Selection device, estimation system, selection method, and program

Abstract

To provide a selection device, an estimation system, etc. that can appropriately select a center frequency for which a trained model should be obtained before collecting data from a real environment.SOLUTION: The selection device includes: signal generation means for generating signal data simulating an arrival signal based on parameters including at least a center frequency and arrival angle information; correlation matrix calculation means for calculating a correlation matrix based on the signal data; inference means for generating an evaluation spectrum based on a lower triangular matrix extracted from the correlation matrix using a trained model; performance evaluation means for evaluating the performance of the trained model based on an estimated arrival angle estimated based on the evaluation spectrum and a correct arrival angle of the arrival angle information; and next frequency determination means for determining a center frequency to be evaluated next based on a result of the performance evaluation. An estimation system according to the present disclosure includes the selection device and an estimation device for estimating the arrival direction of an arriving wave.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to a selection device, an estimation system, a selection method, and a program. [Background technology]

[0002] In wireless communication, it is important to estimate the direction and position of a radio wave transmission source such as a communication terminal, and various techniques related to this have been proposed.

[0003] For example, Patent Document 1 discloses a transmission source estimation device that estimates the position of an unknown transmission source. This transmission source estimation device estimates the direction of arrival for each learning point for each sensor based on the cross-correlation between antennas of the array response for each learning point and the cross-correlation in a location fingerprint database. The transmission source estimation device then estimates the location of the unknown transmission source based on the cross-correlation between antennas of the array response for each learning point, which is derived for each sensor based on the estimated direction of arrival and center frequency, and the cross-correlation between antennas of the received signals. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 2018-017695 Summary of the Invention [Problem to be solved by the invention]

[0005] The MUSIC (Multiple Signal Classification) method, which is based on a subspace method using eigenvalue decomposition of the received data correlation matrix, is widely known as a method for estimating the direction of arrival of a radio wave source when an array antenna is used as a receiving device.

[0006] Recently, a new DOA estimation method using deep learning has been proposed. This method has the advantages of being able to perform estimation with simple calculations when performing estimation (inference) using a trained network, and of being more accurate than methods such as the MUSIC method, which do not require training, under certain conditions. However, when a signal with a center frequency that is not included in the training data set used to generate the trained model used for inference is input, the evaluation spectrum cannot be generated correctly due to differences in the distribution of the input data, and the DOA cannot be accurately estimated.

[0007] To solve this problem, it is conceivable to collect signal data at fine frequency intervals when creating a training data set, and then generate a trained model using that training data set. However, this would result in a new problem of increased costs in terms of training time, storage capacity, etc. The technology described in Patent Document 1 estimates the center frequency, but does not relate to the generation of a training model, and does not disclose any method for solving this problem.

[0008] One of the objectives to be achieved by the embodiments of the present disclosure is to provide a selection device, an estimation system, a selection method, and a program that can appropriately select a center frequency for which a trained model should be obtained before collecting data from a real environment. Note that this objective is only one of multiple objectives to be achieved by multiple embodiments disclosed herein. Other objectives or problems and novel features will become apparent from the description of this specification or the accompanying drawings. [Means for solving the problem]

[0009] A selection device according to one aspect includes: a signal generating means for generating signal data simulating an arrival signal based on parameters including at least a center frequency and arrival angle information; a correlation matrix calculation means for calculating a correlation matrix based on the signal data; an inference means for generating an evaluation spectrum based on a lower triangular matrix extracted from the correlation matrix using the trained model; a performance evaluation means for evaluating the performance of the trained model based on an estimated arrival angle estimated based on the evaluation spectrum and a correct arrival angle of the arrival angle information; next frequency determining means for determining a next center frequency to be evaluated based on the result of the performance evaluation; It is equipped with the following.

[0010] In one aspect, the estimation system comprises: The selection device as described above; an estimation device for estimating the direction of arrival of an incoming wave; Equipped with The estimation device receiving means for receiving an incoming wave; second correlation matrix calculation means for calculating a second correlation matrix based on signal data of the incoming wave received by the receiving means and information on the center frequency to be learned next, obtained from the selection device; a second learning means for generating a second trained model based on a lower triangular matrix extracted from the second correlation matrix; a model selection means for storing the trained model acquired from the selection device and the second trained model; It is equipped with the following.

[0011] In one aspect, the selection method includes: generating signal data simulating the arrival signal based on parameters including at least a center frequency and arrival angle information; Calculating a correlation matrix based on the signal data; Using the trained model, an evaluation spectrum is generated based on a lower triangular matrix extracted from the correlation matrix. Evaluating the performance of the trained model based on the estimated arrival angle estimated based on the evaluation spectrum and the correct arrival angle of the arrival angle information; Based on the results of the performance evaluation, the next center frequency to be evaluated is determined. This is what a computer does.

[0012] In one aspect, the program generating signal data simulating the arrival signal based on parameters including at least a center frequency and arrival angle information; Calculating a correlation matrix based on the signal data; Using the trained model, an evaluation spectrum is generated based on a lower triangular matrix extracted from the correlation matrix. Evaluating the performance of the trained model based on the estimated arrival angle estimated based on the evaluation spectrum and the correct arrival angle of the arrival angle information; Based on the results of the performance evaluation, the next center frequency to be evaluated is determined. This is what causes a computer to execute the above. [Effects of the Invention]

[0013] According to the present disclosure, it is possible to provide a selection device, an estimation system, a selection method, and a program that can appropriately select a center frequency for which a trained model should be obtained before collecting data from a real environment. [Brief explanation of the drawings]

[0014] [Figure 1] FIG. 1 is a block diagram illustrating an example of a selection device according to the present disclosure. [Figure 2] FIG. 2 is a flowchart showing an example of the overall process executed by the selection device shown in FIG. [Figure 3] FIG. 3 is a flowchart illustrating an example of the trained model generation step shown in FIG. 2 . [Figure 4] FIG. 4 is a flowchart showing an example of the next learning frequency determining step shown in FIG. [Figure 5] FIG. 5 is a graph showing an example of the operations in steps S124 to S127 in the flowchart of FIG. [Figure 6] FIG. 6 is a block diagram illustrating an example of an estimation system according to the present disclosure. [Figure 7] FIG. 7 is a flowchart showing an example of the overall processing executed by the estimation system shown in FIG. [Figure 8] FIG. 8 is a flowchart showing an example of the learning model generation step shown in FIG. [Figure 9] FIG. 9 is a flowchart illustrating an example of the inference steps shown in FIG. [Figure 10] FIG. 10 is a block diagram showing an example of the hardware configuration of an information processing device in which the processing of the device or system is executed. DETAILED DESCRIPTION OF THE INVENTION

[0015] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. Note that the following descriptions and drawings in the embodiments have been omitted or simplified as appropriate for clarity of explanation. Furthermore, in this disclosure, unless otherwise specified, when multiple items are defined as "at least one of multiple items," the definition may mean any one item, or any multiple items including all items.

[0016] Each drawing referenced in the embodiments is merely an example for describing one or more embodiments. Each drawing may not relate only to one particular embodiment, but may also relate to one or more other embodiments. As will be understood by those skilled in the art, various features or steps described with reference to any one drawing can be combined with features or steps shown in one or more other drawings to create, for example, an embodiment not explicitly shown or described. Not all features or steps shown in any one drawing are necessarily required to describe an exemplary embodiment, and some features or steps may be omitted. The order of steps described in any drawing may be changed as appropriate. Furthermore, components or processes described with the same reference numeral throughout multiple drawings indicate the same or corresponding components or processes.

[0017] (Embodiment 1) A selection device according to a first embodiment of the present disclosure will be described below with reference to FIGS. 1 to 5. The selection device of the present disclosure assumes that a deep learning method is applied to a real environment and determines the center frequency of signal data to be collected in order to obtain a necessary and sufficient number of learned models in the frequency domain. This selection device performs advance preparation before collecting signal data in the real environment and generates learned models that satisfy required performance at predetermined frequency intervals.

[0018] Specifically, the selection device according to this embodiment first generates simulated arrival signals of a certain center frequency on a computer, generates a trained model using signal data of the generated arrival signals, and then inputs newly generated simulated arrival signal data of a different center frequency into the trained model corresponding to the generated center frequency, and outputs the estimation accuracy of the direction.

[0019] The selection device determines that the generation of trained models can be thinned out if the estimation accuracy of the frequency satisfies a predetermined accuracy, whereas the selection device determines that the generation of trained models cannot be thinned out if the estimation accuracy of the frequency does not satisfy the predetermined accuracy, and determines a center frequency at which the next model should be generated.

[0020] <Configuration of selection device> The configuration of a selection device according to this embodiment will be described with reference to Fig. 1. Fig. 1 is a block diagram showing an example of a selection device according to the present disclosure. As shown in Fig. 1, the selection device 100 includes a signal generation unit 101, a correlation matrix calculation unit 102, a teacher data storage unit 103, a learning unit 104, an inference unit 105, a performance evaluation unit 106, a next frequency determination unit 107, and a determined frequency storage unit 108. Each unit of the selection device 100 is controlled by a hardware controller (not shown).

[0021] When generating a trained model, the signal generation unit 101 generates signal data for training that simulates an arrival signal based on specified parameters. Here, the specified parameters include the arrangement of antenna elements, the center frequency of the signal, and arrival angle information. The arrival angle information is information that indicates a correct arrival angle, which will be described later. Furthermore, when determining the next training frequency, the signal generation unit 101 generates signal data that is different from that used during training based on the specified parameters. The signal generation unit 101 outputs the generated signal data to the correlation matrix calculation unit 102. Note that the correct arrival angle is output to the performance evaluation unit 106, as it is used to evaluate the center frequency.

[0022] The correlation matrix calculation unit 102 calculates a correlation matrix based on the signal data generated by the signal generation unit 101. When generating a trained model, the correlation matrix calculation unit 102 outputs the calculated correlation matrix to the teacher data storage unit 103. When determining the next learning frequency, the correlation matrix calculation unit 102 outputs the calculated correlation matrix to the inference unit 105.

[0023] The teacher data storage unit 103 is a database that stores, as teacher data, a lower triangular matrix extracted from the correlation matrix output from the correlation matrix calculation unit 102, together with arrival angle information corresponding to the teacher data signal related to the correlation matrix. Information on the arrival angle of one or more waves in the teacher data signal is associated with the lower triangular matrix corresponding to the teacher data signal and stored in the teacher data storage unit 103. Here, the teacher data storage unit 103 stores data including multiple sets of a correlation matrix and a vector indicating the characteristics of the signal corresponding to the correlation matrix, as teacher data.

[0024] The learning unit 104 generates a neural network, which is a trained model trained using a deep learning method, in the model training stage. When a correlation matrix related to a signal is input, the learning unit 104 trains the neural network so that an evaluation spectrum (evaluation function) indicating the characteristics of the signal is output as a node output from the output layer of the neural network. That is, the learning unit 104 uses the training data stored in the training data storage unit 103 to train the neural network used by the inference unit 105. In this embodiment, the learning unit 104 generates a trained model for outputting an evaluation spectrum based on a lower triangular matrix corresponding to the training data signal.

[0025] When determining the next learning frequency, the inference unit 105 inputs the correlation matrix calculated by the correlation matrix calculation unit 102 into the neural network, which is a trained model. As a result, the inference unit 105 acquires an evaluation spectrum corresponding to the signal data as an inference result. In this embodiment, the inference unit 105 uses the trained model to perform inference based on a lower triangular matrix extracted from the correlation matrix, and generates an evaluation spectrum. The inference unit 105 outputs the generated evaluation spectrum to the performance evaluation unit 106.

[0026] The performance evaluation unit 106 performs a peak search of the evaluation spectrum based on the evaluation spectrum received from the inference unit 105, and detects peaks present in the evaluation spectrum. The performance evaluation unit 106 estimates the angle of arrival based on this peak search. That is, the performance evaluation unit 106 estimates the angle indicated by the detected peak as the estimated angle of arrival.

[0027] The performance evaluation unit 106 evaluates the performance of the trained model based on the estimated arrival angle and the correct arrival angle. Specifically, the performance evaluation unit 106 calculates the estimation accuracy of the signal data of the center frequency as the performance evaluation. Then, the performance evaluation unit 106 outputs the estimation accuracy to the next frequency determination unit 107.

[0028] The next frequency determination unit 107 determines whether the estimation accuracy received from the performance evaluation unit 106 satisfies a predetermined accuracy. If the next frequency determination unit 107 determines that the estimation accuracy satisfies the predetermined accuracy, it determines that the generation of trained models for the center frequency can be thinned out, and determines the center frequency to be evaluated next. In this case, the next frequency determination unit 107 determines a new center frequency shifted by the difference between the evaluation frequency and the center frequency as the center frequency to be evaluated next. The process of determining a new center frequency will be described later with reference to FIG. 5. The next frequency determination unit 107 outputs information about the determined center frequency to be evaluated next to the determined frequency storage unit 108.

[0029] On the other hand, if the next frequency determination unit 107 determines that the estimation accuracy does not satisfy the predetermined accuracy, it determines that the generation of trained models cannot be thinned out, and determines the center frequency as the frequency at which the next trained model should be generated, i.e., the center frequency to be trained next. The next frequency determination unit 107 outputs the determined center frequency to be trained next to the determined frequency storage unit 108.

[0030] The predetermined accuracy may be set in advance by a user of the selection device 100. When the estimation accuracy exceeds the predetermined accuracy, it is determined that the estimated arrival angle is likely to be accurate. It is also determined that there is no need to learn the signal data for the target center frequency.

[0031] The determined frequency storage unit 108 stores information on the center frequency to be evaluated next, which is determined by the next frequency determination unit 107, and the center frequency to be learned next.

[0032] <Operation of the selection device> Next, the operation of the selection device 100 according to the present embodiment will be described with reference to Fig. 2 to Fig. 4. Fig. 2 is a flowchart showing an example of the overall processing executed by the selection device shown in Fig. 1. Fig. 3 is a flowchart showing an example of the trained model generation step shown in Fig. 2. Fig. 4 is a flowchart showing an example of the next learning frequency determination step shown in Fig. 2.

[0033] In the overall processing, the selection device 100 generates a trained model (step S11), determines the next learning frequency using the generated trained model (step S12), and ends the overall processing.

[0034] First, a detailed description will be given of the operation of the trained model generation step shown in step S11 in Fig. 2. Here, a center frequency A is used to generate a trained model. The signal generation unit 101 generates signal data that simulates an arriving signal based on specified parameters (including the center frequency A and arrival angle information) (step S111), and outputs the generated signal data to the correlation matrix calculation unit 102.

[0035] The correlation matrix calculation unit 102 calculates a correlation matrix based on the received signal data (step S112) and outputs the calculated correlation matrix to the training data storage unit 103. The training data storage unit 103 stores the lower triangular matrix extracted from the received correlation matrix together with the angle of arrival information as training data (step S113).

[0036] The learning unit 104 executes learning of the neural network based on the lower triangular matrix of the training data, and generates a trained model for inferring the evaluation spectrum (step S114). The learning unit 104 stores the generated trained model in a storage unit (not shown) or the determined frequency storage unit 108, and ends the series of processes.

[0037] Next, a detailed description will be given of the operation of the next learning frequency determination step shown in step S12 of Fig. 2. The signal generation unit 101 generates signal data simulating an arrival signal based on specified parameters (including a center frequency B different from that during learning and arrival angle information) (step S121), and outputs the generated signal data to the correlation matrix calculation unit 102.

[0038] The correlation matrix calculation unit 102 calculates a correlation matrix based on the received signal data (step S122) and outputs the calculated correlation matrix to the inference unit 105. The inference unit 105 executes inference processing on the correlation matrix using a trained neural network based on the lower triangular matrix extracted from the received correlation matrix, and generates an evaluation spectrum. The inference unit 105 outputs the generated evaluation spectrum to the performance evaluation unit 106.

[0039] Based on the received evaluation spectrum, the performance evaluation unit 106 executes a peak search of the evaluation spectrum and detects peaks present in the evaluation spectrum. The performance evaluation unit 106 estimates the angle indicated by the detected peak as an estimated arrival angle (estimated arrival direction) (step S123). Then, the performance evaluation unit 106 compares the estimated arrival angle with the correct arrival angle to calculate the estimation accuracy (step S124). The performance evaluation unit 106 outputs the calculated estimation accuracy to the next frequency determination unit 107.

[0040] The next frequency determination unit 107 determines whether the estimation accuracy satisfies a predetermined accuracy (step S125). If it is determined in step S125 that the estimation accuracy satisfies the predetermined accuracy, the next frequency determination unit 107 determines that generation of trained models can be thinned out, and the processing flow proceeds to step S126. Then, the next frequency determination unit 107 determines the center frequency to be evaluated next. The method of this determination will be described later.

[0041] The next frequency determining unit 107 changes the center frequency of the evaluation signal to the determined center frequency (step S126), and the process flow returns to step S121. The signal generating unit 101 generates new signal data that simulates the incoming signal based on the changed center frequency (step S121). The value of the center frequency that is determined to have an estimation accuracy that satisfies the predetermined accuracy may be stored in the determined frequency storage unit 108.

[0042] On the other hand, if it is determined in step S125 that the estimation accuracy does not satisfy the predetermined accuracy, the next frequency determination unit 107 determines this center frequency as the center frequency to be learned next (step S127), and stores the value of this center frequency in the determined frequency storage unit 108. This ends the processing flow.

[0043] As long as the next frequency determining unit 107 determines in step S125 that the estimation accuracy satisfies the predetermined accuracy, the processing flow repeatedly executes the processes from step S121 to step S125.

[0044] Next, an example of the operation of the flowchart in Fig. 4 will be described. Fig. 5 is a graph showing an example of the operation of steps S124 to S127 in the flowchart in Fig. 4. In the graph in Fig. 5, the vertical axis represents the estimation accuracy evaluated by the performance evaluation unit 106 as the square root of the mean squared error (MSE: Mean Squared Error) (RMSE), and the horizontal axis represents the center frequency. In this graph, the larger the value along the vertical axis, the larger the error, and therefore the higher the RMSE value, the worse the accuracy. In other words, the smaller the RMSE value, the better the estimation accuracy.

[0045] In Figure 5, f c1 is the center frequency of the trained model, and Δf is the minimum value of the difference between the evaluation frequency and the center frequency of the trained model. In the flowchart of Figure 4, the center frequency f c1 +Δf is set as the center frequency B, and the estimation accuracy is evaluated in the first loop (step S125). As shown in FIG. 5, the estimation accuracy in the first loop is evaluated based on the predetermined accuracy R r Since it is smaller than c1 There is no need to learn about +Δf.

[0046] So, for the second loop, the next center frequency to evaluate is f c1 In the example of FIG. 5, the RMSE is determined to be +2Δf (step S126). rSince it is smaller than c1 There is no need to learn +2Δf. Therefore, the third loop is performed. As a result, the estimated accuracy RMSE is equal to the specified accuracy R r Since the estimated accuracy is poor, it is determined in step S125 that the predetermined accuracy is not satisfied.

[0047] where the center frequency f c1 +2Δf estimation accuracy RMSE and center frequency f c1 +3Δf and the line connecting the estimated accuracy RMSE and the specified accuracy R r The frequency at the intersection with the line is f, and the center frequency at which learning should be performed next is f C2 In this case, the center frequency f c2 can be determined by the following equation (1): f c2 =f c1 +2×|f x -f c1 | (1)

[0048] In Figure 5, the center frequency f c1 For the center frequency f c2 In contrast, when loop processing is performed in the direction of lower frequencies, the sign before the second term in (Equation 1) is inverted.

[0049] Furthermore, in the process of determining the center frequency to be evaluated described above, the minimum value of the difference between the evaluation frequency and the center frequency of the trained model is fixed as Δf. However, the magnitude (absolute value) of Δf may be changed when the frequency direction in which loop processing is performed is positive (high frequency direction) and when it is negative (low frequency direction). For example, by actively utilizing the property that the frequency range that satisfies the required performance is wide in the high frequency direction, Δf when the frequency direction in which loop processing is performed is positive may be set to a value larger than Δf when it is negative.

[0050] As described above, according to the selection device 100 of this embodiment, the signal generation unit 101 generates signal data simulating an arrival signal based on parameters including at least a center frequency and arrival angle information. The correlation matrix calculation unit 102 calculates a correlation matrix based on this signal data. The inference unit 105 uses a trained model to generate an evaluation spectrum based on a lower triangular matrix extracted from the correlation matrix. The performance evaluation unit 106 evaluates the performance of the trained model based on the estimated arrival angle estimated based on the evaluation spectrum and the correct arrival angle of the arrival angle information. The next frequency determination unit 107 determines the next center frequency to be evaluated based on the results of the performance evaluation. By configuring the selection device 100 in this manner, it is possible to select a center frequency for which a trained model should be obtained before collecting signal data in a real environment. Furthermore, for center frequencies with high performance evaluations, there is no need for training by the learning unit 104. This reduces the costs associated with generating and storing trained models during actual operation to estimate arrival angles.

[0051] The selection device 100 may be configured as a single computer device or as a distributed system having multiple computer devices. In a distributed system, the processing executed by the selection device 100 can be shared and executed by multiple computer devices.

[0052] (Embodiment 2) An estimation system according to a second embodiment of the present disclosure will be described below with reference to Figs. 6 to 10. The estimation system according to this embodiment is a system in which the selection device 100 according to the first embodiment is added to the upstream stage of an estimation device that estimates the arrival direction of a signal. The estimation system according to this embodiment selects a center frequency to be calculated, generates a trained model using actual data, and estimates the arrival direction using the trained model.

[0053] <Configuration of estimation system> The configuration of an estimation system according to the present embodiment will be described with reference to Fig. 6. Fig. 6 is a block diagram showing an example of an estimation system according to the present disclosure. As shown in Fig. 6, the estimation system 1 includes a selection device 100 according to the first embodiment and an estimation device 200. In order to distinguish between the generation of a trained model in the selection device 100 and the generation of a trained model in the estimation device 200, in Fig. 6, the teacher data storage unit 103 is referred to as a first teacher data storage unit 103, and the learning unit 104 is referred to as a first learning unit 104.

[0054] The determined frequency storage unit 108 of the selection device 100 according to this embodiment is configured to output the center frequency to be learned next to the estimation device 200. Note that other configurations of the selection device 100 according to this embodiment are the same as those of the selection device 100 according to the first embodiment, and therefore detailed description thereof will be omitted.

[0055] 6, estimation device 200 includes receiving unit 201, correlation matrix calculation unit 202, second teacher data storage unit 203, second learning unit 204, model selection unit 205, inference unit 206, and arrival angle determination unit 207.

[0056] The receiving unit 201 is connected to an array antenna including a plurality of antenna elements as components. The receiving unit 201 receives a signal (received signal) wirelessly transmitted from a transmitting device (not shown) as an arriving wave via the array antenna. The receiving unit 201 outputs signal data of the received arriving wave to the correlation matrix calculation unit 202. Note that when estimating the direction of arrival of the arriving wave, the signal received by the receiving unit 201 does not need to be limited to the center frequency selected by the selection device 100.

[0057] The correlation matrix calculation unit 202 calculates a correlation matrix based on the signal data output from the receiving unit 201. In this embodiment, when generating a trained model, the correlation matrix calculation unit 102 calculates a correlation matrix based on the center frequency to be trained next, which is acquired from the selection device 100, and outputs the calculated correlation matrix to the teacher data storage unit 203. Furthermore, when estimating the direction of arrival of an arriving wave, the correlation matrix calculation unit 202 outputs the calculated correlation matrix to the model selection unit 205.

[0058] The second teacher data storage unit 203 is a database that stores, as teacher data, a lower triangular matrix extracted from the correlation matrix output from the correlation matrix calculation unit 202, together with arrival angle information corresponding to the teacher data signal related to the correlation matrix. Information on the arrival angle of one or more waves in the teacher data signal is associated with the lower triangular matrix corresponding to the teacher data signal, and is stored in the second teacher data storage unit 203. Here, the second teacher data storage unit 203 stores, as teacher data, data including multiple sets of a correlation matrix and a vector indicating the characteristics of the signal corresponding to the correlation matrix.

[0059] The second learning unit 204 generates a neural network, which is a trained model trained using a deep learning technique, in the model training stage. The second learning unit 204 performs the same operation as the learning unit 104 of the selection device 100. That is, the second learning unit 204 generates a trained model for outputting an evaluation spectrum based on a lower triangular matrix corresponding to the received signal.

[0060] The model selection unit 205 stores the trained models obtained by the training unit 104 and the second training unit 204 of the selection device 100. When estimating the direction of arrival of an incoming wave, the model selection unit 205 selects, as the second trained model, a trained model trained with signal data having a center frequency closest to the frequency of the received signal received by the receiving unit 201. The model selection unit 205 outputs the selected second trained model to the inference unit 206.

[0061] When estimating the direction of arrival of an arriving wave, the inference unit 206 inputs the correlation matrix calculated by the correlation matrix calculation unit 202 to the neural network, which is the second trained model selected by the model selection unit 205. As a result, the inference unit 206 acquires an evaluation spectrum corresponding to the signal data as an inference result. The inference unit 206 outputs the evaluation spectrum obtained by performing inference using the second trained model to the arrival angle determination unit 207.

[0062] The arrival angle determination unit 207 performs a peak search of the evaluation spectrum based on the evaluation spectrum received from the inference unit 206, and detects peaks present in the evaluation spectrum. The arrival angle determination unit 207 sorts the detected peaks in descending order of peak values. The arrival angle determination unit 207 then selects peaks in descending order of peak value, the number of which corresponds to the number of arriving waves output by the wave number estimation unit 228, and outputs the angles indicated by the selected peaks as estimated arrival angles. Note that the arrival angle determination unit 207 may output the estimated arrival angles to a display unit (not shown) connected to the estimation device 200, thereby visualizing information about the arrival angles for the user.

[0063] In Figure 6, the first teacher data storage unit 103 and the second teacher data storage unit 203 are shown as being separately included in the selection device 100 and the estimation device 200, which are different devices, but the teacher data storage unit may also be provided as a single storage.

[0064] Also, in Figure 6, the first learning unit 104 and the second learning unit 204 are shown as being separately included in the selection device 100 and the estimation device 200, which are different devices, but their functions may also be realized by executing processing on a CPU (Central Processing Unit) or GPU (Graphic Processing Unit) provided in a single device.

[0065] <Operation of the estimation system> Next, the operation of the estimation system 1 according to this embodiment will be described with reference to Fig. 7 to Fig. 9. Fig. 7 is a flowchart showing an example of the overall processing executed by the estimation system 1 shown in Fig. 6. Fig. 8 is a flowchart showing an example of the learning model generation step shown in Fig. 7. Fig. 9 is a flowchart showing an example of the inference step shown in Fig. 7.

[0066] In the overall processing, the selection device 100 selects a center frequency to be learned next (step S21) and outputs the selected center frequency to the estimation device 200. The estimation device 200 generates a trained model using actual data based on the selected center frequency (step S22), and estimates the direction of arrival (azimuth) of the received incoming wave using one of the trained models (step S23), thereby terminating this overall processing.

[0067] The step of selecting the next center frequency to be learned, shown as step S21 in FIG. 7, is substantially the same as that described in detail in the first embodiment, and therefore a detailed description thereof will be omitted here.

[0068] Next, with reference to Fig. 8, the detailed operation of the learning model generation step shown in step S22 of Fig. 7 will be described. First, the receiving unit 201 acquires a received signal as a training data signal (step S221). The correlation matrix calculation unit 202 generates a correlation matrix using signal data of the acquired received signal (step S222). The second training data storage unit 203 stores, as training data, a lower triangular matrix extracted from the correlation matrix and arrival angle information corresponding to the training data signal related to the correlation matrix (step S223).

[0069] Next, the second learning unit 204 generates a second trained model for inferring an evaluation spectrum based on the lower triangular matrix of the training data (step S224). In this trained model generation step shown in step S22, this loop continues until the generation of a second trained model corresponding to the number of collected signal data of the center frequency selected in the center frequency selection step to be trained next shown in step S21 is completed. In other words, this loop is repeated the number of times corresponding to the number of signal data of the center frequency to be trained, and a second trained model corresponding to the number of loops is generated, and the processing of step S22 is terminated.

[0070] Next, with reference to Fig. 9, the detailed operation of the inference step shown in step S23 of Fig. 7 will be described. First, the receiver 201 acquires a received signal as a target signal for direction estimation (step S231). The correlation matrix calculator 202 generates a correlation matrix using signal data of the received signal (step S232). The model selector 205 selects a trained model trained with signal data having a center frequency closest to the frequency of the received signal received by the receiver 201 (step S233), and outputs the selected trained model to the inference unit 206.

[0071] The inference unit 206 uses the trained model received from the model selection unit 205 to generate an evaluation spectrum based on the lower triangular matrix extracted from the correlation matrix (step S234), and outputs the generated evaluation spectrum to the arrival angle determination unit 207.

[0072] The arrival angle determining unit 207 executes a peak search on the evaluation spectrum (step S235), detects a peak present in the evaluation spectrum, and outputs the detected peak as an estimated arrival angle (step S236).

[0073] In the inference step shown in step S23, the processing of steps S231 to S236 in the flowchart shown in FIG. 9 may be repeated in a loop as long as the receiving unit 201 continues to receive a signal.

[0074] As described above, the estimation system 1 according to this embodiment includes the selection device 100 and the estimation device 200. The selection device 100 includes the next frequency determination unit 107 that determines whether the estimation accuracy calculated by the performance evaluation unit 106 satisfies a predetermined accuracy. If the next frequency determination unit 107 determines that the estimation accuracy satisfies the predetermined accuracy, it determines a new center frequency shifted by the difference between the evaluation frequency and the center frequency of the signal data simulating the arrival signal as the center frequency to be evaluated next. On the other hand, if the next frequency determination unit 107 determines that the estimation accuracy does not satisfy the predetermined accuracy, it determines this center frequency as the center frequency to be learned next and outputs the determined center frequency to the estimation device 200. Furthermore, in the estimation device 200, the correlation matrix calculation unit 202 calculates a second correlation matrix based on the signal data of the arrival wave received by the receiving unit 201 and information regarding the center frequency to be learned next, acquired from the selection device 100. The second learning unit 204 generates a second trained model based on the lower triangular matrix extracted from the second correlation matrix. By configuring the estimation system 1 in this manner, it is possible to select a center frequency for which a trained model should be obtained before collecting signal data in a real environment, as in the first embodiment. Furthermore, for center frequencies with high performance evaluations, there is no need for training by the second learning unit 204. This reduces the costs associated with generating and storing trained models for actual operation to estimate the angle of arrival, making it possible to perform actual direction-of-arrival estimation using a trained model with reduced costs.

[0075] The estimation system 1, selection device 100, and estimation device 200 according to this embodiment may be configured as a single computer device, similar to the selection device 100 shown in embodiment 1, or may be configured as a distributed system having multiple computer devices.

[0076] The selection device 100 and estimation system 1 according to the present disclosure can be used for applications such as radio wave monitoring, cognitive radio, dynamic frequency sharing, and police and security services, but the applications of the device are not limited to these.

[0077] In the above-described embodiment, the present disclosure has been described as a hardware configuration, but the present disclosure is not limited to this. The present disclosure can also be realized by causing a processor in a computer to execute a computer program to perform the processing of the selection device 100 and each device constituting the estimation system 1 described in the above-described embodiment.

[0078] 10 is a block diagram showing an example of the hardware configuration of an information processing device (in other words, a computer) on which the processing of the system or device described in the present disclosure is executed. Referring to FIG. 10, the information processing device 1000 includes a signal processing circuit 1010, a processor 1020, and a memory 1030.

[0079] The signal processing circuit 1010 is a circuit for processing signals in accordance with the control of the processor 1020. Note that the signal processing circuit 1010 may include a communication circuit for receiving signals from a transmitting device.

[0080] The processor 1020 is connected to the memory 1030, and performs the processing of the system described in the above-described embodiment by reading and executing a computer program from the memory 1030. As an example of the processor 1020, one of a central processing unit (CPU), a micro processing unit (MPU), a field-programmable gate array (FPGA), a demand-side platform (DSP), and an application-specific integrated circuit (ASIC) may be used, or a plurality of these may be used in parallel.

[0081] The memory 1030 is configured with a volatile memory, a nonvolatile memory, or a combination thereof. The memory 1030 is not limited to one, and a plurality of memories may be provided. The volatile memory may be, for example, a random access memory (RAM) such as a dynamic random access memory (DRAM) or a static random access memory (SRAM). The nonvolatile memory may be, for example, a read only memory (ROM) such as a programmable random only memory (PROM) or an erasable programmable read only memory (EPROM), a flash memory, or a solid state drive (SSD).

[0082] The memory 1030 is used to store one or more instructions. Here, the one or more instructions are stored as programs in the memory 1030. The processor 1020 can perform the processing described in the above-mentioned embodiments by reading and executing these programs from the memory 1030.

[0083] The memory 1030 may include memory built into the processor 1020 in addition to memory provided outside the processor 1020. The memory 1030 may also include storage located away from the processors constituting the processor 1020. In this case, the processor 1020 can access the memory 1030 via an I / O (Input / Output) interface.

[0084] As described above, one or more processors included in each device or system in the above-described embodiments execute one or more programs including instructions for causing a computer to execute the algorithms described using the drawings. Execution of the programs enables the information processing described in each embodiment to be realized.

[0085] The program includes instructions or software code that, when loaded into a computer, causes the computer to perform one or more functions described in the above embodiments. The program may be stored in a non-transitory computer-readable medium or a tangible storage medium. By way of example and not limitation, computer-readable media or tangible storage media include random-access memory (RAM), read-only memory (ROM), flash memory, solid-state drive (SSD) or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disk (DVD), Blu-ray disc or other optical disk storage, magnetic cassette, magnetic tape, magnetic disk storage or other magnetic storage devices. The program may also be transmitted on a transient computer-readable medium or communication medium. By way of example and not limitation, transient computer-readable media or communication media include electrical, optical, acoustic, or other forms of propagated signals. The transient computer-readable medium or communication medium may provide the program to the computer via a wired communication path, such as an electric wire or optical fiber, or via a wireless communication path.

[0086] A part or all of the above-described embodiments can be described as, but not limited to, the following supplementary notes. (Appendix 1) a signal generating means for generating signal data simulating an arrival signal based on parameters including at least a center frequency and arrival angle information; a correlation matrix calculation means for calculating a correlation matrix based on the signal data; an inference means for generating an evaluation spectrum based on a lower triangular matrix extracted from the correlation matrix using a trained model; a performance evaluation means for evaluating the performance of the trained model based on an estimated arrival angle estimated based on the evaluation spectrum and a correct arrival angle of the arrival angle information; a next frequency determining means for determining a next center frequency to be evaluated based on the result of the performance evaluation; Equipped with Selection device. (Appendix 2) further comprising a learning means for generating the trained model for outputting the evaluation spectrum; the signal generating means generates learning signal data simulating an arrival signal based on parameters including at least a center frequency different from the center frequency and the arrival angle information; the correlation matrix calculation means calculates a correlation matrix for learning based on the signal data for learning, The learning means generates the trained model based on a lower triangular matrix extracted from the training correlation matrix. 1. A selection device as described in Appendix 1. (Appendix 3) the performance evaluation means calculates estimation accuracy in the signal data of the center frequency based on the estimated arrival angle based on a peak search for the evaluation spectrum and the correct arrival angle. 1. A selection device as described in Appendix 1. (Appendix 4) the next frequency determining means determines whether the estimation accuracy satisfies a predetermined accuracy; If it is determined that the estimation accuracy satisfies the predetermined accuracy, a new center frequency shifted by the difference between the evaluation frequency and the center frequency is determined as the center frequency to be evaluated next; If it is determined that the estimation accuracy does not satisfy the predetermined accuracy, the center frequency is determined to be the center frequency to be learned next. 1. A selection device as described in Appendix 3. (Appendix 5) the next frequency determining means outputs the new center frequency to the signal generating means; the signal generating means generates new signal data simulating the arrival signal based on parameters including at least the new center frequency and the arrival angle information; the correlation matrix calculation means calculates a new correlation matrix based on the new signal data; the inference means generates a new evaluation spectrum based on a new lower triangular matrix extracted from the new correlation matrix using the trained model; the performance evaluation means calculates estimation accuracy for the new signal data of the changed center frequency based on a new estimated arrival angle based on a peak search for the new evaluation spectrum and the correct arrival angle. 1. A selection device as described in Appendix 4. (Appendix 6) a determined frequency storage means for storing information on the new center frequency determined as the next center frequency to be evaluated and information on the frequency determined as the learned frequency, 6. A selection device according to claim 4 or 5. (Appendix 7) A selection device according to any one of Supplementary Notes 4 to 6; an estimation device for estimating the direction of arrival of an incoming wave; Equipped with The estimation device includes: receiving means for receiving the incoming wave; a second correlation matrix calculation means for calculating a second correlation matrix based on the signal data of the arriving wave received by the receiving means and the information on the center frequency to be learned next acquired from the selection device; a second learning means for generating a second trained model based on a lower triangular matrix extracted from the second correlation matrix; A model selection means for saving the trained model acquired from the selection device and the second trained model; Equipped with Estimation system. (Appendix 8) the model selection means selects a trained model trained with signal data having the center frequency closest to the frequency of the arriving wave, The estimation device includes: a second inference means for generating an evaluation spectrum based on a lower triangular matrix extracted from the correlation matrix by using the selected one trained model; an arrival angle determination means for determining an arrival angle of the arrival wave received by the receiving means based on the evaluation spectrum generated by the second inference means; Further provided with 10. The estimation system of claim 7. (Appendix 9) generating signal data simulating the arrival signal based on parameters including at least a center frequency and arrival angle information; calculating a correlation matrix based on the signal data; generating an evaluation spectrum based on a lower triangular matrix extracted from the correlation matrix using the trained model; evaluating the performance of the trained model based on the estimated arrival angle estimated based on the evaluation spectrum and the correct arrival angle of the arrival angle information; determining a next center frequency to be evaluated based on the result of the performance evaluation; A selection method that is performed by a computer. (Appendix 10) generating signal data simulating the arrival signal based on parameters including at least a center frequency and arrival angle information; calculating a correlation matrix based on the signal data; generating an evaluation spectrum based on a lower triangular matrix extracted from the correlation matrix using the trained model; calculating an estimation accuracy of the signal data of the center frequency based on the estimated arrival angle based on a peak search for the evaluation spectrum and the correct arrival angle; If it is determined that the estimation accuracy does not satisfy the predetermined accuracy, the center frequency is determined to be the center frequency to be learned next; Receives incoming waves, calculating a second correlation matrix based on the signal data of the received arriving wave and the information about the center frequency to be learned next; generating a second trained model based on a lower triangular matrix extracted from the second correlation matrix; This is an estimation method performed by a computer. (Appendix 11) generating signal data simulating the arrival signal based on parameters including at least a center frequency and arrival angle information; calculating a correlation matrix based on the signal data; generating an evaluation spectrum based on a lower triangular matrix extracted from the correlation matrix using the trained model; evaluating the performance of the trained model based on the estimated arrival angle estimated based on the evaluation spectrum and the correct arrival angle of the arrival angle information; determining a next center frequency to be evaluated based on the result of the performance evaluation; A program that makes a computer do something. (Appendix 12) generating signal data simulating the arrival signal based on parameters including at least a center frequency and arrival angle information; calculating a correlation matrix based on the signal data; generating an evaluation spectrum based on a lower triangular matrix extracted from the correlation matrix using the trained model; calculating an estimation accuracy of the signal data of the center frequency based on the estimated arrival angle based on a peak search for the evaluation spectrum and the correct arrival angle; If it is determined that the estimation accuracy does not satisfy the predetermined accuracy, the center frequency is determined to be the center frequency to be learned next; Receives incoming waves, calculating a second correlation matrix based on the signal data of the received arriving wave and the information about the center frequency to be learned next; generating a second trained model based on a lower triangular matrix extracted from the second correlation matrix; A program that makes a computer do something.

[0087] Some or all of the elements (e.g., configurations and functions) described in Appendix 2 and Appendix 3 that are dependent on Appendix 1 may also be dependent on Appendix 7 in the same dependency relationship as Appendix 2 and Appendix 3. Also, some or all of the elements (e.g., configurations and functions) described in Appendix 2 and Appendix 3 that are dependent on Appendix 1 may also be dependent on Appendix 11 in the same dependency relationship as Appendix 2 and Appendix 3. Some or all of the elements described in any appendix may be applicable to various hardware, software, recording means for recording software, systems, and methods.

[0088] Although the present disclosure has been described above with reference to the embodiments, the present disclosure is not limited to the above-described embodiments. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present disclosure within the scope of the present disclosure. Furthermore, each embodiment can be combined with other embodiments as appropriate. [Explanation of symbols]

[0089] 1. Estimation System 100 selection device 200 Estimation device 101 signal generation unit 201 Receiving unit 102, 202 Correlation matrix calculation unit 103 (1st) Teacher Data Storage Department 203 Second Teacher Data Storage Unit 104 (1st) Learning Department 204 Second Learning Section 205 Model Selection Section 105, 206 Reasoning part 106 Performance Evaluation Department 107 Next frequency determination unit 108 Determined frequency storage section 207 Arrival angle determination unit 1000 Information Processing Device 1010 Signal Processing Circuit 1020 processor 1030 memory

Claims

1. a signal generating means for generating signal data simulating an arrival signal based on parameters including at least a center frequency and arrival angle information; a correlation matrix calculation means for calculating a correlation matrix based on the signal data; an inference means for generating an evaluation spectrum based on a lower triangular matrix extracted from the correlation matrix using a trained model; a performance evaluation means for evaluating the performance of the trained model based on an estimated arrival angle estimated based on the evaluation spectrum and a correct arrival angle of the arrival angle information; a next frequency determining means for determining a next center frequency to be evaluated based on the result of the performance evaluation; Equipped with Selection device.

2. further comprising a learning means for generating the trained model for outputting the evaluation spectrum; the signal generating means generates learning signal data simulating an arrival signal based on parameters including at least a center frequency different from the center frequency and the arrival angle information; the correlation matrix calculation means calculates a correlation matrix for learning based on the signal data for learning, The learning means generates the trained model based on a lower triangular matrix extracted from the training correlation matrix. The selection device according to claim 1 .

3. the performance evaluation means calculates estimation accuracy in the signal data of the center frequency based on the estimated arrival angle based on a peak search for the evaluation spectrum and the correct arrival angle. The selection device according to claim 1 .

4. the next frequency determining means determines whether the estimation accuracy satisfies a predetermined accuracy; If it is determined that the estimation accuracy satisfies the predetermined accuracy, a new center frequency shifted by the difference between the evaluation frequency and the center frequency is determined as the center frequency to be evaluated next; If it is determined that the estimation accuracy does not satisfy the predetermined accuracy, the center frequency is determined to be the center frequency to be learned next. The selection device according to claim 3 .

5. the next frequency determining means outputs the new center frequency to the signal generating means; the signal generating means generates new signal data simulating the arrival signal based on parameters including at least the new center frequency and the arrival angle information; the correlation matrix calculation means calculates a new correlation matrix based on the new signal data; the inference means generates a new evaluation spectrum based on a new lower triangular matrix extracted from the new correlation matrix using the trained model; the performance evaluation means calculates estimation accuracy for the new signal data of the changed center frequency based on a new estimated arrival angle based on a peak search for the new evaluation spectrum and the correct arrival angle. The selection device according to claim 4.

6. a determined frequency storage means for storing information on the new center frequency determined as the next center frequency to be evaluated and information on the frequency determined as the learned frequency, The selection device according to claim 4.

7. A selection device according to any one of claims 4 to 6; an estimation device for estimating the direction of arrival of an incoming wave; Equipped with The estimation device includes: receiving means for receiving the incoming wave; a second correlation matrix calculation means for calculating a second correlation matrix based on the signal data of the incoming wave received by the receiving means and the information on the center frequency to be learned next obtained from the selection device; a second learning means for generating a second trained model based on a lower triangular matrix extracted from the second correlation matrix; a model selection means for storing the trained model acquired from the selection device and the second trained model; Equipped with Estimation system.

8. the model selection means selects a trained model trained with signal data having the center frequency closest to the frequency of the arriving wave, The estimation device includes: a second inference means for generating an evaluation spectrum based on a lower triangular matrix extracted from the correlation matrix by using the selected one trained model; an arrival angle determining means for determining an arrival angle of the arrival wave received by the receiving means based on the evaluation spectrum generated by the second inference means; Further provided with The estimation system according to claim 7 .

9. generating signal data simulating the arrival signal based on parameters including at least a center frequency and arrival angle information; calculating a correlation matrix based on the signal data; generating an evaluation spectrum based on a lower triangular matrix extracted from the correlation matrix using the trained model; evaluating the performance of the trained model based on the estimated arrival angle estimated based on the evaluation spectrum and the correct arrival angle of the arrival angle information; determining a next center frequency to be evaluated based on the result of the performance evaluation; A selection method that is performed by a computer.

10. generating signal data simulating the arrival signal based on parameters including at least a center frequency and arrival angle information; calculating a correlation matrix based on the signal data; generating an evaluation spectrum based on a lower triangular matrix extracted from the correlation matrix using the trained model; evaluating the performance of the trained model based on the estimated arrival angle estimated based on the evaluation spectrum and the correct arrival angle of the arrival angle information; determining a next center frequency to be evaluated based on the result of the performance evaluation; A program that makes a computer do something.

Images