Signal characteristic identification device, signal characteristic learning device, signal characteristic identification method, control circuit, and storage medium

Abstract

The signal parameter identification device (1) includes a signal acquisition unit (11) that acquires time series data from a signal to be processed and outputs it as input time series data, a signal analysis unit (12) that estimates the center frequency and bandwidth of signal components included in the input time series data, a multiple filter application unit (13) that has multiple digital filters each configured to be able to change filter characteristics based on the estimation result by the signal analysis unit, applies the multiple digital filters to filter the input time series data, and generates multiple filter response time series data, a feature extraction unit (14) that extracts features for each of the multiple filter response time series data, and an inference unit (20) that infers the signal parameters of the signal based on the features.

Description

[Technical Field]

[0001] The present disclosure relates to a signal parameter identification device, a signal parameter learning device, a signal parameter identification method, a control circuit, and a storage medium. [Background technology]

[0002] Signal identification is necessary for analyzing signals in cognitive radio, which improves frequency utilization efficiency, and for analyzing signals that may cause interference to other systems. Signal identification is used to identify the modulation method used for primary modulation, such as QPSK (Quadrature Phase Shift Keying) or FSK (Frequency Shift Keying), the spreading method or multiplexing method used for secondary modulation, such as DSSS (Direct Sequence Spread Spectrum), FHSS (Frequency Hopping Spread Spectrum), or OFDM (Orthogonal Frequency Division Multiplexing), and to identify systems operating under different communication standards, such as Wi-Fi (registered trademark) or Bluetooth (registered trademark). Signal identification is achieved by classifying the acquired signal into one of a set of predefined candidates.

[0003] An example of a technology related to signal parameter identification is described in Patent Document 1. The technology described in Patent Document 1 uses machine learning to classify signals. Specifically, feature quantities of acquired time-series signals are extracted, and a trained model is generated by learning using the extracted feature quantities. Furthermore, signals are classified by performing inference using the trained model. When extracting feature quantities in the learning process and inference process, the technology described in Patent Document 1 extracts feature quantities for each filter response time-series data obtained through multiple randomly selected digital filters, thereby suppressing noise components contained in signals with various frequency characteristics using a set of filters and improving inference accuracy. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 6956913 Summary of the Invention [Problem to be solved by the invention]

[0005] The signal classification method described in Patent Document 1 extracts features for each piece of filter response time-series data obtained through multiple randomly selected digital filters. Therefore, when time-series data in which signals with a bandwidth exist in a specific frequency range, such as signals used in wireless communication or radar, or audio signals, is input, the number of filters that do not contribute to feature extraction increases, resulting in a decrease in inference accuracy.

[0006] The present disclosure has been made in view of the above, and aims to provide a signal parameter identification device that can prevent a decrease in identification accuracy when identifying a signal present in a specific frequency range. [Means for solving the problem]

[0007] In order to solve the above-mentioned problems and achieve the object, the signal parameter identification device according to the present disclosure is characterized by comprising: a signal acquisition unit that acquires time series data from a signal to be processed and outputs it as input time series data; a signal analysis unit that estimates the center frequency and bandwidth of signal components included in the input time series data; a multiple filter application unit that has a plurality of digital filters, each configured to be able to change filter characteristics based on the estimation results by the signal analysis unit, and applies the plurality of digital filters to filter the input time series data and generate a plurality of filter response time series data; a feature extraction unit that extracts features for each of the plurality of filter response time series data; and an inference unit that infers the signal parameters of the signal based on the features. [Effects of the Invention]

[0008] The signal parameter identification device according to the present disclosure has the advantage of being able to prevent a decrease in identification accuracy when identifying a signal present in a specific frequency range. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 is a diagram showing an example of the configuration of a signal parameter identification device according to a first embodiment; [Figure 2] FIG. 1 is a diagram showing an example of hardware for realizing a signal parameter identification device according to a first embodiment. [Figure 3] FIG. 10 is a diagram showing another example of hardware for realizing the signal parameter identification device according to the first embodiment. [Figure 4] FIG. 1 is a diagram illustrating an example of the configuration of a signal analysis unit of a signal parameter identification device according to a first embodiment. [Figure 5] FIG. 1 shows an example of the configuration and operation of a multiple filter application unit according to the first embodiment. [Figure 6] 1 is a flowchart showing an example of the overall operation of the signal parameter identification device according to the first embodiment. [Figure 7] FIG. 1 is a diagram showing an example of the configuration of a signal specification learning device according to a first embodiment; [Figure 8] 1 is a flowchart showing an example of the overall operation of the signal specification learning device according to the first embodiment. [Figure 9] 10 is a flowchart showing another example of the overall operation of the signal specification learning device according to the first embodiment. [Figure 10] FIG. 10 is a diagram showing an example of the configuration of a signal parameter identification device according to a second embodiment; [Figure 11] FIG. 10 is a diagram illustrating a configuration example of a signal analysis unit of a signal parameter identification device according to a second embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0010] A signal parameter identification device, a signal parameter learning device, a signal parameter identification method, a control circuit, and a storage medium according to embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0011] Embodiment 1 1 is a diagram showing an example of the configuration of a signal parameter identification device 1 according to embodiment 1. The signal parameter identification device 1 according to this embodiment includes a time-series feature extraction unit 10 having a signal acquisition unit 11, a signal analysis unit 12, a multiple filter application unit 13, and a feature extraction unit 14, and an inference unit 20. The signal parameter identification device 1 identifies the signal parameters of a signal whose signal parameters are unknown as a processing target.

[0012] Here, a description will be given of the hardware configuration of the signal parameter identification device 1. Fig. 2 is a diagram showing an example of hardware that realizes the signal parameter identification device 1 according to the first embodiment. That is, the signal parameter identification device 1 is realized by a processing device 500 shown in Fig. 2.

[0013] The processing device 500 is dedicated hardware that realizes the signal parameter identification device 1, and may be, for example, a single circuit, a decoding circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or a combination of these.

[0014] 3 is a diagram showing another example of hardware for realizing the signal parameter identification device 1 according to Embodiment 1. That is, the signal parameter identification device 1 is realized by a processor 600 and a memory 700 shown in FIG.

[0015] The processor 600 is configured by, for example, a CPU (Central Processing Unit, also referred to as a central processing unit, processing unit, or arithmetic unit), a microprocessor, a microcontroller, a DSP (Digital Signal Processor), a system LSI (Large Scale Integration), etc. The memory 700 is, for example, a semiconductor memory such as a RAM (Random Access Memory), a ROM (Read Only Memory), a flash memory, an EPROM (Erasable Programmable Read Only Memory), or an EEPROM (registered trademark) (Electrically Erasable Programmable Read Only Memory), a magnetic disk, a flexible disk, etc.

[0016] 1 are realized by software, firmware, or a combination of software and firmware. The software or firmware that realizes each function is written as a program and stored in memory 700. The processor 600 reads and executes the program stored in memory 700, thereby realizing each function. The program stored in memory 700 is provided to a user or the like in a state written on a storage medium such as a CD (Compact Disc)-ROM or a DVD (Digital Versatile Disc)-ROM, and is installed in memory 700 by performing a predetermined operation or the like.

[0017] Next, we will explain the operation of identifying signal parameters by the signal parameter identification device 1. First, we will explain the operation of the signal acquisition unit 11, signal analysis unit 12, multiple filter application unit 13, and feature extraction unit 14 of the time-series feature extraction unit 10.

[0018] The signal acquisition unit 11 acquires time-series data from a signal to be subjected to the classification process and outputs the acquired time-series data as input time-series data D1. The input time-series data D1 is time-series information indicating physical quantities acquired at predetermined time intervals. Specifically, the input time-series data D1 may be a wireless communication signal or a radar signal acquired through an antenna that has been down-converted, quadrature-detected, or otherwise converted into time-series information by a wireless receiver, or an audio signal acquired through a microphone. The input time-series data D1 may be time-series information indicating physical quantities acquired at predetermined time intervals, and is not limited to time-series information obtained by converting a signal output from a wireless receiver or a microphone into time-series information. Note that the predetermined time intervals do not need to be uniform and may be any time interval. For example, the signal acquisition unit 11 acquires the input time-series data D1 by reading the input time-series data D1 from a storage device (not shown). The signal acquisition unit 11 may acquire the input time-series data D1 from any source, as long as it is capable of acquiring the input time-series data D1. Furthermore, the method by which the signal acquiring unit 11 acquires the input time-series data D1 is not limited.

[0019] The signal acquisition unit 11 outputs the acquired input time series data D1 to the signal analysis unit 12 and the multiple filter application unit 13.

[0020] As shown in Fig. 4, the signal analysis unit 12 includes a center frequency estimation unit 121 and a bandwidth estimation unit 122. Fig. 4 is a diagram showing an example of the configuration of the signal analysis unit 12 of the signal parameter identification device 1 according to the first embodiment.

[0021] The center frequency estimation unit 121 of the signal analysis unit 12 estimates the center frequency of a signal present in the input time series data D1 acquired by the signal acquisition unit 11. Possible methods for the center frequency estimation unit 121 to estimate the center frequency include a method in which the input time series data D1 is converted into the frequency domain by Fourier transform and then the center frequency is set to the center of gravity of the amplitude spectrum, a method in which the frequency range in which the amplitude spectrum is equal to or greater than a threshold is regarded as the signal domain and the median value of this range is set to the center frequency, and a method in which the center frequency is directly estimated from the input time series data D1 or the amplitude spectrum using machine learning.

[0022] Furthermore, the bandwidth estimation unit 122 of the signal analysis unit 12 estimates the bandwidth of the signal present in the input time series data D1. Possible methods for the bandwidth estimation unit 122 to estimate the bandwidth include a method in which the input time series data D1 is converted into the frequency domain by a Fourier transform and then the frequency range in which the amplitude spectrum is equal to or greater than a threshold is regarded as the bandwidth, and a method in which the bandwidth is directly estimated from the input time series data D1 or the amplitude spectrum using machine learning.

[0023] The multiple filter application unit 13 has multiple digital filters, applies each digital filter to the input time series data D1 acquired by the signal acquisition unit 11, and outputs the time series data after application (hereinafter referred to as "filter response time series data D5") for each digital filter.

[0024] The filter response time series data D5 output from each of the plurality of digital filters included in the plurality of filter application unit 13 is time series data including time series features or time series data including frequency features.

[0025] Here, the time series data including time series features is time series data obtained by applying a digital filter that recursively uses time series values, such as an IIR (Infinite Impulse Response) filter, to input time series data D1. Also, the time series data including frequency features is time series data obtained by applying a digital filter that acts as a frequency-based filter, such as a low-pass filter, high-pass filter, or band-pass filter, to input time series data D1.

[0026] The multiple filter application unit 13 according to the first embodiment will be described with reference to Fig. 5. Fig. 5 is a diagram showing an example of the configuration and operation of the multiple filter application unit 13 according to the first embodiment. As shown in Fig. 5, the multiple filter application unit 13 includes a filter changing unit 130 and multiple digital filters 131, 132, and 133.

[0027] 5 shows, as an example, multiple filter application unit 13 having three digital filters 131, 132, and 133. The number of digital filters included in multiple filter application unit 13 is not limited to three, and may be two or four or more as long as it is two or more.

[0028] The multiple filter application unit 13 changes the filter characteristics of each of the digital filters 131, 132, and 133 based on the center frequency and bandwidth values ​​output from the signal analysis unit 12, and applies the changes to the input time series data D1 acquired by the signal acquisition unit 11. Each of the multiple digital filters 131, 132, and 133 included in the multiple filter application unit 13 performs a filtering process on the input time series data D1, and the digital filters 131, 132, and 133 output filter response time series data D51, D52, and D53, respectively, which are the time series data after the filtering process.

[0029] The filter characteristics of the digital filters 131, 132, and 133 are changed by the filter change unit 130. In the example shown in FIG.

[0030] As shown in FIG. 5, the filter characteristics may be changed for some or all of the digital filters. In the example shown in FIG. 5, the passband of the digital filter 132 is a frequency domain that does not include the signal components of the input time-series data D1 but includes only noise components. Therefore, the filter change unit 130 changes the filter characteristics of the digital filter 132, and the changed digital filter 132 performs a filtering process on the input time-series data D1 to obtain filter response time-series data D52 in a domain that includes the signal components. Possible methods for changing the filter characteristics include, but are not limited to, a method of aligning the center frequency of the filter passband with the center frequency of the signal to set various passband widths, or a method of aligning the filter passband width within the signal band by dividing the signal bandwidth by an integer multiple. Note that the filter change unit 130 changes the filter characteristics so that each of the multiple digital filters 131, 132, and 133 after the change has a different filter characteristic.

[0031] The filter response time series data D51, D52 and D53 output from each of the plurality of digital filters 131, 132 and 133 included in the plurality of filter application unit 13 is time series data including time series features or time series data including frequency features.

[0032] By configuring as described above, the digital filters 131, 132, and 133 of the multiple filter application unit 13 can be changed to digital filters that pass the signal components included in the input time series data D1, thereby improving the accuracy of signal parameter identification by the signal parameter identification device 1.

[0033] Each of the multiple digital filters 131, 132, and 133 included in the multiple filter application unit 13 is configured by a low-pass filter, a high-pass filter, a band-pass filter, etc. The multiple digital filters 131, 132, and 133 included in the multiple filter application unit 13 are arranged in parallel or in series. Note that Fig. 5 shows, as an example, that the multiple digital filters 131, 132, and 133 included in the multiple filter application unit 13 are arranged in parallel.

[0034] Furthermore, the multiple filter application unit 13 may have, as the multiple digital filters, an RPFB (Random Projection Filter Bank) described in Patent Document 1. The RPFB described in Patent Document 1 is a predetermined number of sets of randomly selected recursive filters and stable filters.

[0035] The feature extraction unit 14 extracts features for each of the plurality of filter response time-series data D5 (D51, D52, D53) output by the plurality of filter application unit 13, and outputs the extracted features as feature data D2. Possible features extracted by the feature extraction unit 14 include, for example, maximum value, minimum value, mean value, median value, variance, standard deviation, quartile, and cyclostationarity. These features may be combined with processing such as moving average or envelope processing, or may be a combination of features.

[0036] As described above, the time-series feature extraction unit 10 performs filtering on the input time-series data D1 by appropriately changing the filter characteristics of multiple digital filters in accordance with the center frequencies and bandwidths of the signal components contained in the input time-series data D1, and extracts features from each of the resulting multiple filter response time-series data D5. The time-series feature extraction unit 10 outputs the extracted features to the inference unit 20 as feature data D2.

[0037] Next, the inference unit 20 will be described. The inference unit 20 uses the feature amount data D2 output from the time-series feature amount extraction unit 10 as input data, performs inference on a predetermined inference target, and outputs inference result information indicating the inference result. Specifically, the inference unit 20 uses the plurality of feature amount data D2 output by the feature amount extraction unit 14 included in the time-series feature amount extraction unit 10 as input data, and performs inference on the predetermined inference target.

[0038] When the input time series data D1 is based on a wireless communication signal or a radar signal, the inference target is identification of the modulation method used for the primary modulation or the secondary modulation, identification of the communication system, identification of the signal strength, identification of the presence or absence of a signal, etc. When the input time series data D1 is based on a voice signal, the inference target is identification of the speaker, identification of the spoken voice, etc. However, the inference target is not limited to these.

[0039] The inference unit 20 inputs the feature data D2 to a trained model corresponding to the learning result obtained by machine learning, acquires inference result information output by the trained model as an inference result, and outputs the acquired inference result information. The inference unit 20 may have the trained model in advance, or may acquire the trained model by reading it from a storage device (not shown) that stores the trained model in advance. A method for generating the trained model will be described later.

[0040] The method by which the inference unit 20 obtains an inference result is not limited to a method of inputting feature data D2 to a trained model and obtaining an inference result from the trained model. For example, the inference unit 20 may obtain an inference result by using a predetermined inference rule to perform inference based on the feature data D2, which is input data. The inference rule is, for example, an if-then rule, an and condition rule, or an or condition rule.

[0041] The overall operation of the signal parameter identification device 1 according to the first embodiment will be described with reference to Fig. 6. Fig. 6 is a flowchart showing an example of the overall operation of the signal parameter identification device 1 according to the first embodiment.

[0042] In the signal parameter identification device 1, first, the signal acquisition unit 11 acquires input time-series data D1 (step S1).

[0043] Next, the signal analysis unit 12 performs signal analysis on the input time-series data D1 and outputs the estimation results of the center frequency and bandwidth (step S2).

[0044] Next, the multiple filter application unit 13 changes the filter characteristics of the digital filters based on the signal analysis result obtained by the signal analysis unit 12 in step S2, specifically based on the estimated center frequency and bandwidth information (step S3), applies multiple digital filters to the input time-series data D1, and outputs filter response time-series data D5 for each digital filter (step S4). Note that depending on the signal analysis result by the signal analysis unit 12, it may be determined that the change in the filter characteristics in step S3 is not necessary, and the change may not be made.

[0045] Next, the feature extracting unit 14 extracts a feature for each piece of filter response time-series data D5 and outputs feature data D2 (step S5).

[0046] Next, the inference unit 20 performs inference on the inference target using the feature amount data D2 output from the time-series feature amount extraction unit 10 as input data, and outputs inference result information indicating the inference result (step S6).

[0047] After step S6, the signal parameter identification device 1 completes the overall operation shown in the flowchart. After completing the overall operation shown in the flowchart, the signal parameter identification device 1 returns to step S1 and repeatedly executes the processes of steps S1 to S6 in the flowchart.

[0048] The inference unit 20 receives as input data the feature data D2 output from the time-series feature extraction unit 10. Therefore, the signal parameter identification device 1 can perform inference with high accuracy because the feature data D2 output from the time-series feature extraction unit 10 contains the desired signal component in the input time-series data D1.

[0049] Next, the signal parameter learning device 3 that generates a learned model that can be used in the inference process by the inference unit 20 will be described.

[0050] 7 is a diagram showing an example of the configuration of the signal parameter learning device 3 according to Embodiment 1. The signal parameter learning device 3 includes a time-series feature quantity extraction unit 10 and a time-series learning unit 30.

[0051] The time-series feature extraction unit 10 of the signal parameter learning device 3 has a configuration similar to that of the time-series feature extraction unit 10 of the signal parameter identification device 1 shown in Fig. 1, and executes similar processing to output feature data D2. Therefore, a detailed description of the time-series feature extraction unit 10 of the signal parameter learning device 3 will be omitted.

[0052] The time series learning unit 30 acquires the feature data D2 output from the time series feature extraction unit 10, generates a trained model using the acquired feature data D2, and outputs the generated trained model to the storage device 40. Note that while FIG. 7 shows an example configuration in which the storage device 40 is provided outside the signal parameter learning device 3, the storage device 40 may be provided inside the signal parameter learning device 3.

[0053] The storage device 40 stores the trained model output from the time series training unit 30.

[0054] The inference unit 20 of the above-described signal parameter identification device 1 acquires the trained model by reading out the trained model stored in the storage device 40, for example.

[0055] The time series learning unit 30 includes a feature acquisition unit 31, a learning unit 32, and a learned model output unit 33.

[0056] The feature acquisition unit 31 acquires feature data D2 output from the time-series feature extraction unit 10. Specifically, the feature acquisition unit 31 acquires a plurality of feature data D2 output from the feature extraction unit 14 included in the time-series feature extraction unit 10. The feature acquisition unit 31 generates and outputs learning data D8 based on the feature data D2.

[0057] The feature acquisition unit 31 may acquire the feature data D2 output from the time-series feature extraction unit 10 via the storage device 40. For example, the time-series feature extraction unit 10 outputs the feature data D2 to the storage device 40 and writes the feature data D2 to the storage device 40, thereby pre-storing the feature data D2 in the storage device 40. The feature acquisition unit 31 acquires the feature data D2 by reading, from the storage device 40, the feature data D2 pre-stored in the storage device 40.

[0058] In addition, the signal parameter learning device 3 may include a time series feature extraction unit 10 within the time series learning unit 30, and the feature acquisition unit 31 may directly acquire the feature data D2 output from the time series feature extraction unit 10 provided within the time series learning unit 30 from the time series feature extraction unit 10.

[0059] The learning unit 32 uses the learning data D8 generated based on the feature amount data D2 acquired by the feature amount acquisition unit 31 to cause the learning model to learn the relationship between the feature amounts of the filter response time-series data D5 and the predetermined inference object. That is, the learning unit 32 causes the learning model to learn, thereby generating a learned model that outputs an inference result inferred about a predetermined inference object as inference result information. The inference object has been described above, so a description thereof will be omitted.

[0060] Specifically, the learning unit 32 generates a trained model by, for example, causing the learning model to learn a predetermined number of times, terminating the learning after the repeated learning has been performed that number of times, and treating the trained learning model as a trained model. The learning unit 32 may generate a trained model by instructing the user to terminate learning by operating an operation input device (not shown), and by having the learning unit 32 terminate learning when it acquires operation information indicating the operation instructing the user to terminate learning.

[0061] The initial learning model is, for example, stored in advance in the storage device 40, and the learning unit 32 acquires the initial learning model stored in advance in the storage device 40 by reading it from the storage device 40. The learning unit 32 repeatedly trains the acquired initial learning model to generate a trained model.

[0062] For example, the learning unit 32 uses the feature amount data D2 acquired by the feature amount acquisition unit 31 and the teacher data D3 corresponding to the feature amount data D2 as learning data D8 to cause the learning model to undergo supervised learning. Fig. 7 shows, as an example, the time-series learning unit 30 in which the learning unit 32 performs supervised learning on the learning model.

[0063] Specifically, the teacher data D3 is stored in advance in, for example, the storage device 40. The feature acquisition unit 31 acquires the teacher data D3 by reading out the teacher data D3 stored in advance in the storage device 40. The learning unit 32 uses the feature data D2 and the teacher data D3 acquired by the feature acquisition unit 31 as learning data D8 to perform supervised learning on the learning model.

[0064] The feature acquisition unit 31 may specify the teacher data D3 by a user operating an operation input device (not shown), and may acquire operation information indicating the operation of specifying the teacher data D3 from the operation input device, thereby reading the teacher data D3 specified by the operation indicated by the operation information from the storage device 40. Alternatively, the feature acquisition unit 31 may input the teacher data D3 by a user operating an operation input device (not shown), and may acquire operation information indicating the input teacher data D3 from the operation input device.

[0065] The learning unit 32 performs supervised learning on the learning model by using a known supervised learning algorithm such as linear regression, logistic regression, support vector machine, decision tree, random forest, gradient boosting tree, neural network, naive Bayes, AR (Auto Regressive) model, MA (Moving Average) model, ARIMA (Auto Regressive Integrated Moving Average) model, state space model, clustering, or ensemble learning.

[0066] The trained model output unit 33 outputs the trained model generated by the learning unit 32. Specifically, the trained model output unit 33 outputs the trained model generated by the learning unit 32 to the storage device 40, and writes the trained model to the storage device 40, thereby causing the storage device 40 to store the trained model.

[0067] With the above configuration, the time series learning unit 30 can generate a learned model that enables the inference unit 20 to perform inference with high accuracy.

[0068] The hardware configuration of the signal parameter learning device 3 according to the first embodiment is the same as that of the signal parameter identification device 1. That is, each unit of the signal parameter learning device 3, like each unit of the signal parameter identification device 1, is realized by the hardware shown in FIG. 2 or FIG. 3.

[0069] The overall operation of the signal parameter learning device 3 according to the embodiment 1 will be described with reference to Fig. 8. Fig. 8 is a flowchart showing an example of the overall operation of the signal parameter learning device 3 according to the embodiment 1. The flowchart in Fig. 8 shows the operation of the signal parameter learning device 3 that performs supervised learning to generate a learned model.

[0070] In the signal parameter learning device 3 that performs supervised learning to generate a learned model, first, the feature acquisition unit 31 acquires the feature data D2 output from the time-series feature extraction unit 10 (step S11).

[0071] Next, the feature amount acquiring unit 31 acquires the training data D3 corresponding to the feature amount data D2 (step S12).

[0072] Next, the learning unit 32 performs supervised learning on the learning model using the feature amount data D2 and the teacher data D3 as learning data D8 (step S13).

[0073] Next, the learning unit 32 determines whether or not the learning has ended (step S14).

[0074] If the learning unit 32 determines that learning has not ended (step S14: No), the signal parameter learning device 3 returns to the processing of step S11 and repeatedly executes the processing from step S11 to step S14 until the learning unit 32 determines that learning has ended.

[0075] If the learning unit 32 determines that the learning has ended (step S14: Yes), the learning unit 32 generates a trained model (step S15).

[0076] Next, the trained model output unit 33 outputs the trained model to the storage device 40 (step S16).

[0077] After step S16, the signal specification learning device 3 ends the entire operation shown in the flowchart.

[0078] The signal parameter learning device 3 may include a learning unit 32 that performs unsupervised learning on a learning model using the feature data D2 acquired by the feature acquisition unit 31 as learning data D8. In this case, the learning unit 32 performs unsupervised learning on the learning model by using a known unsupervised learning algorithm such as clustering, principal component analysis, self-organizing map, vector quantization, or neural network.

[0079] With reference to Fig. 9, the overall operation of the signal parameter learning device 3 when the signal parameter learning device 3 performs unsupervised learning on a learning model will be described. Fig. 9 is a flowchart showing another example of the overall operation of the signal parameter learning device 3 according to embodiment 1. The flowchart in Fig. 9 shows the operation of the signal parameter learning device 3 that performs unsupervised learning to generate a learned model.

[0080] In the signal parameter learning device 3 that performs unsupervised learning to generate a learned model, first, the feature acquisition unit 31 acquires the feature data D2 output from the time-series feature extraction unit 10 (step S21).

[0081] Next, the learning unit 32 performs unsupervised learning on the learning model using the feature amount data D2 as learning data D8 (step S22).

[0082] Next, the learning unit 32 determines whether or not the learning has ended (step S23).

[0083] If the learning unit 32 determines that learning has not ended (step S23: No), the signal parameter learning device 3 returns to the processing of step S21 and repeatedly executes the processing from step S21 to step S23 until the learning unit 32 determines that learning has ended.

[0084] If the learning unit 32 determines that the learning has ended (step S23: Yes), the learning unit 32 generates a learned model (step S24).

[0085] Next, the trained model output unit 33 outputs the trained model to the storage device 40 (step S25).

[0086] After step S25, the signal specification learning device 3 ends the overall operation shown in the flowchart.

[0087] As described above, the signal parameter learning device 3 includes a time-series feature extraction unit 10 that applies multiple digital filters to the time-series data acquired from the signal to be processed, the filter characteristics of which change depending on the location on the frequency axis of signal components included in the time-series data acquired from the signal to be processed, specifically, the center frequency and bandwidth of the signal components, and extracts feature parameters from the multiple filter response time-series data obtained by the filtering process. Because the feature data D2 output from the time-series feature extraction unit 10 includes the desired signal component in the input time-series data D1, the signal parameter learning device 3 can generate a trained model that enables the signal parameter identification device 1 to perform inference with high accuracy. This makes it possible to realize a signal parameter identification device 1 that can prevent a decrease in identification accuracy when identifying the signal parameters of signals existing in a specific frequency range.

[0088] Embodiment 2 In the above-described first embodiment, the signal analysis unit 12 of the time-series feature extraction unit 10 is configured to estimate the center frequency and bandwidth of the signal included in the input time-series data. Next, an embodiment will be described that makes it possible to accurately estimate that a no-signal interval exists when a section containing no signal exists.

[0089] Fig. 10 is a diagram showing an example of the configuration of a signal parameter identification device 1a according to embodiment 2. In Fig. 10, components similar to those of the signal parameter identification device 1 according to embodiment 1 shown in Fig. 1 are assigned the same reference numerals as in Fig. 1. Explanation of components assigned the same reference numerals as in Fig. 1 will be omitted.

[0090] As shown in FIG. 10, a signal parameter identification device 1a according to the second embodiment includes a time-series feature extraction unit 10a and an inference unit 20a.

[0091] The time-series feature quantity extracting unit 10a is configured by replacing the signal analyzing unit 12 of the time-series feature quantity extracting unit 10 constituting the signal characteristic identifying device 1 according to the first embodiment with a signal analyzing unit 12a.

[0092] 11, the signal analysis unit 12a includes a center frequency estimation unit 121, a bandwidth estimation unit 122, and a long-term feature calculation unit 123. Note that Fig. 11 is a diagram illustrating an example of the configuration of the signal analysis unit 12a of the signal parameter identification device 1a according to the second embodiment.

[0093] The center frequency estimation unit 121 and the bandwidth estimation unit 122 of the signal analysis unit 12a are similar to the center frequency estimation unit 121 and the bandwidth estimation unit 122 of the signal analysis unit 12 of the signal parameter identification device 1 according to Embodiment 1. That is, the signal analysis unit 12a according to Embodiment 2 has a configuration in which a long-term feature calculation unit 123 is added to the signal analysis unit 12 according to Embodiment 1. Description of the center frequency estimation unit 121 and the bandwidth estimation unit 122 will be omitted.

[0094] The long-term feature calculation unit 123 calculates long-term features for the input time series data D1. Specifically, the long-term feature calculation unit 123 first calculates features of the input time series data D1. The calculated features are statistical values ​​such as the mean, median, variance, standard deviation, maximum value, and minimum value of the input time series data D1. The long-term feature calculation unit 123 records the calculated features. The long-term feature calculation unit 123 further calculates the change in feature, i.e., the change from the past feature to the current feature, as a long-term feature based on past feature values ​​calculated and recorded in the past and newly calculated current feature values, and outputs the calculated long-term feature data D4. The long-term feature calculated by the long-term feature calculation unit 123, i.e., the change in feature of the input time series data D1, can be a difference obtained by subtracting the value of the past feature from the value of the current feature, a ratio of the current feature to the value of the past feature, a combination of these (a combination of a difference and a ratio), etc. Furthermore, the value of the past feature amount is not limited to a single point, and may be a statistical value such as the average value, median value, variance, standard deviation, maximum value, or minimum value of a plurality of feature amount values.

[0095] The inference unit 20a performs inference on the inference target using the long-term feature data D4 in addition to the feature data D2 output from the feature extraction unit 14, and outputs the inference result. The learning for generating a trained model used for inference by the inference unit 20a is the same as in the first embodiment except that the long-term feature data D4 is added to the training data, and therefore a description thereof will be omitted.

[0096] As described above, the signal parameter identification device 1a according to the second embodiment infers signal parameters using the long-term parameter data D4, which is the amount of change in the parameters of the input time-series data D1, in addition to the parameter data D2 used for inference by the signal parameter identification device 1 according to the first embodiment, and therefore can achieve highly accurate signal parameter identification.

[0097] The configurations shown in the above embodiments are merely examples, and may be combined with other known technologies, or different embodiments may be combined with each other. It is also possible to omit or modify parts of the configurations as long as they do not deviate from the gist of the invention. [Explanation of symbols]

[0098] 1,1a Signal characteristic identification device, 3 Signal characteristic learning device, 10,10a Time series feature extraction unit, 11 Signal acquisition unit, 12,12a Signal analysis unit, 13 Multiple filter application unit, 14 Feature extraction unit, 20,20a Inference unit, 30 Time series learning unit, 31 Feature acquisition unit, 32 Learning unit, 33 Trained model output unit, 40 Storage device, 121 Center frequency estimation unit, 122 Bandwidth estimation unit, 123 Long-term feature calculation unit, 130 Filter change unit, 131,132,133 Digital filter.

Claims

1. a signal acquisition unit that acquires time series data from a signal to be processed and outputs the time series data as input time series data; a signal analysis unit that estimates a center frequency and a bandwidth of a signal component included in the input time series data; a multiple filter application unit having a plurality of digital filters each configured to be able to change filter characteristics based on the estimation result by the signal analysis unit, filtering the input time series data by applying the plurality of digital filters to generate a plurality of filter response time series data; a feature extraction unit that extracts a feature from each of the plurality of filter response time series data; an inference unit that infers signal parameters of the signal based on the feature amount; A signal parameter identification device comprising:

2. the inference unit infers the signal parameters of the signal using a trained model generated by performing training for inferring the signal parameters of the signal from the feature amount; 2. The signal specification identification device according to claim 1.

3. The signal analysis unit further calculates a change amount of a feature amount of the input time series data acquired by the signal acquisition unit as a long-term feature amount; the inference unit infers the signal parameters of the signal using a trained model generated by performing training for inferring the signal parameters of the signal from the feature amount and the long-term feature amount.

2. The signal specification identification device according to claim 1.

4. a signal acquisition unit that acquires time series data from a signal to be processed and outputs the time series data as input time series data; a signal analysis unit that estimates a center frequency and a bandwidth of a signal component included in the input time series data; a multiple filter application unit having a plurality of digital filters each configured to be able to change filter characteristics based on the estimation result by the signal analysis unit, filtering the input time series data by applying the plurality of digital filters to generate a plurality of filter response time series data; a feature extraction unit that extracts a feature from each of the plurality of filter response time series data; a learning unit that causes a learning model to learn the relationship between the feature amount and the signal parameters of the signal, and generates a learned model for inferring the signal parameters of the signal from the feature amount; A signal parameter learning device comprising:

5. a signal acquisition unit that acquires time series data from a signal to be processed and outputs the time series data as input time series data; a signal analysis unit that estimates a center frequency and a bandwidth of a signal component included in the input time series data and calculates a change in a feature of the input time series data acquired by the signal acquisition unit as a long-term feature; a multiple filter application unit that has a plurality of digital filters each configured to be able to change filter characteristics based on the estimation results of the center frequency and the bandwidth by the signal analysis unit, and that applies the plurality of digital filters to filter the input time series data to generate a plurality of filter response time series data; a feature extraction unit that extracts a feature from each of the plurality of filter response time series data; a learning unit that causes a learning model to learn the relationship between the feature amounts and the long-term feature amounts extracted by the feature amount extraction unit and signal specifications of the signal, and generates a trained model for inferring the signal specifications of the signal from the feature amounts and the long-term feature amounts extracted by the feature amount extraction unit; A signal parameter learning device comprising:

6. A signal attribute identification method executed by a signal attribute identification device for identifying a signal attribute, comprising: a signal acquisition step of acquiring time series data from a signal to be processed and outputting the time series data as input time series data; a signal analysis step of estimating a center frequency and a bandwidth of a signal component included in the input time series data; a multiple filter applying step, which has a plurality of digital filters each configured to be able to change filter characteristics based on the estimation result of the signal analyzing step, and filters the input time series data by applying the plurality of digital filters to generate a plurality of filter response time series data; a feature extraction step of extracting a feature from each of the plurality of filter response time series data; an inference step of inferring signal parameters of the signal based on the feature amount; A signal parameter identification method comprising:

7. A control circuit for controlling a signal parameter identification device for identifying signal parameters, a signal acquisition step of acquiring time series data from a signal to be processed and outputting the time series data as input time series data; a signal analysis step of estimating a center frequency and a bandwidth of a signal component included in the input time series data; a multiple filter applying step, which has a plurality of digital filters each configured to be able to change filter characteristics based on the estimation result of the signal analyzing step, and filters the input time series data by applying the plurality of digital filters to generate a plurality of filter response time series data; a feature extraction step of extracting a feature from each of the plurality of filter response time series data; an inference step of inferring signal parameters of the signal based on the feature amount; A control circuit that causes the signal parameter identification device to execute the above.

8. A storage medium for storing a program for controlling a signal parameter identification device for identifying signal parameters, The program a signal acquisition step of acquiring time series data from a signal to be processed and outputting the time series data as input time series data; a signal analysis step of estimating a center frequency and a bandwidth of a signal component included in the input time series data; a multiple filter applying step, which has a plurality of digital filters each configured to be able to change filter characteristics based on the estimation result of the signal analyzing step, and filters the input time series data by applying the plurality of digital filters to generate a plurality of filter response time series data; a feature extraction step of extracting a feature from each of the plurality of filter response time series data; an inference step of inferring signal parameters of the signal based on the feature amount; A storage medium that causes the signal parameter identification device to execute the above.

Images