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

Abstract

To provide a radio wave anomaly detection system, a radio wave anomaly detection method, and a radio wave anomaly detection program that can accurately detect radio wave anomaly.SOLUTION: A radio wave anomaly detection system 10 includes a determining unit 11 that determines an extraction period on the basis of a transmission duration estimated from first reception data of radio waves received at a first timing, a learning unit 12 that generates a trained model by machine learning feature quantities extracted during an extraction period from second reception data of radio waves received at a second timing, and a detection unit 13 that detects a radio wave anomaly using the learned model and the feature quantities extracted during the extraction period from the third reception data of the radio wave received at a third timing.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] This disclosure relates to a radio wave anomaly detection system, a radio wave anomaly detection method, and a radio wave anomaly detection program. [Background technology]

[0002] Radio communication using radio waves is utilized in various fields, but among them, communications such as police radio, fire department radio, aviation radio, and railway radio are considered particularly important and are called critical radio communications. If interference occurs with critical radio communications, it could develop into a situation that endangers human lives, so it is extremely important to detect (monitor) radio interference and disruptions to the radio waves emitted for these communications.

[0003] In these wireless communication systems, some systems, like data communication, have a predetermined transmission duration, while others, like voice communication, have a transmission duration that varies each time depending on the content of the message.

[0004] Typically, when detecting radio wave transmission conditions or interference / faults in wireless systems, a method is used where a threshold is set for the reception level of a particular radio wave, and any radio wave whose reception level exceeds the threshold is judged as abnormal. The reception level threshold defined for each radio wave frequency is called the spectrum mask. However, in this case, since only reception levels exceeding the spectrum mask value are judged as abnormal, there is a problem that low reception levels or abnormalities in features other than the reception level will not be judged as abnormal.

[0005] To solve these problems, Patent Document 1 proposes a technique for determining the presence or absence of radio interference and its causes by extracting amplitude features at predetermined time intervals from sampled data of received radio waves and calculating the similarity between the extracted amplitude features and multiple training data. [Prior art documents] [Patent Documents]

[0006] [Patent Document 1] Patent No. 5760899 [Overview of the project] [Problems that the invention aims to solve]

[0007] However, since Patent Document 1 extracts amplitude features at predetermined time intervals, the amplitude features will differ if the proportion of the transmission duration within those predetermined time intervals is different. Therefore, related technologies such as Patent Document 1 may not be able to accurately detect radio anomalies such as radio interference.

[0008] In view of the above-mentioned problems, this disclosure aims to provide a radio wave anomaly detection system, a radio wave anomaly detection method, and a radio wave anomaly detection program that can detect radio wave anomalies with high accuracy. [Means for solving the problem]

[0009] The radio wave anomaly detection system according to this disclosure comprises: a determination unit that determines an extraction period based on the transmission duration estimated from first reception data of radio waves received at a first timing; a learning unit that generates a trained model by machine learning the features extracted during the extraction period from second reception data of radio waves received at a second timing; and a detection unit that detects radio wave anomalies using the features extracted during the extraction period from third reception data of radio waves received at a third timing and the trained model.

[0010] The radio wave anomaly detection method described herein determines an extraction period based on the transmission duration estimated from first received data of radio waves received at a first timing, generates a trained model by machine learning using the features extracted during the extraction period from second received data of radio waves received at a second timing, and detects radio wave anomalies using the features extracted during the extraction period from third received data of radio waves received at a third timing and the trained model.

[0011] The radio wave anomaly detection program described herein is a radio wave anomaly detection program that causes a computer to execute a process to detect radio wave anomalies. This process involves determining an extraction period based on the transmission duration estimated from first received data of radio waves received at a first timing, generating a trained model by machine learning the features extracted during the extraction period from second received data of radio waves received at a second timing, and using the features extracted during the extraction period from third received data of radio waves received at a third timing and the trained model. [Effects of the Invention]

[0012] This disclosure provides a radio wave anomaly detection system, a radio wave anomaly detection method, and a radio wave anomaly detection program that can accurately detect radio wave anomalies. [Brief explanation of the drawing]

[0013] [Figure 1] This is a block diagram outlining the radio wave anomaly detection system according to an embodiment. [Figure 2] This is a block diagram showing an example of the overall configuration of a radio wave anomaly detection system according to the first embodiment. [Figure 3] This flowchart shows an overview of the overall processing flow of the radio wave anomaly detection system according to the first embodiment. [Figure 4] This flowchart shows a specific example of the feature parameter determination process according to the first embodiment. [Figure 5A] This is a diagram illustrating the steps of the feature parameter determination process according to the first embodiment. [Figure 5B] This is a diagram illustrating the steps of the feature parameter determination process according to the first embodiment. [Figure 5C] This is a diagram illustrating the steps of the feature parameter determination process according to the first embodiment. [Figure 5D] This is a diagram illustrating the steps of the feature parameter determination process according to the first embodiment. [Figure 6]It is a flowchart showing a specific example of learning processing according to the first embodiment. [Figure 7] It is a flowchart showing a specific example of determination processing according to the first embodiment. [Figure 8] It is a block diagram showing an example of the overall configuration of a radio wave abnormality detection system according to the second embodiment. [Figure 9] It is a flowchart showing an outline of the overall processing flow of a radio wave abnormality detection system according to the second embodiment. [Figure 10] It is a flowchart showing a specific example of the determination process of feature quantity parameters according to the second embodiment. [Figure 11A] It is a diagram for explaining the steps of the determination process of feature quantity parameters according to the second embodiment. [Figure 11B] It is a diagram for explaining the steps of the determination process of feature quantity parameters according to the second embodiment. [Figure 11C] It is a diagram for explaining the steps of the determination process of feature quantity parameters according to the second embodiment.

Modes for Carrying Out the Invention

[0014] Hereinafter, embodiments will be described with reference to the drawings. In each drawing, the same elements are denoted by the same reference numerals, and redundant explanations are omitted as necessary.

[0015] [Outline of the Embodiment] As described above, in Patent Document 1, since the amplitude feature amount is extracted at each predetermined time, if the ratio of the transmission continuation time occupied by the predetermined time is different, the amplitude feature amount will also be different. Therefore, in a wireless system where the transmission continuation time is different each time, such as voice communication, each communication will have a different amplitude feature amount. Therefore, a communication with a transmission continuation time that did not appear when collecting the received data that is the basis of the teacher data is determined as an interference wave. This is a problem that occurs when the predetermined time set when calculating the amplitude feature amount is larger than the transmission continuation time.

[0016] One way to circumvent this problem is to set a sufficiently short time interval for calculating amplitude features, so that normal radio signals are transmitted continuously during that interval. However, if the time interval is too short, not only normal radio signals but also interfering waves will be detected continuously during that interval, and the features will no longer represent the difference in temporal characteristics between normal radio signals and interfering waves. The purpose of an anomaly detection system is to distinguish between normal radio signals and interfering waves, and as a means to achieve this, it is desirable to combine as many characteristics as possible, such as intensity, frequency, and time.

[0017] Therefore, in this embodiment, even when the transmission duration of the regular wireless system differs each time, the characteristics are compared, including the difference in the temporal characteristics of the regular wireless signal and the interfering wave, making it possible to detect radio wave anomalies caused by the interfering wave.

[0018] Figure 1 shows an overview of the radio wave anomaly detection system according to the embodiment. As shown in Figure 1, the radio wave anomaly detection system 10 according to the embodiment includes a determination unit 11, a learning unit 12, and a detection unit 13. The radio wave anomaly detection system 10 may be composed of one device or multiple devices.

[0019] The determination unit 11 determines the extraction period based on the transmission duration estimated from the first received data of the radio waves received at the first timing. In one example, the determination unit 11 may determine the extraction period based on the minimum transmission duration estimated from the first received data. In another example, the determination unit 11 may determine the extraction period based on the maximum transmission duration estimated from the first received data. The determination unit 11 may also estimate the type of transmission duration and determine the extraction period from the minimum or maximum transmission duration according to the estimated type.

[0020] The learning unit 12 generates a trained model by machine learning the features extracted during the extraction period from the second received data of the radio waves received at the second timing (the timing after the first timing). The detection unit 13 uses the features extracted during the extraction period from the third received data of the radio waves received at the third timing (the timing after the second timing) and the trained model to detect radio wave anomalies.

[0021] In this embodiment, the extraction period for generating a trained model and extracting features used for detecting radio anomalies is determined based on the transmission duration estimated from the received data. This allows for the appropriate determination of the feature extraction period, enabling accurate detection of radio anomalies.

[0022] In this embodiment, for example, the extraction period is determined based on the minimum transmission duration estimated from the received data, or based on the type of estimated transmission duration. This allows for the detection of interference waves without being affected by fluctuations in the transmission duration of the regular wireless system, even when the transmission duration of the regular wireless system is not constant, and by comparing the characteristics, including the differences in the temporal characteristics of the regular wireless system and the interference wave.

[0023] [First Embodiment] Next, the configuration of the radio wave anomaly detection system according to the first embodiment will be described. Figure 2 is a block diagram showing an example of the overall configuration of the radio wave anomaly detection system according to the first embodiment.

[0024] As shown in Figure 2, the radio wave anomaly detection system 100 according to this embodiment includes an extraction period determination unit 110, a feature extraction unit 120, a learning unit 130, a determination unit 140, and databases DB1 and DB2. The radio wave anomaly detection system 100 is equipped with a receiver (not shown) that receives radio waves, either inside or outside the system, and received data (sample data) is input from the receiver to the extraction period determination unit 110 and the feature extraction unit 120.

[0025] The extraction period determination unit 110 determines the extraction period for the feature extraction unit 120 to extract features based on the acquired sample data (received data) before the learning process and the judgment process (first timing). For example, the extraction period determination unit 110 determines the extraction period based on the transmission duration estimated from the received data (first received data) of the first radio wave (radio wave for extraction period determination) received at the first timing, and in particular, determines the extraction period based on the minimum transmission duration estimated from the received data. The transmission duration to be estimated is the transmission duration in normal (regular) wireless communication under normal circumstances.

[0026] The extraction period determination unit 110 includes a spectrogramming unit 111, a signal level estimation unit 112, a transmission duration calculation unit 113, and a minimum transmission duration estimation unit 114. The spectrogramming unit 111 calculates a spectrum or spectrogram from sample data (received data). The signal level estimation unit 112 uses the calculated spectrum or spectrogram to estimate the signal level of the transmitted signal included in the received data, based on the value of the noise floor inside the receiver measured during the time when no signal is being transmitted, and the received power intensity (signal level) of the signal measured during the time when the signal is being transmitted.

[0027] The transmission duration calculation unit 113 calculates the duration (transmission duration) for which the signal level of each signal estimated by the signal level estimation unit 112 is continuously transmitted. The minimum transmission duration estimation unit 114 estimates the minimum transmission duration from the transmission duration data of multiple signals calculated by the transmission duration calculation unit 113. Then, based on this minimum transmission duration, the minimum transmission duration estimation unit 114 determines a predetermined time length for calculating the feature quantities described later, and outputs the determined time length (extraction period) as a feature quantity parameter.

[0028] Database DB1 is a feature parameter storage (memory) unit that stores the feature calculation time (extraction period) output by the extraction period determination unit 110 as feature parameters.

[0029] The feature extraction unit 120 extracts features from the received data (second received data) of the second radio wave (learning radio wave) received during the learning process (second timing) for an extraction period determined by the extraction period determination unit 110, and also extracts features from the received data (third received data) of the third radio wave (monitored radio wave) received during the judgment process (third timing) for an extraction period determined by the extraction period determination unit 110.

[0030] The feature extraction unit 120 comprises a spectrogramming unit 121 and a feature extraction unit 122. The spectrogramming unit 121 calculates a spectrum or spectrogram from the received data. The feature extraction unit 122 uses the calculated spectrum or spectrogram to extract features from the received data for the extraction period stored in the database DB1.

[0031] During the learning process, the learning unit 130 uses the features extracted by the feature extraction unit 120 from the received data samples to perform machine learning on a model (trained model) that indicates whether or not the received data contains radio anomalies (such as radio interference or malfunctions). In other words, the learning unit 130 generates a trained model by machine learning the features extracted from the second radio wave received data during the extraction period. The database DB2 is a trained model storage (memory) unit that stores the trained model generated by the learning unit 130 through machine learning.

[0032] During the determination process, the determination unit 140 uses the features output from the received data by the feature extraction unit 120 and the trained model to determine whether the received data contains abnormal radio waves (radio wave anomalies) and outputs the determination result. In other words, the determination unit (detection unit) 140 detects radio wave anomalies using the features extracted from the received data of the third radio wave during the extraction period and the trained model.

[0033] The learning process involves first training the model using sample data to learn the normal state, and then performing machine learning on the model to determine whether or not actual received data contains radio anomalies. The determination process, on the other hand, involves using the actual received data to determine whether or not the received data contains radio anomalies.

[0034] Furthermore, the features extracted from the received data are statistical information that includes both frequency and time-domain information of the received data over a predetermined extraction period. For example, the features are assumed to be statistical features extracted at predetermined intervals, such as frequency distributions (histograms), probability density functions (PDF), cumulative distribution functions (CDF), and amplitude probability distributions (APD). Note that the features may be extracted directly from the time-domain received data, or they may be extracted for each frequency after being converted to frequency-domain data through spectral analysis or spectrogramming.

[0035] Next, the operation (radio wave anomaly detection method) of the radio wave anomaly detection system according to the first embodiment will be described. Figure 3 is a flowchart showing an overview of the overall processing flow of the radio wave anomaly detection system 100 according to this embodiment.

[0036] As shown in Figure 3, the processing by the radio wave anomaly detection system 100 includes the steps of feature parameter determination (S100), learning (S200), and judgment (S300). The radio wave anomaly detection system 100 performs the feature parameter determination (S100) and learning (S200) in advance or periodically, and the judgment (S300) sequentially during operation. For example, the received data of the first radio wave (for determining the extraction period) and the received data of the second radio wave (for learning) are received data of radio waves received in advance or at periodic timings (first and second timings), while the received data of the third radio wave (the radio wave to be monitored) is received data of radio waves received sequentially during operation (third timing).

[0037] In the feature parameter determination process (S100), the extraction period determination unit 110 determines the time length (extraction period) for extracting features (S101). In the learning process (S200), the feature extraction unit 120 collects features extracted from a large number of sample data based on a predetermined period (extraction period) (S201), and the learning unit 130 generates a trained model for classifying normal and abnormal states using machine learning (S202). Then, in the judgment process (S300), the feature extraction unit 120 extracts features from data received during operation (S301), and the judgment unit 140 determines whether or not the received data contains radio wave anomalies using the trained model generated in the learning process (S200) (S302).

[0038] The feature parameter determination process (S100) and the learning process (S200) are assumed to be performed by collecting received data as sample data during normal conditions (when no abnormal radio signals are being transmitted). The collection period for the received data (sample data) used for learning only needs to include the communication conditions expected during normal conditions (when no abnormal radio signals are being transmitted). Therefore, it is not a problem to use data from a few seconds to tens of seconds on any given day, or data totaling tens of seconds or several hours extracted randomly or at regular intervals over several days or a week.

[0039] Figure 4 is a flowchart showing a specific example of the feature parameter determination process (S100) according to this embodiment shown in Figure 3. Figures 5A to 5D are diagrams illustrating each step of the feature parameter determination process shown in Figure 4.

[0040] As shown in Figure 4, first, in the spectrogramming step (S111), the spectrogramming unit 111 spectrally analyzes or spectrograms the received data collected for extraction period determination or learning, converting it into data that shows the received level for each time or frequency. Figure 5A shows an example of acquired received data (sample data). In the example of Figure 5A, the received data is time-series data of received power, but it is not limited to this. Figure 5B shows an example of data that has been spectrogrammed from the received data. In the example of Figure 5B, the horizontal axis represents time, the vertical axis represents frequency, and the color or brightness at each time and frequency represents the signal level.

[0041] Next, in the signal level estimation step (S112), the signal level estimation unit 112 first creates a histogram of received power for each frequency bin of the spectrogram, as shown in Figure 5C, with the horizontal axis representing received power and the vertical axis representing frequency. If the target frequency includes both time periods when radio waves are transmitted and time periods when they are not, peaks corresponding to each period will appear in the histogram. Peaks with low intensity represent measurement data from time periods when radio waves are not transmitted and usually correspond to the noise floor caused by internal noise in the receiver. Peaks with high intensity indicate the received power of signals observed in this frequency band.

[0042] Then, the signal level estimation unit 112 identifies the time period during which radio waves are being transmitted, and in the histogram of received power, it starts from the peak with the lowest intensity corresponding to the noise floor and determines the received level L that is the first minimum point. t Calculate L t Let's call this the signal presence / absence determination threshold, L t If a higher received power is observed, it is assumed that some radio wave is being transmitted. In other words, the signal presence / absence threshold is the received power at which the frequency of occurrence first becomes minimum, starting from the received power corresponding to the noise floor of the receiver receiving the radio wave in the histogram showing the frequency of occurrence for each received power.

[0043] Next, in the step of calculating the transmission duration (S113), the transmission duration calculation unit 113 first determines the signal presence / absence threshold L t The transmission duration is calculated as the time length (duration) at which a greater received power can be obtained. This process is performed for all the sample data obtained for each frequency bin of the spectrogram. Then, as shown in Figure 5D, the transmission duration calculation unit 113 creates a histogram of the transmission duration for each frequency bin.

[0044] Next, in the step of calculating the minimum transmission duration (S114), the minimum transmission duration estimation unit 114 first determines the threshold r of the occurrence frequency r, as shown in Figure 5D. t Set the frequency of occurrence r to r <r t It is assumed that communications with a transmission duration of this magnitude rarely occur with normal signals. Next, the minimum transmission duration estimation unit 114 calculates the maximum frequency in the transmission duration histogram and obtains the corresponding transmission duration t1. Then, t <t1でr<r t We calculate the maximum transmission duration t2 that results in this. This t2 is defined as the minimum transmission duration in the sample data.

[0045] In the example in Figure 5D, r > r also occurs in the range t < 1 second. t Signals like the one described above have been observed, but such short-duration signals are thought to be caused by noise interference, errors in radio communication, or transmission cancellations, and it is desirable to exclude such signals. Therefore, by calculating the minimum transmission duration using the method described above, it is possible to calculate the minimum transmission duration for normal radio communication. In other words, the minimum transmission duration is the transmission duration that is shorter than the transmission duration with the highest frequency in the histogram showing the frequency of occurrence for each transmission duration, and is the longest transmission duration among those whose frequency is less than a predetermined frequency threshold.

[0046] Finally, the minimum transmission duration estimation unit 114 determines the feature extraction period based on the minimum transmission duration t2. The feature extraction period may be t2 itself, or a value close to t2. For example, a round number close to t2 may be used. The minimum transmission duration estimation unit 114 stores the determined feature extraction period in the database DB1.

[0047] Figure 6 is a flowchart showing a specific example of the learning process (S200) according to the embodiment shown in Figure 3. As shown in Figure 6, first, the spectrogramming unit 121 performs spectral or spectrogramming on the received data collected for learning, similar to the feature parameter determination process, and converts it into data that shows the received level for each time or frequency (S211).

[0048] Next, the feature extraction unit 122 refers to the database DB1, divides the spectrogram according to the feature extraction period determined in the feature parameter determination step (S114), and converts each of them into features (S212).

[0049] Next, the learning unit 130 uses the features extracted and collected by the feature extraction unit 122 to generate a trained model for classifying normal and abnormal states using machine learning (S213). Then, the learning unit 130 stores the trained model generated by machine learning in the database DB2.

[0050] For machine learning, methods such as One-class SVM (One-class Support Vector Machine), Hotelling's theory, lof (Local Outlier Factor) method, and k-nearest neighbors can be used. Alternatively, machine learning of models such as Deep Anomaly Detection using deep learning is also acceptable. Furthermore, the machine learning employed in the learning unit 130 could employ methods such as learning a threshold for statistical anomaly detection using Mahalanobis distance or variance (standard deviation) from the distribution of high-dimensional (multi-dimensional) feature vectors as described above, or predicting trends in change using state filters such as Kalman filters or particle filters, and detecting deviations from those trends. There is no problem with using such methods.

[0051] Figure 7 is a flowchart showing a specific example of the determination process (S300) according to this embodiment shown in Figure 3. This determination process acquires received data from the monitored radio wave environment during actual operation and sequentially performs detection and determination of radio interference and faults.

[0052] As shown in Figure 7, first, similar to the learning process, the spectrogramming unit 121 spectrally converts or spectrograms the collected received data into data that shows the received level for each time period or frequency (S311).

[0053] Next, similar to the learning process, the feature extraction unit 122 refers to the database DB1, divides the spectrogram according to the feature extraction period determined in the feature parameter determination step (S114), and converts each of them into features (S312).

[0054] Next, the determination unit 140 refers to the database DB2 and uses the trained model generated in the training process step (S213) to determine whether the features extracted by the feature extraction unit 122 are normal or abnormal (S313).

[0055] If the trained model was generated by machine learning such as One-class SVM, which is a type of unsupervised learning, the positional relationship and distance of the feature vectors with respect to the discrimination boundary configured within the trained model are converted into a normality (or abnormality) score and output. Based on this calculation, the determination unit 140 outputs a determination result indicating whether or not the received data contains radio wave anomalies.

[0056] In this way, by setting the feature extraction period to be shorter than the transmission duration of normal radio, fluctuations in the transmission duration of normal radio are not reflected in the features. However, if the feature extraction period is longer than the transmission duration of normal radio, for example, if it is set as t3 in the transmission duration histogram in Figure 5D, then changes in the transmission duration of normal radio will be reflected in the features. For example, in the case of a radio system where the transmission duration differs each time and the characteristics of the speaker are reflected, such as in aviation radio voice calls, it is difficult to collect all possible samples of normal radio transmission durations during training. If the training samples are insufficient, signals with transmission durations that have not been trained may be judged as abnormal even if they are from a normal radio system. Therefore, by setting the feature extraction period to be shorter than the transmission duration of a normal radio system, the features will appear as if normal radio is being transmitted throughout the entire extraction period, thus avoiding the influence of fluctuations in the transmission duration of normal radio.

[0057] On the other hand, if the feature extraction period is set too short, abnormal signals will also be detected throughout the entire extraction period, making it impossible to classify normal and abnormal signals based on transmission duration. Therefore, it is desirable to set the feature extraction period as long as possible while maintaining the condition that it is shorter than the transmission duration of normal radio signals. In the radio wave anomaly detection system of this embodiment, the feature extraction period can be set so as to be shorter than the transmission duration of normal radio signals and so as to be able to detect abnormal signals with transmission durations shorter than those.

[0058] [Second Embodiment] Next, the configuration of the radio wave anomaly detection system according to the second embodiment will be described. The radio wave anomaly detection system according to this embodiment has an added function that observes normal radio waves in the target frequency band and determines whether it is effective to apply the radio wave anomaly detection system according to the first embodiment.

[0059] Figure 8 is a block diagram showing an example of the overall configuration of the radio wave anomaly detection system according to this embodiment. The differences between the radio wave anomaly detection system 100 according to this embodiment and the overall configuration of the first embodiment shown in Figure 2 are that the extraction period determination unit 110 is equipped with a transmission duration type estimation unit 115, and the minimum transmission duration estimation unit 114 is replaced with a transmission duration range estimation unit 116. The configuration and functions of the other parts are the same as those of the radio wave anomaly detection system of the first embodiment.

[0060] The transmission duration type estimation unit 115 estimates the type of transmission duration calculated by the transmission duration calculation unit 113. The transmission duration range estimation unit 116 estimates the range (extraction period) of transmission duration based on the transmission duration type estimation unit 115. For example, if the estimated type is a type indicating a constant transmission duration, the transmission duration range estimation unit 116 determines the extraction period based on the minimum transmission duration, and if the estimated type is a type indicating an indeterminate transmission duration, it determines the extraction period based on the maximum transmission duration.

[0061] Figure 9 is a flowchart illustrating the overall processing flow of the radio wave anomaly detection system according to this embodiment. As shown in Figure 9, the processing by the radio wave anomaly detection system 100 includes the steps of feature parameter determination (S100), learning (S200), and judgment (S300), similar to the radio wave anomaly detection system of the first embodiment.

[0062] In the step of the determination process (S100) of the feature quantity parameter, the extraction period determination unit 110 analyzes the trend of the transmission duration of normal radio waves in the target frequency band, and estimates whether there is a regularity in the transmission duration (the type of transmission duration) (S121). Then, based on the result, the extraction period of the feature quantity is determined (S101). The steps of the learning process (S200) and the determination process (S300) are the same as those in the first embodiment.

[0063] FIG. 10 is a flowchart showing a specific example of the determination process (S100) of the feature quantity parameter according to the present embodiment shown in FIG. 9. FIGS. 11A to 11C are diagrams for explaining the process (S115) of estimating the transmission duration type in the determination process of the feature quantity parameter in FIG. 10.

[0064] First, in the spectrogram generation step (S111), similar to the radio wave abnormality detection system of the first embodiment, the spectrogram generation unit 111 performs spectrogram generation or spectrogram conversion on the received data collected for extraction period determination or learning, and converts it into data indicating the reception level for each time or each frequency.

[0065] Next, in the signal level estimation step (S112), similar to the radio wave abnormality detection system of the first embodiment, the signal level estimation unit 112 estimates the reception power level of the observed received signal for each frequency bin of the spectrogram.

[0066] Next, in the step of calculating the transmission duration (S113), similar to the radio wave abnormality detection system of the first embodiment, the transmission duration calculation unit 113 calculates, for each frequency bin of the spectrogram, the time length of all sample data for which a reception power greater than the signal presence / absence determination threshold L t is obtained as the transmission duration, and creates a histogram of the transmission duration. <0者の電波異常検知システムと同様に、送信継続時間算出部113は、スペクトログラムの各周波数ビンについて、信号有無判定閾値Lよりも大きい受信電力が得られるすべてのサンプルデータの時間長を送信継続時間として算出し、送信継続時間のヒストグラムを作成する。

[0067] Next, in the transmission duration type estimation step (S115), the transmission duration type estimation unit 115 estimates the type of transmission duration of normal radio transmissions in the target frequency band. In this embodiment, the transmission duration types are classified into two types: "constant" and "variable". Then, it is determined whether the transmission duration type of the normal radio transmission is "constant" or "variable" (S116).

[0068] In a wireless system with a constant transmission duration, it is assumed that the transmission duration is distributed around a specific value. The histogram of the transmission duration is fitted with a predetermined function, and if the correlation coefficient (correlation value) is greater than or equal to a predetermined threshold (i.e., the distribution of transmission duration correlates with the predetermined function), it is determined that the transmission duration is constant. Examples of the "pre-defined function" include the normal distribution, log-normal distribution, and Rayleigh distribution. The threshold for the correlation coefficient is between -1 and 1, and the closer it is to 1, the stronger the correlation between the measured value of the transmission duration and the predetermined function. For example, if the calculated threshold for the correlation coefficient is set to 0.8, and the calculated correlation coefficient is 0.8 or higher, it is estimated that the transmission duration of a normal wireless system in the target frequency band is constant.

[0069] Figure 11A shows an example of a histogram when the transmission duration is constant, where the transmission duration of normal radio signals is distributed around a single value, t5, in this frequency band. In this case, it can be inferred that the transmission duration of systems using this frequency band is specified to be t5. Figure 11B shows another example of a histogram when the transmission duration is constant, where the transmission duration of normal radio signals is distributed around two values, t6 and t7, in this frequency band. In this case, two radio systems use the same frequency band, and their transmission durations are different, but even in this example, the transmission duration of normal radio signals can be considered constant.

[0070] In cases like these, where the transmission duration of the normal radio is designed to be constant, setting the feature extraction period to a value longer than those transmission durations t5, t6, and t7 makes it easier to detect the interference by taking advantage of the difference in temporal characteristics between the normal radio and the interference.

[0071] On the other hand, if the correlation coefficient (correlation value) between the histogram of measured transmission duration and the set function is smaller than the threshold, it is estimated that the transmission duration of normal radio in the target frequency band is indeterminate. Figure 11C is an example of a histogram when the transmission duration is indeterminate, and the histogram of transmission duration has a distribution that is clearly different from a normal distribution, and the correlation coefficient is small.

[0072] In this case, it can be assumed that the transmission duration of normal radio waves is not specified to take a specific value. Therefore, if the feature extraction period is made longer than the transmission duration of normal radio waves, for example as t9 in the figure, there is a possibility that transmission durations not extracted during training may be detected during the judgment process, in which case it will be incorrectly judged as abnormal. On the other hand, if the feature extraction period is made shorter than the transmission duration of normal radio waves, for example as t8 in the figure, fluctuations in the transmission duration of normal radio waves will not be reflected in the features, thus avoiding incorrectly judging normal radio waves as abnormal. At the same time, by making the feature extraction period as long as possible within a range shorter than the transmission duration of normal radio waves, it becomes possible to detect interference waves by utilizing the difference in transmission duration when the transmission duration of abnormal radio waves is shorter than that.

[0073] Examples of radio systems with a fixed transmission duration include, in the case of critical aviation radio, radar and systems that transmit data of a predetermined length, such as SSR (Secondary Surveillance Radar). Examples of radio systems with an indefinite transmission duration include radiotelephones used in air traffic control and around airport facilities.

[0074] Next, if the transmission duration type is estimated to be undefined, the transmission duration range estimation unit 116 calculates the minimum transmission duration in the sample data using the same method as the radio wave anomaly detection system of the first embodiment (S114). In the example of Figures 11A to 11C, Figure 11C corresponds to this example, and the minimum transmission duration is t8. Then, the transmission duration range estimation unit 116 determines the feature extraction period based on this minimum transmission duration. The feature extraction period may be the minimum transmission duration itself, or it may be a value close to it.

[0075] Furthermore, if the transmission duration type is estimated to be constant, the transmission duration range estimation unit 116 calculates the maximum transmission duration in the sample data (S117). Figures 11A and 11B correspond to examples of this case, and the transmission duration range estimation unit 116 only needs to obtain the maximum value in these distributions. That is, in this case, the transmission duration range estimation unit 116 determines the extraction period based on the maximum transmission duration estimated from the received data. The maximum transmission duration is the transmission duration with the highest occurrence rate in the histogram showing the frequency of occurrence for each transmission duration.

[0076] As described above, the radio wave anomaly detection system of this embodiment first classifies the types of radio transmission durations observed under normal conditions into two types: constant and variable. When the transmission duration of normal radio (normal radio) is constant, interference waves can be effectively detected by setting the feature extraction period based on its maximum value. On the other hand, when the transmission duration of normal radio is variable, interference waves can be detected without being affected by fluctuations in the transmission duration of normal radio by setting the feature extraction period based on its minimum transmission duration.

[0077] This disclosure is not limited to the embodiments described above, and can be modified as appropriate without departing from its intent. For example, this disclosure can be applied to applications such as illegal radio monitoring systems and radio equipment fault detection systems, but it may also be applied to other systems.

[0078] Each configuration in the above-described embodiment may consist of hardware, software, or both, and may consist of one piece of hardware or software, or multiple pieces of hardware or software. The functions (processing) of each device may be realized by a computer having a CPU (Central Processing Unit), FPGA (Field Programmable Gate Array), memory, etc. For example, a program for performing the method in the embodiment may be stored in a storage device, and each function may be realized by executing the program stored in the storage device with the CPU.

[0079] These programs, when loaded into a computer, include a set of instructions (or software code) for causing the computer to perform one or more of the functions described in the embodiments. The programs may be stored on non-temporary computer-readable media or tangible storage media. Examples, but not limited to, include random-access memory (RAM), read-only memory (ROM), flash memory, solid-state drives (SSDs), or other memory technologies, CD-ROMs, digital versatile discs (DVDs), Blu-ray® discs, or other optical disc storage, magnetic cassettes, magnetic tapes, magnetic disk storage, or other magnetic storage devices. The programs may be transmitted over temporary computer-readable media or communication media. Examples, but not limited to, include electrical, optical, acoustic, or other forms of propagating signals.

[0080] Some or all of the above embodiments may also be described as follows, but are not limited to the following: (Note 1) A determination unit that determines the extraction period based on the transmission duration estimated from the first received data of the radio waves received at the first timing, A learning unit that generates a trained model by machine learning the features extracted during the extraction period from the second reception data of the radio waves received at the second timing, A detection unit that detects radio anomalies using the feature quantities extracted during the extraction period from the third reception data of the radio waves received at the third timing and the trained model, A radio wave anomaly detection system equipped with the following features. (Note 2) The determination unit determines the extraction period based on the minimum transmission duration estimated from the first received data. The radio wave anomaly detection system described in Appendix 1. (Note 3) The minimum transmission duration is, in the histogram showing the frequency of occurrence for each transmission duration, a transmission duration that is shorter than the transmission duration at which the frequency of occurrence is highest, and is the longest transmission duration among those transmission durations whose frequency of occurrence is lower than a predetermined frequency threshold. The radio wave anomaly detection system described in Appendix 2. (Note 4) The determination unit determines the extraction period based on the maximum transmission duration estimated from the first received data. The radio wave anomaly detection system described in Appendix 1. (Note 5) The maximum transmission duration is the maximum transmission duration in the histogram showing the frequency of occurrence for each transmission duration. The radio wave anomaly detection system described in Appendix 4. (Note 6) The determination unit estimates the type of transmission duration and determines the extraction period based on the estimated type. The radio wave anomaly detection system described in Appendix 1. (Note 7) The determination unit estimates that the type of transmission duration is constant when the distribution of the transmission duration correlates with a predetermined function. If the correlation value between the distribution of transmission duration and the predetermined function is smaller than a preset threshold, it is estimated that the type of transmission duration is undefined. The radio wave anomaly detection system described in Appendix 6. (Note 8) The determination unit determines the extraction period based on the maximum transmission duration if it is estimated that the type of transmission duration is constant, and determines the extraction period based on the minimum transmission duration if it is estimated that the type of transmission duration is indeterminate. The radio wave anomaly detection system described in Appendix 7. (Note 9) The transmission duration is the duration for which a received power greater than the signal presence / absence threshold can be obtained. The signal presence / absence determination threshold is the received power at which the frequency of occurrence first reaches a minimum, starting from the received power corresponding to the noise floor of the receiver receiving the radio waves in the histogram showing the frequency of occurrence for each received power. A radio wave anomaly detection system as described in any one of the appendices 1 to 8. (Note 10) The extraction period is determined based on the normal transmission duration estimated from the first received data of the radio waves received at the first timing. A trained model is generated by machine learning, which uses the features extracted during the extraction period from the second reception data of the radio waves received at the second timing. Using the features extracted during the extraction period from the third received data of the radio waves received at the third timing, and the trained model, radio wave anomalies are detected. Method for detecting radio wave anomalies. (Note 11) The above determination determines the extraction period based on the minimum transmission duration estimated from the first received data. The radio wave anomaly detection method described in Appendix 10. (Note 12) The extraction period is determined based on the normal transmission duration estimated from the first received data of the radio waves received at the first timing. A trained model is generated by machine learning, which uses the features extracted during the extraction period from the second reception data of the radio waves received at the second timing. Using the features extracted during the extraction period from the third received data of the radio waves received at the third timing, and the trained model, radio wave anomalies are detected. A radio wave anomaly detection program that causes a computer to perform a process. (Note 13) The above determination determines the extraction period based on the minimum transmission duration estimated from the first received data. The radio wave anomaly detection program described in Appendix 13. [Explanation of symbols]

[0081] 10. Radio wave anomaly detection system 11. Decision Section 12. Learning Department 13 Detection unit 100 Radio wave anomaly detection system 110 Extraction period determination unit 111 Spectrogramming section 112 Signal level estimation unit 113 Transmission duration calculation unit 114 Minimum transmission duration estimation unit 115 Transmission duration type estimation unit 116 Transmission duration range estimation unit 120 Feature Extraction Unit 121 Spectrogramming section 122 Feature conversion part 130 Learning Department 140 Judgment section

Claims

1. A determination unit that determines the extraction period based on the transmission duration estimated from the first reception data of the radio waves received at the first timing, A learning unit that generates a trained model by machine learning the features extracted during the extraction period from the second reception data of the radio waves received at the second timing, A detection unit that detects radio wave anomalies using the feature quantities extracted during the extraction period from the third reception data of the radio waves received at the third timing and the trained model, Equipped with, The determination unit determines the extraction period based on the minimum transmission duration estimated from the first received data. The determination unit estimates the minimum transmission duration to be the largest transmission duration among those transmission durations that are shorter than the transmission duration at which the frequency of occurrence is maximized, and whose frequency of occurrence is less than a predetermined frequency threshold, in the histogram showing the frequency of occurrence for each transmission duration. Radio wave anomaly detection system.

2. A determination unit that determines the extraction period based on the transmission duration estimated from the first reception data of the radio waves received at the first timing, A learning unit that generates a trained model by machine learning the features extracted during the extraction period from the second reception data of the radio waves received at the second timing, A detection unit that detects radio wave anomalies using the feature quantities extracted during the extraction period from the third reception data of the radio waves received at the third timing and the trained model, Equipped with, The determination unit estimates the type of transmission duration and determines the extraction period based on the estimated type. The determination unit estimates that the type of transmission duration is constant when the distribution of the transmission duration correlates with a predetermined function. If the correlation value between the distribution of transmission duration and the predetermined function is smaller than a preset threshold, it is estimated that the type of transmission duration is undefined. Radio wave anomaly detection system.

3. The determination unit determines the extraction period based on the maximum transmission duration if it is estimated that the type of transmission duration is constant, and determines the extraction period based on the minimum transmission duration if it is estimated that the type of transmission duration is indeterminate. The radio wave anomaly detection system according to claim 2.

4. The extraction period is determined based on the transmission duration estimated from the first received data of the radio waves received at the first timing. A trained model is generated by machine learning using the features extracted during the extraction period from the second reception data of the radio waves received at the second timing. Using the features extracted during the extraction period from the third reception data of the radio waves received at the third timing, and the trained model, a radio wave anomaly is detected. In the above determination, the extraction period is determined based on the minimum transmission duration estimated from the first received data. In the above determination, the minimum transmission duration is estimated to be the largest transmission duration among those transmission durations that are shorter than the transmission duration at which the frequency of occurrence is maximized, and whose frequency of occurrence is less than a predetermined frequency threshold, in the histogram showing the frequency of occurrence for each transmission duration. Method for detecting radio wave anomalies.

5. An extraction period is determined based on the transmission duration estimated from the first reception data of the radio waves received at the first timing, A trained model is generated by machine learning using the features extracted during the extraction period from the second reception data of the radio waves received at the second timing. Using the features extracted during the extraction period from the third reception data of the radio waves received at the third timing, and the trained model, a radio wave anomaly is detected. In the above decision, the type of the transmission duration is estimated, and the extraction period is determined based on the estimated type. In the above determination, if the distribution of the transmission duration correlates with a predetermined function, it is estimated that the type of transmission duration is constant. If the correlation value between the distribution of transmission duration and the predetermined function is smaller than a preset threshold, it is estimated that the type of transmission duration is undefined. Method for detecting radio wave anomalies.

6. The extraction period is determined based on the transmission duration estimated from the first received data of the radio waves received at the first timing. A trained model is generated by machine learning using the features extracted during the extraction period from the second reception data of the radio waves received at the second timing. Using the features extracted during the extraction period from the third reception data of the radio waves received at the third timing, and the trained model, a radio wave anomaly is detected. In the above determination, the extraction period is determined based on the minimum transmission duration estimated from the first received data. In the above determination, the minimum transmission duration is estimated to be the largest transmission duration among those transmission durations that are shorter than the transmission duration at which the frequency of occurrence is maximized, and whose frequency of occurrence is less than a predetermined frequency threshold, in the histogram showing the frequency of occurrence for each transmission duration. A radio wave anomaly detection program that causes a computer to perform a process.

7. An extraction period is determined based on the transmission duration estimated from the first reception data of the radio waves received at the first timing, A trained model is generated by machine learning using the features extracted during the extraction period from the second reception data of the radio waves received at the second timing. Using the features extracted during the extraction period from the third reception data of the radio waves received at the third timing, and the trained model, a radio wave anomaly is detected. In the above decision, the type of the transmission duration is estimated, and the extraction period is determined based on the estimated type. In the above determination, if the distribution of the transmission duration correlates with a predetermined function, it is estimated that the type of transmission duration is constant. If the correlation value between the distribution of transmission duration and the predetermined function is smaller than a preset threshold, it is estimated that the type of transmission duration is undefined. A radio wave anomaly detection program that causes a computer to perform a process.

Images