Radio wave abnormality detection system, radio wave abnormality detection method, and radio wave abnormality detection program

Abstract

To provide a radio wave abnormality detection system, a radio wave abnormality detection method, and a radio wave abnormality detection program that can accurately detect a radio wave abnormality contained in received data.SOLUTION: A radio wave abnormality detection system includes: a first feature quantity extraction unit that extracts a feature quantity of received data for each first predetermined period as a first feature quantity; a second feature quantity extraction unit that extracts a feature quantity of the received data for each second predetermined period longer than the first predetermined period as a second feature quantity; and an abnormality detection unit that uses the first feature quantity extracted by the first feature quantity extraction unit and the second feature quantity extracted by the second feature quantity extraction unit to detect a radio wave abnormality contained in the received data.SELECTED DRAWING: Figure 11

Description

Technical Field

[0001] The present invention relates to a radio wave abnormality detection system, a radio wave abnormality detection method, and a radio wave abnormality detection program.

Background Art

[0002] Wireless communication using radio waves is utilized in various fields. Among them, communications such as police radio, fire radio, aviation radio, and railway radio, which are particularly important, are called critical wireless communications. When interference occurs in critical wireless communications, it may develop into a situation involving human life. Therefore, it is very important to detect (monitor) radio wave interference and faults in the emission status of radio waves used in these communications.

[0003] In recent years, as a type of private radio, private LTE (Long Term Evolution) and local 5G, which are communication standards for mobile phones and have evolved from 4G / 5G (the fourth generation / fifth generation mobile communication system), have also attracted attention. In local 5G and the like, it is possible to obtain a license and construct a wireless communication system for each private area or operator. Therefore, similar to the above-mentioned critical wireless communications, by automatically detecting (monitoring) radio wave interference from adjacent areas and the like to a wireless communication system such as local 5G, and faults in the self-system, it is also expected to improve the efficiency of preventing the degradation of the wireless performance of the system and analyzing the cause of the degradation. Here, a fault includes a radio wave fault, a failure or fault of a system, equipment, etc. (hereinafter simply referred to as a fault).

[0004] Normally, when detecting the emission status of radio waves and radio wave interference and faults in a wireless communication system, a method of setting a threshold value for a reception level of a certain radio wave and determining that a radio wave with a reception level exceeding the threshold value is abnormal can be considered. The threshold value of the reception level determined for each frequency of radio waves is called a spectrum mask. In this case, since only the case where the reception level exceeds the value of the spectrum mask is determined as abnormal, there is a problem that it is not determined as abnormal when the reception level is low or when an abnormality occurs in a feature amount other than the reception level.

[0005] A method for identifying abnormal radio wave environments and interference factors is also disclosed in, for example, Patent Document 1.

[0006] Patent Document 1 proposes a technique for determining the presence or absence of radio wave interference and its causes by extracting amplitude feature amounts for each predetermined time from received sampling data and calculating the degree of similarity between the extracted amplitude feature amounts and a plurality of teacher data.

Prior Art Documents

Patent Documents

[0007]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0008] In Patent Document 1, since the amplitude feature amount is extracted for each predetermined time, there is a problem that it is impossible to detect the occurrence of radio wave interference having a period exceeding the predetermined time. That is, in Patent Document 1, there is a problem that it is impossible to accurately detect radio wave abnormalities included in received data.

[0009] One object of the present disclosure is to provide a radio wave abnormality detection system, a radio wave abnormality detection method, and a radio wave abnormality detection program that solve the above-described problems.

Means for Solving the Problems

[0010] According to one embodiment, a radio wave abnormality detection system outputs a detection result regarding a radio wave abnormality included in the received data, which is detected using a first feature amount that is a feature amount extracted from the received data for each first predetermined period and a second feature amount that is a feature amount extracted from the received data for each second predetermined period longer than the first predetermined period.

[0011] According to one embodiment, a radio wave abnormality detection method includes: a first feature amount extraction step of extracting, as a first feature amount, a feature amount for each first predetermined period of received data; a second feature amount extraction step of extracting, as a second feature amount, a feature amount for each second predetermined period of the received data that is longer than the first predetermined period; and an abnormality detection step of detecting a radio wave abnormality included in the received data by using the first feature amount and the second feature amount.

[0012] According to one embodiment, a radio wave abnormality detection program causes a computer to execute: a first feature amount extraction process of extracting, as a first feature amount, a feature amount for each first predetermined period of received data; a second feature amount extraction process of extracting, as a second feature amount, a feature amount for each second predetermined period of the received data that is longer than the first predetermined period; and an abnormality detection process of detecting a radio wave abnormality included in the received data by using the first feature amount and the second feature amount.

Advantages of the Invention

[0013] According to the above-described embodiment, it is possible to provide a radio wave abnormality detection system, a radio wave abnormality detection method, and a radio wave abnormality detection program that can accurately detect a radio wave abnormality included in received data.

Brief Description of the Drawings

[0014]

Figure 1

Figure 2

Figure 3

Figure 4A

Figure 4B

Figure 4C

Figure 5

Figure 6A

Figure 6B

Figure 6C

Figure 6D

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13A

Figure 13B

Figure 13C

Figure 14

Figure 15

Figure 16

Figure 17A

Figure 17B

Figure 17C

Figure 18

Embodiments for Carrying Out the Invention

[0015] Hereinafter, embodiments will be described with reference to the drawings. Note that since the drawings are simplified, the technical scope of the embodiments should not be narrowly interpreted based on the description of these drawings. Also, the same elements are denoted by the same reference numerals, and redundant descriptions are omitted.

[0016] Also, in the following embodiments, when necessary for convenience, they are divided and described in a plurality of sections or embodiments. However, unless otherwise explicitly stated, they are not unrelated to each other, and one is related to a partial or entire modification example, application example, detailed description, supplementary description, etc. of the other. Also, in the following embodiments, when referring to the number of elements, etc. (including the number, numerical value, quantity, range, etc.), unless otherwise explicitly stated and except for cases where it is clearly limited to a specific number in principle, it is not limited to that specific number, and it may be more than or less than the specific number.

[0017] Furthermore, in the following embodiments, the components (including operation steps, etc.) are not necessarily essential unless specifically stated or considered clearly essential in principle. Similarly, in the following embodiments, when referring to the shape, positional relationship, etc. of components, etc., unless specifically stated or considered clearly not so in principle, those substantially approximating or similar to the shape, etc. shall be included. This also applies to the above numbers (including the number, numerical value, quantity, range, etc.).

[0018] <Overview of Embodiment> Hereinafter, with reference to FIGS. 1 to 18, the details of the radio wave abnormality detection system according to Embodiments 1 to 3 will be described.

[0019] First, in Embodiment 1, a first feature amount extraction unit that extracts a first feature amount, which is a short-term feature amount, provided in the radio wave abnormality detection system, a second feature amount extraction unit that performs re-feature quantification using a plurality of first feature amounts extracted by the first feature amount extraction unit to extract a second feature amount, which is a long-term feature amount, an extraction period determination unit that determines the extraction period of the second feature amount by the second feature amount extraction unit, and a detection processing unit that performs learning of a model representing the presence or absence of radio wave abnormality in the received data and determination of the presence or absence of radio wave abnormality in the received data using the first feature amount and the second feature amount will be described in detail in terms of their respective basic configurations, features, and operations. The details of the feature amounts will be described later.

[0020] Also, in Embodiment 2, an example will be described in which the detection processing unit provided in the radio wave abnormality detection system has a function of learning and determining using the first feature amount (short-term feature amount) extracted by the first feature amount extraction unit and a function of learning and determining using the second feature amount (long-term feature amount) extracted by the second feature amount extraction unit, respectively.

[0021] Furthermore, in Embodiment 3, in the detection determination using the second feature amount (long-term feature amount) by the detection processing unit provided in the radio wave abnormality detection system, instead of determining the presence or absence of radio wave abnormality using a learned model, an example in which the presence or absence of radio wave abnormality is determined by comparing the maximum continuous transmission time determined during learning with the transmission continuous time of the transmission signal included in the received data will be described. Here, the transmission signal included in the received data is, for example, a signal transmitted from a radio device other than the receiver or a radio wave interference source.

[0022] <Embodiment 1> FIG. 1 is a block diagram showing an example of the overall configuration of the radio wave abnormality detection system 100 according to Embodiment 1. Further, FIG. 2 is a block diagram showing a modified example of the radio wave abnormality detection system 100 as the radio wave abnormality detection system 101. In FIG. 1, a part of each of the learning process and the determination process is realized using a common processing circuit, whereas in FIG. 2, the learning process and the determination process are realized using separate processing circuits. Since any of the configurations shown in FIGS. 1 and 2 can achieve the same effect, hereinafter, mainly the configuration shown in FIG. 1 will be described.

[0023] The radio wave abnormality detection system 100 includes, as an example, a feature amount extraction unit (first feature amount extraction unit) 20 that extracts a first feature amount that is a short-term feature amount, a re-feature quantification unit (second feature amount extraction unit) 30 that performs re-feature quantification using a plurality of first feature amounts extracted by the feature amount extraction unit 20 to extract a second feature amount that is a long-term feature amount, an extraction period determination unit 40 that determines the extraction period of the second feature amount by the re-feature quantification unit 30, and a detection processing unit (abnormality detection unit) 50 that performs learning of a model representing the presence or absence of radio wave abnormality in the received data and determination of the presence or absence of radio wave abnormality in the received data using the first feature amount and the second feature amount.

[0024] The feature extraction unit 20 includes at least a spectrogram conversion unit 26, a signal extraction unit 27, and a feature quantization unit 28. The spectrogram conversion unit 26 performs at least one of spectrum conversion and spectrogram conversion on the received data. The signal extraction unit 27 extracts the signal region of the transmission signal included in the received data and the reception level of the transmission signal using a threshold value of the noise level or the like from the spectrogram or the like output from the spectrogram conversion unit 26. The feature quantization unit 28 performs feature quantization for each short extraction period (predetermined period) T1 using the extraction result of the signal extraction unit 27. As a result, in the feature extraction unit 20, a plurality of first features, which are short-term features, are extracted for each short extraction period T1.

[0025] The re-feature quantization unit 30 includes a feature addition unit 37 and a feature quantization unit 38. The feature addition unit 37 adds (summates) a plurality of first features for a long extraction period (predetermined period) T2. The feature quantization unit 38 performs re-feature quantization using the result of adding a plurality of first features for each long extraction period T2.

[0026] The extraction period determination unit 40 includes a signal level estimation unit 46 and a maximum continuous transmission time estimation unit 47. The signal level estimation unit 46 estimates (or sets) the signal level of the transmission signal included in the received data (the reception level exceeding the reception level estimated as the noise floor level among the reception levels of the received data) using a plurality of first features extracted by the feature extraction unit 20. The maximum continuous transmission time estimation unit 47 estimates the maximum continuous transmission time (such as the maximum value of the time during which the signal level is maintained) using the signal level estimated by the signal level estimation unit 46 and a plurality of first features extracted by the feature extraction unit 20. Note that the long extraction period T2 is determined from the estimated maximum continuous transmission time and output to the re-feature quantization unit 30.

[0027] The detection processing unit 50 includes a learning unit 60, a database 70, and a determination unit 80. During the learning process, the learning unit 60 performs machine learning on a model that represents whether radio wave anomalies (radio wave interference or obstacles) are included in the received data, using the first feature amount (first sample feature amount) and the second feature amount (second sample feature amount) extracted from the sample data. The database 70 stores a learned model (hereinafter also referred to as the learned model 70) generated by the machine learning of the learning unit 60 and the like. During the determination process, the determination unit 80 determines whether radio wave anomalies are included in the received data, using the first feature amount output from the feature amount extraction unit 20, the second feature amount output from the re-feature quantification unit 30, and the learned model 70. Note that the learning process refers to a process of performing machine learning on a model that represents whether radio wave anomalies are included in the actual received data to be performed later, using sample data in advance, and the determination process refers to a process of determining whether radio wave anomalies are included in the received data, using the actual received data.

[0028] Note that the feature amount is statistical information including information in both the frequency direction and the time direction of the received data during a predetermined extraction period. For example, the feature amount is assumed to be statistical feature amounts extracted for each predetermined period, such as a frequency distribution (histogram), a probability density function (PDF), a cumulative distribution function (CDF), and an amplitude probability distribution (APD). Note that the feature amount may be directly extracted from the received data on the time axis, or may be extracted for each frequency after being converted into frequency-axis data by spectralization or spectrogramming.

[0029] Note that, compared with the radio wave abnormality detection system 100 shown in FIG. 1, the radio wave abnormality detection system 101 shown in FIG. 2 separately includes a feature amount extraction unit 20_1 for learning processing and a feature amount extraction unit 20_2 for determination processing instead of the feature amount extraction unit 20, and separately includes a re-feature quantization unit 30_1 for learning processing and a re-feature quantization unit 30_2 for determination processing instead of the re-feature quantization unit 30. Each of the feature amount extraction units 20_1 and 20_2 has the same function as the feature amount extraction unit 20, and each of the re-feature quantization units 30_1 and 30_2 has the same function as the re-feature quantization unit 30. Further, compared with the radio wave abnormality detection system 100 shown in FIG. 1, the radio wave abnormality detection system 101 shown in FIG. 2 includes a detection processing unit 50_1 for learning processing and a detection processing unit 50_2 for determination processing instead of the detection processing unit 50. The detection processing unit 50_1 has a learning unit 60 and a database 70, and the detection processing unit 50_2 has a determination unit 80. Since the other configurations and operations of the radio wave abnormality detection system 101 are the same as those of the radio wave abnormality detection system 100, the description thereof is omitted. Note that the database 70 may be provided in the detection processing unit 50_2 or may be provided independently.

[0030] (Operation of Radio Wave Abnormality Detection System 100) Subsequently, the operation of the radio wave abnormality detection system 100 according to Embodiment 1 will be described. FIG. 3 is a flowchart showing an overview of the overall processing flow by the radio wave abnormality detection system 100.

[0031] As shown in FIG. 3, the processing by the radio wave abnormality detection system 100 is divided into a learning process step performed in advance or periodically and a determination process step sequentially performed during operation. The learning process mainly includes a process of extracting a first feature amount which is a short-term feature amount (steps S11 to S14), a process of determining an extraction period for extracting a second feature amount which is a long-term feature amount (steps S21 to S23), a process of extracting (re-feature quantifying) the second feature amount (steps S31 to S33), and a learning process (step S41). The determination process mainly includes a process of sequentially extracting the first feature amount and the second feature amount (steps S61 to S67), and a process of determining whether the received data contains a radio wave abnormality using the extracted first and second feature amounts and the learned model (step S68).

[0032] (Details of the learning process by the radio wave abnormality detection system 100) FIGS. 4A to 4C are flowcharts showing the flow of the learning process by the radio wave abnormality detection system 100 according to Embodiment 1.

[0033] It is assumed that the learning process shown in FIGS. 4A to 4C is performed by collecting received data in a normal state (normal time) as sample data. The collection period of the received data (sample data) used for learning may generally include the communication situation assumed in the normal state (normal time). Therefore, even if the data is for several seconds to several tens of seconds on any one day, or data for several tens of seconds or several hours extracted randomly or at regular intervals within several days or one week, there is no problem at all.

[0034] First, among the learning processes, the process of extracting a first feature amount (first sample feature amount) which is a short-term feature amount in the feature amount extraction unit 20 (steps S11 to S14) will be described.

[0035] First, the spectrogram conversion unit 26 performs spectrogram conversion or spectrogramming on the received data collected for learning, and converts it into data indicating the reception level for each time or each frequency (step S11). Then, the feature quantification unit 28 extracts the first feature amount, which is a short-term feature amount, for each extraction period T1 (for example, 5 ms, etc.) determined by the radio standard, multiplexing period, etc. of the wireless communication system to be monitored from the data converted by the spectrogram conversion unit 26 (step S12) (steps S13→S14). At this time, in order to monitor the radio wave state at the signal level instead of the noise level and detect interference and obstacles at the signal level, in the signal extraction unit 27, signals below the noise floor are degenerate to a certain level to extract the signal level (step S13), and it may be passed to the feature quantification unit 28. However, a part of the signals (noise) at the noise floor level is intentionally extracted without being degenerate to a certain level. Then, the feature quantification unit 28 extracts, as the first feature amount, statistical feature amounts such as a frequency distribution (histogram), probability density function (PDF), cumulative distribution function (CDF), or amplitude probability distribution (APD) for the reception level for each predetermined extraction period T1 (step S14).

[0036] Next, in the learning process, the process of determining the extraction period T2 of the second feature amount (second sample feature amount), which is a long-term feature amount, in the extraction period determination unit 40 (steps S21 to S23) will be described.

[0037] First, the signal level estimation unit 46 estimates the signal level (noise floor level) using the first feature amount, which is a short-term feature amount extracted by the feature amount extraction unit 20 for each extraction period T1 (step S21). Then, the maximum continuous transmission time estimation unit 47 sequentially extracts the transmission continuous time (the time during which the signal level is maintained) of the transmission signal included in the received data based on the plurality of extracted first feature amounts and the information on the estimated signal level (step S22). Thereafter, the maximum continuous transmission time estimation unit 47 estimates the maximum continuous transmission time based on the distribution of each of the extracted transmission continuous times (step S23), and determines the extraction period T2 corresponding to the maximum continuous transmission time. The details of the operation of the extraction period determination unit 40 will be described later.

[0038] Next, among the learning processes, the process of extracting the second feature amount (second sample feature amount), which is a long-term feature amount, in the re-feature quantization unit 30 (steps S31 to S33) will be described.

[0039] First, the feature amount addition unit 37 adds (summates) a plurality of first feature amounts for each long-term extraction period T2 (steps S31 → S32). Thereafter, the feature quantization unit 38 extracts (re-feature quantizes) the addition result of the plurality of first feature amounts for each long-term extraction period T2 as the second feature amount (step S33).

[0040] Finally, among the learning processes, the machine learning process (step S41) will be described.

[0041] The learning unit 60 performs machine learning of a model representing whether the received data includes radio wave anomalies (radio wave interference or obstacles) using a high-dimensional (multi-dimensional) feature vector formed by concatenating the first feature amount and the second feature amount (step S41).

[0042] In this embodiment, it is assumed that the machine learning adopted by the learning unit 60 is unsupervised learning in order to detect unknown interference and obstacles that are difficult to define in advance as teacher data. As an example, it is assumed that it is One-class SVM (One-class Support Vector Machine). However, the machine learning adopted by the learning unit 60 is not limited to One-class SVM. For example, it may be machine learning of a model such as a deep anomaly detection model (Deep Anomaly Detection) using deep learning or the like. Alternatively, the machine learning adopted by the learning unit 60 is a method of learning a threshold for statistically detecting anomalies using Mahalanobis distance, variance (standard deviation), etc. from the distribution of high-dimensional (multidimensional) feature vectors as described above, or a method of predicting a change trend using a state filter such as a Kalman filter or a particle filter and detecting whether it deviates from the change trend. There is no problem even if such a method is used. Furthermore, the machine learning adopted by the learning unit 60 is not limited to unsupervised learning, and may be supervised learning. Note that in unsupervised learning, machine learning is performed without giving an output (correct answer) for the input, whereas in supervised learning, machine learning is performed with an output (correct answer) for the input given.

[0043] The learned model generated by the machine learning of the learning unit 60, the threshold for anomaly detection, the coefficient indicating the change trend, etc. are stored in the database 70.

[0044] (Details of the determination process by the radio wave anomaly detection system 100) FIG. 5 is a flowchart showing the flow of the determination process by the radio wave anomaly detection system 100 according to Embodiment 1.

[0045] The determination process shown in FIG. 5 is a process of acquiring reception data from the radio wave environment to be monitored during actual operation and sequentially performing detection determination of radio wave interference and obstacles.

[0046] First, in the determination process, similar to the learning process, the process of extracting the first feature quantity, which is a short-term feature quantity, and the second feature quantity, which is a long-term feature quantity, is performed using the feature quantity extraction unit 20 and the re-feature quantization unit 30 (S61 to S67). Since the extraction processes of the first feature quantity and the second feature quantity in the determination process are the same as the extraction processes of the first feature quantity (first sample feature quantity) and the second feature quantity (second sample feature quantity) in the learning process, the description thereof is omitted.

[0047] Note that, for the short-term extraction period and the long-term extraction period during the determination process, the same values of the short-term extraction period T1 and the long-term extraction period T2 as those during the learning process are used. Specifically, for the short-term extraction period during the determination process, the value of the extraction period set based on the radio standard and multiplexing period of the wireless communication system to be monitored is used, and for the long-term extraction period during the determination process, the value of the extraction period determined by the process of the extraction period determination unit 40 during the learning process is used.

[0048] Next, the detection determination process (step S68) in the determination process will be described. The determination unit 80 performs a detection determination process (step S68) using the first feature quantity and the second feature quantity sequentially extracted from the received data and the learned model 70 generated by machine learning during the learning process.

[0049] For example, when the learned model 70 is generated by machine learning such as One-class SVM, which is a type of unsupervised learning, when a high-dimensional (multi-dimensional) feature quantity vector formed by concatenating the first feature quantity and the second feature quantity is input, the determination unit 80 converts the positional relationship and the distance of the feature quantity vector with respect to the discrimination boundary configured in the learned model 70 into a normality (or abnormality) and outputs it. Then, the determination unit 80 outputs a determination result as to whether the received data includes a radio wave abnormality based on the calculation result.

[0050] In addition, when performing statistical anomaly detection using the Mahalanobis distance, variance (standard deviation), etc., the determination unit 80 may calculate the distance of the feature vector from the learned threshold value learned during the learning process and output the distance as a determination result. Alternatively, when predicting the change trend using a state filter such as a Kalman filter or a particle filter, the determination unit 80 may calculate the degree of deviation of the feature vector from the coefficient indicating the change trend and output it as a determination result.

[0051] (Details of the operations of the feature quantity extraction unit 20, the re-feature quantization unit 30, and the extraction period determination unit 40) Next, the details of the operations of the feature quantity extraction unit 20, the re-feature quantization unit 30, and the extraction period determination unit 40 will be described.

[0052] Figures 6A to 6D are schematic diagrams showing an example of statistical feature quantities extracted from received data by the radio wave anomaly detection system 100. As already described, in this embodiment, the feature quantity extraction unit 20 extracts the first feature quantity for each short-term extraction period T1, and the re-feature quantization unit 30 extracts the second feature quantity for each long-term extraction period T2.

[0053] In the feature quantity extraction unit 20, first, the spectrogram conversion unit 26 converts received data such as a time-axis IQ signal into signals for each frequency by FFT (Fast Fourier Transform) processing or the like (step S11). Then, the signal extraction unit 27 extracts the signal level for each frequency based on the noise floor threshold value for each extraction period T1 (step S12) from the data converted by the spectrogram conversion unit 26 (step S13). Then, the feature quantization unit 28 extracts the first feature quantity from the signal levels for each frequency extracted by the signal extraction unit 27 (step S14).

[0054] Here, the feature quantity (including both the first and second feature quantities) may be, for example, a frequency distribution (histogram) for each received level extracted for each extraction period as shown in FIG. 6A, or may be a probability density function (PDF) obtained by functionalizing the unit of frequency as a probability density instead of the absolute value of the frequency as shown in FIG. 6B. However, when the PDF is used as the feature quantity, the feature quantity is not the probability density function itself, but a discrete value integrated for each fixed received level, and is approximately equal to the value obtained by normalizing the frequency distribution as the probability density. Alternatively, the feature quantity may be, for example, a cumulative distribution function (CDF) or an amplitude probability distribution (APD) obtained by cumulative probability conversion of the frequency distribution in FIG. 6A as shown in FIGS. 6C and 6D. The CDF indicates the probability that a signal below the received level occurs for each received level, and the APD indicates the probability that a signal above the received level occurs for each received level. Note that when the CDF is used as the feature quantity, the feature quantity is not the distribution function itself, but a discrete value integrated for each fixed received level, that is, approximately equal to the value obtained by normalizing after accumulating the frequency distribution. Similarly, when the APD is used as the feature quantity, the feature quantity is not the distribution function itself, but a discrete value integrated for each fixed received level, that is, approximately equal to the value obtained by normalizing after accumulating the frequency distribution.

[0055] Note that regardless of which form (type) of feature quantity the first feature quantity extracted by the feature quantity extraction unit 20 is among those shown in FIGS. 6A to 6D, in principle, in the subsequent processing, the first feature quantity can be converted into and used as any other form of feature quantity shown in FIGS. 6A to 6D. Therefore, different forms (types) of feature quantities may be used for the first feature quantity and the second feature quantity.

[0056] Furthermore, the feature quantity may be composed of, for example, the occurrence probability for each reception level as shown in FIGS. 6A to 6D, extracted for each spectrogrammed frequency (sub-carrier of FFT processing), or may be composed of the integrated value of the occurrence probability for each reception level as shown in FIGS. 6A to 6D, extracted for several frequencies (sub-carriers). That is, the feature quantity may be a high-dimensional (multi-dimensional) feature quantity vector composed of a matrix of the frequency dimension and the reception level dimension for each extraction period.

[0057] FIG. 7 is a diagram for explaining the first feature quantity extracted by the radio wave abnormality detection system 100 and a method for detecting radio wave abnormality using the same. FIG. 7 shows an example of the first feature quantity at an arbitrary frequency, extracted for each extraction period T1 (for example, 5 ms, etc.) set based on the radio standard and multiplexing period of the wireless communication system to be monitored. In the example of FIG. 7, the probability density distribution (PDF) shown in FIG. 6B is used as the first feature quantity.

[0058] Here, depending on important radios for monitoring radio wave emission status or the wireless communication system to be monitored, radio waves may be transmitted on demand, such as in the case of local 5G or wireless LAN (Local Area Network). In this case, as the "first feature quantity in the normal state" (where there is no radio wave interference or failure), feature quantities in various forms as shown in FIGS. 7(a) to 7(d) are extracted.

[0059] Specifically, as the "first feature quantity in the normal state", feature quantities in various forms such as (a) no transmission (no transmission signal is included in the received data for all periods of the received data in the extraction period T1), (b) transmission only for a part of the time (no transmission signal is included in the received data for a part of the periods of the received data in the extraction period T1), (c) continuous transmission (a transmission signal with a low reception level is included in the received data for all periods of the received data in the extraction period T1), and (d) continuous transmission (a transmission signal with a high reception level is included in the received data for all periods of the received data in the extraction period T1) are extracted.

[0060] For example, during the learning process, when the learning unit 60 performs machine learning using only four first feature amounts as shown in FIGS. 7(a) to 7(d) to generate a learned model 70, during the determination process, the determination unit 80 uses the learned model 70 to determine that it is normal for the first feature amounts similar to the above-mentioned four first feature amounts, and determines that it is abnormal for the first feature amounts deviating from the above-mentioned four first feature amounts.

[0061] That is, during the learning process, when machine learning is performed using only four first feature amounts as shown in FIGS. 7(a) to 7(d), during the determination process, the determination unit 80 determines "normal" as expected for the first feature amounts similar to the above-mentioned four first feature amounts in the normal state.

[0062] Next, the first feature amounts extracted when interference occurs will be described. As the "first feature amounts when interference is occurring", for example, the feature amounts when a large power interference continuously occurs as shown in FIG. 7(e), or the feature amounts when a small power interference continuously occurs as shown in FIG. 7(f) are extracted. In FIGS. 7(e) and 7(f), since interference continuously occurs, apparently, there is no reception level without transmission.

[0063] For example, for the first feature amounts when a large power interference continuously occurs as shown in FIG. 7(e), since they are different from any of the first feature amounts shown in FIGS. 7(a) to 7(d) etc., the determination unit 80 determines "abnormal" as expected. On the other hand, for the first feature amounts when a small power interference continuously occurs as shown in FIG. 7(f), since they are almost the same as the first feature amount shown in FIG. 7(c) among the first feature amounts shown in FIGS. 7(a) to 7(d) etc., the determination unit 80 may determine "normal" contrary to expectations.

[0064] Therefore, in the present embodiment, by performing the learning process and the determination process using not only the first feature amounts which are short-term feature amounts but also the second feature amounts which are long-term feature amounts, the abnormal detection accuracy by the determination unit 80 is improved.

[0065] FIG. 8 is a diagram for explaining the second feature amount extracted by the radio wave abnormality detection system 100 and a method for detecting radio wave abnormality using the same. In FIG. 8, examples of the second feature amount at an arbitrary frequency, which are re-extracted every extraction period T2 (for example, 5 seconds) determined by the extraction period determination unit 40, are shown. In the example of FIG. 8, the probability density distribution (PDF) shown in FIG. 6B is used as the second feature amount.

[0066] The re-feature quantification unit 30 groups a plurality of first feature amounts extracted by the feature amount extraction unit 20 every extraction period T1 (for example, 5 ms) into groups every extraction period T2 (for example, 5 seconds), and performs a process of adding frequencies and probabilities for each reception level for the plurality of first feature amounts grouped every extraction period T2 (step S31). Then, the re-feature quantification unit 30 outputs the result of the addition process as the second feature amount every extraction period T2 (step S32) (step S33).

[0067] Here, a wireless communication system that transmits radio waves on demand does not continuously output radio waves all the time. Therefore, the extraction period determination unit 40 calculates an extraction period that exceeds the maximum continuous transmission time (the time during which the signal level is maintained) in the normal state and sets it as the extraction period T2. As a result, in the re-feature quantification unit 30, a feature amount having a form as shown in (g) of FIG. 8 is extracted as the "second feature amount in the normal state" (without radio wave interference or obstacles). That is, as the "second feature amount in the normal state", for example, a feature amount in which a reception level without transmission always occurs (in other words, a feature amount in which there is a period in the extraction period T2 during which the received data does not contain a transmission signal) as shown in (g) of FIG. 8 is extracted. This is because the second feature amount as shown in (g) of FIG. 8 is the result of cumulative addition so as to include the first feature amount as shown in (a) or (b) of FIG. 7.

[0068] After that, in the learning unit 60, not only the first feature amount but also the second feature amount is learned together (for example, made into a higher dimension or concatenated), so that for the dimension of the second feature amount, as shown in Fig. 8(g), "a feature amount in which a reception level without transmission always occurs (in other words, a feature amount in which there is a period in the extraction period T2 during which the received data does not contain a transmission signal)" is used as the second feature amount in the normal state, and a learned model 70 is constructed.

[0069] Here, when received data in which small power interference continuously occurs is acquired, as the second feature amount, for example, as shown in Fig. 8(h), "a feature amount in which a reception level without transmission does not occur (in other words, a feature amount in which there is no period in the extraction period T2 during which the received data does not contain a transmission signal)" is extracted. As a result, for the second feature amount extracted from the received data in which small power interference continuously occurs, as shown in Fig. 8(h), regarding the presence or absence of the occurrence of the reception level without transmission, it is different from the second feature amount shown in Fig. 8(g), so the determination unit 80 determines "abnormal" as expected. That is, for received data that would be misjudged as "normal" in the determination process using only the first feature amount, the determination unit 80 can perform the determination process using the first feature amount and the second feature amount and determine "abnormal" as expected.

[0070] Finally, the details of the method for determining the extraction period T2 by the extraction period determination unit 40 will be described. Fig. 9 is a diagram for explaining the method for determining the long-term extraction period T2 by the extraction period determination unit 40.

[0071] In the present embodiment, as described above, an extraction period exceeding the maximum continuous transmission time in the normal state is set for the extraction period T2, and the determination process is performed using the second feature amount extracted for each extraction period T2, aiming to be able to detect continuous interference. Here, it is desirable that the extraction period T2 be as short as possible within the long period in which continuous interference can be detected.

[0072] First, in the extraction period determination unit 40, the signal level estimation unit 46 estimates the boundary between the "no transmission" reception level, which is the reception level not including the transmission signal among the reception levels, and the "transmission" reception level, which is the reception level including the transmission signal. Then, the signal level estimation unit 46 estimates the reception levels above the boundary among the reception levels as the reception levels (signal levels) including the transmission signal (step S21). Note that the extraction period determination unit 40 may receive and set the information on the signal level estimated by the signal level estimation function provided outside instead of including the signal level estimation unit 46.

[0073] More specifically, first, as shown in FIGS. 9(a) and 9(b), using each first feature amount extracted from the reception data collected for several seconds to several tens of seconds or more for learning, the distribution of the reception level is created for each arbitrary frequency. Then, the boundary between the noise level and the signal level is estimated from the distribution of the reception level. In the estimation of the boundary, for example, assuming that only the peak (cluster) of the distribution on the lowest reception level side among several peaks (clusters) of the distribution is the noise level, the point where the first minimum value (concave downward) is reached when viewed from the lower reception level side is calculated as the boundary. Note that as the minimum value, it may be the point where the occurrence frequency changes from monotonically decreasing to monotonically increasing when viewed from the lower reception level side, or the point where the slope changes from negative to positive after differentiating once. Alternatively, the entire distribution may be clustered, and the boundary may be estimated between the distribution cluster with the lowest reception level and the next distribution cluster with the lower reception level. Then, the reception levels above the estimated boundary among the reception levels are estimated as the reception levels (signal levels) including the transmission signal.

[0074] Next, in the extraction period determination unit 40, the maximum continuous transmission time estimation unit 47 extracts the transmission continuous time of each transmission signal in the normal state (step S22). For example, as shown in FIG. 9(c), the maximum continuous transmission time estimation unit 47 uses a plurality of first feature amounts from several seconds to several tens of seconds or more for each arbitrary frequency, and from the first feature amount where the reception level of "no transmission" occurs to the first feature amount where only the reception level of "transmission" occurs, and then until it switches back to the first feature amount where the reception level of "no transmission" occurs again, the time (extraction period T1×n (n is an integer of 0 or more)) is extracted and frequency-distributed. In other words, the maximum continuous transmission time estimation unit 47 extracts and frequency-distributes the time from the first feature amount having an untransmitted time band (distribution of reception levels at a noise level lower than the signal level) to the first feature amount having no untransmitted time band, and then until it switches back to the first feature amount having an untransmitted time band again. More specifically, the maximum continuous transmission time estimation unit 47 extracts and frequency-distributes the time from when the received data starts to include a transmission signal to when it no longer includes a transmission signal for each arbitrary frequency. Then, the maximum continuous transmission time estimation unit 47 estimates the maximum continuous transmission time based on the frequency distribution of the transmission continuous period (step S23).

[0075] For example, as shown in FIG. 9(d), the maximum value on the frequency distribution may be directly estimated as the maximum continuous transmission time, or a value obtained by adding a margin to the maximum value on the frequency distribution may be estimated as the maximum continuous transmission time. The margin here is, for example, a value of 5 to 10% of the maximum value on the frequency distribution. Alternatively, the maximum continuous transmission time may be estimated statistically or probabilistically from the frequency distribution. For example, assuming that the frequency distribution follows a Gaussian distribution or a normal distribution, after calculating its mean and variance (regression analysis), the maximum continuous transmission time may be estimated based on points such as a probability of 95.5% (2σ), 99.7% (3σ), or 99.99% (4σ). For example, if this estimated value is larger than the maximum value on the frequency distribution, the estimated value is adopted as the maximum continuous transmission time.

[0076] As described above, the radio wave anomaly detection system 100(101) according to this embodiment uses both the first feature amount extracted by the feature amount extraction unit 20 every short extraction period T1 and the second feature amount extracted by the re-feature quantification unit 30 every long extraction period T2 to perform learning and determination in the detection processing unit 50. Then, in order for the radio wave anomaly detection system 100(101) according to this embodiment to detect interference (anomaly) from received data in which small interference in power continuously occurs, the extraction period determination unit 40 estimates the maximum continuous transmission time in the normal state and determines a long extraction period T2 according to the maximum continuous transmission time. Thereby, the radio wave anomaly detection system 100 according to this embodiment can also perform learning and determination using not only the first feature amount but also the second feature amount for "received data when small interference in power continuously occurs" that cannot be determined as "abnormal" in the determination process using only the first feature amount, which is a short-term feature amount, and can be determined as "abnormal" as expected.

[0077] That is, the radio wave anomaly detection system 100(101) according to this embodiment can also detect radio wave interference and obstacles having characteristics with a period different from the extraction period T1 (for example, a period longer than the extraction period T1) that cannot be detected as abnormal only by the first feature amount extracted every short extraction period T1. In addition, since the radio wave anomaly detection system 100(101) according to this embodiment performs a determination process on whether the received data contains a radio wave anomaly using a learned model generated by unsupervised learning, for example, it can also detect unknown radio wave interference and obstacles with a low reception level and not included in the teacher data.

[0078] In the determination process using only the second feature quantity, which is a long-term feature quantity, there is a possibility that radio wave interference or obstacles that occur instantaneously, as shown in Fig. 7(e), cannot be detected. The reason is that the first feature quantity extracted every short extraction period T1, such as 5 ms, from the received data including radio wave interference or obstacles that occur instantaneously, has a peculiar form that does not occur in the normal state as shown in Fig. 7(e). However, the second feature quantity extracted every long extraction period T2, such as 5 seconds, has a feature quantity in which the influence of instantaneously occurring radio wave interference or the like is averaged.

[0079] Therefore, the radio wave abnormality detection system 100(101) according to the present embodiment performs learning and determination using both the first feature quantity, which is a short-term feature quantity considering the radio standards and multiplexing period of the wireless communication system, and the second feature quantity, which is a long-term feature quantity considering the maximum continuous transmission time. As a result, for any of the received data when small interference in power continuously occurs and the received data when radio wave interference or obstacles occur instantaneously, it can be determined as "abnormal" as expected (that is, abnormalities can be detected with high accuracy).

[0080] In the present embodiment, the case where the radio wave abnormality detection system 100(101) has both a learning processing function and a determination processing function has been described as an example. However, the present invention is not limited to this, and it may have only a determination processing function. This will be briefly described below with reference to Fig. 10.

[0081] FIG. 10 is a block diagram showing a modified example of the radio wave abnormality detection system 100 as the radio wave abnormality detection system 102. The radio wave abnormality detection system 102 has no learning processing function and includes only a determination processing function as compared with the radio wave abnormality detection system 100. Since other configurations of the radio wave abnormality detection system 102 are the same as those of the radio wave abnormality detection system 100, the description thereof is omitted. Note that the radio wave abnormality detection system 102 realizes the determination processing function using a pre-prepared learned model 70. The radio wave abnormality detection system 102 may include an extraction period determination unit 40 as shown in FIG. 10, or may acquire information on the extraction period T2 determined by an extraction period determination unit 40 provided separately, etc., and extract the second feature amount in the re-feature quantification unit 30. FIG. 11 is a block diagram showing a modified example of the radio wave abnormality detection system 102 as the radio wave abnormality detection system 102a. The radio wave abnormality detection system 102a does not include the extraction period determination unit 40 as compared with the radio wave abnormality detection system 102, and acquires information on the extraction period T2 determined by an extraction period determination unit 40 (not shown) provided separately, etc., and extracts the second feature amount in the re-feature quantification unit 30.

[0082] Also, in the present embodiment, the case where the re-feature quantification unit 30 re-feature quantifies using a plurality of first feature amounts extracted by the feature amount extraction unit 20 to generate a second feature amount has been described as an example, but it is not limited thereto. If the re-feature quantification unit 30 can generate a similar second feature amount, the second feature amount may be directly extracted from the received data. Further, in the present embodiment, the case of detecting that a small interference in power is continuously occurring using the second feature amount has been described as an example, but of course, regardless of the magnitude of the power, it is possible to detect that the interference is continuously occurring.

[0083] <Embodiment 2> FIG. 12 is a block diagram showing a configuration example of the radio wave abnormality detection system 103 according to Embodiment 2. In the radio wave abnormality detection system 103, the detection processing unit has a function of learning and determining using a first feature amount which is a short-term feature amount, and a function of learning and determining using a second feature amount which is a long-term feature amount, respectively. Note that, in the radio wave abnormality detection system 103, part of each of the learning process and the determination process is realized using a common processing circuit, but the learning process and the determination process may be realized using separate processing circuits, respectively.

[0084] The radio wave abnormality detection system 103 includes, as an example, a feature amount extraction unit 20 that extracts a first feature amount which is a short-term feature amount, a re-feature quantification unit (second feature amount extraction unit) 30 that performs re-feature quantification using a plurality of first feature amounts extracted by the feature amount extraction unit 20 and extracts a second feature amount which is a long-term feature amount, an extraction period determination unit 40 that determines an extraction period of the second feature amount by the re-feature quantification unit 30, and a detection processing unit (abnormality detection unit) 52 that performs learning of a model representing the presence or absence of radio wave abnormality in the received data and determination of the presence or absence of radio wave abnormality in the received data using the first feature amount and the second feature amount.

[0085] Here, since the configurations of the feature amount extraction unit 20, the re-feature quantification unit 30, and the extraction period determination unit 40 are the same as those of the feature amount extraction unit 20, the re-feature quantification unit 30, and the extraction period determination unit 40 provided in the radio wave abnormality detection system 100, the description thereof is omitted. Hereinafter, mainly, the detection processing unit 52 will be described.

[0086] The detection processing unit 52 includes a learning unit 61, a learning unit 62, a database 71, a database 72, a determination unit 81, and a determination unit 82.

[0087] During the learning process, the learning unit 61 performs machine learning on a model that indicates whether the received data contains radio wave anomalies, using a plurality of first feature amounts extracted by the feature amount extraction unit 20. The learning unit 62 performs machine learning on a model that indicates whether the received data contains radio wave anomalies, using a plurality of second feature amounts extracted by the re-feature quantification unit 30 during the learning process. The database 71 stores a learned model (hereinafter also referred to as the learned model 71) generated by the machine learning of the learning unit 61 and the like. The database 72 stores a learned model (hereinafter also referred to as the learned model 72) generated by the machine learning of the learning unit 62 and the like.

[0088] During the determination process, the determination unit 81 determines whether the received data contains radio wave anomalies, using the first feature amount extracted from the received data and the learned model 71. During the determination process, the determination unit 82 determines whether the received data contains radio wave anomalies, using the second feature amount extracted from the received data and the learned model 72. The determination results of the determination units 81 and 82 are output from the detection processing unit 52 separately as the first and second determination results, respectively.

[0089] Note that the feature amounts are assumed to be statistical feature amounts extracted at predetermined intervals, such as a frequency distribution (histogram), a probability density function (PDF), a cumulative distribution function (CDF), and an amplitude probability distribution (APD). Note that the feature amounts may be directly extracted from the received data on the time axis, or may be extracted for each frequency after being converted into data on the frequency axis by spectralization or spectrogramming.

[0090] (Operation of the radio wave anomaly detection system 103) Subsequently, the operation of the radio wave anomaly detection system 103 according to the second embodiment will be described. Note that the overall processing flow by the radio wave anomaly detection system 103 is the same as the overall processing flow by the radio wave anomaly detection system 100 shown in FIG. 3.

[0091] (Details of the learning process by the radio wave anomaly detection system 103) FIG. 13A to FIG. 13C are flowcharts showing the flow of learning processing by the radio wave abnormality detection system 103 according to Embodiment 2.

[0092] The learning processing shown in FIGS. 13A to 13C is assumed to be performed by collecting reception data in a normal state (normal time) as sample data. Note that, among the learning processing, the processing of extracting the first feature amount by the feature amount extraction unit 20, the processing of determining the extraction period T2 of the second feature amount by the extraction period determination unit 40, and the processing of extracting the second feature amount by the re-feature quantification unit 30 are the same as those in the case of the radio wave abnormality detection system 100, and thus the description thereof is omitted.

[0093] Hereinafter, among the learning processing, the characteristic machine learning processing (steps S41a and S41b) in the radio wave abnormality detection system 103 will be described.

[0094] First, the learning unit 61 performs machine learning of a model representing whether or not the received data includes radio wave abnormalities (radio wave interference or obstacles) using a feature vector representing the first feature amount extracted for each short-term extraction period T1 in a high dimension (multiple dimensions) (step S41a). Further, the learning unit 62 performs machine learning of a model representing whether or not the received data includes radio wave abnormalities (radio wave interference or obstacles) using a feature vector representing the second feature amount extracted for each long-term extraction period T2 in a high dimension (multiple dimensions) (step S41b).

[0095] In this embodiment, it is assumed that the machine learning employed in the learning units 61 and 62 is unsupervised learning in order to detect unknown interference and obstacles that are difficult to define in advance as teacher data. As an example, it is assumed that it is One-class SVM. However, the machine learning employed in the learning units 61 and 62 is not limited to One-class SVM. For example, it may be machine learning of a model such as an anomaly detection model (Deep Anomaly Detection) using deep learning or the like. Alternatively, the machine learning employed in the learning units 61 and 62 may be a method of learning a threshold for statistically detecting anomalies using Mahalanobis distance, variance (standard deviation), etc. from the distribution of high-dimensional (multidimensional) feature vectors as described above, or a method of predicting a change trend using a state filter such as a Kalman filter or a particle filter and detecting whether it deviates from the change trend. Furthermore, the machine learning employed in the learning units 61 and 62 is not limited to unsupervised learning and may be supervised learning.

[0096] The learned models generated by the machine learning of the learning units 61 and 62, the thresholds for anomaly detection, the coefficients indicating the change trends, etc. are stored in the databases 71 and 72, respectively.

[0097] (Details of the determination process by the radio wave anomaly detection system 103) FIG. 14 is a flowchart showing the flow of the determination process by the radio wave anomaly detection system 103 according to Embodiment 2.

[0098] The determination process shown in FIG. 14 is a process of acquiring reception data from the radio wave environment to be monitored during actual operation and sequentially performing detection determination of radio wave interference and obstacles. Note that, among the determination processes, the process of extracting the first feature amount by the feature amount extraction unit 20 and the process of extracting the second feature amount by the re-feature quantization unit 30 are the same as those in the case of the radio wave anomaly detection system 100, and thus the description thereof is omitted.

[0099] Hereinafter, among the determination processes, the detection determination processes (steps S68a and S68b), which are characteristic in the radio wave anomaly detection system 103, will be described.

[0100] First, the determination unit 81 performs a first detection determination process using the first feature amount sequentially extracted from the received data and the learned model 71 generated by the machine learning of the learning unit 61 during the learning process (step S68a).

[0101] For example, when the learned model 71 is generated by machine learning such as One-class SVM which is a type of unsupervised learning, when a feature vector representing the first feature amount in a high-dimensional (multi-dimensional) manner is input, the determination unit 81 converts the positional relationship and distance of the feature vector with respect to the discrimination boundary configured in the learned model 71 into normality (or abnormality) and outputs it. Then, the determination unit 81 outputs a first determination result as to whether the received data includes a radio wave abnormality based on the calculation result.

[0102] Also, the determination unit 82 performs a second detection determination process using the second feature amount extracted for each extraction period T2 and the learned model 72 generated by the machine learning of the learning unit 62 during the learning process (step S68a).

[0103] For example, when the learned model 72 is generated by machine learning such as One-class SVM which is a type of unsupervised learning, when a feature vector representing the second feature amount in a high-dimensional (multi-dimensional) manner is input, the determination unit 82 converts the positional relationship and distance of the feature vector with respect to the discrimination boundary configured in the learned model 72 into normality (or abnormality) and outputs it. Then, the determination unit 82 outputs a second determination result as to whether the received data includes a radio wave abnormality based on the calculation result.

[0104] In addition, when performing statistical anomaly detection using the Mahalanobis distance, variance (standard deviation), etc., each determination unit 81, 82 may calculate the distance of the feature vector from the learned threshold value learned during the learning process and output the distance as a determination result. Alternatively, when predicting a change trend using a state filter such as a Kalman filter or a particle filter, each determination unit 81, 82 may calculate the degree of deviation of the feature vector from the coefficient indicating the change trend and output it as a determination result.

[0105] As described above, the radio wave anomaly detection system 103 according to the present embodiment can achieve an effect equivalent to that of the radio wave anomaly detection system 100. Further, the radio wave anomaly detection system 103 separately performs a process of learning and determining using the first feature amount extracted every short extraction period T1, and a process of learning and determining using the second feature amount extracted every long extraction period T2. Thereby, the radio wave anomaly detection system 103 can separately output a first determination result obtained by performing a determination process using the first feature amount which is a short-term feature amount, and a second determination result obtained by performing a determination process using the second feature amount which is a long-term feature amount. Therefore, for example, the user can more accurately specify the cause and state of radio wave interference and obstacles from the combination of the first and second determination results.

[0106] Specifically, for example, when a short extraction period such as 5 ms is set as the extraction period T1 based on the radio standards and multiplexing methods of important wireless signals and wireless communication systems to be monitored, as the first determination result, when interference or obstacles with large instantaneous fluctuations occur on the order of the extraction period T1, a determination result indicating "abnormal" is output. On the other hand, when a long extraction period such as 5 seconds is set as the extraction period T2 based on the estimated maximum continuous transmission time, as the second determination result, when interference or obstacles continuously occur during the extraction period T2, a determination result indicating "abnormal" is output. The user can easily estimate the magnitude of the fluctuation of the interference or obstacle, whether the interference or obstacle is instantaneous or continuous, etc. from the combination of these determination results.

[0107] In addition, in this embodiment, the radio wave abnormality detection system 103 has been described by taking the case where it has both a learning processing function and a determination processing function as an example. However, the present invention is not limited to this, and as shown in the configuration of FIG. 10, it may have only a determination processing function.

[0108] Further, in this embodiment, the case where the re-feature quantification unit 30 generates a second feature amount by re-feature quantifying using a plurality of first feature amounts extracted by the feature amount extraction unit 20 has been described as an example. However, the present invention is not limited to this. If the re-feature quantification unit 30 can generate a similar second feature amount, it may directly extract the second feature amount from the received data. Furthermore, in this embodiment, the case where it is detected that interference with low power continuously occurs using the second feature amount has been described as an example. However, of course, regardless of the magnitude of the power, it is possible to detect that interference continuously occurs.

[0109] <Embodiment 3> FIG. 15 is a block diagram showing a configuration example of a radio wave abnormality detection system 104 according to Embodiment 3. In the radio wave abnormality detection system 104, the detection processing unit does not have a function of learning a model using a second feature amount which is a long-term feature amount. Further, in the radio wave abnormality detection system 104, instead of the function of determining whether or not the received data includes a radio wave abnormality using the second feature amount, by comparing the maximum continuous transmission time estimated during the learning process with the transmission continuous time of the transmission signal included in the received data, it has a function of determining whether or not the received data includes a radio wave abnormality.

[0110] Note that in the radio wave abnormality detection system 104, although a part of each of the learning process and the determination process is realized using a common processing circuit, the learning process and the determination process may be realized using separate processing circuits.

[0111] The radio wave abnormality detection system 103 includes, as an example, a feature quantity extraction unit 20 that extracts a first feature quantity which is a short-term feature quantity, a re-feature quantification unit (second feature quantity extraction unit) 30 that performs re-feature quantification using a plurality of first feature quantities extracted by the feature quantity extraction unit 20 and extracts a second feature quantity which is a long-term feature quantity, an extraction period determination unit 40 that determines the extraction period of the second feature quantity by the re-feature quantification unit 30, and a detection processing unit (abnormality detection unit) 53 that performs learning of a model representing the presence or absence of radio wave abnormality in received data and determination of the presence or absence of radio wave abnormality in the received data, and includes at least these components.

[0112] Here, regarding the configurations of the feature quantity extraction unit 20, the re-feature quantification unit 30, and the extraction period determination unit 40, since they are the same as the feature quantity extraction unit 20, the re-feature quantification unit 30, and the extraction period determination unit 40 provided in the radio wave abnormality detection system 100, the description thereof is omitted. However, during the learning process, the maximum continuous transmission time (extraction period T2) estimated by the extraction period determination unit 40 is used by the re-feature quantification unit 30 and the transmission continuous time determination unit 83 described later during the determination process. Hereinafter, mainly the detection processing unit 53 will be described.

[0113] The detection processing unit 53 includes a learning unit 61, a database 71, a determination unit 81, and a transmission continuous time determination unit 83.

[0114] During the learning process, the learning unit 61 performs machine learning of a model representing whether the received data contains radio wave abnormality using a plurality of first feature quantities extracted by the feature quantity extraction unit 20. The database 71 stores a learned model (hereinafter, also referred to as the learned model 71) generated by the machine learning of the learning unit 61 and the like. During the determination process, the determination unit 81 determines whether the received data contains radio wave abnormality using the first and second feature quantities extracted from the received data and the learned model 71. The determination result of the determination unit 81 is output from the detection processing unit 53 as a first determination result.

[0115] The transmission duration determination unit 83 compares and determines the maximum continuous transmission time (extraction period T2) estimated by the extraction period determination unit 40 and the transmission duration of each transmission signal obtained from a plurality of second feature amounts extracted by the re-feature quantification unit 30. The determination result of the transmission duration determination unit 83 is output from the detection processing unit 53 as a second determination result separately from the first determination result.

[0116] (Operation of the radio wave abnormality detection system 104) Next, the operation of the radio wave abnormality detection system 104 according to Embodiment 3 will be described. FIG. 16 is a flowchart showing an outline of the overall processing flow by the radio wave abnormality detection system 104. Note that the overall processing flow by the radio wave abnormality detection system 104 is mostly the same as the overall processing flow by the radio wave abnormality detection system 100 shown in FIG. 3. However, during the learning process, the process of extracting the second feature amount by the re-feature quantification unit 30 (steps S31 to S33) may be omitted.

[0117] Specifically, the processing by the radio wave abnormality detection system 100 is divided into a learning process step performed in advance or periodically and a determination process step performed sequentially during operation. The learning process mainly includes a process of extracting the first feature amount, which is a short-term feature amount (steps S11 to S14), a process of determining the extraction period for extracting the second feature amount, which is a long-term feature amount (steps S21 to S23), and a learning process (step S41a). The determination process mainly includes a process of sequentially extracting the first feature amount and the second feature amount (steps S61 to S66, S67c) and a process of determining whether the received data includes a radio wave abnormality (steps S68a, S68c).

[0118] (Details of the learning process by the radio wave abnormality detection system 104) FIGS. 17A to 17C are flowcharts showing the learning process flow by the radio wave abnormality detection system 104 according to Embodiment 3.

[0119] The learning process shown in FIGS. 17A to 17C is assumed to collect the received data in the normal state (normal time) as sample data. Among the learning processes, the process of extracting the first feature amount by the feature amount extraction unit 20 and the process of determining the extraction period T2 of the second feature amount by the extraction period determination unit 40 are the same as those in the radio wave abnormality detection system 100, and thus the description thereof is omitted. However, during the learning process, the signal level and the maximum continuous transmission time (extraction period T2) estimated by the extraction period determination unit 40 are used by the re-feature quantification unit 30 and the transmission continuous time determination unit 83 described later during the determination process.

[0120] Hereinafter, among the learning processes, the machine learning process (step S41a), which is characteristic in the radio wave abnormality detection system 103, will be described.

[0121] First, the learning unit 61 performs machine learning of a model that indicates whether the received data includes radio wave abnormalities (radio wave interference or obstacles) using a feature vector that represents the first feature amount extracted for each short-term extraction period T1 in a high dimension (multiple dimensions) (step S41a).

[0122] In the present embodiment, it is assumed that the machine learning adopted by the learning unit 61 is unsupervised learning in order to detect unknown interference and obstacles that are difficult to define in advance as teacher data. As an example, it is assumed that the machine learning is One-class SVM. However, the machine learning adopted by the learning unit 61 is not limited to One-class SVM, and for example, machine learning of a model such as an anomaly detection model (Deep Anomaly Detection) using deep learning or the like may be used. Alternatively, the machine learning adopted by the learning unit 61 is not limited to unsupervised learning, and supervised learning may also be used. Methods such as learning a threshold value for statistically detecting anomalies using Mahalanobis distance or variance (standard deviation) from the distribution of high-dimensional (multiple-dimensional) feature vectors as described above, or predicting a change trend using a state filter such as a Kalman filter or a particle filter and detecting whether the change trend deviates may also be used without any problem. Furthermore, the machine learning adopted by the learning unit 61 is not limited to unsupervised learning and may be supervised learning.

[0123] The learned model generated by the machine learning of the learning unit 61, the threshold value for anomaly detection, the coefficient indicating the change trend, etc. are stored in the database 71.

[0124] In this embodiment, during the learning process, the process of performing machine learning using the second feature amount is omitted. In this embodiment, during the determination process, instead of the learned model obtained by performing machine learning using the second feature amount, the signal level and the maximum continuous transmission time estimated by the extraction period determination unit 40 are used.

[0125] (Details of the determination process by the radio wave anomaly detection system 104) FIG. 18 is a flowchart showing the flow of the determination process by the radio wave anomaly detection system 104 according to Embodiment 3.

[0126] The determination process shown in FIG. 18 is a process of acquiring reception data from the radio wave environment to be monitored during actual operation and sequentially performing detection determination of radio wave interference and failures. Note that, among the determination processes, the process of extracting the first feature amount by the feature amount extraction unit 20 and the process of extracting the second feature amount by the re-feature quantification unit 30 are the same as those in the case of the radio wave anomaly detection system 100, and thus the description thereof is omitted.

[0127] However, the second feature amount extracted by the re-feature quantification unit 30 only needs to include information necessary for estimating the transmission duration of the transmission signal by the transmission duration determination unit 83. Therefore, for example, as in the case of Embodiment 1, various statistical feature amounts as shown in FIGS. 6A to 6D may be used, or only the information on the presence or absence of the non-transmission time zone (distribution of reception levels at noise levels lower than the signal level) for each extraction period T2 may be included. Alternatively, the second feature amount may directly include the information of the first feature amount.

[0128] Hereinafter, among the determination processes, the detection determination processes (steps S68a and S68c), which are characteristic of the radio wave anomaly detection system 104, will be described.

[0129] First, the determination unit 81 performs a first detection determination process using the first feature amount sequentially extracted from the received data and the learned model 71 generated by the machine learning of the learning unit 61 during the learning process (step S68a).

[0130] For example, when the learned model 71 is generated by machine learning such as One-class SVM which is a kind of unsupervised learning, when a feature vector representing the first feature amount in a high-dimensional (multi-dimensional) manner is input, the determination unit 81 converts the positional relationship and the distance of the feature vector with respect to the discrimination boundary configured in the learned model 71 into a normality (or abnormality degree) and outputs it. Then, the determination unit 81 outputs a first determination result as to whether the received data includes a radio wave abnormality based on the calculation result.

[0131] When statistically detecting an abnormality using the Mahalanobis distance, variance (standard deviation), etc., the determination unit 81 may calculate the distance of the feature vector from the learned threshold value learned during the learning process and output the distance as the determination result. Alternatively, when predicting a change trend using a state filter such as a Kalman filter or a particle filter, the determination unit 81 may calculate the deviation degree of the feature vector from the coefficient indicating the change trend and output it as the determination result.

[0132] Also, the transmission duration determination unit 83 performs a determination process of the transmission duration of each transmission signal included in the received data based on the second feature amount extracted for each long-term extraction period T2 and the information on the signal level and the maximum continuous transmission time estimated by the extraction period determination unit 40 during the learning process (step S68C).

[0133] Here, when the re-feature quantification unit 30 extracts the second feature amount every long extraction period T2 as in the case of the first embodiment, the transmission duration determination unit 83 only needs to detect the presence or absence of an untransmitted time band (distribution of reception levels of noise levels less than the signal level) among the second feature amounts in the extraction period T2. For example, when there is an untransmitted time band, the transmission duration determination unit 83 outputs a determination result indicating "normal" (a determination result indicating that there is no radio wave abnormality in the received data), and when there is no untransmitted time band, it outputs a determination result indicating "abnormal" (a determination result indicating that there is a radio wave abnormality in the received data).

[0134] Alternatively, the transmission duration determination unit 83 uses at least one of the first feature amount and the second feature amount to extract the time from when the feature amount with an untransmitted time band switches to the feature amount without an untransmitted time band until it switches back to the feature amount with an untransmitted time band as the transmission duration of the transmission signal, and compares the transmission duration of the transmission signal with the maximum continuous transmission time estimated by the extraction period determination unit 40, and outputs it as a determination result. For example, when the transmission duration of the transmission signal does not exceed the maximum continuous transmission time, the transmission duration determination unit 83 outputs a determination result indicating "normal", and when it exceeds the maximum continuous transmission time, it outputs a determination result indicating "abnormal". Note that in the re-feature quantification unit 30, when the detection determination process for the presence or absence of an untransmitted time band has already been completed and only the determination result is input to the transmission duration determination unit 83, the transmission duration determination unit 83 may output a determination result indicating "normal" when there is an untransmitted time band, for example, and a determination result indicating "abnormal" when there is no untransmitted time band based on the determination result.

[0135] The determination result by the transmission duration determination unit 83 is output from the detection processing unit 53 as a second determination result separately from the first determination result by the first determination unit 81.

[0136] Thus, the radio wave anomaly detection system 104 according to this embodiment can achieve an effect equivalent to that of the radio wave anomaly detection system 100. In addition, the radio wave anomaly detection system 104 performs a process of learning and determination using the first feature amount extracted every short extraction period T1, and for the second feature amount extracted every long extraction period T2, without requiring learning, the maximum continuous transmission time determined during the learning process and the transmission continuous time of the transmission signal obtained from the second feature amount are compared to determine whether the received data contains a radio wave anomaly. Thereby, similar to the case of the radio wave anomaly detection system 103, the radio wave anomaly detection system 104 can separately output the first determination result obtained by performing the determination process using the first feature amount, which is a short-term feature amount, and the second determination result obtained by performing the determination process using the second feature amount, which is a long-term feature amount. Therefore, for example, the user can more accurately identify the causes and states of radio wave interference and obstacles from the combination of the first and second determination results.

[0137] Furthermore, since the radio wave anomaly detection system 104 according to this embodiment does not need to perform machine learning using the second feature amount, which is a long-term feature amount, the entire system can be simplified and made more efficient compared to the case of the radio wave anomaly detection system 103. This leads to, for example, an effect of reducing the circuit scale and learning time. Also, for the second determination process using the second feature amount, which is a long-term feature amount, when it is determined as an anomaly, the cause of the anomaly can be determined as "anomaly due to the transmission continuous time". However, in the radio wave anomaly detection system 103 according to Embodiment 2, for the interference and obstacles that could be detected using the second feature amount in the long extraction period T2, there is a possibility that they cannot be detected in the radio wave anomaly detection system 104 according to Embodiment 3. Therefore, it is considered that there is a trade-off between detection performance and efficiency improvement.

[0138] In addition, in this embodiment, the case where the radio wave anomaly detection system 104 has both a learning process function and a determination process function has been described as an example, but it is not limited to this. As shown in the configuration of FIG. 10, it may have only a determination process function.

[0139] In addition, in this embodiment, a case where the re-feature quantification unit 30 generates a second feature amount by performing re-feature quantification using a plurality of first feature amounts extracted by the feature amount extraction unit 20 has been described as an example. However, the present invention is not limited to this. If the re-feature quantification unit 30 can generate a similar second feature amount, the second feature amount may be directly extracted from the received data. Furthermore, in this embodiment, a case where it is detected that a small interference in power is continuously occurring using the second feature amount has been described as an example. However, of course, regardless of the magnitude of the power, it is possible to detect that an interference is continuously occurring.

[0140] As described above, the radio wave abnormality detection system according to the present disclosure can achieve the following effects.

[0141] First, the radio wave abnormality detection system according to the present disclosure uses both the first feature amount extracted by the feature amount extraction unit every short extraction period T1 and the second feature amount extracted by the re-feature quantification unit every long extraction period T2 to perform learning and determination in the detection processing unit. Here, the radio wave abnormality detection system according to the present disclosure estimates the maximum continuous transmission time in the normal state in the extraction period determination unit and determines a long extraction period T2 according to the maximum continuous transmission time. Thereby, the radio wave abnormality detection system according to the present disclosure can also perform learning and determination using not only the first feature amount but also the second feature amount for the "received data in the case where a small interference in power is continuously occurring" that cannot be determined as "abnormal" in the determination process using only the first feature amount, which is a short-term feature amount, and can be determined as "abnormal" as expected.

[0142] In addition, since the radio wave abnormality detection system according to the present disclosure performs a determination process on whether the received data includes a radio wave abnormality using a learned model generated by unsupervised learning, for example, it is also possible to detect unknown radio wave interference or obstacles such that the received level is low and not included in the teacher data.

[0143] In addition, the radio wave abnormality detection system according to the present disclosure separately performs a process of learning and determination using the first feature quantity extracted every short extraction period T1 and a process of learning and determination using the second feature quantity extracted every long extraction period T2. As a result, it is possible to separately output a first determination result obtained by performing a determination process using the first feature quantity, which is a short-term feature quantity, and a second determination result obtained by performing a determination process using the second feature quantity, which is a long-term feature quantity. For example, a user can more accurately identify the cause and state of radio wave interference or a failure from the combination of the first and second determination results.

[0144] For example, if both the first and second determination results are normal, the user can determine that the received data does not contain radio wave abnormalities. Also, if the first determination result is abnormal and the second determination result is normal, it can be estimated that "an interference or failure with a large instantaneous fluctuation has occurred". Further, if the first determination result is normal and the second determination result is abnormal, it can be estimated that "an interference or failure with a small fluctuation is continuously occurring". Furthermore, if both the first and second determination results are abnormal, it can be estimated that "an interference or failure with a large fluctuation is continuously occurring even instantaneously".

[0145] In addition, the radio wave abnormality detection system according to the present disclosure generates the second feature quantity by re-feature quantifying a plurality of first feature quantities, which are short-term feature quantities extracted by the feature quantity extraction unit, for each long extraction period in the re-feature quantification unit. As a result, compared with the case where the first and second feature quantity extraction units are provided in parallel to extract the first and second feature quantities, it is possible to share the process of spectrogramming (spectrum conversion) in the frequency axis direction and the process of extracting the reception level for each time and creating a frequency distribution. Therefore, the re-feature quantification unit can realize the extraction of the second feature quantity only by performing addition. Since the FFT process for spectrogramming (spectrum conversion, frequency conversion) and the process of creating a frequency distribution are more complex processes than addition, it is expected to improve the efficiency and simplify the system configuration by sharing the processes.

[0146] Furthermore, in the determination process using the second feature quantity extracted every long extraction period T2, in the radio wave anomaly detection system according to the present disclosure, instead of the determination process using the learned model, the maximum continuous transmission time determined during the learning process and the transmission continuous time of the transmission signal obtained from the second feature quantity are compared to determine whether the received data contains radio wave anomalies. As a result, a learning unit and a learned model for performing machine learning using the second feature quantity are not required, so the entire system can be simplified and made more efficient. This leads to, for example, a reduction effect in circuit scale and learning time.

[0147] As described above, the embodiments of the present disclosure have been described in detail with reference to the drawings. However, the specific configuration is not limited to the above, and various design changes and the like are possible without departing from the gist of the present disclosure. Also, the contents described in the embodiments can be used in combination.

[0148] In the above-described embodiment, the present disclosure has been described as a hardware configuration. However, the present disclosure is not limited to this. The present disclosure can also be realized by causing a Central Processing Unit (CPU) to execute a computer program for the control process by the radio wave anomaly detection system.

[0149] Also, when the above-described program is loaded into a computer, it includes a set of instructions (or software code) for causing the computer to perform one or more functions described in the embodiments. The program may be stored in a non-transitory computer-readable medium or a tangible storage medium. By way of example and not limitation, the computer-readable medium or tangible storage medium includes Random-Access Memory (RAM), Read-Only Memory (ROM), flash memory, Solid-State Drive (SSD), or other memory technologies, CD-ROM, Digital Versatile Disc (DVD), Blu-ray (registered trademark) disc, or other optical disk storage, magnetic cassette, magnetic tape, magnetic disk storage, or other magnetic storage devices. The program may be transmitted on a transitory computer-readable medium or a communication medium. By way of example and not limitation, the transitory computer-readable medium or communication medium includes electrical, optical, acoustic, or other forms of propagated signals.

Explanation of Signs

[0150] 20 Feature extraction unit 26 Spectrogram generation unit 27 Signal extraction unit 28 Feature quantization unit 30 Re-feature quantization unit 37 Feature addition unit 38 Feature quantization unit 40 Extraction period determination unit 46 Signal level estimation unit 47 Maximum continuous transmission time estimation unit 50 Detection processing unit 52 Detection processing unit 53 Detection processing unit 60 Learning unit 61 Learning unit 62 Learning unit 70 Database of learned models, etc. 71 Database of learned models, etc. 72 Database of learned models, etc. 80 Judgment unit 81 Determination Unit 82 Determination Unit 83 Transmission Duration Determination Unit 100 Radio Wave Abnormality Detection System 101 Radio Wave Abnormality Detection System 102 Radio Wave Abnormality Detection System 102a Radio Wave Abnormality Detection System 103 Radio Wave Abnormality Detection System 104 Radio Wave Abnormality Detection System

Claims

1. Output a detection result for a radio wave abnormality included in the received data, which is detected using a first feature amount that is a feature amount extracted from the received data every first predetermined period, and a second feature amount that is a feature amount extracted from the received data every second predetermined period longer than the first predetermined period. Radio wave abnormality detection system.

2. Output a detection result for a first radio wave abnormality that is a part of the radio wave abnormality included in the received data, which is detected using the first feature amount extracted from the received data every first predetermined period, and output a detection result for a second radio wave abnormality that is another part of the radio wave abnormality included in the received data, which is detected using the second feature amount extracted from the received data every second predetermined period. The radio wave abnormality detection system according to claim 1.

3. A first feature amount extraction unit that extracts the feature amount of the received data every first predetermined period as the first feature amount. A second feature amount extraction unit that extracts the feature amount of the received data every second predetermined period longer than the first predetermined period as the second feature amount. An abnormality detection unit that detects a radio wave abnormality included in the received data using the first feature amount extracted by the first feature amount extraction unit and the second feature amount extracted by the second feature amount extraction unit. Comprising The radio wave abnormality detection system according to claim 1.

4. The second feature amount extraction unit is configured to extract the second feature amount for each second predetermined period by adding up a plurality of the first feature amounts for each second predetermined period. The radio wave abnormality detection system according to claim 3.

5. Further comprising an extraction period determination unit that estimates the maximum continuous transmission time of a transmission signal included in the received data from a plurality of the first feature amounts extracted by the first feature amount extraction unit and determines the second predetermined period according to the estimated maximum continuous transmission time. The radio wave abnormality detection system according to claim 3 or 4.

6. The extraction period determination unit A signal level estimation unit that estimates the signal level of the transmission signal included in the received data from a plurality of the first feature amounts extracted by the first feature amount extraction unit. A maximum continuous transmission time estimation unit that estimates the maximum continuous transmission time, which is the maximum value of the time during which the signal level is maintained, based on the occurrence status of the signal level in each of the plurality of the first feature amounts. Having The radio wave abnormality detection system according to claim 5.

7. The abnormality detection unit uses the first feature amount extracted by the first feature amount extraction unit to detect a first radio wave abnormality that is part of the radio wave abnormality included in the received data, and is detected from the second feature amount extracted by the second feature amount extraction unit. By comparing the transmission duration of the transmission signal included in the received data with the maximum continuous transmission time, a second radio wave abnormality that is another part of the radio wave abnormality included in the received data is detected. The radio wave abnormality detection system according to claim 5 or 6.

8. The abnormality detection unit uses the first feature amount extracted by the first feature amount extraction unit to detect a first radio wave abnormality that is part of the radio wave abnormality included in the received data, and uses the second feature amount extracted by the second feature amount extraction unit. To detect a second radio wave abnormality that is another part of the radio wave abnormality included in the received data. The radio wave abnormality detection system according to any one of claims 3 to 6.

9. Further comprising a learning unit, The first feature amount extraction unit is configured to extract the feature amount for each first predetermined period of the sample data as a first sample feature amount. The second feature amount extraction unit is configured to extract the feature amount for each second predetermined period of the sample data as a second sample feature amount. The learning unit uses the plurality of first sample feature amounts extracted by the first feature amount extraction unit and the plurality of second sample feature amounts extracted by the second feature amount extraction unit to determine whether the received data contains radio wave abnormalities. It is configured to perform machine learning of a model representing whether or not. The abnormality detection unit is configured to detect radio wave abnormalities included in the received data based on the plurality of first feature amounts, the plurality of second feature amounts, and the learned model generated by machine learning of the learning unit. The radio wave abnormality detection system according to any one of claims 3 to 8.

10. The first feature amount is statistical information including information in both the frequency direction and the time direction of the received data in the first predetermined period. The second feature amount is statistical information including information in both the frequency direction and the time direction of the received data in the second predetermined period. The radio wave abnormality detection system according to any one of claims 1 to 9.

11. A first feature amount extraction step of extracting the feature amount for each first predetermined period of the received data as a first feature amount, A second feature amount extraction step of extracting, as a second feature amount, a feature amount of the received data for each second predetermined period longer than the first predetermined period; An abnormality detection step of detecting a radio wave abnormality included in the received data using the first feature amount and the second feature amount; A radio wave abnormality detection method comprising the above.

12. A first feature amount extraction process of extracting, as a first feature amount, a feature amount of received data for each first predetermined period; A second feature amount extraction process of extracting, as a second feature amount, a feature amount of the received data for each second predetermined period longer than the first predetermined period; An abnormality detection process of detecting a radio wave abnormality included in the received data using the first feature amount and the second feature amount; A radio wave abnormality detection program for causing a computer to execute the above.

Images