Direction of arrival estimation device, direction of arrival estimation method, and program

Abstract

To provide a direction-of-arrival estimation device which offers high estimation accuracy for a direction-of-arrival of a radio wave.SOLUTION: A direction-of-arrival estimation device provided herein is configured to derive a first spectrum of a radio wave signal using the MUSIC method based on the radio wave signal received by an array antenna, derive a second spectrum of the radio wave signal using a deep learning method based on the radio wave signal, and estimate a direction of arrival of the radio wave signal using a peak search method based on the first spectrum and the second spectrum.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to an arrival direction estimation device, an arrival direction estimation method, and a program. [Background technology]

[0002] Techniques for estimating the direction of arrival of radio waves using an array antenna composed of multiple antennas have been disclosed. For example, Non-Patent Documents 1 and 2 disclose techniques using deep learning. [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] Y. Kase and 5 others, “Fundamental Trial on DOA Estimation with Deep Learning,” IEICE Trans. Commun., Vol.E103-B, Oct. 2020. [Non-patent document 2] W. Zhu and three others, “Broadband Direction of Arrival Estimation Based on Convolutional Neural Network,” IEICE Trans. Commun., Vol.E103-B, Mar. 2020. Summary of the Invention [Problem to be solved by the invention]

[0004] In the technology for estimating the direction of arrival of radio waves, there is a demand for a technology with higher estimation accuracy.

[0005] Therefore, an object of the present invention is to provide an arrival direction estimation device, an arrival direction estimation method, and a program that solve the above-mentioned problems. [Means for solving the problem]

[0006] According to a first aspect of the present invention, an arrival direction estimation device includes a first radio wave characteristic calculation means for calculating a first radio wave characteristic indicating a characteristic of a radio wave signal to be calculated, the first radio wave characteristic being a characteristic of a radio wave signal and its reception strength depending on the arrival direction of the radio wave, using an estimation method other than a deep learning method; a second radio wave characteristic calculation means for calculating a second radio wave characteristic indicating the characteristic corresponding to the radio wave signal to be calculated, using an estimation model obtained by deep learning of the relationship between the radio wave signal and the characteristic of the reception strength depending on the arrival direction of the radio wave; and a peak search means for estimating the arrival direction of the radio wave signal using a peak search method for the radio wave characteristic identified based on the first radio wave characteristic and the second radio wave characteristic.

[0007] According to a second aspect of the present invention, a method for estimating direction of arrival includes calculating a first radio wave characteristic indicating the characteristics of a radio wave signal and its reception strength depending on the direction of arrival of the radio wave using an estimation method other than a deep learning method, calculating a second radio wave characteristic indicating the characteristics corresponding to the radio wave signal to be calculated using an estimation model obtained by deep learning of the relationship between the radio wave signal and the characteristics of the reception strength depending on the direction of arrival of the radio wave, and estimating the direction of arrival of the radio wave signal using a peak search method for the radio wave characteristic identified based on the first radio wave characteristic and the second radio wave characteristic.

[0008] According to a third aspect of the present invention, a program causes a computer of an arrival direction estimation device to function as a first radio wave characteristic calculation means that calculates a first radio wave characteristic, which is a characteristic of a radio wave signal and its reception strength depending on the arrival direction of the radio wave, for a radio wave signal to be calculated using an estimation method other than a deep learning method; a second radio wave characteristic calculation means that calculates a second radio wave characteristic, which is a characteristic corresponding to the radio wave signal to be calculated, using an estimation model obtained by deep learning of the relationship between the radio wave signal and the characteristic of the reception strength depending on the arrival direction of the radio wave; and a peak search means that estimates the arrival direction of the radio wave signal using a peak search method for radio wave characteristics identified based on the first radio wave characteristic and the second radio wave characteristic. [Effects of the Invention]

[0009] According to the present invention, the direction of arrival of radio waves can be estimated with higher accuracy. [Brief explanation of the drawings]

[0010] [Figure 1] 1 is a diagram showing a schematic configuration of an arrival direction estimation system according to an embodiment of the present invention. [Figure 2] FIG. 1 is a hardware configuration diagram of an arrival direction estimation device according to an embodiment of the present invention. [Figure 3] FIG. 1 is a functional block diagram of an arrival direction estimation device according to a first embodiment. [Figure 4] FIG. 2 is a diagram showing a processing flow of the direction-of-arrival estimation device according to the first embodiment. [Figure 5] FIG. 10 is a functional block diagram of an arrival direction estimation device according to a second embodiment. [Figure 6] FIG. 10 is a diagram showing a processing flow of the direction of arrival estimation device according to the second embodiment. [Figure 7] FIG. 1 is a diagram showing a minimum configuration of an arrival direction estimation device according to the present embodiment. [Figure 8] FIG. 10 is a diagram showing a processing flow of an arrival direction estimation device with a minimum configuration according to the present embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, an arrival direction estimation device according to an embodiment of the present invention will be described with reference to the drawings. Fig. 1 is a diagram showing the schematic configuration of an arrival direction estimation system including an arrival direction estimation device according to this embodiment. As shown in Fig. 1, as an example, an arrival direction estimation system 100 is configured by connecting an arrival direction estimation device 1 and an array antenna 2 via a communication network. The arrival direction estimation device 1 is a computer server that acquires radio wave signals received by the array antenna 2 and estimates the arrival direction of the radio waves using the received signals.

[0012] FIG. 2 is a hardware configuration diagram of the direction of arrival estimation device according to this embodiment. As shown in this figure, the direction of arrival estimation device 1 is a computer equipped with various hardware components such as a CPU (Central Processing Unit) 101, a ROM (Read Only Memory) 102, a RAM (Random Access Memory) 103, an HDD (Hard Disk Drive) 104, a communication module 105, and a database 106.

[0013] The direction of arrival estimation device 1 according to this embodiment calculates a radio wave signal and a characteristic of the reception strength according to the direction of arrival of the radio wave (the spectrum of the evaluation function). The direction of arrival estimation device 1 calculates a first spectrum (first radio wave characteristic) indicating the spectrum of the evaluation function for the radio wave signal to be calculated using an estimation method other than deep learning. The direction of arrival estimation device 1 calculates a second spectrum (second radio wave characteristic) indicating the spectrum of the evaluation function for the radio wave signal to be calculated using an estimation model obtained by deep learning of the relationship between the radio wave signal and the characteristic of the reception strength according to the direction of arrival of the radio wave. The direction of arrival estimation device 1 estimates the direction of arrival of the radio wave signal using a peak search method for the radio wave characteristic identified based on the first spectrum (first radio wave characteristic) and the second spectrum (second radio wave characteristic).

[0014] (First embodiment) FIG. 3 is a functional block diagram of the direction of arrival estimation device according to the first embodiment. The direction-of-arrival estimation device 1 executes a first direction-of-arrival estimation program, thereby providing the functions of a receiver 11, a correlation matrix calculator 12, a first spectrum calculator 13, a preprocessor 14, a second spectrum calculator 15, a signal measurer 16, a spectrum synthesizer 17, a peak searcher 18, and a learning unit 19.

[0015] The receiving unit 11 receives the signals received by each antenna of the array antenna. The correlation matrix calculation unit 12 calculates a correlation matrix. The first spectrum calculation unit 13 calculates a first spectrum. The first spectrum is the spectrum of an evaluation function calculated by an estimation method other than a deep learning method, such as the MUSIC (Multiple Signal Classification) method. The spectrum of the evaluation function is, specifically, an arrival evaluation function for determining whether radio waves are arriving, and is a function that indicates that the amplitude corresponding to the direction from which the radio waves arrive increases (peaks) in a graph showing the relationship between the arrival direction and amplitude and representing the function. The amplitude corresponding to the direction from which the radio waves arrive increases (peaks) due to the characteristics of the reception strength according to the arrival direction of the radio waves. The preprocessing unit 14 is a processing unit that performs preprocessing before calculating the second spectrum. The second spectrum calculation unit 15 calculates the second spectrum, which is the spectrum of the evaluation function, by a calculation method different from that for the first spectrum. The second spectrum is the spectrum of the evaluation function calculated using the deep learning method described above. The signal measurement unit 16 calculates weighting coefficients for the first spectrum and the second spectrum using the received radio waves, using the SNR (Signal to Noise Ratio) of the received signal, the RSS (Received Signal Strength) and other received power indicators of the received radio waves. The spectrum synthesis unit 17 synthesizes the first spectrum and the second spectrum. The peak search unit 18 performs a peak search on the combined spectrum obtained by combining the first spectrum and the second spectrum, and estimates the direction of arrival of the radio wave. The learning unit 19 calculates a second spectrum estimation model used to calculate the second spectrum using a machine learning technique.

[0016] 4 is a diagram showing a processing flow of the direction-of-arrival estimation device according to the first embodiment. In the following, the present embodiment will be described taking as an example a case where the MUSIC method is used as a method for calculating the first spectrum, but the first spectrum may be calculated using other estimation methods such as the beamformer method, the Capon method, the linear prediction method, or the ESPRIT (Estimation of Signal Parameters via Rotational Invariance Techniques) method. First, the receiver 11 receives a radio signal wirelessly transmitted from a transmitter via the array antenna 2 (step S100). Below, we will explain an example in which k plane waves with wavelength λ and complex amplitude s_k(t) arrive at an angle θ from an L-element equally-spaced linear array antenna with an interval d, which is made up of identical antennas. In this case, if the additive noise of the antenna is represented by n_l(t), the received data of the l-th antenna can be expressed as in equation (1).

[0017]

number

[0018] Here, when the received data of the entire array antenna is expressed in vector form, it can be expressed as in equation (2).

[0019]

number

[0020] The correlation matrix calculation unit 12, preprocessing unit 14, and signal measurement unit 16 acquire the signal received by the receiving unit 11. The correlation matrix calculation unit 12 calculates the correlation matrix of the received signal (step S101). Specifically, the correlation matrix (covariance matrix) is a matrix that represents the degree of correlation between arriving waves. For example, the correlation matrix Rxx is shown in the following equation (3). The correlation matrix calculation unit 12 outputs information about the correlation matrix to the first spectrum calculation unit 13.

[0021]

number

[0022] The first spectrum calculation unit 13 acquires a correlation matrix. The first spectrum calculation unit 13 performs eigenvalue decomposition on the correlation matrix (step S102). The first spectrum calculation unit 13 calculates a MUSIC spectrum S using the result of the eigenvalue decomposition. MU (θ) is calculated (step S103). The calculated MUSIC spectrum S MU (θ). The first spectrum calculation unit 13 calculates the MUSIC spectrum S MU (θ) is output to the spectrum synthesis unit 17. MUSIC spectrum S MU (θ) is one aspect of the first spectrum. In equation (4), H is the Hermitian transpose, a(θ) is the array response vector, and E N denotes a matrix consisting of the thermal noise elements of the eigenvectors obtained by eigenvalue decomposition.

[0023]

number

[0024] The MUSIC spectrum S shown in equation (4) MU In the equation (θ), a(θ) represents the mode vector of the array antenna. The mode vector is a vector of coefficients determined by the arrangement of antenna elements, and is also called the array factor or array response vector. In addition, in equation (4), E N denotes a matrix consisting of the thermal noise elements of the eigenvectors obtained by eigenvalue decomposition.

[0025] On the other hand, when the preprocessing unit 14 acquires the correlation matrix, it calculates a lower triangular matrix of the correlation matrix (step S104). The preprocessing unit 14 outputs the calculated lower triangular matrix to the second spectrum calculation unit 15. When the second spectrum calculation unit 15 acquires the lower triangular matrix, it inputs the lower triangular matrix to a neural network and calculates the resulting DL spectrum S DL (θ) is calculated (step S105). DL (θ) is one aspect of the second spectrum.

[0026]

number

[0027] Note that z shown in equation (5) (M) is the output of the output layer of the neural network used in deep learning.

[0028] The neural network used by the second spectrum calculation unit 15 to calculate the second spectrum is a neural network constructed by a second spectrum estimation model. The second spectrum estimation model is calculated by the learning unit 19 acquiring multiple relationships between the output of the pre-processing unit 14 and the spectrum indicating the relationship between the arrival direction and amplitude of the radio waves actually measured at that time in the case of that output, and performing machine learning (deep learning) on ​​the multiple relationships. The second spectrum estimation model includes information specifying the weighting coefficients of the neural network, etc. The DL spectrum S is calculated using this second spectrum estimation model. DL To calculate (θ), the above MUSIC spectrum S MU (θ) indicates the spectrum of the evaluation function calculated using the MUSIC method, and the DL spectrum S DL (θ) indicates the spectrum of the evaluation function calculated using deep learning.

[0029] In calculating the second spectral estimation model described above, the learning unit 19 uses a rectified linear unit (ReLU) as the activation function of the units in the intermediate layer of the deep neural network obtained by the second spectral estimation model. The learning unit 19 sets the activation function of the units in the output layer of the deep neural network to an identity function (linear). The rectified linear unit (ReLU) is given by equation (6).

[0030]

number

[0031] The learning unit 19 inputs Rxx (Equation (3)), which is an autocorrelation function (covariance matrix), to the input y in the deep neural network. When the values ​​of the units in the output layer are expressed as vectors, Equation (5) is obtained.

[0032] The second spectrum calculation unit 15 calculates the DL spectrum S DL The signal measurement unit 16 acquires the received signal from the receiver 11. The signal measurement unit 16 identifies a received power index such as SNR (Signal to Noise Ratio) or RSS (Received Signal Strength) of the received signal. The signal measurement unit 16 evaluates the performance of each received power index such as SNR or RSS by using the MUSIC spectrum S obtained by the MUSIC technique. MU (θ) and the DL spectrum S obtained by the deep learning method. DL (θ) and (θ). Here, performance evaluation for each received power index corresponds to determining the estimation success rate and RMSE for each received power index through simulation or the like. As a result, for example, in a high SNR region above a threshold, a higher estimation success rate may be obtained by selecting the spectrum of the evaluation function calculated using the MUSIC method, while in a low SNR region below the threshold, a higher estimation success rate may be obtained by selecting the spectrum of the evaluation function calculated using deep learning. Such performance evaluation results (estimation success rate and RMSE) are used to calculate the weighting coefficients described below.

[0033] The signal measurement unit 16 generates a table including values ​​of the estimation success rate or mean squared error (MSE) of the estimation error angle corresponding to the SNR for each received power index calculated during the performance evaluation. The mean squared error value may be the root mean squared error (RMSE). Performance evaluation refers to evaluating the performance of the estimation algorithm, for example, by the estimation success rate of the arrival angle or the mean squared error of the estimation error angle. For example, when evaluating the performance of the estimation algorithm using the estimation success rate, the signal measurement unit 16 identifies the correct arrival angle of the radio wave as θk and the estimated arrival angle of the radio wave as θk^. The signal measurement unit 16 calculates the estimated arrival angle θk^n for each reception timing n of the radio wave. The signal measurement unit 16 counts the arrival angle of the radio wave as successful if the condition shown in equation (7) is satisfied.

[0034]

number

[0035] The signal measurement unit 16 calculates the estimated success rate using equation (8).

[0036]

number

[0037] When the performance of the estimation algorithm is evaluated by the mean square error of the estimated error angle, the signal measuring unit 16 calculates the root mean square error (RMSE) of the estimated error angle using equation (9).

[0038]

number

[0039] For example, the signal measurement unit 16 calculates the first weighting coefficient W in the following equation (10) using the estimated success rate: MU (step S106), and the second weighting factor W is calculated using equation (11). DL(Step S107). Specifically, P MU is the MUSIC spectrum S obtained by the MUSIC method. MU The estimated success rate obtained by performance evaluation of (θ) is shown. DL is the DL spectrum S obtained by the deep learning method. DL The estimation success rate obtained by the performance evaluation of (θ) is shown. MU is the weighting factor multiplied by the MUSIC spectrum (first spectrum). The second weighting factor W DL is a weighting coefficient by which the DL spectrum (second spectrum) is multiplied. The signal measurement unit 16 measures the estimation success rates P MU and P DL Based on this, the weighting coefficients W are normalized so that the sum of the ratios of each estimated success rate becomes 1 using Equation (10) and Equation (11). MU , W DL This process is one aspect of the process in which the signal measurement unit 16 calculates the value of the weighting coefficient for each of the first and second radio wave characteristics, normalized based on the evaluation values ​​of the radio wave characteristics calculated using the first spectrum and the second spectrum, so that the sum of the ratios of the evaluation values ​​becomes 1. The evaluation values ​​of the radio wave characteristics are the estimation success rate and RMSE described above. The signal measurement unit 16 calculates the value of the first weighting coefficient W MU and the second weighting factor W DL and output to the spectrum synthesis unit 17.

[0040]

number

[0041]

number

[0042] In the above description, the first weighting factor W MU and the second weighting factor W DL and the estimated success rate P MU and P DLHowever, the signal measurement unit 16 uses the root mean square error (RMSE) of the estimation error angle and the power ratio (DUR) of the desired wave and the undesired wave to calculate the estimation success rate P MU and P DL may be calculated.

[0043] The spectrum synthesis unit 17 receives the MUSIC spectrum S from the first spectrum calculation unit 13. MU (θ) is obtained, and the DL spectrum S DL (θ) is acquired from the signal measurement unit 16, and the first weighting coefficient W MU and the second weighting factor W DL The spectrum synthesis unit 17 uses equation (12) to obtain the MUSIC spectrum S MU (θ) and the second spectrum, the DL spectrum S DL (θ) and are weighted and added together to form a composite spectrum S Comb (θ) is calculated (step S108).

[0044]

number

[0045] The spectrum synthesis unit 17 synthesizes the synthesized spectrum S Comb (θ) is output to the peak search unit 18. Comb (θ) is one aspect of the composite radio wave characteristic. The peak search unit 18 calculates the composite spectrum S Comb (θ) to perform a peak search to calculate the estimated direction of arrival of the radio wave (step S109). Comb Based on the relationship between the angle of the radio wave arrival direction indicated by (θ) and the magnitude of the radio wave arrival strength (db), the peak point of the angle of the radio wave arrival direction with the greatest radio wave arrival strength is identified. The peak search unit 18 calculates the angle of the radio wave arrival direction corresponding to that peak point as the estimated radio wave arrival direction.

[0046] According to the above process, the direction-of-arrival estimation device 1 uses a deep learning method such as a neural network to generate the second spectrum, the DL spectrum S DL (θ) is calculated, and the DL spectrum S DL (θ) and MUSIC spectrum S MU (θ) are weighted and added together. According to this process, the direction of arrival estimation device 1 calculates the weighting coefficients for each spectrum from the estimation success rate or root mean square error (RMSE). Then, the direction of arrival estimation device 1 uses these weighting coefficients to generate the composite spectrum S Comb (θ) is calculated. Here, the weighting coefficients can be used as an index to determine which spectrum is more likely to derive an erroneous estimated angle (which spectrum is more reliable) by calculating the ratio of the error rates. Each weighting coefficient is then multiplied by the corresponding spectrum. Therefore, the amplitude of one spectrum (height in the direction of arrival indicated by the evaluation function) becomes relatively smaller than the amplitude of the other spectrum. In other words, as the spectrum with the smaller weighting coefficient becomes smaller, the spectrum with the relatively larger weighting coefficient becomes dominant over the synthesized spectrum, resulting in a higher reliability of the synthesized spectrum. As a result, compared to when the first and second spectra are synthesized without weighting using the weighting coefficients of these spectra, the spectrum with the higher reliability is given more weight, and the reliability of the calculated direction of arrival estimation result is higher.

[0047] In the above-described deep learning, it is known that the spectrum of an evaluation function estimated using an estimation model using a sigmoid function as the activation function of the output layer has large peaks. In a situation where the spectrum of the estimated evaluation function is too peaked, there is a possibility that an erroneously calculated large peak will appear in a certain direction of arrival. In such a case, the influence of the erroneously calculated peak will be too great, and the benefits of synthesis will not be obtained. Therefore, it is difficult to synthesize a spectrum estimated using an estimation model generated using such a deep learning method with a spectrum estimated using another method. However, when an estimation model is calculated using the deep learning method of the processing of this embodiment described above, an identity function (linear) is used as the activation function of the unit in the output layer of deep learning. As a result, the peaks of the radio wave reception strength of the spectrum estimated using the deep learning method tend to be gentler. As a result, the DL spectrum S generated by this embodiment DL (θ) and MUSIC spectrum S MU (θ) can be synthesized.

[0048] According to the above process, the direction-of-arrival estimation device 1 combines a radio wave spectrum estimated using the MUSIC method and a radio wave spectrum estimated using a deep learning method. This allows the direction-of-arrival estimation to take into account the directions of arrival of each combined spectrum, thereby improving the accuracy of the direction-of-arrival estimation. The direction-of-arrival estimation device 1 is characterized by improving the accuracy of the direction-of-arrival estimation by changing the weighting coefficients for weighting each spectrum according to the radio wave conditions before the combination.

[0049] Second Embodiment In the above description of the first embodiment, the DL spectrum S DL (θ) and MUSIC spectrum S MU (θ) and the weighted sum of the synthesized spectrum S Comb The estimated direction of arrival of the radio wave is calculated by peak search based on (θ). However, the direction of arrival of the radio wave may be estimated by other methods.

[0050] FIG. 5 is a functional block diagram of an arrival direction estimation device according to the second embodiment. The direction-of-arrival estimation device 1 according to the second embodiment executes a second direction-of-arrival estimation program, which provides the functions of a receiver 11, a correlation matrix calculator 12, a first spectrum calculator 13, a preprocessor 14, a second spectrum calculator 15, a spectrum evaluator 21, a spectrum selector 22, a peak searcher 18, and a learning unit 19.

[0051] The functional units of the receiving unit 11, the correlation matrix calculation unit 12, the first spectrum calculation unit 13, the pre-processing unit 14, the second spectrum calculation unit 15, the peak search unit 18, and the learning unit 19, which are some of the functions performed by the direction-of-arrival estimation device 1 according to the second embodiment, are the same as those in the first embodiment.

[0052] The spectrum evaluation unit 21 evaluates the first spectrum, MUSIC spectrum S MU Calculate the standard deviation from the variance of the main lobe of (θ). The spectrum selector 22 selects the first spectrum, MUSIC spectrum S MU The first spectrum, the MUSIC spectrum S, is calculated based on the standard deviation of (θ). MU Select whether to use (θ) or the second spectrum to estimate the radio wave direction.

[0053] FIG. 6 is a diagram showing a processing flow of the direction of arrival estimation device according to the second embodiment. In the second embodiment, the MUSIC spectrum S in the first embodiment MU (θ) and DL spectrum S DL After the processing of steps S100 to S105 is completed, the first spectrum calculation unit 13 calculates the MUSIC spectrum S MU (θ) to the spectrum selection unit 22, and the second spectrum calculation unit 15 outputs the DL spectrum S DL (θ) to the spectrum selection unit 23. The first spectrum calculation unit 13 also outputs the MUSIC spectrum S MU (θ) is output to the spectrum evaluation unit 21.

[0054] The spectrum evaluation unit 21 calculates the MUSIC spectrum S obtained from the first spectrum calculation unit 13. MU The spectrum evaluation unit 21 calculates the standard deviation from the variance of the main lobe of the MUSIC spectrum S (θ) (step S201). MU The standard deviation calculated from the variance of the main lobe of (θ) is output to the spectrum selection unit 22.

[0055] If the standard deviation of the main lobe variance is less than a predetermined threshold, the spectrum selector 22 selects the MUSIC spectrum S MU (θ) is selected, and the MUSIC spectrum S MU If it can be determined that the main lobe of (θ) is broad, the DL spectrum S DL (θ) is selected (step S202). The predetermined threshold is calculated by the learning unit 19 that learns the second spectrum estimation model calculated by machine learning. MU The main lobe spread of (θ) is the DL spectrum S DL The spectrum selector 22 selects the MUSIC spectrum S MU The spectrum to be selected is determined based on the relationship between the standard deviation of the main lobe variance of (θ) and the threshold value.

[0056] MUSIC Spectrum S MU A large spread of (θ), that is, a large standard deviation of the spectrum, means that the estimation by the MUSIC method is ambiguous. Therefore, if the standard deviation is less than a predetermined threshold, the MUSIC spectrum S MU If it is determined that the spread of (θ) is large, the direction-of-arrival estimation device 1 selects the direction-of-arrival estimation result from the deep learning side. With the above configuration, the direction-of-arrival estimation device 1 can obtain the result of improving the estimation accuracy by selecting the direction-of-arrival estimation result from the deep learning in a situation where the estimation accuracy of the MUSIC method is low (= the spread of the MUSIC spectrum is large).

[0057] The peak search unit 18 calculates the MUSIC spectrum S based on the selection by the spectrum selection unit 22. MU (θ) or DL ​​spectrum S DL The peak search unit 18 obtains one of the MUSIC spectrum S MU (θ) or DL ​​spectrum S DL Using one of the selected spectra (θ), the estimated direction of arrival of the radio wave is calculated by the above-mentioned peak search technique (step S203).

[0058] In the above process, the direction of arrival estimation device 1 uses a linear activation function for the output layer of the neural network, and outputs the DL spectrum S DL The arrival direction estimation device 1 calculates the MUSIC spectrum S MU (θ) and DL spectrum S DL (θ) are appropriately weighted and synthesized to form the composite spectrum S Comb The arrival direction is estimated by performing peak search processing on (θ).

[0059] The MUSIC (Multiple Signal Classification) method, which is based on a subspace method using eigenvalue decomposition of a received data correlation matrix, is widely known as a direction of arrival estimation method using an array antenna, but direction of arrival estimation methods using deep learning have also been proposed, such as those in Non-Patent Document 1 and Non-Patent Document 2, which allow estimation using a trained network with simple calculations. According to the processing of each of the above-mentioned embodiments, the direction of arrival estimation device 1 discloses a method for combining the estimation results of the MUSIC method and the deep learning method in order to improve the accuracy of direction of arrival estimation.

[0060] (Minimum Configuration Embodiment) FIG. 7 is a diagram showing the minimum configuration of the direction of arrival estimation device according to this embodiment. FIG. 8 is a diagram showing a processing flow of the direction of arrival estimation device with the minimum configuration according to this embodiment. The direction of arrival estimation device 1 at least performs the functions of a first radio wave characteristic calculation means 71 (corresponding to the first spectrum calculation unit 13), a second radio wave characteristic calculation means 72 (corresponding to the second spectrum calculation unit 15), and a peak search means 73. The first radio wave characteristic calculation means 71 calculates the first radio wave characteristic, which is a characteristic of the radio wave signal and the reception strength depending on the direction of arrival of the radio wave, and indicates the characteristics of the radio wave signal to be calculated, using an estimation method other than a deep learning method (step S301). The second radio wave characteristic calculation means 72 calculates the second radio wave characteristic, which indicates the characteristics corresponding to the radio wave signal to be calculated, using an estimation model obtained by deep learning of the relationship between the radio wave signal and the characteristics of the reception strength according to the direction of arrival of the radio wave (step S302). The peak search means 73 estimates the direction of arrival of the radio wave signal by using a peak search method for the radio wave characteristics identified based on the first radio wave characteristics and the second radio wave characteristics (step S303).

[0061] The direction-of-arrival estimation device 1 described above has an internal computer system. The steps of the above-described processes are stored in the form of a program on a computer-readable recording medium, and the computer reads and executes this program to perform the above-described processes. Here, the computer-readable recording medium refers to a magnetic disk, a magneto-optical disk, a CD-ROM, a DVD-ROM, a semiconductor memory, etc. Alternatively, the computer program may be distributed to a computer via a communication line, and the computer that receives the program may execute the program.

[0062] The program may also be a program for realizing some of the functions described above, or may be a so-called differential file (differential program) that can realize the functions described above in combination with a program already recorded in the computer system. [Explanation of symbols]

[0063] 1... Direction of arrival estimation device 2. Array antenna 11. Receiving unit 12. Correlation matrix calculation section 13...First spectrum calculation unit 14. Pretreatment section 15 Second spectrum calculation unit 16 Signal measurement section 17 Spectral synthesis section 18. Peak search section 19. Learning Department 21 Spectral evaluation unit 22 Spectral selection unit 100 Direction of Arrival Estimation System

Claims

1. a first radio wave characteristic calculation means for calculating a first radio wave characteristic indicating a characteristic of a radio wave signal and its reception strength depending on the direction of arrival of the radio wave, the first radio wave characteristic indicating the characteristic of the radio wave signal to be calculated, using an estimation method other than a deep learning method; a second radio wave characteristic calculation means for calculating a second radio wave characteristic indicating the characteristic corresponding to the radio wave signal to be calculated using an estimation model obtained by deep learning of the relationship between the radio wave signal and the characteristics of the reception strength according to the direction of arrival of the radio wave; a peak search means for estimating the direction of arrival of the radio wave signal by using a peak search method for the radio wave characteristics identified based on the first radio wave characteristics and the second radio wave characteristics; An arrival direction estimation device comprising:

2. a composite radio wave characteristic calculation means for calculating a composite radio wave characteristic based on the first radio wave characteristic and the second radio wave characteristic, The peak search means estimates the direction of arrival of the radio wave signal by using the peak search method for the synthesized radio wave characteristic that is the specified radio wave characteristic. The direction of arrival estimation device according to claim 1 .

3. a signal measurement means for calculating weighting coefficient values ​​for the first radio wave characteristic and the second radio wave characteristic, based on evaluation values ​​of radio wave characteristics calculated using the first radio wave characteristic and the second radio wave characteristic, normalized so that the sum of the ratios of the evaluation values ​​becomes 1; The composite radio wave characteristic calculation means weights and combines the first radio wave characteristic and the second radio wave characteristic based on the weighting coefficients corresponding to the first radio wave characteristic and the second radio wave characteristic calculated using the received power index of the radio wave signal. The direction of arrival estimation device according to claim 2 .

4. a radio wave characteristic selection means for selecting the first radio wave characteristic when a standard deviation calculated from a variance of a main lobe of the first radio wave characteristic is less than a predetermined threshold, and for selecting the second radio wave characteristic when the standard deviation is equal to or greater than the predetermined threshold, The peak search means estimates the direction of arrival of the radio wave signal by using the peak search method for the radio wave characteristic selected by the radio wave characteristic selection means, which is the specified radio wave characteristic. The direction of arrival estimation device according to claim 1 .

5. calculating a first radio wave characteristic, which is a characteristic of a radio wave signal and a reception strength depending on the direction of arrival of the radio wave, for the radio wave signal to be calculated, using an estimation method other than the deep learning method; calculating a second radio wave characteristic indicating the characteristic corresponding to the radio wave signal to be calculated using an estimation model obtained by deep learning of a relationship between the radio wave signal and a characteristic of reception strength according to the direction of arrival of the radio wave; Estimating the direction of arrival of the radio wave signal using a peak search method for the radio wave characteristics identified based on the first radio wave characteristics and the second radio wave characteristics. Direction of arrival estimation method.

6. The computer of the direction of arrival estimation device a first radio wave characteristic calculation means for calculating a first radio wave characteristic indicating a characteristic of a radio wave signal and its reception strength depending on the direction of arrival of the radio wave, the first radio wave characteristic indicating the characteristic of the radio wave signal to be calculated, using an estimation method other than a deep learning method; a second radio wave characteristic calculation means for calculating a second radio wave characteristic indicating the characteristic corresponding to the radio wave signal to be calculated using an estimation model obtained by deep learning of the relationship between the radio wave signal and the characteristics of the reception strength according to the direction of arrival of the radio wave; a peak search means for estimating the direction of arrival of the radio wave signal by using a peak search method for the radio wave characteristics identified based on the first radio wave characteristics and the second radio wave characteristics; A program that functions as a