JP7557861B2 - Learning data generation method, signal type classification system, data collection system, and program - Google Patents

Description

The present invention relates to a signal type classification technology that enables understanding of the radio wave environment.

IoT (Internet of Things) devices are increasingly being introduced into indoor environments such as factories, hospitals, and commercial facilities for the purpose of grasping and controlling the operating status of various equipment. In such environments, wireless communication traffic is likely to be concentrated in terms of location and time, and collisions and delays of wireless communication packets are likely to occur. Even under such conditions, stable communication is desired. In addition, in order to grasp the operating status of the entire system, a system is desired that can grasp information such as the congestion level and radio wave strength in the monitored frequency band, monitor whether there is a problem with the wireless quality, and, if a problem occurs, provide information that leads to identifying the cause. In order to realize such a system, it is important to establish a signal type classification technology that enables the status of the radio wave environment to be grasped.

Technologies have been developed to identify noise sources. For example, Patent Document 1 discloses technology for detecting noise (single pulses and burst-like pulses) generated by lightning discharges.

JP 2001-4731 A

Conventional technologies such as those described in Patent Document 1 are based on the premise that the nature of the source of the wireless signal or electromagnetic noise is known to some extent, and the frequency spectrum of the wireless signal or electromagnetic noise generated from the source and the time waveform pattern of the signal (or electromagnetic noise) (for example, a single pulse or a burst-like pulse) can be predicted to some extent, making it possible to identify the source.

However, in a wireless environment where there are many sources of wireless signals and electromagnetic noise and the frequency spectrum and waveform patterns of the wireless signals and electromagnetic noise are unpredictable, it is extremely difficult to grasp the state of the wireless environment using conventional technology.

Therefore, there is a demand for technology that can properly grasp the state of the wireless environment, even in an environment where there are many wireless signals and electromagnetic noise whose frequency spectra and waveform patterns are difficult to predict. In order to realize such technology, it is effective to establish technology (signal type classification technology) that properly classifies the types of wireless signals and electromagnetic noise whose frequency spectra and waveform patterns are difficult to predict. In other words, by using this technology, it is possible to grasp what types of wireless signals and electromagnetic noise are present in the wireless environment, and as a result, the state of the wireless environment can be properly grasped.

In view of the above problems, the present invention aims to provide a signal type classification system that appropriately classifies the types of wireless signals and electromagnetic noise whose frequency spectra and waveform patterns are difficult to predict, as well as a learning data generation method, data collection system, and program for constructing a classifier to be installed in the signal type classification system through machine learning.

To solve the above problem, the first invention is a method for generating learning data for signal type classification processing, which includes a radio wave waveform acquisition processing step, a feature amount acquisition processing step, a clustering processing step, and a learning data acquisition step.

The radio wave waveform acquisition processing step acquires radio wave waveform data of the radio waves received by the antenna.

The feature acquisition process step acquires feature data, which is data that indicates the characteristics of the time domain or frequency domain of the radio wave waveform data and has a smaller amount of data than the radio wave waveform data.

The clustering process classifies the feature data into clusters.

The learning data acquisition step assigns labels to the feature data classified by the clustering process, thereby acquiring data in which the feature data and the labels are associated as learning data.

This learning data generation method acquires feature data with a reduced amount of data while maintaining the characteristics of the radio wave waveform, so that even if a large amount of feature data is acquired and transmitted, the load on the communication path is small. As a result, this learning data generation method can collect a large amount of data and generate a large amount of learning data using the collected data. Furthermore, this learning data generation method classifies the collected feature data by clustering processing and assigns labels to the classified clusters, so that appropriate type classification can be performed regardless of the type of radio wave waveform data. As a result, this learning data generation method can generate learning data that can perform learning processing of a signal type classifier (e.g., a neural network model) that corresponds to a variety of radio wave waveform data with high accuracy and effectiveness.

The second invention is the first invention, in which the feature data is data extracted from the cepstrum.

As a result, this learning data generation method can generate learning data using data extracted from the cepstrum as feature data.

The third invention is the first invention, in which the feature data is data obtained by extracting or processing low-order components of the cepstrum.

As a result, in this training data generation method, it is possible to perform processing to generate training data using data obtained by extracting or processing low-order components of the cepstrum as feature data. Data obtained by extracting low-order components of the cepstrum is data in which white noise components in the frequency domain are effectively suppressed while maintaining the characteristics of the radio wave waveform. Therefore, in this training data generation method, by performing processing to generate training data using data obtained by extracting low-order components of the cepstrum as feature data, it is possible to generate training data with extremely high accuracy.
Note that "processing" the low-order components of the cepstrum is a concept that includes, for example, performing a predetermined process (for example, filter process) on the low-order components of the cepstrum. Note that the "processing" may be any process that can obtain data in which white noise components in the frequency domain are effectively suppressed while maintaining the characteristics of the radio wave waveform.

For example, when all cepstrum data is expressed as an N1-dimensional vector, it is preferable to set the low-dimensional array before convergence, such as data of N1 x 0.1 or less (low-dimensional data), because the cepstrum array converges to 0 toward higher dimensions. For example, when all cepstrum data is expressed as a 1000-dimensional (N1 = 1000) vector, some or all of the 1- to 100-dimensional data (100 = N1 x 0.1) may be extracted as low-dimensional components from the cepstrum. Also, when all cepstrum data is expressed as a 1500-dimensional (N1 = 1500) vector, 1- to 16-dimensional data (16 dimensions' worth of data) may be extracted as low-dimensional components from the cepstrum.

The fourth invention is any one of the first to third inventions, further comprising an optimal cluster number acquisition step for acquiring the optimal number of classifications in the clustering process.

As a result, this learning data generation method makes it possible to perform clustering processing using the optimal number of clusters.

The fifth invention is the fourth invention, and the optimal cluster number acquisition step detects the number of clusters for which the classification status of the clusters does not change when the number of clusters is changed while changing the initial settings (e.g., random number seeds) of the clustering process, and acquires the optimal number of classifications based on the detected number of clusters.

As a result, in this learning data generation method, a number that does not change the number of clusters even if the initial setting (e.g., random number seed) is changed (e.g., the maximum number at which the number of clusters does not change even if the initial setting (e.g., random number seed) is changed) can be set as the optimal number of classifications (optimal number of clusters).

The sixth invention is a signal type classification system that performs a signal type classification process using a trained model of a signal type classifier obtained by performing a learning process using training data generated by the training data generation method that is any one of the first to fifth inventions, and includes a wireless communication device and a host device.

The wireless communication device is
a radio wave waveform acquisition processing step for acquiring radio wave waveform data of radio waves received by an antenna;
a feature acquisition processing step of acquiring feature data which is data indicating features in a time domain or a frequency domain of the radio wave waveform data and has a smaller amount of data than the radio wave waveform data;
a transmitting step of transmitting the feature data to a host device;
Execute.

The host device
a receiving step of receiving feature data transmitted from a wireless communication device;
a label acquisition step of acquiring a label of the feature data by inputting the feature data acquired by the wireless communication device into a trained model of the signal type classifier;
Execute.

In this signal type classification system, the amount of data can be significantly reduced by performing feature data acquisition processing in the wireless communication device, so that a large amount of feature data with small individual data amounts can be acquired while maintaining the characteristics of wireless signals and electromagnetic noise, and the acquired large amount of feature data can be used to perform signal type classification processing with the trained model of the signal type classifier. Therefore, with this signal type classification system, even wireless signals and electromagnetic noise whose frequency spectra and waveform patterns are difficult to predict can be appropriately classified.

The seventh invention is the sixth invention, in which the feature data is data extracted from the cepstrum.

As a result, this signal type classification system can perform processing using the data extracted from the cepstrum as feature data.

The eighth invention is the sixth invention, in which the feature data is data obtained by extracting or processing low-order components of the cepstrum.

As a result, in this signal type classification system, processing can be performed using data obtained by extracting or processing low-order components of the cepstrum as feature data.
Note that "processing" the low-order components of the cepstrum is a concept that includes, for example, performing a predetermined process (for example, filter process) on the low-order components of the cepstrum. Note that the "processing" may be any process that can obtain data in which white noise components in the frequency domain are effectively suppressed while maintaining the characteristics of the radio wave waveform.

The ninth invention is a data collection system including N (N: natural number) signal type classification systems according to any one of the sixth to eighth inventions, and a server that collects signal type data from the N signal type classification systems, the signal type data being data obtained by the N signal type classification systems and associating feature data with labels.

As a result, this data collection system can aggregate data acquired by the signal type classification system on a server, where it can perform a variety of processes (data analysis, data visualization, etc.).

The tenth invention is a program for causing a computer to execute the learning data generation method according to any one of the first to fifth inventions.

This makes it possible to realize a program for causing a computer to execute a learning data generation method that has the same effect as any one of the first to fifth inventions.

The present invention provides a signal type classification system that appropriately classifies the types of wireless signals and electromagnetic noise whose frequency spectra and waveform patterns are difficult to predict, as well as a learning data generation method, data collection system, and program for constructing a classifier to be installed in the signal type classification system through machine learning.

1 is a schematic configuration diagram of a signal type classification system 100 according to a first embodiment. FIG. 4 is a sequence diagram for explaining a process (learning data generation process) during learning of the signal type classification system 100. 3 is a diagram for explaining a feature amount acquisition process executed in the signal type classification system 100; 3 is a diagram for explaining a feature amount acquisition process executed in the signal type classification system 100; 3 is a diagram for explaining a feature amount acquisition process executed in the signal type classification system 100; 13 is a flowchart of a process for obtaining an optimal number of clusters (number of classifications) in a clustering process. 11 is a diagram for explaining a process of obtaining an optimal number of clusters (number of classifications) in a clustering process. FIG. 2 is a schematic diagram of the signal type classification system 100 at the time of prediction. FIG. 4 is a sequence diagram for explaining a process (signal type classification process) during prediction by the signal type classification system 100. 1 is a schematic diagram of a data collection system 1000 including n signal type classification systems 100_1 to 100_n. FIG. 1 is a schematic configuration diagram of a signal type classification system 100A according to a first modified example of the first embodiment. FIG. 11 is a schematic configuration diagram of a signal type classification system 100B according to a second modified example of the first embodiment. FIG. 11 is a schematic configuration diagram of a data collection system 1000A according to a third modified example of the first embodiment. FIG. 2 is a diagram showing a CPU bus configuration.

[First embodiment]
The first embodiment will be described below with reference to the drawings.

<1.1: Configuration of signal type classification system>
FIG. 1 is a schematic configuration diagram of a signal type classification system 100 according to the first embodiment.

As shown in FIG. 1, the signal type classification system 100 includes N (N: natural number) wireless communication devices S1_1 to S1_N, a host device H1, and N wireless communication devices S1_1 to S1_N (N: natural number).

(1.1.1: Wireless Communication Device)
As shown in FIG. 1, the wireless communication device S1_1 includes an antenna Ant_S1, an RF processing unit 11, an IQ data acquisition unit 12, a feature acquisition unit 13, a first communication interface 14, and a wireless communication device control unit 15.

The antenna Ant_S1 is an antenna for receiving radio waves (RF signals) radiated (transmitted) from outside.

The RF processing unit 11 receives an RF signal from the outside via the antenna Ant_S1, performs RF processing for reception (RF demodulation processing, AD conversion, etc.) on the received RF signal, and acquires a post-RF processing signal Sig0 (RF demodulated signal Sig0 (e.g., baseband OFDM signal)). The RF processing unit 11 then outputs the post-RF processing signal Sig0 to the IQ data acquisition unit 12. The RF processing unit 11 inputs a control signal CTL1 from the wireless communication device control unit 15 and performs processing according to the control signal CTL1. The RF processing unit 11 also sets the frequency band (e.g., center frequency fo, frequency band Δf) to be subjected to RF processing according to the control signal CTL1, and acquires an RF demodulated signal Sig0 of the RF signal in the frequency domain set above by performing RF processing for reception.

The IQ data acquisition unit 12 inputs the control signal CTL2 output from the wireless communication device control unit 15 and the signal Sig0 output from the RF processing unit 11. In accordance with the control signal CTL2, the IQ data acquisition unit 12 acquires data (I component data) of the I component signal (in-phase component signal) and data (Q component data) of the Q component signal (quadrature component signal) from the signal Sig0 output from the RF processing unit 11, and outputs the acquired data as data D1 to the feature acquisition unit 13.

The feature acquisition unit 13 inputs a control signal CTL3 output from the wireless communication device control unit 15 and data D1 output from the IQ data acquisition unit 12. In accordance with the control signal CTL3, the feature acquisition unit 13 executes feature acquisition processing on the data D1 output from the IQ data acquisition unit 12 to acquire predetermined features (details will be described later). The feature acquisition unit 13 then outputs data including the acquired features as data D2 to the first communication interface 14.

The first communication interface 14 is a communication interface for transmitting and receiving data to and from an external device, for example, via a wired network (including a communication path that complies with a specific serial bus standard (e.g., USB or PCI Express)) or a wireless network. The first communication interface 14 is a communication interface for communicating commands, data, etc. between the wireless communication device S1_1 and the host device H1 (for example, a communication interface for communication via USB).

The first communication interface 14 inputs data D2 output from the feature acquisition unit 13, converts the input data into data in a format that can be communicated via a wired or wireless network, and transmits the data to the host device H1.

The first communication interface 14 also outputs the command Cmd1 received from the host device H1 to the wireless communication device control unit 15.

The wireless communication device control unit 15 is a functional unit for controlling each functional unit of the wireless communication device S1_1, and generates a predetermined control signal according to the command Cmd1 received from the host device H1 via the first communication interface 14, and controls each functional unit by outputting the generated control signal to each functional unit. For example, the wireless communication device control unit 15 generates a control signal CTL1 for controlling the RF processing unit 11, and outputs the control signal CTL1 to the RF processing unit 11. The wireless communication device control unit 15 also generates a control signal CTL2 for controlling the IQ data acquisition unit 12, and outputs the control signal CTL2 to the IQ data acquisition unit 12. The wireless communication device control unit 15 also generates a control signal CTL3 for controlling the feature acquisition unit 13, and outputs the control signal CTL3 to the feature acquisition unit 13.

The wireless communication devices S1_2 to S1_N each have the same configuration as described above (the same configuration as the wireless communication device S1_1). Note that the signal type classification system 100 may include one or more wireless communication devices.

(1.1.2: Host Device)
1, the host device H1 includes a second communication interface 21, a host device control unit 22, a ROM 23, a RAM 24, an acquired data buffer 25, a storage unit 26, a third communication interface 27, and a bus Bus_H1. The second communication interface 21, the host device control unit 22, the ROM 23, the RAM 24, the acquired data buffer 25, the storage unit 26, and the third communication interface 27 are connected to the bus Bus_H1, and can transmit and receive data, commands, and the like to each other via the bus Bus_H1. Note that some or all of the above-mentioned functional units of the host device H1 do not necessarily need to be bus-connected, and may be directly connected.

The second communication interface 21 is, for example, a communication interface for transmitting and receiving data with an external device via a wired or wireless network (including a communication path that complies with a specific serial bus standard (e.g., USB or PCI Express)). The second communication interface 21 is also a communication interface for communicating commands, data, etc. between the wireless communication device S1_1 and the host device H1 (for example, a communication interface for communication via USB).

The second communication interface 21 receives data transmitted from the wireless communication devices S1_1 to S1_N. Then, the second communication interface 21 outputs the received data to the acquired data buffer 25. The second communication interface 21 also generates transmission data for transmitting the command Cmd1 generated by the host device control unit 22 to a destination specified by the host device control unit 22 (one or more of the wireless communication devices S1_1 to S1_N), and transmits the transmission data to the destination.

The host device control unit 22 is a functional unit for controlling each functional unit of the host device H1, and controls each functional unit by generating a control signal for controlling each functional unit and outputting the generated control signal to the corresponding functional unit. The host device control unit 22 also generates a command Cmd1 for controlling one or more of the wireless communication devices S1_1 to S1_N, and transmits the generated command Cmd1 to the corresponding (specified destination) wireless communication device via the second communication interface 21.

The host device control unit 22 also has a mode switching function that switches between learning mode and prediction mode, and the host device control unit 22 sets the operating mode of the host device H1 to the learning mode or prediction mode.

In the learning mode, the host device control unit 22 controls each functional unit so that the host device H1 executes a learning process (or a process for generating learning data). In the prediction mode, the host device control unit 22 controls each functional unit so that the host device H1 executes a prediction process (e.g., a signal type classification process).

The host device control unit 22 also reads out data (instructions, codes, etc.) stored in ROM 23 and executes programs, etc. stored in ROM. The host device control unit 22 also stores predetermined data, codes, etc. in RAM 24 and executes predetermined processing (programs, etc.) by reading out the data, codes, etc. stored in RAM 24 at a predetermined timing.

The ROM 23 is a read-only memory that stores programs, libraries, instructions, codes, etc. executed by the host device H1. The data stored in the ROM 23 is read by the host device control unit 22.

RAM 24 is a random access memory that stores data, commands, codes, etc. for executing a specific process in the host device H1. The data, commands, codes, etc. stored in RAM 24 are read by the host device control unit 22.

The acquired data buffer 25 is a buffer for storing transmission data from the wireless communication devices S1_1 to S1_N received by the second communication interface 21. The acquired data buffer 25 may include the same number of buffers as the number of wireless communication devices (e.g., buffers Buf_1 to Buf_N in FIG. 1) in order to separately store and manage data for each of the wireless communication devices S1_1 to S1_N. The acquired data buffer 25 may also include a single buffer, and the memory area in this single buffer may be divided into memory areas in the same number as the number of wireless communication devices for memory management.

The memory unit 26 is a device for storing data. For example, the memory unit 26 stores learning data acquired by the host device H1.

The third communication interface 27 is a communication interface for transmitting and receiving data with an external device, for example, via a wired or wireless network (e.g., a LAN). The host device H1 performs data communication with, for example, an externally installed server, via the third communication interface 27.

<1.2: Operation of the signal type classification system>
The operation of the signal type classification system 100 configured as above will be described below with reference to the drawings.

Figure 2 is a sequence diagram for explaining the process during learning (learning data generation process) of the signal type classification system 100.

Figure 3 is a diagram for explaining the feature acquisition process performed by the signal type classification system 100.

Figure 4 is a diagram for explaining the feature acquisition process executed by the signal type classification system 100, and shows the spectrum and cepstrum of a wireless signal (narrowband communication signal), electromagnetic noise (type 1), and electromagnetic noise (type 2) (processing targeting the 920 MHz band).

Figure 5 is a diagram for explaining the feature acquisition process executed by the signal type classification system 100, showing the spectrum and cepstrum (cepstrum expanded from the low-dimensional region (region equivalent to 16 dimensions)) of a wireless signal (narrowband communication signal), electromagnetic noise (type 1), and electromagnetic noise (type 2).

Figure 6 is a flowchart showing the process of obtaining the optimal number of clusters (number of classifications) in the clustering process.

Figure 7 is a diagram explaining the process of obtaining the optimal number of clusters (number of classifications) in the clustering process.

For ease of explanation, the following describes a case (one example) where N=3, three wireless communication devices S1_1, S1_2, S1_3, and a host device H1 are installed in a small space (e.g., inside a factory), and (1) wireless communication device S1_1 acquires wireless signal and electromagnetic noise data in the 920 MHz band, (2) wireless communication device S1_2 acquires signal and electromagnetic noise data in the 2.4 GHz band, and (3) wireless communication device S1_3 acquires signal and electromagnetic noise data in the 5.2 GHz band.

The operation of the signal type classification system 100 will be explained below, divided into (1) processing during learning and (2) processing during prediction.

(1.2.1: Processing during learning)
First, the learning process of the signal type classification system 100 will be described.

(Step S1):
In step S1, the host device H1 executes an initialization process. Specifically, the process is executed as follows.

The host device control unit 22 of the host device H1 generates a command Cmd1 (fo=920MHz, Δf=15MHz) for the wireless communication device S1_1 to acquire data on wireless signals and electromagnetic noise in a frequency band (920MHz band) with a center frequency fo=920MHz and a frequency band Δf=15MHz, and transmits data including the command Cmd1 (fo=920MHz, Δf=15MHz) via the second communication interface 21 to the wireless communication device S1_1, which is the destination of the command.

The host device control unit 22 of the host device H1 also generates a command Cmd1 (fo=2.4 GHz, Δf=15 MHz) for the wireless communication device S1_2 to acquire data on wireless signals and electromagnetic noise in a frequency band (2.4 GHz band) with a center frequency fo=2.4 GHz and a frequency band Δf=15 MHz, and transmits data including the command Cmd1 (fo=2.4 GHz, Δf=15 MHz) to the wireless communication device S1_2, which is the destination of the command, via the second communication interface 21.

The host device control unit 22 of the host device H1 also generates a command Cmd1 (fo=5.2 GHz, Δf=15 MHz) for the wireless communication device S1_3 to acquire data on wireless signals and electromagnetic noise in a frequency band (5.2 GHz band) with a center frequency fo=5.2 GHz and a frequency band Δf=15 MHz, and transmits data including the command Cmd1 (fo=5.2 GHz, Δf=15 MHz) to the wireless communication device S1_3, which is the destination of the command, via the second communication interface 21.

(Step S1.1):
The wireless communication device S1_1 receives communication data including the command Cmd1 (fo=920 MHz, Δf=15 MHz) transmitted from the host device H1. The first communication interface 14 of the wireless communication device S1_1 extracts the command Cmd1 (fo=920 MHz, Δf=15 MHz) transmitted from the host device H1 from the received communication data, and outputs the extracted command Cmd1 (fo=920 MHz, Δf=15 MHz) to the wireless communication device control unit 15.

In accordance with the command Cmd1 (fo=920 MHz, Δf=15 MHz), the wireless communication device control unit 15 of the wireless communication device S1_1 generates a control signal CTL1 that instructs the RF processing unit 11 to receive a wireless signal (and electromagnetic noise) in the frequency band of fo=920 MHz, Δf=15 MHz, and outputs the control signal CTL1 to the RF processing unit 11.

The RF processing unit 11 of the wireless communication device S1_1 sets the receivable frequency band to a frequency band of fo = 920 MHz, Δf = 15 MHz in accordance with the control signal CTL1.

After performing the above-mentioned frequency band setting process, the wireless communication device control unit 15 of the wireless communication device S1_1 transmits an Ack signal indicating that the process corresponding to the command Cmd1 received from the host device H1 has been completed to the host device H1 via the first communication interface 14.

(Step S1.2):
The wireless communication device S1_2 receives communication data including the command Cmd1 (fo=2.4 GHz, Δf=15 MHz) transmitted from the host device H1. The first communication interface 14 of the wireless communication device S1_2 extracts the command Cmd1 (fo=2.4 GHz, Δf=15 MHz) transmitted from the host device H1 from the received communication data, and outputs the extracted command Cmd1 (fo=2.4 GHz, Δf=15 MHz) to the wireless communication device control unit 15.

In accordance with the command Cmd1 (fo=2.4 GHz, Δf=15 MHz), the wireless communication device control unit 15 of the wireless communication device S1_2 generates a control signal CTL1 that instructs the RF processing unit 11 to receive a wireless signal (and electromagnetic noise) in the frequency band of fo=2.4 GHz, Δf=15 MHz, and outputs the control signal CTL1 to the RF processing unit 11.

The RF processing unit 11 of the wireless communication device S1_2 sets the receivable frequency band to a frequency band of fo = 2.4 GHz, Δf = 15 MHz in accordance with the control signal CTL1.

After performing the above-mentioned frequency band setting process, the wireless communication device control unit 15 of the wireless communication device S1_2 transmits an Ack signal indicating that the process corresponding to the command Cmd1 received from the host device H1 has been completed to the host device H1 via the first communication interface 14.

(Step S1.3):
The wireless communication device S1_3 receives communication data including the command Cmd1 (fo=5.2 GHz, Δf=15 MHz) transmitted from the host device H1. The first communication interface 14 of the wireless communication device S1_3 extracts the command Cmd1 (fo=5.2 GHz, Δf=15 MHz) transmitted from the host device H1 from the received communication data, and outputs the extracted command Cmd1 (fo=5.2 GHz, Δf=15 MHz) to the wireless communication device control unit 15.

In accordance with the command Cmd1 (fo=5.2 GHz, Δf=15 MHz), the wireless communication device control unit 15 of the wireless communication device S1_3 generates a control signal CTL1 that instructs the RF processing unit 11 to receive a wireless signal (and electromagnetic noise) in the frequency band of fo=5.2 GHz, Δf=15 MHz, and outputs the control signal CTL1 to the RF processing unit 11.

The RF processing unit 11 of the wireless communication device S1_3 sets the receivable frequency band to a frequency band of fo = 5.2 GHz, Δf = 15 MHz in accordance with the control signal CTL1.

After performing the above-mentioned frequency band setting process, the wireless communication device control unit 15 of the wireless communication device S1_3 transmits an Ack signal indicating that the process corresponding to the command Cmd1 received from the host device H1 has been completed to the host device H1 via the first communication interface 14.

When the host device H1 receives an Ack signal in response to the transmitted command Cmd1 from the wireless communication devices S1_1 to S1_3, it determines that the settings according to the transmitted command Cmd1 have been completed in the wireless communication devices S1_1 to S1_3, and proceeds to step S2.

(Step S2):
In step S2, the host device H1 executes a data acquisition instruction process. Specifically, the process is executed as follows.

The host device control unit 22 of the host device H1 generates a command Cmd1 (Data_get_start) that instructs each of the wireless communication devices S1_1 to S1_3 to start acquiring data. The host device control unit 22 then transmits communication data including the generated command Cmd1 (Data_get_start) to each of the wireless communication devices S1_1 to S1_3 via the second communication interface 21.

The wireless communication device S1_1 receives communication data including the command Cmd1 (Data_get_start) sent from the host device H1. The first communication interface 14 of the wireless communication device S1_1 extracts the command Cmd1 (Data_get_start) sent from the host device H1 from the received communication data, and outputs the extracted command Cmd1 (Data_get_start) to the wireless communication device control unit 15.

The wireless communication device control unit 15 of the wireless communication device S1_1 generates control signals CTL1 to CTL3 for executing (starting) the data acquisition process in accordance with the command Cmd1 (Data_get_start), and outputs the generated control signal CTL1 to the RF processing unit 11, the generated control signal CTL2 to the IQ data acquisition unit 12, and the generated control signal CTL3 to the feature acquisition unit 13.

The RF processing unit 11, IQ data acquisition unit 12, and feature acquisition unit 13 of the wireless communication device S1_1 execute (start) data acquisition processing according to the control signals CTL1 to CTL3 input from the wireless communication device control unit 15.

The same process as described above is also executed in wireless communication device S1_2 and wireless communication device S1_3, and data acquisition processing is executed (started).

(Step S3):
In step S3, a feature amount data acquisition process is executed in each of the wireless communication devices S1_1 to S1_3. Specifically, the process is executed as follows.

The RF processing unit 11 of the wireless communication device S1_1 receives an RF signal from the outside via the antenna Ant_S1, performs RF processing for reception (920 MHz band RF demodulation processing, AD conversion, etc.) on the received RF signal, and acquires the RF processed signal Sig0 (RF demodulated signal Sig0 (e.g., baseband OFDM signal)). The RF processing unit 11 then outputs the RF demodulated signal Sig0 to the IQ data acquisition unit 12.

The IQ data acquisition unit 12 of the wireless communication device S1_1 acquires data (I component data) of the I component signal (in-phase component signal) and data (Q component data) of the Q component signal (quadrature component signal) from the signal Sig0 output from the RF processing unit 11, and outputs the acquired data as data D1 to the feature acquisition unit 13.

The feature acquisition unit 13 of the wireless communication device S1_1 executes feature acquisition processing on the data D1 output from the IQ data acquisition unit 12 to acquire a predetermined feature. Specifically, the feature acquisition unit 13 calculates a cepstrum from samples (a predetermined number of samples) of the data D1 for a predetermined period, and acquires data of only the low-order components of the calculated cepstrum as a feature.

In other words, the feature acquisition unit 13 acquires k samples (k samples of I component data and/or Q component data) included in a predetermined period from the data D1 as an Rx signal sample set x _n (1≦n≦Nmax), and acquires cepstrum data (set) c _n by performing processing equivalent to the following formula on the acquired Rx signal sample set x _n (see Figure 3).
s _n =log(|DFT(x _n )|)
c _n =DFT ^-1 (s _n )
DFT(): Discrete Fourier Transform DFT ⁻¹ (): Inverse Discrete Fourier Transform |x|: Absolute value of x log(): Function to take logarithm Furthermore, the feature acquisition unit 13 executes a process of extracting only low-dimensional elements from the elements of the cepstrum data (set) c _n acquired as described above, that is, a process equivalent to the following formula, to acquire feature data c _fn (see FIG. 3 ).
c _fn =LPF_cep(c _n )
LPF_cep(): function for extracting low-order component data (low-dimensional elements) of the cepstrum For example, when the cepstrum data (set) _cn is a 15000-dimensional vector, the feature acquisition unit 13 acquires 16-dimensional data of the cepstrum data (set) _cn (data obtained by extracting 1st to 16th dimensional data of the cepstrum data (vector) (data obtained by extracting the 1st element to the 16th element of the 15000-dimensional vector)) as feature data. Note that the feature acquisition unit 13 may acquire data processed on the low-order component data of the cepstrum (for example, data subjected to a predetermined filter) as feature data.

By performing the feature data acquisition process in this manner, the amount of data can be significantly reduced, and therefore the amount of data transmitted from the wireless communication devices S1_1 to S1_3 to the host device H1 can be significantly reduced. Furthermore, the amount of data used in the clustering process in the host device H1 is also significantly reduced. This makes it possible to speed up the clustering process in the host device H1.

Furthermore, by performing the feature data acquisition process as described above, the accuracy of the clustering process in the host device H1 can be improved. This will be explained using Figures 4 and 5.

FIG. 4 shows the spectrum s n (=log(|DFT(x n )|)) of (A) Rx signal sample set x _n (corresponding to a set of IQ samples) of (1) wireless signal (narrowband communication signal), (2) electromagnetic noise (type 1), and (3) electromagnetic _noise (type 2) when the ₉₂₀ MHz band is processed (data shown as Sptr() in the upper part of FIG. 4 ),
(B) Cepstrum c _n (=DFT ⁻¹ (s _n )) (data shown as Cepstrum() in the lower part of FIG. 4), and
(C) Lifted spectrum (data shown as Lifted_sptr() in the upper part of FIG. 4)
FIG.

The lifted spectrum is data (spectrum) obtained by extracting only the lower 16 dimensions of the cepstrum and setting the data from the 17th dimension onwards to "0" (low-pass cepstrum) and applying a discrete Fourier transform to the data.

As can be seen from FIG. 4, the lifted spectra of (1) wireless signal (narrowband communication signal), (2) electromagnetic noise (type 1), and (3) electromagnetic noise (type 2) are clearly distinguishable from each other (graphs). In other words, the feature data obtained by extracting only the 16 dimensions on the lower dimensional side of the cepstrum distinguishes between (1) wireless signal (narrowband communication signal), (2) electromagnetic noise (type 1), and (3) electromagnetic noise (type 2), and can be clearly classified as different types of data. As can be seen from the data converted to a lifted spectrum (see the upper graph of FIG. 4), the data on the lower dimensional side of the cepstrum (for example, the 16-dimensional data on the lower dimensional side) is data that effectively extracts (reflects) the characteristic parts of the wireless signal or electromagnetic noise while suppressing white noise components (noise components that occur approximately uniformly in the frequency domain). Even in the 16-dimensional data on the lower dimensional side of the cepstrum in the lower part of Figure 5, characteristic parts of the wireless signal and electromagnetic noise (e.g., the wireless signal, electromagnetic noise (type 1), and electromagnetic noise (type 2) in Figure 5) are reflected, and it can be seen that it is quite possible to classify (distinguish) the wireless signal and electromagnetic noise (e.g., the wireless signal, electromagnetic noise (type 1), and electromagnetic noise (type 2) in Figure 5).

In other words, the feature data obtained by extracting only the lower 16 dimensions of the cepstrum can suppress the influence of white noise components that make it difficult to distinguish between wireless signals and electromagnetic noise, and is data that effectively extracts (reflects) the characteristic parts of wireless signals or electromagnetic noise.

Therefore, as described above, the process of obtaining feature data is performed by extracting only the low-dimensional data of the cepstrum (the 16-dimensional data on the low-dimensional side), and the feature data obtained by this process is used to perform clustering processing, thereby improving the accuracy of the clustering process.

The feature acquisition unit 13 of the wireless communication device S1_1 outputs the feature data acquired as described above to the first communication interface 14 as data D2. The first communication interface 14 transmits communication data including the data D2 acquired by the feature acquisition unit 13 (referred to as transmission data Dtx(D2, S1_1)) to the host device H1.

The same process as above is performed in the wireless communication devices S1_2 and S1_3, and communication data (transmission data Dtx(D2, S1_2), Dtx(D2, S1_3)) including the feature data D2 acquired by the wireless communication devices S1_2 and S1_3 are transmitted to the host device H1.

(Step S4):
In step S4, the host device H1 receives the communication data Dtx (D2, S1_1), Dtx (D2, S1_2), and Dtx (D2, S1_3) transmitted from each of the wireless communication devices S1_1 to S1_3 through the second communication interface 21. Then, the second communication interface 21 acquires the feature amount data D2 acquired by each wireless communication device from the received communication data, and outputs the acquired feature amount data D2 to the acquired data buffer 25. At this time, the host device control unit 22 of the host device H1 stores and manages the acquired feature amount data D2 in the acquired data buffer 25 so that it is possible to know which wireless communication device acquired the data. For example, the host device control unit 22 of the host device H1 may store feature data D2 acquired from wireless communication device S1_1 in buffer Buf_1 of the acquired data buffer 25, store feature data D2 acquired from wireless communication device S1_2 in buffer Buf_2 of the acquired data buffer 25, and store feature data D2 acquired from wireless communication device S1_3 in buffer Buf_3 of the acquired data buffer 25.

In addition, in order to acquire a large amount of feature data in the wireless communication devices S1_1 to S1_3, the processes of steps S3 and S4 may be executed repeatedly.

(Step S5):
In step S5, the host device H1 performs clustering processing on the feature data acquired by the wireless communication devices S1_1 to S1_3 and stored in the acquired data buffer 25. Specifically, the host device control unit 22 of the host device H1 executes processing to classify the feature data stored in the acquired data buffer 25 into clusters by, for example, the k-means++ method.

Here, the process (method) of obtaining the optimal number of clusters (number of classifications) in the clustering process will be explained using the flowchart in FIG. 6 and the explanatory diagram in FIG. 7.

(Steps SA101 to SA104):
The host device H1 sets a random number seed to seed1 (step SA101), and sets the initial value of the number of clusters k to "2" (step SA102).

In step SA103, the host device H1 performs a process (clustering process) to classify the feature data stored in the acquired data buffer 25 into clusters, for example, by the k-means++ method.

Then, in step SA104, the host device H1 acquires the data resulting from the clustering in step SA103 as data Rst(k, seed1).

(Steps SB101 to SB104):
The host device H1 sets the random number seed to seed2 (a value different from seed1) (step SB101), and sets the initial value of the number of clusters k to "2" (step SB102).

In step SB103, the host device H1 performs a process (clustering process) to classify the feature data stored in the acquired data buffer 25 (the same data as the data to be processed in step SA103) into clusters, for example, using the k-means++ method.

Then, in step SB104, the host device H1 acquires the data of the clustering result of step SB103 as data Rst(k, seed2).

(Step S106)
In step S106, the host device H1 judges whether the clustering result data Rst(k, seed1) when the random number seed is set to seed1 is the same as the clustering result data Rst(k, seed2) when the random number seed is set to seed2. If the result of the judgment is that the clustering result data Rst(k, seed1) and the clustering result data Rst(k, seed2) are the same, the host device assigns the currently set number of clusters k to the variable k0 (step S106), and further performs a process of incrementing the value of k by +1 (step S107). Then, the process returns to SA103 and SB103. Then, the processes of steps SA103 to SA104, SB103 to SB104, and S105 are repeatedly executed.

On the other hand, if the result of the determination is that the clustering result data Rst(k, seed1) and the clustering result data Rst(k, seed2) are not identical, the host device H1 sets the optimal number of clusters k_opt to the number of clusters k0 that is set immediately before the currently set number of clusters. As a result, when the random number seed is changed, the maximum number of clusters that results in the same clustering results is set as the optimal number of clusters. In other words, the above process makes it possible to obtain the optimal number of clusters k_opt.

FIG. 7 shows an example of a process for determining the optimum number of clusters.
(A) Clustering result when the random number seed is set to “35” (A1) number of clusters = 3 (k = 3)
(A2) Clustering result when the number of clusters is 3 (k=4),
(A3) Clustering result when the number of clusters is 3 (k=5), and
(B) Clustering result when the random number seed is set to “171” (B1) number of clusters=3 (k=3)
(B2) Clustering result when the number of clusters is 3 (k=4),
(B3) Clustering result when the number of clusters is 3 (k=5),
In the graph of Fig. 7, the horizontal axis represents time and the vertical axis represents cluster number.

In the case of Figure 7, when the number of clusters is 4 (k=4), the clustering result (Rst(k=4, seed1=35)) of random number seed = 35 and the clustering result (Rst(k=4, seed2=171)) of random number seed = 171 are the same data, but when the number of clusters is 5 (k=5), the clustering result (Rst(k=5, seed1=35)) of random number seed = 35 and the clustering result (Rst(k=5, seed2=171)) of random number seed = 171 are not the same data (the clustering results of the part surrounded by the dotted line and the part surrounded by the dashed line in the figure are different).

Therefore, in the case of Figure 7, the maximum number of clusters that will give the same clustering result when the random number seed is changed is "4", and this value is set as the optimal number of clusters k_opt.

In other words, the host device H1 obtains the optimal number of clusters k_opt through the above process, and performs clustering processing using the obtained optimal number of clusters k_opt, for example, by the k-means++ method. This allows the host device H1 to perform clustering processing with high accuracy.

The process of obtaining the optimal number of clusters k_opt is not limited to the procedure of the flowchart shown in FIG. 6, and for example, the number of random number seeds to be changed may be "3" or more. In this case, the same process as SA101 to SA104 (the random number seeds are set to values different from seed1 and seed2) is executed in parallel for the number of random number seeds, and when the number of random number seeds set is N1, in step S105, it is determined whether or not all N1 clustering result data are the same. Also, the steps (SA101 to SA104, SB101 to SB104) executed in the parallel process of FIG. 6 may be processed in series (serial processing).

(Step S6):
In step S6, the host device H1 executes a teacher data generation process. Specifically, the host device H1 generates training data (teacher data) by (automatically or manually) assigning the same label to data classified into the same cluster based on the result of the clustering process in step S5. In other words, the host device H1 generates data (training data (teacher data)) for training a neural network model of a classifier that classifies signal types from feature data by using data in which feature data and labels assigned to clusters into which the feature data is classified as training data.

Using the learning data generated in step S6, a classifier that classifies signal types from feature data is trained using a classification model such as a neural network, thereby acquiring a trained model of the classifier that classifies signal types from feature data. Note that this learning process may be executed by the host device H1, or may be executed by a computer with high processing power such as a server.

The trained model of the classifier that classifies the signal type from the feature data obtained by the above learning process is constructed in the ROM 23 or RAM 24 of the host device H1. In other words, the program, code, parameter data, etc. for executing processing using the trained model of the classifier that classifies the signal type from the feature data obtained by the above learning process are stored in the ROM 23 or RAM 24 of the host device H1.

(1.2.2: Processing during prediction)
Next, the prediction process of the signal type classification system 100 will be described.

Figure 8 is a schematic diagram of the signal type classification system 100 during prediction.

As shown in FIG. 8, the trained model of the classifier that classifies the signal type from the feature data, obtained by the above learning process, is constructed in the ROM 23 of the host device H1. In other words, the program, code, parameter data, etc. for executing the process using the trained model of the classifier that classifies the signal type from the feature data, obtained by the above learning process, are stored in the ROM 23 of the host device H1.

Figure 9 is a sequence diagram to explain the processing (signal type classification processing) during prediction by the signal type classification system 100.

Figure 10 is a schematic diagram of a data collection system 1000 including n signal type classification systems 100_1 to 100_n.

(Steps S21 to S24):
In steps S21 to S24, the signal type classification system 100 executes the same processes as those in steps S1 to S4 during learning.

(Step S25):
In step S25, the host device H1 executes a signal type classification process. Specifically, the host device H1 inputs the feature data stored in the acquired data buffer 25 to a trained model of a signal type classifier constructed in the host device H1, and acquires data output from the trained model of the signal type classifier. The data output from the trained model of the signal type classifier is data in which a label is added to the input feature data, and the assigned label makes it possible to identify the type of wireless signal or electromagnetic noise from which the input feature data was acquired (to classify the signal type).

The host device H1 may transmit the data (labeled data) of the signal types acquired as described above to, for example, an external server via the third communication interface 27.

For example, as shown in FIG. 10, the host device H1 may be connected to the server Svr1 via a network NW1 (e.g., a LAN), and the host device H1 may transmit data (labeled data) that classifies the signal types acquired as described above to the server Svr1 via the third communication interface 27 and the network NW1.

Also, as shown in FIG. 10, a data collection system 1000 may be constructed by providing n (n: natural number of 2 or more) signal type classification systems 100_1 to 100_n and connecting the host devices H1 to Hn of each of the signal type classification systems 100_1 to 100_n to a server Svr1 via a network NW1.

In the data collection system 1000 constructed in this manner, for example, by installing the signal type classification systems 100_1 to 100_n at different positions in a narrow space, a large amount of data (labeled data) in which the signal types are classified using a trained model (classifier) from feature data of wireless signals and electromagnetic noise flying through the narrow space can be acquired, and the acquired large amount of data can be collected in the server Svr1. Note that the server Svr1 may be connected to a data storage unit Strg that can store and hold a large amount of data, as shown in FIG. 10, and in this case, the server Svr1 stores the data (signal type classification data) collected from the signal type classification systems 100_1 to 100_n in the data storage unit Strg.

Furthermore, the server Svr1 analyzes the data (signal type classification data) collected as described above, and, for example, performs a process to visualize the analysis results, thereby enabling the radio wave conditions in the narrow space of the subject from which the data was acquired to be properly understood.

In addition, in the data collection system 1000, by setting the frequency bands of the data acquisition targets of the wireless communication devices S1_1 to S1_N, S2_1 to S2_N, ..., Sn_1 to Sn_N included in each of the signal type classification systems 100_1 to 100_n signal type classification systems 100_1 to 100_n to various values, it is possible to acquire data on wireless signals and electromagnetic noise in various frequency bands in the narrow space from which the data is acquired. As a result, in the data collection system 1000, it is possible to appropriately grasp the radio wave conditions in various frequency bands in the narrow space from which the data is acquired.

<Summary>
As described above, in the signal type classification system 100, the amount of data can be significantly reduced by performing the feature data acquisition process in each wireless communication device, and therefore the amount of data transmitted from each wireless communication device to the host device H1 can be significantly reduced. In addition, in the signal type classification system 100, the amount of data used in the clustering process in the host device H1 is also significantly reduced, so that the clustering process in the host device H1 can be performed at a high speed. Furthermore, in the signal type classification system 100, the optimal number of clusters for the clustering process is acquired, and the clustering process is performed mainly based on the optimal number of clusters, so that the clustering process can be performed with high accuracy. In addition, in the signal type classification system 100, each wireless communication device performs a process of acquiring feature data by extracting only data on the low-dimensional side of the cepstrum (16-dimensional data on the low-dimensional side), and the host device H1 performs a clustering process using the feature data acquired by the process, so that the accuracy of the clustering process can be improved.

Then, in the signal type classification system 100, labels are assigned to the data classified by the above-mentioned high-precision clustering process, and data in which feature data is associated with the labels assigned to the clusters into which the feature data is classified is used as training data, thereby generating data (training data (teacher data)) for training a neural network model of a classifier that classifies signal types from feature data.

In this way, the signal type classification system 100 can perform a method of generating learning data for constructing a classifier (a model realized using a neural network that can be the subject of machine learning, and a classifier model that classifies signal types from feature data) that is installed in the signal type classification system 100 and that appropriately classifies the types of wireless signals and electromagnetic noise whose frequency spectra and waveform patterns are difficult to predict, by machine learning.

In addition, by using the training data generated as described above to train a classifier model that classifies signal types from feature data (a model that is realized using a neural network that can be the subject of machine learning, and is a classifier model that classifies signal types from feature data), a trained model of the classifier that classifies signal types from feature data can be acquired. Then, in the signal type classification system 100, by using the acquired trained model (signal type classifier) to perform signal type classification processing on the feature data acquired by each wireless communication device, it is possible to appropriately classify the type of even wireless signals or electromagnetic noise whose frequency spectrum or waveform pattern is difficult to predict.

In other words, in the signal type classification system 100, by performing feature data acquisition processing in each wireless communication device, the amount of data can be significantly reduced, so that during learning and prediction, a large amount of feature data with a small amount of individual data can be acquired while maintaining the characteristics of wireless signals and electromagnetic noise, and the acquired large amount of feature data can be used to generate learning data, to learn a classifier (a model realized using a neural network that can be the subject of machine learning, which is a classifier model that classifies signal types from feature data), and to perform signal type classification processing using the trained model of the signal type classifier. Therefore, in the signal type classification system 100, even wireless signals and electromagnetic noise whose frequency spectra and waveform patterns are difficult to predict can be appropriately classified into their types.

First Modified Example
Next, a first modified example of the first embodiment will be described. Note that the same reference numerals are used to designate the same parts as those in the above embodiment, and detailed descriptions thereof will be omitted.

Figure 11 is a schematic diagram of a signal type classification system 100A according to a first modified example of the first embodiment.

In the signal type classification system 100A of this modified example, as shown in FIG. 11, one antenna Ant_S1 is connected to N wireless communication devices S1A_1 to S1A_N. In other words, in the signal type classification system 100A, the received signal is processed by the N wireless communication devices S1A_1 to S1A_N using one antenna Ant_S1 (a common antenna), so that the phase difference and level difference of the received signal can be adjusted among the N wireless communication devices S1A_1 to S1A_N.

The signal type classification system 100A has a configuration in which an oscillator Osc1 and a distributor Dist1 are added to the signal type classification system 100 of the first embodiment, the wireless communication device S1_1 is replaced with a wireless communication device S1A_1, and the wireless communication devices S1_2 to S1_N are replaced with wireless communication devices S1A_1 to S1_N, respectively.

The oscillator Osc1 is an oscillator for generating a reference signal sig_ref. The reference signal sig_ref generated by the oscillator Osc1 is output to the wireless communication device S1A_1.

As shown in FIG. 11, the wireless communication device S1A_1 has a configuration in which a synchronization signal generating unit 16 is added to the wireless communication device S1_1 of the first embodiment.

The synchronization signal generating unit 16 inputs the reference signal sig_ref output from the oscillator Osc1, and generates a synchronization signal sig_sync (e.g., a clock signal) synchronized with the reference signal sig_ref. The synchronization signal generating unit 16 then outputs the generated synchronization signal sig_ref to each functional unit of the wireless communication device S1A_1, and each functional unit performs processing (including sampling processing) based on the synchronization signal sig_ref.

In addition, the synchronization signal generation unit 16 outputs the generated synchronization signal sig_sync to the distributor Dist1.

The distributor Dist1 receives the synchronization signal sig_sync output from the synchronization signal generating unit 16 of the wireless communication device S1A_1, and outputs (distributes) the synchronization signal sig_sync to each of the wireless communication devices S1A_2 to S1A_N.

Each of the wireless communication devices S1A_2 to S1A_N receives the synchronization signal sig_sync output from the distributor Dist1, and executes processing in each functional unit based on the synchronization signal sig_sync (synchronized with the synchronization signal sig_sync).

In this way, in the signal type classification system 100A, the synchronization signal sig_sync generated from the common reference signal sig_ref is used in the N wireless communication devices S1A_1 to S1A_N, so that the sampling timing can be aligned in the N wireless communication devices S1A_1 to S1A_N, and processing can be performed in synchronization with the common synchronization signal sig_sync. This further enables the signal type classification system 100A to perform highly accurate processing.

<Second Modified Example>
Next, a second modification of the first embodiment will be described. Note that the same parts as those in the above embodiment are given the same reference numerals and detailed description will be omitted.

Figure 12 is a schematic diagram of a signal type classification system 100B according to a second modified example of the first embodiment.

As shown in FIG. 12, the signal type classification system 100B of this modified example has a configuration in which the host device H1 of the signal type classification system 100 of the first embodiment is replaced with a server Svr1 having the functions of the host device H1.

In the signal type classification system 100B, for example, the server Svr1 can be made more functional than the host device H1, so that the feature data transmitted from each wireless communication device can be processed using the more functional server Svr1.

<Third Modified Example>
Next, a third modified example of the first embodiment will be described. Note that the same reference numerals are used to designate the same parts as those in the above embodiment, and detailed description thereof will be omitted.

Figure 13 is a schematic diagram of a data collection system 1000A according to a third modified example of the first embodiment.

The data collection system 1000A of this modified example has a configuration similar to that of the data collection system 1000 of the first embodiment (see FIG. 10), in which the server Svr1 is further connected to a network NW2 to which other servers can be connected.

In this way, in the data collection system 1000A, for example, data acquired by the server Svr1 can be transmitted to another server (for example, server Svr2 in FIG. 13) via the network NW2, and the multiple servers connected to the network NW2 can perform analysis processing, data collection processing, data storage management processing, etc., on the data acquired by the signal type classification system. Note that the data collection system 1000A is not limited to the case shown in FIG. 13, and may have a configuration in which multiple servers (three or more servers) are connected to the network NW2. Also, in the data collection system 1000A, data may be aggregated in a specified server, or data may be distributed among multiple servers. Also, processing of the acquired data (for example, analysis processing, visualization processing), etc. may be distributed among multiple servers (edge processing, cloud processing may be performed).

In this way, the signal type classification system can acquire large amounts of feature data with small individual data volumes while maintaining the characteristics of wireless signals and electromagnetic noise, making it easy to build systems with diverse configurations that include multiple signal type classification systems. In other words, because the signal type classification system can acquire feature data with small individual data volumes, even if a large amount of that feature data is transmitted to other devices via communication, the communication load is dramatically smaller than in the past. For this reason, systems capable of handling large amounts of feature data can be built with diverse configurations. In other words, by using the present invention, the scalability of the system configuration becomes diverse, and the system configuration can be easily expanded and modified.

[Other embodiments]
In the above embodiment (including the modified example), the wireless communication device used in the signal type classification system and the data collection system may be realized by an SDR (software defined radio). In this case, some or all of the functional units of the wireless communication device may be realized by, for example, an FPGA.

In addition, some or all of the above embodiments (including variations) may be combined to configure a signal type classification system, a data collection system, a wireless communication device, a host device, a server, etc.

In the above embodiment (including the modified example), the case where data obtained by extracting the low-dimensional part of the cepstrum data is obtained as the feature data has been described, but the present invention is not limited to this. For example, if the feature data reflects the signal characteristics of the time domain of the wireless signal and/or electromagnetic noise, or the spectrum characteristics in the frequency domain, other data may be adopted as the feature data and the feature data acquisition process may be performed. For example, histogram data of the values of IQ data (IQ samples) acquired during a specified period (specified time) may be adopted as the feature data. In this case, for example, regardless of the timing of the occurrence of impulse-like noise during the specified period, the histogram data will be similar data, so that it is possible to acquire feature data that prominently represents the characteristics of the impulse-like noise.

In addition, in the signal type classification system, data collection system, wireless communication device, host device, and server described in the above embodiments (including modified examples), each block may be individually integrated into a single chip using a semiconductor device such as an LSI, or may be integrated into a single chip to include some or all of the blocks.

Note that although we refer to it as an LSI here, it may also be called an IC, system LSI, super LSI, or ultra LSI depending on the level of integration.

In addition, the method of integration is not limited to LSI, but may be realized by a dedicated circuit or a general-purpose processor. It is also possible to use a field programmable gate array (FPGA) that can be programmed after LSI manufacturing, or a reconfigurable processor that can reconfigure the connections and settings of circuit cells inside the LSI.

In addition, some or all of the processing of each functional block in each of the above embodiments may be realized by a program. And some or all of the processing of each functional block in each of the above embodiments is performed by a central processing unit (CPU) in a computer. Also, the programs for performing each process are stored in a storage device such as a hard disk or ROM, and are read out and executed in the ROM or RAM.

The processes in the above embodiments may be implemented by hardware or software (including cases where they are implemented together with an operating system (OS), middleware, or a specified library). Furthermore, they may be implemented by a combination of software and hardware.

For example, when each functional unit of the above embodiment (including the modified example) is realized by software, each functional unit may be realized by software processing using the hardware configuration shown in FIG. 14 (for example, a hardware configuration in which a CPU, ROM, RAM, input unit, output unit, etc. are connected by a bus).

In addition, when each functional unit of the above embodiment (variation example) is realized by software, the software may be realized by using a single computer having the hardware configuration shown in FIG. 14, or may be realized by distributed processing using multiple computers.

In addition, the execution order of the processing methods in the above embodiment is not necessarily limited to that described in the above embodiment, and the execution order can be changed without departing from the spirit of the invention.

The scope of the present invention includes a computer program that causes a computer to execute the above-mentioned method and a computer-readable recording medium on which the program is recorded. Here, examples of computer-readable recording media include flexible disks, hard disks, CD-ROMs, MOs, DVDs, DVD-ROMs, DVD-RAMs, large-capacity DVDs, next-generation DVDs, and semiconductor memories.

The computer program is not limited to one recorded on the recording medium, but may be one transmitted via a telecommunications line, a wireless or wired communication line, a network such as the Internet, etc.

In addition, in the above embodiment (including the modified examples), "same" is a concept that includes being roughly the same. "Simultaneous" is a concept that includes being roughly the same. "Matched" is a concept that includes being roughly the same.

The specific configuration of the present invention is not limited to the above-described embodiment, and various changes and modifications are possible without departing from the spirit of the invention.

100, 100A, 100B Signal type classification systems S1_1 to S1_N, S1A_1 to S1A_N Wireless communication device H1 Host device 1000 Data collection systems Svr1, Svr2 Server

Claims (8)

a radio wave waveform acquisition processing step for acquiring radio wave waveform data of radio waves received by an antenna;
a feature acquisition processing step of acquiring feature data which indicates a feature in a time domain or a frequency domain of the radio wave waveform data and is smaller in amount than the radio wave waveform data;
a clustering process step of classifying the feature amount data into clusters;
a learning data acquisition step of acquiring data in which the feature amount data and the labels are associated with each other as learning data by assigning labels to the feature amount data classified by the clustering process;
an optimal cluster number acquisition step for acquiring an optimal number of classifications in the clustering process;
Equipped with
The optimal cluster number acquisition step includes:
While changing the number of clusters, a number of clusters is detected that does not change the classification state of the clusters even if the initial setting of the clustering process is changed, and an optimal number of classifications is obtained based on the detected number of clusters.
A method for generating learning data for signal type classification processing. The feature data is data extracted from a cepstrum.
The learning data generating method according to claim 1 . The feature data is data obtained by extracting or processing low-order components of a cepstrum.
The learning data generating method according to claim 1 . A signal type classification system that performs a signal type classification process using a trained model of a signal type classifier acquired by performing a learning process using training data generated by the training data generation method according to any one of claims 1 to 3 ,
A wireless communication device;
A host device;
Equipped with
The wireless communication device includes:
a radio wave waveform acquisition processing step for acquiring radio wave waveform data of radio waves received by an antenna;
a feature acquisition processing step of acquiring feature data which indicates a feature in a time domain or a frequency domain of the radio wave waveform data and is smaller in amount than the radio wave waveform data;
a transmitting step of transmitting the feature amount data to a host device;
Run
The host device is
a receiving step of receiving the feature data transmitted from the wireless communication device;
a label acquisition step of acquiring a label of the feature data by inputting the feature data acquired by the wireless communication device into a trained model of the signal type classifier;
Execute
Signal type classification system. The feature data is data extracted from a cepstrum.
5. The signal type classification system according to claim 4 . The feature data is data obtained by extracting or processing low-order components of a cepstrum.
5. The signal type classification system according to claim 4 . N signal type classification systems according to any one of claims 4 to 6 (N: natural number);
A server that collects signal type data from the N signal type classification systems, the signal type data being data obtained by the N signal type classification systems and including feature amount data associated with a label;
A data collection system comprising: A program for causing a computer to execute the learning data generating method according to any one of claims 1 to 3 .

Images