input data output by the input data generation unit 14 and provides the captured input data to the input layer of the trained learning model 32. The inference processing unit 30 obtains output data output from the output layer of the trained learning model 32 by providing input data. The inference processing unit 30 outputs the acquired output data to the angle of arrival calculation unit 19 (step Sc5).

The angle of arrival calculation unit 19 captures the output data output by the inference processing unit 30, regards the captured output data as data indicating a spectrum related to the angle of arrival of an incoming radio wave, and performs a peak search on the spectrum. The estimated arrival angle of the radio wave is calculated by The angle of arrival calculation unit 19 outputs data indicating the calculated estimated angle of arrival to the output device. When the output device receives the data indicating the estimated angle of arrival output by the angle of arrival calculation unit 19, it displays the data indicating the estimated angle of arrival on the screen (step Sc6).

While the receiving unit 11 is receiving radio waves, the processes of steps Sc2 to Sc6 are repeatedly performed (loop Lc1s to Lc1e). When the receiving unit 11 no longer receives radio waves, the subroutine of the inference process ends, and the process shown in FIG. 4 also ends.

In the first embodiment described above, the signal generation unit 15 generates a transmission signal vector such that the signal strength of the transmission signal corresponding to the specified arrival angle is stronger than the transmission signal corresponding to other arrival angles. The correlation matrix generation unit 13 generates a correlation matrix based on the transmission signal vector and the mode vector of the real environment array antenna used in the real environment, that is, the array antenna 10. The input data generation unit 14 generates input data from the elements of the correlation matrix. The learning unit 17 receives the input data and generates the learning model 22 by performing a learning process using teacher data whose correct label is data indicating the angle of arrival corresponding to the input data.

By having the above-described configuration, it is possible to generate correlation matrices R _xx for each of a plurality of angles of arrival at fine angular intervals by using the mode vector of the array antenna 10 used in a real environment. The generated correlation matrix R _xx for each of the plurality of arrival angles with fine angular intervals becomes a correlation matrix R _xx that reflects the characteristics of the array antenna 10 used in the real environment. By generating training data from the correlation matrix _Rxx , it is possible to generate training data in which each of a plurality of angles of arrival at small angular intervals is used as a correct label, without actually measuring the received signal. By performing a learning process using the teacher data, it becomes possible to generate a trained learning model 32 with high estimation accuracy for various angles of arrival. Since the trained learning model 32 reflects the characteristics of the array antenna 10 used in the real environment in the training data, there is no need to perform relearning that is required when using the theoretical value mode vector, and it can be used in the real world. It becomes possible to estimate the arrival angle of radio waves in the environment with high accuracy.

When using theoretical mode vectors as in Non-Patent Documents 1 and 2, it is necessary to perform re-learning in order to adapt the generated learning model to the actual environment. The training data required for this relearning needs to be generated by actually measuring a large number of received signals for each of multiple angles of arrival at small angular intervals in a real environment, so the amount of training data is also large. . In contrast, the trained learning model 32 generated in the first embodiment is already suitable for the actual environment. Therefore, when retraining the trained learning model 32 to make it more suitable for the real environment using training data generated using actual received signals, it is necessary for retraining when using the mode vector of the theoretical value. Compared to the number of training data, it is possible to further improve the estimation accuracy by relearning a small number of training data. That is, by using the arrival angle estimating device 1 of the first embodiment, in order to generate training data for relearning, the number of received signals actually measured in the real environment can be changed by using the mode vector of the theoretical value. It is possible to reduce the amount by comparison.

(Second embodiment)
In general, when the distance between devices that transmit and receive radio waves is shorter than the distance assumed at the time of design, the diagonal component values of the correlation matrix _Rxx become large, etc. It is thought that the distribution of input data generated by will change. Similarly, if the distance between devices that transmit and receive radio waves is longer than the distance assumed at the time of design, the distribution of input data generated by the input data generation unit 14 of the first embodiment is considered to change. . In the general machine learning field, it is known that estimation accuracy decreases when the distribution of input data differs between learning processing and inference processing. On the other hand, Non-Patent Documents 1 and 2 evaluate performance using the signal-to-noise ratio as an index, and do not disclose measures against the effects that occur depending on the distance between devices that transmit and receive radio waves. do not have. The second embodiment includes a configuration that prevents the accuracy of estimating the angle of arrival from decreasing even when the distance between devices that transmit and receive radio waves deviates from the distance assumed at the time of design.

FIG. 8 is a block diagram showing the configuration of an angle of arrival estimation device 1a according to the second embodiment. In the second embodiment, the same configurations as those in the first embodiment are given the same reference numerals, and different configurations will be described below. The angle of arrival estimation device 1a includes a plurality of antenna elements 10-1 to 10-K, a receiving section 11, a mode vector storage section 12, a correlation matrix generation section 13, an input data generation section 14a, a signal generation section 15, and a teacher data storage section. 16, a learning section 17, an inference section 18, and an angle of arrival calculation section 19.

The input data generation unit 14a has the same configuration as the input data generation unit 14 of the first embodiment except for the configuration shown below. The input data generation unit 14a extracts the elements of the lower triangular matrix of the correlation matrix _Rxx , generates a vector y using equation (7), and shows the following equation (11) for each element of the generated vector y. Perform normalization.

The normalization shown in equation (11) above is normalization called Min-MaX-Normalization. In equation (11), “v _j ” on the right side is a variable indicating each element of vector y. Here, j is the index value of the element of vector y. max(v) on the right side is the maximum value of the elements of vector y, and min(v) is the minimum value of the elements of vector y. v′ _j on the left side of equation (11) is the value of the element of vector y after normalization.

The angle-of-arrival estimating device 1b of the second embodiment has the same hardware configuration as the angle-of-arrival estimating device 1 of the first embodiment shown in FIG. 3, except for the configuration shown below. . In the second embodiment, the application program stored in advance in the ROM 203 or the auxiliary storage device 204 is executed by the CPU 201, so that the functional unit of the input data generation unit 14a is activated instead of the input data generation unit 14. configured. The input data generation unit 14a configured in this manner takes in data provided from an input device via the input/output interface 206, and outputs the data to the output device via the input/output interface 206.

The processing in the second embodiment is the same as that shown in FIGS. 4 to 7 in the first embodiment except for the processing described below. However, in this processing, the processing that was performed by the input data generation section 14 is now performed by the input data generation section 14a. The difference in processing between the first embodiment and the second embodiment is the processing in step Sa5 in FIG. 5 and step Sc4 in FIG. Instead of generating input data from the correlation matrix _Rxx generated by the generation unit 13 using equation (7), the following processing is performed. That is, the input data generation unit 14a generates a vector y according to equation (7) from the correlation matrix _Rxx , and normalizes the elements of the generated vector y by applying equation (11). The subsequent vector y is used as input data.

As a result, in the second embodiment, by applying the normalization shown in equation (11) to the vector y, input data can be generated even if the distance between devices that transmit and receive radio waves changes. Since the difference in distribution in the input data generated by the unit 14a becomes smaller, it becomes possible to alleviate the decrease in the accuracy of estimating the angle of arrival.

Note that in the second embodiment described above, the input data generation unit 14a does not perform the normalization shown in equation (11) during the process of step Sa5 in FIG. Alternatively, input data may be generated from the correlation matrix R _xx using equation (7), and the generated input data may be written and stored in the teacher data storage unit 16. In this case, in the process of step Sb2 of the learning process in FIG. 6, the learning processing unit 20 does not directly read the teacher data from the teacher data storage unit 16, but performs the following process, for example. The learning processing section 20 outputs a teacher data read instruction signal to the input data generation section 14a. When the input data generating section 14a receives the teaching data read instruction signal from the learning processing section 20, it reads out any one teaching data that is not a target of learning processing from the teaching data storage section 16. The input data generation unit 14a uses, as input data, the input data included in the read teacher data, that is, the vector y that is normalized by applying equation (11) to the elements of the vector y. The input data generation unit 14a outputs the input data, which is the normalized vector y, and the identification information and specified arrival angle data included in the read teacher data to the learning processing unit 20.

In the first and second embodiments described above, the plurality of arrival angles included in the signal generation parameters given to the signal generation section 15 are the arrival angles θ indicated in the mode vector data stored in the mode vector storage section 12. ₁ to _θN . Therefore, when the mode vector data to be stored in the mode vector storage unit 12 is specified, the arrival angles θ ₁ to θ _N to be included in the signal generation parameters are also specified accordingly. Therefore, when writing mode vector data to the mode vector storage section 12, the maximum number of angles of arrival to be specified is determined, and then the signal generation parameters are stored in the internal storage section of the signal generation section 15. Good too. In this case, the signal generation unit 15 receives, for example, an instruction signal to start the teacher data processing given from the outside, starts the processing of step Sa2, and in the started processing of step Sa2, the signal generation section 15 receives the signal generation parameter given from the outside. Instead of reading the signal generation parameters stored in the internal storage area.

(Third embodiment)
FIG. 9 is a block diagram showing the overall configuration of the angle of arrival estimation system 100 and the internal configuration of the operating device 3 according to the third embodiment. In the third embodiment, the same configurations as those in the first embodiment are given the same reference numerals, and different configurations will be described below.

The angle of arrival estimation system 100 includes an angle of arrival estimation device 1b, a learning device 2, an operating device 3, and a communication network 4. The communication network 4 interconnects the angle of arrival estimation device 1b, the learning device 2, and the operating device 3. The operating device 3 includes a communication processing section 51, an operating section 52, an output section 53, a control section 54, and a storage section 55. The communication processing unit 51 connects to the communication network 4 and sends and receives data. The operation unit 52 includes input devices such as a mouse and a keyboard, and receives operations from an operator. The output unit 53 includes an output device such as a liquid crystal display, and outputs data. The control unit 54 performs various control processes in the operating device 3. The storage unit 55 stores in advance data of the mode vector of the array antenna 10, which is generated in advance by the method described in the first embodiment and is used in the actual environment provided by the angle of arrival estimating device 1b. However, in the third embodiment, unlike the first and second embodiments, it is possible to switch the wavelength, that is, the center frequency, of the radio wave for which the angle of arrival is estimated in the angle of arrival estimating device 1b. . Therefore, the storage unit 55 stores in advance mode vector data corresponding to each center frequency in association with each of a plurality of center frequencies selected in advance.

FIG. 10 is a block diagram showing the internal configuration of the angle of arrival estimation device 1b and the learning device 2 in the angle of arrival estimation system 100. The angle of arrival estimation device 1b includes a plurality of antenna elements 10-1 to 10-K, a reception section 11a, a correlation matrix generation section 13-1, an input data generation section 14-1, an inference section 18a, an angle of arrival calculation section 19, and It includes a communication processing section 41-1.

The receiving unit 11a has the same configuration as the receiving unit 11 of the first embodiment except for the configuration described below. The receiving unit 11a selects radio waves of a wavelength corresponding to the center frequency specified in the reception parameters given from the outside via the communication processing unit 41-1 from among the radio waves reaching the antenna elements 10-1 to 10-K. Extract. The receiving unit 11a acquires K electric digital reception signals from the extracted radio waves, and generates a reception signal vector including the acquired K reception signals as elements. The correlation matrix generation unit 13-1 has the same configuration as the correlation matrix generation unit 13 of the first embodiment in a state where a processing type signal indicating the type of “inference” is provided. In other words, the correlation matrix generation unit 13-1 does not receive the processing type signal, the acquisition source is fixed to the reception unit 11a, and the correlation matrix generation unit 13-1 generates the correlation matrix _Rxx based on the received signal vector.

The input data generation unit 14-1 has the same configuration as the input data generation unit 14 of the first embodiment in a state where a processing type signal indicating the type of “inference” is applied. In other words, the input data generation section 14-1 does not receive the processing type signal and has the same configuration as the input data generation section 14 whose output destination is fixed to the inference processing section 30.

The inference section 18a includes an inference processing section 30a and a function approximator 31. The inference processing unit 30a has the same configuration as the inference processing unit 30 of the first embodiment except for the configuration shown below. The inference processing unit 30a does not receive a processing type signal like the inference processing unit 30 of the first embodiment, but when given learned model data from the outside via the communication processing unit 41-1, The given trained learning model data is applied to the function approximator 31. The communication processing unit 41-1 connects to the communication network 4 and sends and receives data.

The learning device 2 includes a mode vector storage section 12, a correlation matrix generation section 13-2, an input data generation section 14-2, a signal generation section 15, a teacher data storage section 16, a learning section 17a, and a communication processing section 41-2. Be prepared. The mode vector storage unit 12 stores mode vector data corresponding to the center frequency specified in the reception parameter, among the plurality of mode vector data stored in the storage unit 55 of the operating device 3.

The correlation matrix generation unit 13-2 has the same configuration as the correlation matrix generation unit 13 of the first embodiment in a state where a processing type signal indicating the type of “teacher data generation” is given. Be prepared. In other words, the correlation matrix generation unit 13-2 does not receive the processing type signal, the acquisition source is fixed to the signal generation unit 15, and the transmission signal vector and the mode vector stored in the mode vector storage unit 12 are A correlation matrix _Rxx is generated based on the data.

The input data generation section 14-2 has the same configuration as the input data generation section 14 of the first embodiment in a state where a processing type signal indicating the type of "teacher data generation" is given. Be prepared. In other words, the input data generation section 14-1 does not receive the processing type signal and has the same configuration as the input data generation section 14 whose output destination is fixed to the teacher data storage section 16.

The learning section 17a includes a learning processing section 20a, a function approximator 21, and a learning model data storage section 23. The learning processing unit 20a has the same configuration as the learning processing unit 20 of the first embodiment except for the configuration shown below. The learning processing section 20a does not receive a processing type signal like the learning processing section 20 of the first embodiment, but when receiving a learning processing start instruction signal from the outside via the communication processing section 41-1, the learning processing section 20a performs the process shown in FIG. The subroutine of the learning process shown in FIG. 1 starts with the learning processing section 20 replaced by the learning processing section 20a. The communication processing unit 41-2 connects to the communication network 4 and sends and receives data.

Note that in the first embodiment, the input data generation section 14, the learning processing section 20, and the arrival angle calculation section 19 output the data to the output device, but the arrival of the third embodiment The angle of arrival calculation unit 19 of the angle estimating device 1b outputs data indicating the estimated angle of arrival to the communication processing unit 41-1 instead of the output device. The input data generation unit 14-2 of the learning device 2 outputs the teacher data generation process completion notification signal to the communication processing unit 41-2 instead of the output device, and the learning processing unit 20a outputs the learning process completion notification signal to the output device. Instead, it is output to the communication processing unit 41-2.

FIG. 11 is a diagram showing the hardware configuration of the angle of arrival estimation device 1b, the learning device 2, and the operating device 3. Note that functional units indicated by the same reference numerals in FIG. 3 and FIG. 11 showing the hardware configuration of the first embodiment do not mean the same hardware, but the same type of hardware. For example, the CPUs 201 included in the angle-of-arrival estimating device 1 in FIG. 3, the angle-of-arrival estimating device 1b in FIG. Neither does it mean that the CPUs have the same specs.

The angle of arrival estimation device 1b is, for example, a computer including a CPU 201, a RAM 202, a ROM 203, an auxiliary storage device 204, a wireless communication module 205, and a communication module 206. The CPU 201, RAM 202, ROM 203, auxiliary storage device 204, wireless communication module 205, and communication module 206 are interconnected by a bus. The wireless communication module 205 corresponds to the hardware portion of the antenna elements 10-1 to 10-K and the receiving section 11a. The communication module 206 corresponds to the hardware part of the communication processing section 41-1. A software portion that generates a received signal vector from a received signal of the receiving unit 11a by executing an application program stored in advance in the ROM 203 or the auxiliary storage device 204 by the CPU 201, a correlation matrix generating unit 13-1, Functional units of the software portion include the input data generation unit 14-1, the inference processing unit 30a, the function approximator 31, the arrival angle calculation unit 19, and the communication processing unit 41-1.

The learning device 2 is, for example, a computer including a CPU 201, a RAM 202, a ROM 203, an auxiliary storage device 204, and a communication module 206. The CPU 201, RAM 202, ROM 203, auxiliary storage device 204, and communication module 206 are interconnected by a bus. The communication module 206 corresponds to the hardware part of the communication processing section 41-2. By executing the application program stored in advance in the ROM 203 or the auxiliary storage device 204 by the CPU 201, the correlation matrix generation section 13-2, the input data generation section 14-2, the signal generation section 15, and the learning processing section 20a are executed. , the function approximator 21, and the communication processing unit 41-2. A corresponding storage area is allocated to each of the storage units 23.

The operating device 3 is, for example, a computer including a CPU 201, a RAM 202, a ROM 203, an auxiliary storage device 204, a communication module 206, an input device 207, and an output device 208. The input device 207 is, for example, a mouse, a keyboard, or the like. The output device 208 is, for example, a liquid crystal display. The communication module 206 corresponds to the hardware part of the communication processing section 51. By executing an application program stored in advance in the ROM 203 or the auxiliary storage device 204 by the CPU 201, the operation unit 52 including the input device 207, the output unit 53 including the output device 208, the control unit 54, and communication processing A functional unit of the software portion of the unit 51 is configured, and a storage area corresponding to the storage unit 55 is allocated in the RAM 203 or the auxiliary storage device 204.

The communication module 206 of the angle of arrival estimation device 1b, the communication module 206 of the learning device 2, and the communication module 206 of the operating device 3 are interconnected by the communication network 4. Note that the operating device 3 does not include an input device 207 and an output device 208, but instead includes an input/output interface 206 shown in FIG. The configuration may be such that they are connected.

(Processing in the third embodiment)
The flow of processing performed by the angle of arrival estimation system 100 of the third embodiment will be described with reference to FIG. 12. Before the process of the sequence diagram shown in FIG. 12 is started, the mode vector generated in advance by the method described in the first embodiment is stored in the storage unit 55 of the operating device 3, and is provided in the angle of arrival estimation device 1b. Each mode vector data for each center frequency of the array antenna 10 used in the actual environment is written in association with the corresponding center frequency. Note that the range of center frequencies that can be set by reception parameters in the reception unit 11a of the angle of arrival estimating device 1b is predetermined by specifications or the like. Therefore, data of mode vectors corresponding to some center frequencies selected in advance from the range are stored in the storage unit 55 in advance. The operator selects one of the center frequencies corresponding to the plurality of mode vector data stored in the storage unit 55. Further, the initial value of the learning model data is written into the learning model data storage section 23 of the learning device 2.

The operator specifies a center frequency on the operating unit 52 of the operating device 3 and performs an operation to transmit mode vector data to the learning device 2. Since the received operation involves reading the mode vector from the storage unit 55, the operation unit 52 specifies the transmission destination as the learning device 2 and outputs a mode vector transmission instruction signal including the specified center frequency to the control unit 54. . When the control unit 54 receives the mode vector transmission instruction signal from the operation unit 52, it reads data of the mode vector corresponding to the center frequency included in the received mode vector transmission instruction signal from the storage unit 55. The control unit 54 includes the read mode vector data in the mode vector transmission instruction signal received from the operation unit 52 and outputs the mode vector transmission instruction signal to the communication processing unit 51 . The communication processing unit 51 takes in the mode vector transmission instruction signal output by the control unit 54. The communication processing unit 51 includes the data of the mode vector included in the mode vector transmission instruction signal that has been taken in, sets the learning device 2 specified as the transmission destination in the mode vector transmission instruction signal as the destination, and sets the type of data to " mode vector" is generated. The communication processing unit 51 sends the generated transmission data to the communication network 4. The communication network 4 transfers the transmission data to the learning device 2 according to the destination included in the transmission data.

The communication processing unit 41-2 of the learning device 2 receives the transmission data transmitted by the operating device 3 from the communication network 4. Since the type of data included in the transmission data is a "mode vector", the communication processing unit 41-2 writes the data of the mode vector included in the transmission data into the mode vector storage unit 12 and stores it (step Sd1).

The operator performs an operation on the operation unit 52 of the operating device 3 to designate a signal generation parameter, and also performs an operation to set the destination of the designated signal generation parameter to the learning device 2 . Here, the signal generation parameters specified by the operator include the angle of arrival θ ₁ to θ _N indicated in the mode vector data transmitted to the learning device 2 in the process of step Sd1, and the maximum angle of arrival specified. The number will be included. The operation unit 52 outputs to the communication processing unit 51 a signal generation parameter transmission instruction signal that includes the specified signal generation parameter and specifies the transmission destination as the learning device 2 . The communication processing section 51 takes in the signal generation parameter transmission instruction signal output from the operation section 52. The communication processing unit 51 includes the signal generation parameters included in the imported signal generation parameter transmission instruction signal, sets the learning device 2 specified as the transmission destination in the signal generation parameter transmission instruction signal as the destination, and determines the type of data. Transmission data to be used as "signal generation parameters" is generated. The communication processing unit 51 sends the generated transmission data to the communication network 4. The communication network 4 transfers the transmission data to the learning device 2 according to the destination included in the transmission data.

The communication processing unit 41-2 of the learning device 2 receives the transmission data transmitted by the operating device 3 from the communication network 4. Since the type of data included in the transmission data is a "signal generation parameter," the communication processing unit 41-2 outputs the signal generation parameter included in the transmission data to the signal generation unit 15 (step Sd2). As a result, the same process as steps Sa2 to Sa7 of the subroutine of the teacher data generation process shown in FIG. 5 is started as the subroutine of the teacher data generation process of the third embodiment. However, in the subroutine of the teacher data generation process of the third embodiment, the correlation matrix generation section 13 is replaced with the correlation matrix generation section 13-2 in the processing of steps Sa2 to Sa7 in FIG. The data generation unit 14-2 performs the processing that is replaced by the data generation unit 14-2. In addition, in the process of step Sa4, when the correlation matrix generation unit 13-2 first takes in the transmission signal vector to which the identification information outputted by the signal generation unit 15 is attached, the mode vector is stored in the mode vector storage unit 12. Data reading processing is added (step Sd3).

In the process of step Sa7 of the teacher data generation process of step Sd3, the input data generation unit 14-2 outputs a teacher data generation process completion notification signal to the communication processing unit 41-2. Upon receiving the teacher data generation process completion notification signal from the input data generation unit 14-2, the communication processing unit 41-2 sends a message with the destination as the operating device 3 and the data type as "teacher data generation process completion notification signal." It generates data and sends the generated transmission data to the communication network 4. The communication network 4 transfers the transmission data to the operating device 3 according to the destination included in the transmission data. The communication processing unit 51 of the operating device 3 receives the transmission data transmitted by the learning device 2 from the communication network 4 . Since the type of data included in the transmission data is a “teacher data generation process completion notification signal”, the communication processing unit 51 outputs a teacher data generation process completion notification signal to the output unit 53. When the output unit 53 receives the teacher data generation processing completion notification signal from the communication processing unit 51, it displays a message indicating the completion of the teacher data generation process on the screen of its own output device such as a liquid crystal display (step Sd4).

When the operator confirms the message indicating completion of the teacher data generation process displayed on the screen, the operator operates the operation unit 52 of the operation device 3 to transmit a learning process start instruction signal to the learning device 2. The operation unit 52 outputs a learning process start instruction signal specifying the learning device 2 as the destination to the communication processing unit 51. The communication processing unit 51 takes in the learning process start instruction signal output by the operation unit 52, sets the learning device 2 specified as the transmission destination in the received signal generation parameter transmission instruction signal as the destination, and sets the data type to “learning process”. The transmission data is generated as a "start instruction signal". The communication processing unit 51 sends the generated transmission data to the communication network 4. The communication network 4 transfers the transmission data to the learning device 2 according to the destination included in the transmission data. The communication processing unit 41-2 of the learning device 2 receives the transmission data transmitted by the operating device 3 from the communication network 4. Since the type of data included in the transmission data is a "learning process start instruction signal", the communication processing unit 41-2 outputs a learning process start instruction signal to the learning process unit 20a (step Sd5). As a result, the same process as the learning process subroutine shown in FIG. 6 is started as the learning process subroutine of the third embodiment. However, in the subroutine of the learning process of the third embodiment, the learning process section 20 in the process of FIG. 6 is replaced with the learning process section 20a (step Sd6).

In the processing of step Sb6 of the learning processing subroutine of step Sd6, the learning processing section 20a outputs a learning processing completion notification signal to the communication processing section 41-2. When the communication processing unit 41-2 receives the learning processing completion notification signal from the learning processing unit 20a, it generates transmission data with the destination as the operating device 3 and the data type as “learning processing completion notification signal”. The transmission data is sent to the communication network 4. The communication network 4 transfers the transmission data to the operating device 3 according to the destination included in the transmission data. The communication processing unit 51 of the operating device 3 receives the transmission data transmitted by the learning device 2 from the communication network 4 . Since the type of data included in the transmission data is a “learning process completion notification signal”, the communication processing unit 51 outputs a learning process completion notification signal to the output unit 53. When the output unit 53 receives the learning process completion notification signal from the communication processing unit 51, it displays a message indicating the completion of the learning process on the screen (step Sd7).

When the communication processing unit 41-2 receives the learning processing completion notification signal from the learning processing unit 20a, it reads out the learned learning model data stored in the learning model data storage unit 23. The communication processing unit 41-2 generates transmission data that includes the read learning model data, has the destination as the arrival angle estimation device 1b, and has the data type as “learning model data,” and transmits the generated transmission data to the communication network 4. Send to. The communication network 4 transfers the transmission data to the arrival angle estimating device 1b according to the destination included in the transmission data. The communication processing unit 41-1 of the angle of arrival estimation device 1b receives the transmission data transmitted by the learning device 2 from the communication network 4. Since the type of data included in the transmission data is "learning model data," the communication processing unit 41-1 reads the learned learning model data from the received transmission data and infers the read learning model data. It is output to the processing section 30a. The inference processing unit 30a takes in the trained learning model data output by the communication processing unit 41-1, and applies the learned learning model data to the function approximator 31. As a result, a trained learning model 32 is generated (step Sd8).

Note that the processing in step Sd7 and the processing in step Sd8, which are performed when the communication processing unit 41-2 receives a learning process completion notification signal from the learning processing unit 20a, are the same as those in step Sd7, as shown in FIG. The process of step Sd8 may be performed after the process, or the process of step Sd7 may be performed after the process of step Sd8.

The operator performs an operation on the operation unit 52 of the operation device 3 to generate reception parameters including a center frequency that matches the center frequency specified in step Sd1, and also selects the destination of the reception parameters from the angle of arrival estimation device 1b. Perform the operation specified as . The operation unit 52 outputs to the communication processing unit 51 a reception parameter transmission instruction signal including the generated reception parameters and data indicating that the transmission destination is the arrival angle estimation device 1b. The communication processing section 51 takes in the reception parameter transmission instruction signal output from the operation section 52. The communication processing unit 51 includes the reception parameters included in the received reception parameter transmission instruction signal, sets the arrival angle estimation device 1b specified as the transmission destination in the reception parameter transmission instruction signal as the destination, and sets the type of data as " Generates transmission data as "reception parameters". The communication processing unit 51 sends the generated transmission data to the communication network 4. The communication network 4 transfers the transmission data to the arrival angle estimating device 1b according to the destination included in the transmission data.

The communication processing unit 41-1 of the angle of arrival estimation device 1b receives the transmission data transmitted by the operating device 3 from the communication network 4. Since the type of data included in the transmission data is "reception parameter," the communication processing unit 41-1 outputs the reception parameter included in the transmission data to the reception unit 11a. The receiving section 11a takes in the receiving parameters output by the communication processing section 41-1. The receiving unit 11a reads the center frequency included in the received reception parameters, and sets the read center frequency in the wireless communication module 205, which is its own hardware part (step Sd9). As a result, the same process as the loop Lc1s to Lc1e of the inference process subroutine shown in FIG. 7 is started as the inference process subroutine of the third embodiment. However, in the inference process of the third embodiment, the following process is performed in the process of step Sc2. That is, the receiving unit 11a extracts radio waves having a wavelength corresponding to the center frequency specified in the received reception parameters from among the radio waves reaching the antenna elements 10-1 to 10-K. The receiving unit 11a acquires K electric digital reception signals from the extracted radio waves, and generates a reception signal vector including the acquired K reception signals as elements. Regarding the processing of steps Sc3 to Sc6 other than step Sc2, the correlation matrix generation section 13 is replaced with the correlation matrix generation section 13-1, the input data generation section 14 is replaced with the input data generation section 14-1, and the inference processing section 30 is replaced with the input data generation section 14-1. The replaced process is performed by the inference processing unit 30a (step Sd10).

In the repeated processing of the loops Lc1s to Lc1e of the inference processing subroutine of step Sd10, each time the processing of step Sc6 is performed, the angle of arrival calculation unit 19 sends data indicating the calculated estimated angle of arrival to the communication processing unit 41-1. Output. Every time the communication processing unit 41-1 captures the data indicating the estimated angle of arrival output by the angle of arrival calculation unit 19, it includes the data indicating the captured estimated angle of arrival, sets the destination to the operating device 3, and sets the type of data to “ It generates transmission data "estimated arrival angle" and sends the generated transmission data to the communication network 4. The communication network 4 transfers the transmission data to the operating device 3 according to the destination included in the transmission data. The communication processing unit 51 of the operating device 3 receives the transmission data transmitted by the learning device 2 from the communication network 4 . Since the type of data included in the transmission data is “estimated angle of arrival,” the communication processing unit 51 reads data indicating the estimated angle of arrival from the transmission data, and outputs the read data indicating the estimated angle of arrival to the output unit 53. do. When the output unit 53 receives the data indicating the estimated angle of arrival from the communication processing unit 51, it displays the data indicating the estimated angle of arrival on the screen (step Sd11).

By configuring the angle-of-arrival estimation system 100 of the third embodiment described above, for example, the angle-of-arrival estimation device 1b, the learning device 2, and the operating device 3 can be installed at different positions apart from each other. be able to. For example, the angle of arrival estimation device 1b is installed at a position where radio waves are actually received, the learning device 2 is installed at a data center, etc., the operating device 3 is installed at a location where an operator is present, and the operating device 3 allows It becomes possible to operate and use the learning device 2 and the angle of arrival estimation device 1b remotely.

For example, if it is required to install each of the plurality of angle of arrival estimation devices 1b at different positions, the angle of arrival estimation system 100 includes the plurality of angle of arrival estimation devices 1b, and each of the plurality of angle of arrival estimation devices 1b is The configuration may be such that the communication network 4 is connected to the communication network 4. In this case, data of a plurality of mode vectors corresponding to the array antennas 10 included in each of the plurality of angle of arrival estimating devices 1b is stored in advance in the storage unit 55 of the operating device 3 for each center frequency. It will be remembered. The learning device 2 generates training data based on the mode vector data of the angle of arrival estimation device 1b and the center frequency specified by the operating device 3 in response to the operator's operation, and the signal generation parameter corresponding to the data of the mode vector. generate. The learning device 2 uses the generated teacher data to generate trained learning model data to be applied to the function approximator 31 of the designated arrival angle estimation device 1b. When the learning device 2 completes the generation of the learning model data, the learning device 2 transmits the generated learning model data to the specified arrival angle estimation device 1b. A center frequency corresponding to the mode vector data used by the learning device 2 when the learning model data was generated is set in the receiving section 11a of the angle of arrival estimation device 1b to which the learning model data is transmitted, using a reception parameter. That will happen.

The second embodiment may be applied to the angle of arrival estimation system 100 of the third embodiment described above. That is, the input data generation unit 14a of the second embodiment performs the following operations on the input data generation unit 14-1 of the angle of arrival estimation device 1b of the angle of arrival estimation system 100 and the input data generation unit 14-2 of the learning device 2. A configuration that performs normalization may be applied.

(Other configuration examples (Part 1))
In the first to third embodiments described above, the wavelength of radio waves received by the array antenna 10 is determined to be one, and mode vector data of the one wavelength is used. On the other hand, if there are a plurality of wavelengths of radio waves scheduled to be received by the array antenna 10, data of a plurality of mode vectors corresponding to each of the plurality of wavelengths will be generated. In the first and second embodiments, the mode vector storage unit 12 stores data on a plurality of mode vectors corresponding to each of the plurality of generated wavelengths. In the third embodiment, the storage unit 55 of the operating device 3 stores data on a plurality of mode vectors corresponding to each of the plurality of generated wavelengths. Data of a plurality of mode vectors stored in the storage unit 55 will be written to the mode vector storage unit 12 of the learning device 2 in the process of step Sd1 in FIG.

In this case, in the first to third embodiments, when the loops La1s to La1e of the subroutine of the teacher data generation process shown in FIG. 5 are performed, the correlation matrix generation units 13 and 13-2 , and a correlation matrix _Rxx for each mode vector data is generated. The input data generation units 14, 14a, and 14-2 generate input data for each correlation matrix _Rxx generated by the correlation matrix generation units 13 and 13-2. Therefore, the teacher data storage unit 16 stores teacher data for each arrival angle and for each mode vector data, in other words, for each arrival angle and for each radio wave wavelength, and in other words, for each arrival angle and for each radio wave center frequency. will be remembered. By performing a learning process using this teacher data, the learning units 17 and 17a can generate a learning model 22 that estimates the arrival angles of radio waves having various arrival angles and various center frequencies.

(Other configuration examples (Part 2))
In the first to third embodiments described above, in the flowchart shown in FIG. 6, the learning processing units 20 and 20a update the learning model 22 by calculating new learning model data for each teacher data. ing. This is a learning processing method called so-called online learning. On the other hand, the learning processing units 20 and 20a may update the learning model 22 using a so-called mini-batch learning method or a batch learning method.

For example, in the case of the mini-batch learning method, the learning processing units 20 and 20a perform the following processing. The learning processing units 20 and 20a divide the teacher data set stored in the teacher data storage unit 16 into a plurality of subsets such that each subset includes a certain number of teacher data. The learning processing units 20 and 20a repeatedly perform the processing of steps Sb2 and Sb3 on the teacher data included in the subset. The learning processing units 20, 20a generate a plurality of correct labels generated from the plurality of output data obtained through the processing, the designated angle of arrival data included in the teacher data targeted for the processing in steps Sb2, Sb3, and a plurality of correct labels that are predetermined. New learning model data is calculated based on the loss function stored in the learning model data storage section 23 and the learning model data stored in the learning model data storage section 23 at that time. Here, the predetermined loss function is a loss function that calculates one loss value for a combination of a plurality of output data and a plurality of correct labels. The learning processing units 20 and 20a update the learning model 22 using the calculated new learning model data. In this way, the learning processing units 20 and 20a repeatedly perform the process of updating the learning model 22 for each subset. That is, in the mini-batch learning method, the learning model 22 is updated a number of times corresponding to the number of subsets.

For example, when a batch learning method is adopted, the learning processing units 20 and 20a perform the following processing. The learning processing units 20 and 20a repeatedly perform the processing of steps Sb2 and Sb3 on all the teacher data stored in the teacher data storage unit 16. The learning processing units 20 and 20a generate a plurality of output data obtained through the processing, a plurality of correct labels generated from specified arrival angle data included in all the teacher data, a predetermined loss function, and , new learning model data is calculated based on the learning model data stored in the learning model data storage unit 23. Here, the predetermined loss function is a loss function that calculates one loss value for a combination of a plurality of output data and a plurality of correct labels. The learning processing units 20 and 20a update the learning model 22 using the calculated new learning model data. That is, in the batch learning method, the learning model 22 is updated only once for one teacher data set.

Regardless of the online learning method, mini-batch learning method, or batch learning method, the learning processing units 20, 20a update the learning model 22 a predetermined number of times or less. The process may be repeated until the following conditions are met. For example, before calculating new learning model data, the learning processing units 20 and 20a calculate a loss value by substituting the output data and the correct label corresponding to the output data into a loss function, and the calculated loss The learning model 22 may be updated repeatedly until the value satisfies the condition that the value becomes less than a predetermined threshold.

(Other supplementary configuration examples)
In the first to third embodiments described above, the input data generation units 14, 14a, 14-1, and 14-2 receive input from the upper triangular matrix of the correlation matrix _Rxx instead of the lower triangular matrix of the correlation matrix _Rxx . Data may also be generated. Further, the input data generation units 14, 14a, 14-1, and 14-2 may generate input data from all elements of the correlation matrix _Rxx instead of the lower triangular matrix and the upper triangular matrix. When generating input data from an upper triangular matrix or a correlation matrix _Rxx , the input data generating units 14, 14a, 14-1, 14-2 input data from a lower triangular matrix of the correlation matrix _Rxx based on equation (7). In the same way as data is generated, for complex number elements, input data is generated by separating the real component value and the imaginary component value.

In the first to third embodiments described above, the function approximators 21 and 31 are multilayer neural networks, but the multilayer neural network may be a deep neural network. Other neural networks may also be used. In addition, the learning processing units 20 and 20a employ a supervised machine learning method other than the supervised learning method using a neural network, so that the function approximators 21 and 31 detect the presence of radio waves for each angle of arrival. The learning process may be performed so that the learning models 22 and 32 approximate an evaluation function for evaluating probability.

In the first to third embodiments described above, the learning processing units 20 and 20a calculate new learning model data using the error backpropagation method, but in the first to third embodiments, the learning processing units 20 and 20a calculate new learning model data using a method other than the error backpropagation method. Alternatively, learning model data may be calculated.

In the first to third embodiments described above, the loss function used by the learning processing units 20 and 20a may be, for example, a loss function that calculates a mean square error, or a loss function other than a loss function that calculates a mean square error. You may also use a loss function of

In the first to third embodiments described above, the learning processing units 20 and 20a perform learning processing using all the teacher data stored in the teacher data storage unit 16. In contrast, the learning processing units 20 and 20a perform a learning process using some of the teacher data stored in the teacher data storage unit 16, and use the teacher data not used in the learning process as test data. You may also use it.

In the first to third embodiments described above, the array antenna 10 formed by the plurality of antenna elements 10-1 to 10-K is, for example, an array antenna of a circular array type with K elements shown in FIG. However, any array antenna may be used as long as it is a linear array antenna.

In the first and second embodiments described above, the receiving unit 11a of the third embodiment is provided instead of the receiving unit 11, and before the process of step Sc2 is performed, Alternatively, reception parameters may be given to the reception unit 11a from the outside. In this case, the mode vector data to be stored in the mode vector storage section 12 is the mode vector data corresponding to the center frequency specified in the reception parameters.

In the first to third embodiments described above, a functional unit that demodulates a transmitted signal from a received signal generated by the receiving unit 11, 11a is connected to the receiving unit 11, 11a, and the arrival angle estimation device 1, 1a , 1b may function as a wireless receiving device.

FIG. 13 is a block diagram showing the configuration of a learning device 500 according to an embodiment of the present invention, and FIG. 14 is a flowchart showing the flow of processing performed by the learning device 500. As shown in FIG. 13, the learning device 500 includes a signal generation means 501, a correlation matrix generation means 502, an input data generation means 503, and a learning means 504. As shown in FIG. 14, the signal generation means 501 generates a transmission signal vector so that the signal strength of the transmission signal corresponding to the specified arrival angle is stronger than the transmission signal corresponding to other arrival angles (step S511 ). The correlation matrix generation means 502 generates a correlation matrix based on the transmission signal vector and the mode vector of the real environment array antenna (step S512). The input data generation means 503 generates input data from the elements of the correlation matrix (step S513). The learning unit 504 receives the input data and generates a learning model by performing a learning process using teacher data whose correct label is data indicating the angle of arrival corresponding to the input data (step S514).

FIG. 15 is a block diagram showing the configuration of an angle of arrival estimation device 600 according to an embodiment of the present invention, and FIG. 16 is a flowchart showing the flow of processing performed by the angle of arrival estimation device 600. As shown in FIG. 15, the arrival angle estimation device 600 includes an array antenna 601, a receiving means 602, a correlation matrix generation means 603, an input data generation means 604, an inference means 605, and an arrival angle calculation means 606. As shown in FIG. 16, the receiving means 602 generates a received signal vector from the radio waves arriving at the array antenna 601 (step S611). Correlation matrix generation means 603 generates a correlation matrix based on the received signal vector (step S612). The input data generation means 604 generates input data from the elements of the correlation matrix (step S613). The inference means 605 is a learning model generated to be an evaluation function for evaluating the existence probability of radio waves for each angle of arrival at the array antenna 601, and is based on training data generated using the mode vector of the array antenna 601. Input data is given to the learning model generated according to the learning process to obtain output data (step S614). The arrival angle calculating means 606 calculates the estimated arrival angle of the radio wave from the output data (step S615).

The above-mentioned arrival angle estimating devices 1, 1a, 1b, 600, learning device 2,500, and operating device 3 have a computer system inside, and the software part of the receiving units 11, 11a, the mode vector storage unit 12, correlation matrix generation unit 13, 13-1, 13-2, input data generation unit 14, 14a, 14-1, 14-2, signal generation unit 15, teacher data storage unit 16, learning unit 17, 17a, inference parts 18 and 18a, the angle of arrival calculation part 19, the software part of the communication processing parts 41-1, 41-2, and 51, the software part of the operation part 52 and the output part 53, the control part 54, the storage part 55, and the signal generation The means 501, the correlation matrix generation means 502, the input data generation means 503, the learning means 504, the software part of the reception means 602, the correlation matrix generation means 603, the input data generation means 604, the inference means 605, and the angle of arrival calculation means 606. 4 to 7, FIG. 12, FIG. 14, The processing described with reference to FIG. 16 may also be performed. Note that the "computer system" herein includes hardware such as an OS and peripheral devices. Furthermore, the term "computer system" includes a WWW system equipped with a home page providing environment (or display environment). Furthermore, the term "computer-readable recording medium" refers to portable media such as flexible disks, magneto-optical disks, ROMs, and CD-ROMs, and storage devices such as hard disks built into computer systems. Furthermore, "computer-readable recording medium" refers to volatile memory (RAM) inside a computer system that serves as a server or client when a program is transmitted via a network such as the Internet or a communication line such as a telephone line. This also includes programs that are retained for a certain period of time.

Further, the program may be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by a transmission wave in a transmission medium. Here, the "transmission medium" that transmits the program refers to a medium that has a function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line. Moreover, the above program may be for realizing a part of the above-mentioned functions. Furthermore, it may be a so-called difference file (difference program) that can realize the above-mentioned functions in combination with a program already recorded in the computer system.

Although the embodiments of the present invention have been described above in detail with reference to the drawings, the specific configuration is not limited to these embodiments, and includes designs within the scope of the gist of the present invention.

Part or all of the above embodiments may be described as in the following additional notes, but are not limited to the following.

(Additional Note 1) Signal generation means (for example, signal generation unit 15) that generates a transmission signal vector such that the signal strength of the transmission signal corresponding to a specified arrival angle is stronger than transmission signals corresponding to other arrival angles; , correlation matrix generation means (for example, correlation matrix generation units 13 and 13-2) that generates a correlation matrix based on the transmission signal vector and a mode vector of a real environment array antenna used in a real environment; Input data generation means (for example, input data generation units 14, 14a, 14-2) that generates input data from elements of a matrix; A learning device comprising learning means (for example, learning units 17 and 17a) that generates a learning model by performing learning processing using teacher data used as correct answer labels.

(Supplementary note 2) The input data generation means (for example, input data generation units 14, 14a, 14-2) performs a process of generating the input data from elements of a triangular matrix of the correlation matrix, or The learning device according to (Additional Note 1), which performs one of the steps of normalizing elements of a matrix to generate the input data.

(Additional Note 3) The learning device according to (Additional Note 2), wherein the triangular matrix is a lower triangular matrix.

(Appendix 4) The learning device according to any one of (Appendix 1) to (Appendix 3), wherein the mode vector is generated for each wavelength of the radio wave that the real environment array antenna is scheduled to receive.

(Additional Note 5) An array antenna (for example, array antenna 10), a receiving means (for example, receiving section 11, 11a) that generates a received signal vector from radio waves arriving at the array antenna, and a correlation based on the received signal vector. Correlation matrix generation means (for example, correlation matrix generation units 13, 13-1) that generate a matrix, and input data generation means (for example, input data generation units 14, 14a, 14) that generate input data from the elements of the correlation matrix. -1) and a learning model generated to be an evaluation function for evaluating the probability of existence of a radio wave for each angle of arrival at the array antenna, which is based on training data generated using the mode vector of the array antenna. An inference unit (for example, inference unit 18, 18a) that provides the input data to a learning model generated according to a learning process to obtain output data; and an arrival angle calculation unit that calculates an estimated arrival angle of radio waves from the output data. (For example, an angle of arrival calculation unit 19).

(Additional Note 6) Signal generation means (for example, signal generation unit 15) that generates a transmission signal vector such that the signal strength of the transmission signal corresponding to the specified arrival angle is stronger than the transmission signal corresponding to other arrival angles. , the input data generation means (for example, input data generation section 13) generates the correlation matrix from the elements of the correlation matrix generated by the correlation matrix generation means (for example, the correlation matrix generation section 13) based on the transmission signal vector and the mode vector of the array antenna. learning means for generating the learning model by performing a learning process using the input data generated by the unit 14, 14a) as an input and using teacher data having data indicating the angle of arrival corresponding to the input data as a correct answer label; (For example, the learning unit 17).

(Additional Note 7) The input data generation means (14, 14a, 14-1) performs a process of generating the input data from the elements of the triangular matrix of the correlation matrix, or normalizes the elements of the triangular matrix of the correlation matrix. The arrival angle estimating device according to (Appendix 5) or (Appendix 6), which performs one of the processes of generating the input data.

(Additional Note 8) The arrival angle estimation device according to (Additional Note 7), wherein the triangular matrix is a lower triangular matrix.

(Additional Note 9) The arrival angle estimation device according to any one of (Appendix 5) to (Appendix 8), wherein the mode vector is generated for each wavelength of the radio wave that the array antenna is scheduled to receive.

(Appendix 10) The learning device according to (Appendix 1) and the angle of arrival estimation device according to (Appendix 5) are provided, and the real environment array antenna in the learning device the learning model generated by the learning means of the learning device is the learning model of the inference means of the angle of arrival estimation device;
Arrival angle estimation system.

(Additional Note 11) A plurality of the angle of arrival estimation devices exist, the real environment array antenna in the learning device is each of the array antennas of the plurality of angle of arrival estimation devices, and the learning means The angle of arrival estimation system according to (Appendix 10), which generates the learning model corresponding to each of the angle of arrival estimation devices.

(Additional Note 12) The input data generation means may perform a process of generating the input data from the elements of a triangular matrix of the correlation matrix, or a process of generating the input data by normalizing the elements of the triangular matrix of the correlation matrix. The arrival angle estimation system according to (Appendix 10) or (Appendix 11), which performs either one of the processes.

(Additional Note 13) The arrival angle estimation system according to (Additional Note 12), wherein the triangular matrix is a lower triangular matrix.

(Additional Note 14) The arrival angle estimation system according to any one of (Appendix 10) to (Appendix 13), wherein the mode vector is generated for each wavelength of the radio wave that the array antenna is scheduled to receive.

(Additional Note 15) A transmission signal vector is generated so that the signal strength of a transmission signal corresponding to a specified arrival angle is stronger than transmission signals corresponding to other arrival angles, and the generated transmission signal vector and A correlation matrix is generated based on the mode vector of the actual environment array antenna to be used, input data is generated from the elements of the generated correlation matrix, the generated input data is input, and the A learning model generation method that generates a learning model by performing learning processing using teacher data whose correct label is data indicating the angle of arrival.

(Additional Note 16) Generate a received signal vector from radio waves arriving at the array antenna, generate a correlation matrix based on the generated received signal vector, generate input data from the elements of the generated correlation matrix, A learning model generated to be an evaluation function for evaluating the probability of existence of radio waves for each angle of arrival at A method for estimating an angle of arrival, in which the input data is given to a model, output data is obtained, and an estimated arrival angle of a radio wave is calculated from the obtained output data.

(Additional Note 17) A procedure for generating a transmission signal vector such that the signal strength of a transmission signal corresponding to a specified arrival angle is stronger than transmission signals corresponding to other arrival angles, and , a procedure for generating a correlation matrix based on the mode vector of a real-environment array antenna used in a real environment; a procedure for generating input data from the elements of the generated correlation matrix; A program for executing a procedure for generating a learning model by performing a learning process using teacher data in which data indicating the angle of arrival corresponding to input data is used as a correct label.

(Additional note 18) A procedure for generating a received signal vector from radio waves arriving at the array antenna, a procedure for generating a correlation matrix based on the generated received signal vector, and a procedure for generating input data from the elements of the generated correlation matrix. a learning process based on training data that is a learning model generated to be an evaluation function for evaluating the probability of existence of radio waves for each angle of arrival at the array antenna, and that is generated using the mode vector of the array antenna; A program for executing a procedure of providing the input data to a learning model generated according to the above to obtain output data, and a procedure of calculating an estimated arrival angle of a radio wave from the obtained output data.

It can be used to estimate the arrival angle of radio waves.

DESCRIPTION OF SYMBOLS 1... Angle of arrival estimation device, 10... Array antenna, 10-1 to 10-K... Antenna element, 11... Receiving section, 12... Mode vector storage section, 13... Correlation matrix generation section, 14... Input data generation section, 15 ... Signal generation section, 16... Teacher data storage section, 17... Learning section, 18... Inference section, 19... Arrival angle calculation section, 20... Learning processing section, 21... Function approximator, 22... Learning model, 23... Learning model Storage unit, 30... Inference processing unit, 31... Function approximator, 32... Learning model

Claims (10)

signal generation means for generating a transmission signal vector such that the signal strength of the transmission signal corresponding to the specified arrival angle is stronger than the transmission signal corresponding to other arrival angles;
Correlation matrix generation means for generating a correlation matrix based on the transmission signal vector and a mode vector of a real environment array antenna used in a real environment;
input data generation means for generating input data from the elements of the correlation matrix;
a learning means that generates a learning model by performing a learning process using teacher data in which the input data is input and data indicating the angle of arrival corresponding to the input data is used as a correct answer label;
A learning device equipped with. The input data generation means includes:
performing either a process of generating the input data from the elements of a triangular matrix of the correlation matrix, or a process of generating the input data by normalizing the elements of the triangular matrix of the correlation matrix;
The learning device according to claim 1. array antenna,
receiving means for generating a received signal vector from radio waves arriving at the array antenna;
Correlation matrix generation means for generating a correlation matrix based on the received signal vector;
input data generation means for generating input data from the elements of the correlation matrix;
A learning model generated to be an evaluation function for evaluating the existence probability of radio waves for each angle of arrival at the array antenna, and generated according to a learning process based on training data generated using the mode vector of the array antenna. inference means for obtaining output data by giving the input data to a learning model that is
Arrival angle calculation means for calculating an estimated arrival angle of radio waves from the output data;
An angle of arrival estimation device comprising: signal generation means for generating a transmission signal vector such that the signal strength of the transmission signal corresponding to the specified arrival angle is stronger than the transmission signal corresponding to other arrival angles;
input data generated by the input data generation means from the elements of a correlation matrix generated by the correlation matrix generation means based on the transmission signal vector and the mode vector of the array antenna; learning means for generating the learning model by performing a learning process using teaching data in which data indicating the angle of arrival is used as a correct label;
The arrival angle estimation device according to claim 3, comprising: The learning device according to claim 1;
An angle of arrival estimation device according to claim 3,
The real environment array antenna in the learning device is the array antenna of the angle of arrival estimation device, and the learning model generated by the learning means of the learning device is the learning model of the inference means of the angle of arrival estimation device. be,
Arrival angle estimation system. A plurality of the arrival angle estimating devices exist,
The real environment array antenna in the learning device is each of the array antennas of the plurality of angle of arrival estimation devices,
The learning means generates the learning model corresponding to each of the plurality of angle of arrival estimation devices,
The arrival angle estimation system according to claim 5. Generating a transmission signal vector so that the signal strength of the transmission signal corresponding to the specified arrival angle is stronger than transmission signals corresponding to other arrival angles,
Generate a correlation matrix based on the generated transmission signal vector and a mode vector of a real environment array antenna used in a real environment,
Generating input data from the elements of the generated correlation matrix,
generating a learning model by performing a learning process using teacher data in which the generated input data is input and data indicating the angle of arrival corresponding to the input data is used as a correct label;
How to generate a learning model. Generates a received signal vector from radio waves arriving at the array antenna,
generating a correlation matrix based on the generated received signal vector;
Generating input data from the elements of the generated correlation matrix,
A learning model generated to be an evaluation function for evaluating the existence probability of radio waves for each angle of arrival at the array antenna, and generated according to a learning process based on training data generated using the mode vector of the array antenna. giving the input data to a learning model to obtain output data;
calculating an estimated arrival angle of radio waves from the acquired output data;
Arrival angle estimation method. to the computer,
a procedure for generating a transmission signal vector such that the signal strength of the transmission signal corresponding to a specified arrival angle is stronger than transmission signals corresponding to other arrival angles;
a step of generating a correlation matrix based on the generated transmission signal vector and a mode vector of a real environment array antenna used in a real environment;
a step of generating input data from the generated elements of the correlation matrix;
A step of generating a learning model by performing a learning process using training data in which the generated input data is input and data indicating the arrival angle corresponding to the input data is used as a correct label;
A program to run. to the computer,
A procedure for generating a received signal vector from radio waves arriving at an array antenna,
a step of generating a correlation matrix based on the generated received signal vector;
a step of generating input data from the generated elements of the correlation matrix;
A learning model generated to be an evaluation function for evaluating the probability of existence of a radio wave for each angle of arrival at the array antenna, and generated according to a learning process based on teacher data generated using the mode vector of the array antenna. a step of giving the input data to a learning model to obtain output data;
a step of calculating an estimated arrival angle of radio waves from the acquired output data;
A program to run.

Images