Spectrum monitoring device, spectrum monitoring method and program

The spectrum monitoring device uses GAN-based image reconstruction to differentiate noise from interference, accurately identifying and visualizing interference locations in wireless communication systems.

JP2024171082APending Publication Date: 2024-12-11NEC CORP +1

Patent Information

Application Number
JP2023087962
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-05-29
Publication Date
2024-12-11

AI Technical Summary

Technical Problem

Existing methods fail to accurately identify the frequency bin and location of radio wave interference in wireless communication systems, often misattributing noise fluctuations to interference signals, and lack detailed visualization of interference occurrence.

Method used

A spectrum monitoring device and method using generative adversarial networks (GAN) for reconstructing spectrum images, calculating pixel and frequency bin differences to distinguish noise from interference, and visually highlighting interference locations.

Benefits of technology

Enables precise identification of interference locations by reducing noise influence, allowing for detailed visualization of interference signals in wireless communication systems.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024171082000001_ABST
    Figure 2024171082000001_ABST
Patent Text Reader

Abstract

To identify the time and location of radio interference without being affected by fluctuations due to noise when monitoring radio wave abnormalities (radio wave interference) in wireless communication.SOLUTION: A spectrum monitoring device receives radio waves transmitted wirelessly to generate a received signal, calculates spectrum information from the received signal to generate a received spectrum image, reconstructs the received spectrum image through learning to generate a reconstructed spectrum image, determines the presence or absence of an interfering signal on the basis of the difference between the received spectrum image and the reconstructed spectrum image, and visualizes the frequency band in which the interfering signal exists when the presence of an interfering signal is determined.SELECTED DRAWING: Figure 1
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to a spectrum monitoring device, a spectrum monitoring method, and a program for monitoring radio interference in wireless communication. [Background technology]

[0002] Important wireless communications such as police radio, fire radio, aviation radio, and railway radio require high security. If interference occurs in wireless communications, it may develop into a life-threatening situation, so it is necessary to detect (monitor) radio interference and communication failures in wireless communications. In addition to wireless LANs (Local Area Networks) such as Wi-Fi, 4th generation mobile communication systems (4G), 5th generation mobile communication systems (5G), and communication systems such as private LTE (Long Term Evolution) and local 5G that apply cellular communications are attracting attention as wireless communication technologies used by individuals and self-employed people. In private LTE and local 5G, it is expected that systems will be built for each area or each operator. Therefore, there is a possibility that radio interference will occur between adjacent areas of systems installed by different operators. It is expected that detecting interference to one's own system will prevent deterioration of wireless communication performance and analyze the cause of deterioration highly efficiently.

[0003] Various techniques have been developed as a method for detecting radio interference and communication failure in wireless communication, and one method for detecting the reception state of radio waves is to monitor spectrum (frequency distribution). Patent Document 1 discloses a radio wave anomaly detection system for accurately detecting radio wave anomalies contained in received data in wireless communication, and detects radio wave anomalies contained in the received data using a first feature amount for each first period of the received data and a second feature amount for each second period longer than the first period. Patent Document 2 discloses a transmission device matching device for matching a transmission device that performs wireless transmission, and receives a signal wirelessly transmitted from a transmission device, generates a spectrogram (a three-dimensional graph showing the calculation result of a frequency spectrum), converts the received signal with parameters according to the signal region, extracts a feature amount, and matches the transmission device based on the similarity with a pre-stored feature amount. Patent Document 3 discloses an anomaly determination device for determining an anomaly in the emission state of radio waves, and receives wireless radio waves, extracts and stores a feature amount in a predetermined frequency band, and generates an anomaly determination mask in which a threshold value is set to detect an anomaly in wireless radio waves based on the result of determining whether the feature amount is within a pre-set normal range.

[0004] Meanwhile, machine learning and deep learning for data analysis have been developed and are used to detect anomalies in various images and radio wave anomalies in wireless communication. In deep learning, an autoencoder is an algorithm that compresses the dimensions of input data to leave only important features, and then restores the data to the original dimensions. Generative Adversarial Networks (GANs) use two networks, a generator and a discriminator, to generate new pseudo data from input data, and are applied to generating voices and images. Non-Patent Document 1 discloses a technology for detecting radio spectrum anomalies in the digital television reception frequency band using a GAN. Non-Patent Document 2 discloses an anomaly detection technology using a GAN, which is applied to the analysis of X-ray images used in baggage inspection at airports. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] JP 2022-182844 A [Patent Document 2] JP 2021-180388 A [Patent Document 3] International Publication No. 2021 / 176805 [Non-patent literature]

[0006] [Non-Patent Document 1] L. Zhang, CX Huang, H. Tang, JJ Yang, and M. Huang, “DTV radio spectrum anomaly detection based on an improved GAN”, 2020 XXXIIIrd General Assembly and Scientific Symposium of the International Union of Radio Science, 2020, pp. 1-4 [Non-Patent Document 2] S. Akcay, AA-Abarghouei, TP Breckon, “GANomarly: Semi-Supervised Anomaly Detection via Adversarial Training”, ACCV2018, Dec. 2018 Summary of the Invention [Problem to be solved by the invention]

[0007] When detecting radio interference in a wireless communication system, a method is adopted in which the radio waves are judged to be abnormal if the radio wave reception level exceeds a predetermined threshold. When an abnormality in the reception level is detected, it is desirable not only to record the time of occurrence of the radio wave interference in a log, but also to visualize the frequency bin where the radio wave interference (radio wave abnormality) occurred in order to implement measures against the radio wave interference depending on the purpose of monitoring wireless communication.

[0008] In Patent Document 1, although it is possible to identify the time when radio interference occurs, it does not identify the frequency bin where radio wave abnormality occurs at the time of occurrence, and it is not possible to identify (visualize) the location where radio wave interference occurs in detail. Patent Document 2 detects a specific signal region based on a spectrogram generated from a received signal when checking a transmission source using radio waves wirelessly transmitted from a transmitting device (wireless terminal), and identifies standardized communication methods, non-standardized communication methods, and devices that unintentionally transmit radio waves, but does not identify the time when radio wave interference occurs or the frequency bin where radio wave abnormality occurs. Patent Document 3 generates an abnormality determination mask in which a threshold value is set for a feature amount of a specific frequency band to deal with interference with important wireless communication, and detects a decrease in the reception level in the frequency spectrum, but although it can detect radio wave abnormality in a specific frequency band, it does not identify the frequency bin where radio wave abnormality occurs in association with the time when radio wave interference occurs.

[0009] In Non-Patent Document 1, a reconstructed spectrum image is generated from an original spectrum image using a GAN model, and although it is possible to identify detailed locations of radio interference by visualizing the locations of the spectrum that have not been reconstructed, it is difficult to distinguish the locations of radio interference signals because frequency fluctuation components due to simple noise are also visualized. For this reason, a spectrum monitoring method capable of identifying the locations of radio interference signals without being affected by frequency fluctuations due to noise is required. Non-Patent Document 2 introduces an anomaly determination model using a GAN to learn the generation of a high-dimensional image space and an interference delay space, and presents an anomaly detection method using an autoencoder. In particular, since an autoencoder trained with normal data cannot correctly restore abnormal data, it uses the characteristic that the error (reconstruction error) between the input data (input image) and the restored data (reconstructed image) becomes large, and determines that the input data is abnormal if the reconstruction error is equal to or greater than a predetermined threshold. Non-Patent Document 2 only performs an anomaly determination according to the error between the input image and the reconstructed image, and is unable to detect radio anomalies using a spectrum image in wireless communication.

[0010] An object of the present invention is to provide a spectrum monitoring device, a spectrum monitoring method, and a program that solve the above-mentioned problems. [Means for solving the problem]

[0011] A first aspect of the present invention is a spectrum monitoring device comprising: a receiving unit that receives radio waves transmitted by wireless and generates a received signal; a spectrum image generating unit that calculates spectrum information from the received signal and generates a received spectrum image; a spectrum reconstruction unit that reconstructs the received spectrum image by learning to generate a reconstructed spectrum image; a determination unit that determines the presence or absence of an interference signal based on a difference between the received spectrum image and the reconstructed spectrum image; and a visualization unit that visualizes a frequency band in which the interference signal exists when the determination unit determines the presence of an interference signal.

[0012] A second aspect of the present invention is a spectrum monitoring method that receives radio waves transmitted wirelessly to generate a received signal, calculates spectrum information from the received signal to generate a received spectrum image, reconstructs the received spectrum image by learning to generate a reconstructed spectrum image, determines the presence or absence of an interfering signal based on a difference between the received spectrum image and the reconstructed spectrum image, and, when the presence of an interfering signal is determined, visualizes a frequency band in which the interfering signal exists.

[0013] A third aspect of the present invention is a program implemented in a computer of a spectrum monitoring device, which receives wirelessly transmitted radio waves to generate a received signal, calculates spectrum information from the received signal to generate a received spectrum image, reconstructs the received spectrum image by learning to generate a reconstructed spectrum image, determines the presence or absence of an interfering signal based on a difference between the received spectrum image and the reconstructed spectrum image, and, when the presence of an interfering signal is determined, visualizes a frequency band in which the interfering signal exists. Effect of the Invention

[0014] According to the present invention, when monitoring radio interference (radio wave abnormality) in wireless communication, it is possible to identify the time and location of occurrence of radio interference without being affected by fluctuations due to noise. [Brief description of the drawings]

[0015] [Figure 1] 1 is a block diagram of a spectrum monitoring device according to a first embodiment. [Diagram 2] 4 is a flowchart showing a processing procedure of the spectrum monitoring device according to the first embodiment. [Diagram 3] 4 is a flowchart showing an abnormality determination process of the spectrum monitoring device according to the first embodiment. [Figure 4] 13 is a graph showing a frequency distribution of a first difference value for each pixel of a received spectrum image and a reconstructed spectrum image. [Diagram 5] 13 is a graph showing a frequency distribution of second difference values ​​in the level direction between a received spectrum image and a reconstructed spectrum image. [Figure 6] FIG. 11 is a block diagram of a spectrum monitoring device according to a second embodiment. [Figure 7] 10 is a flowchart showing a processing procedure of a spectrum monitoring device according to a second embodiment. [Figure 8] 10 is a flowchart showing details of model learning in the spectrum monitoring device according to the second embodiment. [Figure 9] FIG. 13 is a block diagram showing a configuration of a spectrum monitoring system according to a third embodiment. [Figure 10] FIG. 11 is a sequence diagram showing a processing procedure of the spectrum monitoring system according to the third embodiment. [Figure 11] FIG. 2 is a block diagram showing a minimum configuration of a spectrum monitoring device. [Figure 12] 4 is a flowchart showing a processing procedure of the spectrum monitoring device. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0016] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A spectrum monitoring device, a spectrum monitoring method, and a program for monitoring radio interference (radio wave abnormality) in wireless communication will be described in detail with reference to the accompanying drawings along with embodiments.

[0017] [First embodiment] FIG. 1 is a block diagram of a spectrum monitoring device 1 according to a first embodiment. The spectrum monitoring device 1 is configured with an information processing device such as a personal computer (PC) having a communication function and a display function, and a processor executes software (such as a spectrum monitoring program) stored in a memory to implement a desired function. As the processor, in addition to a CPU (Central Processing Unit) that controls the entire device, a GPU (Graphics Processing Unit), a GPGPU (General-Purpose GPU), a TPU (Tensor Processing Unit), etc. that perform image analysis by machine learning and deep learning are applied. As the software, in addition to C language and embedded programming languages, Python, CuPy, etc. for machine learning and deep learning are applied. The spectrum monitoring device 1 includes a receiving unit 11, a spectrum image generating unit 12, a spectrum reconstruction unit 13, a determining unit 14, and a visualizing unit 15. The receiving unit 11 receives radio waves wirelessly communicated from the outside via an antenna, and sends the received signal to the spectrum image generating unit 12. The spectrum image generating unit 12 performs a discrete Fourier transform (DFT) on the received signal from the receiving unit 11 to obtain frequency components of the measurement data, calculates spectrum information, and generates a spectrum image from the spectrum information. The spectrum image visualizes the relationship between the frequency and amplitude of radio waves. The spectrum image related to the received radio waves is called a received spectrum image. The spectrum reconstructing unit 13 reconstructs the received spectrum image to generate a reconstructed spectrum image. As a reconstruction method of the spectrum image, for example, deep learning reconstruction using a generative adversarial network (GAN) or a convolutional neural network (CNN) may be used. Note that various algorithms have been developed for the generation and reconstruction methods of spectrum images.

[0018] The determination unit 14 calculates the sum of the difference values ​​(first difference values) for each pixel between the received spectrum image and the reconstructed spectrum image, and determines that there is an abnormality when the sum of the first difference values ​​is greater than a first threshold value. The determination unit 14 also calculates the difference in the level direction (second difference value) between the received spectrum image and the reconstructed spectrum image for each frequency bin, and determines that a frequency bin in which the second difference value is smaller than the second threshold value is a fluctuation due to noise. The first threshold value is related to the pixel difference, and the second threshold value is related to the noise fluctuation, and is set in advance as a design value, a measured value, or the like. The visualization unit 15 colors pixels in which the first difference value exists for frequency bins of the received spectrum image that are determined to be abnormal by the determination unit 14 and are not determined to be a fluctuation due to noise. That is, the visualization unit 15 colors and displays an area in which a pixel difference exists in a frequency bin that is not determined to be a noise fluctuation, in which the pixel difference between the spectrum images is relatively large and the level difference is relatively small. This allows a user of the spectrum monitoring device 1 (such as an operator monitoring a spectrum related to wireless communication) to check the location of the radio wave anomaly by looking at the colored frequency bin area in the spectrum image.

[0019] Next, an example of the operation of the spectrum monitoring device 1 will be described with reference to the flowcharts of Fig. 2 and Fig. 3. Fig. 2 shows the overall processing procedure (steps S11 to S15) of the spectrum monitoring device 1, and Fig. 3 shows the abnormality determination process (steps S141 to S144). In Fig. 2, the receiver 11 of the spectrum monitoring device 1 acquires a reception signal from radio waves received via an antenna (S11). The spectrum image generator 12 calculates spectrum information from the reception signal to generate a reception spectrum image (S12). The spectrum reconstructor 13 reconstructs the reception spectrum image to generate a reconstructed spectrum image (S13). The determiner 14 executes the abnormality determination process according to the difference between the reception spectrum image and the reconstructed spectrum image (S14). The visualizer 15 visualizes the abnormality occurrence location (the area of ​​the frequency bin determined to be abnormal) by coloring it (S15).

[0020] FIG. 3 shows details of the abnormality determination process in step S14, which is executed by the determination unit 14 of the spectrum monitoring device 1. In the abnormality determination process, the sum of the differences (first difference values) for each pixel between the received spectrum image and the reconstructed spectrum image is calculated (S141), and it is determined whether the sum of the first difference values ​​is greater than a first threshold value (S142). If the sum of the first difference values ​​is less than the first threshold value, no abnormality is determined. If the sum of the first difference values ​​is greater than the first threshold value, the difference in the level direction (second difference value) between the received spectrum image and the reconstructed spectrum image is calculated for each frequency bin (S143), and it is determined whether the second difference value is greater than a second threshold value (S144). If the second difference value is less than the second threshold value, even if an abnormality is detected, it is determined to be due to noise fluctuation, and therefore visualization is not performed. If the second difference value is greater than the second threshold, an abnormality other than noise fluctuation is detected, and the frequency bin region related to the abnormality is colored. In other words, if the first difference value detected from the spectrum image is greater than the first threshold and the second difference value is greater than the second threshold, an abnormality is determined, and the process proceeds from step S14 to step S15 to visualize the location of the abnormality.

[0021] Next, the abnormality determination process by the determination unit 14 will be described with reference to Figs. 4 and 5 together with Equations 1 and 2. In step S141 of Fig. 3, the sum of the difference values ​​(first difference values) for each pixel between the received spectrum image and the reconstructed spectrum image is calculated. i , the i-th reconstructed spectrum image is denoted by ^s i Then, the first difference value is expressed by the following formula (1).

[0022]

number

[0023] In formula 1, the symbol r i indicates the sum of the first difference value corresponding to the i-th received spectrum image and the reconstructed spectrum image, and the symbol (m, n) indicates the coordinates of the spectrum image composed of M × N pixels (M and N are natural numbers).

[0024] Step S142 in FIG. 3 determines whether the first difference value (r i ) between pixels of the spectrum images is greater than the first threshold value (Th1), thereby determining the occurrence of abnormalities in the received radio waves due to the presence or absence of interference signals, etc. Specifically, when r i ≧ Th1, it is determined that an abnormality has occurred, and when r i < Th1, it is determined that the received radio wave is normal.

[0025] FIG. 4 illustrates an example of a method for setting the first threshold value Th1 and is a graph showing the frequency distribution of the first difference value. The horizontal axis of the graph in FIG. 4 represents the first difference value r, and the vertical axis represents the frequency of the first difference value r. Two data sets are prepared for machine learning (deep learning), and waveforms indicated by solid lines showing the frequency distribution of normal data and waveforms indicated by dotted lines showing the frequency distribution of abnormal data appear in the frequency distribution. The first difference value r of the abnormal data is higher than that of the normal data. That is, the frequency distribution D1 of the first difference value when the received spectrum data is normal and the frequency distribution D2 of the first difference value when the received spectrum data is abnormal are created on the graph. The first difference value r when the frequency distribution D1 related to the normal data and the frequency distribution D2 related to the abnormal data intersect (overlap) is read and set as the first threshold value Th1. The method for setting the first threshold value Th1 is not limited to the graph in FIG. 4, but it is required to specify a value that can distinguish between normal data and abnormal data and set it as the first threshold value Th1.

[0026] Step S143 in FIG. 3 calculates the difference value (second difference value) in the level direction between the received spectrum image and the reconstructed spectrum image for each frequency bin. When the data sequence of the i-th received spectrum is X i and the data sequence of the i-th reconstructed spectrum is ^X i , the second difference value is represented by Equation (2).

[0027]

Equation

[0028] In Equation 2, the symbol d i represents the second difference value corresponding to the data sequence of the i-th received spectrum and the data sequence of the i-th reconstructed spectrum, and the symbol l represents the frequency bin number. When the received spectrum data is generated from the received signal by discrete Fourier transform of L points (L is a natural number), the frequency bin number is represented by l = 0, 1, …, L−1.

[0029] Step S144 in FIG. 3 determines, for each frequency bin, whether the abnormality of the received radio wave is caused by noise by determining whether the second difference value (d i ) in the level direction between the spectrum images is greater than the second threshold value (Th2). Specifically, it is determined that it is noise fluctuation when d i (l)<Th2, and it is determined that it is an abnormality of the received radio wave caused by an interference signal or the like rather than noise fluctuation when d i (l)≧Th2.

[0030] FIG. 5 illustrates an example of a method for setting the second threshold value Th2 and is a graph showing the frequency distribution of the second difference value. The horizontal axis of the graph in FIG. 5 represents the second difference value d, and the vertical axis represents the frequency of the second difference value d. Two data sets are prepared for machine learning (deep learning), and waveforms indicated by solid lines showing the frequency distribution of data including only noise fluctuations and waveforms indicated by dotted lines showing the frequency distribution of data including interference signals appear in the frequency distribution. The data with only noise fluctuations has a lower second difference value d than the data including interference signals. That is, the frequency distribution D3 of the second difference value of the received spectrum data with only noise fluctuations and the frequency distribution D4 of the second difference value of the received spectrum data including interference signals are created on the graph. The second threshold value Th2 is set by reading the second difference value d when the frequency distribution D3 related to noise fluctuations and the frequency distribution D4 related to interference signals intersect (overlap). The method for setting the second threshold value Th2 is not limited to the graph in FIG. 5, but it is required to specify a value that can distinguish the data related to noise fluctuations from the data related to interference signals and set it as the second threshold value Th2.

[0031] [Second embodiment] Next, a second embodiment will be described with reference to Fig. 6 to Fig. 8. Fig. 6 is a block diagram of a spectrum monitoring device 2 according to the second embodiment. The spectrum monitoring device 2 shown in Fig. 6 further includes a storage unit 21 and a learning unit 22 for generating a learning model used in the spectrum reconstruction unit 13, in addition to the configuration of the spectrum monitoring device 1 shown in Fig. 1. The spectrum monitoring device 2 includes a receiving unit 11 that realizes the same functions as those of the spectrum monitoring device 1, a spectrum image generating unit 12, a spectrum reconstruction unit 13, a determining unit 14, and a visualizing unit 15. Therefore, the storage unit 21 and the learning unit 22 will be described in detail.

[0032] The storage unit 21 stores the received spectrum image generated by the spectrum image generation unit 12. The learning unit 22 learns data of the received spectrum image to generate a learned model and outputs it to the spectrum reconstruction unit 13.

[0033] Next, an example of the operation of the spectrum monitoring device 2 will be described with reference to Fig. 7 and Fig. 8. Fig. 7 is a flowchart (steps S21, S22) showing the processing procedure of the spectrum monitoring device 2, and Fig. 8 is a flowchart (S211 to S216) showing details of the model learning in step S22 of Fig. 7. As shown in Fig. 7, the spectrum monitoring device 2 performs model learning (S21) and inference (anomaly detection processing) (S22) using the learned model.

[0034] 8 shows details of model learning (S21) of the spectrum monitoring device 2. In the spectrum monitoring device 2, the receiver 11 acquires a received signal from radio waves of wireless communication (S211), the spectrum image generator 12 generates a received spectrum image from the received signal (S212), and the storage unit 21 stores the received spectrum image (S213). After that, the learning unit 22 reads out the received spectrum image from the storage unit 21 as learning data (S214), performs learning using the learning data (S215), generates a trained model, and outputs it to the spectrum reconstruction unit 13 (S216).

[0035] The inference (anomaly detection process) using the trained model in the spectrum monitoring device 2 according to the second embodiment is similar to the anomaly determination process (steps S11 to S15 in FIG. 2) in the spectrum monitoring device 1 according to the first embodiment, but is different in that the spectrum reconstruction unit 13 reconstructs a received spectrum image using the trained model generated by the learning unit 22. In addition, as a machine learning algorithm for generating the trained model in the learning unit 22, a principal component analysis used for reducing the dimension of features obtained from image analysis may be used, or an algorithm for compressing input data to leave only important features and then restoring the data to the original dimension may be used. Alternatively, an algorithm having an autoencoder structure such as GANomaly (Non-Patent Document 2), which is known as an anomaly detection method based on a generative adversarial network (GAN), may be used.

[0036] [Third embodiment] The spectrum monitoring device 2 according to the second embodiment includes a storage unit 21 and a learning unit 22 in addition to the configuration of the spectrum monitoring device 1 according to the first embodiment, but when spectrum monitoring devices are installed at multiple sites, it is not reasonable to provide a learning unit at each site in terms of site space and cost. Therefore, it is possible to collect received spectrum images from the spectrum monitoring device installed at each site, generate a learning model corresponding to each site individually, and distribute the learning model adapted to each of the multiple sites.

[0037] In the third embodiment, the storage unit 21 and the learning unit 22 that realize the function of generating a learning model used in the spectrum reconstruction unit 13 in the spectrum monitoring device 2 shown in Fig. 6 are separated as a model learning device. This is an effective configuration when the model learning device is installed in one place and multiple spectrum monitoring devices are installed at multiple sites where spectrum monitoring of wireless communication is required. In other words, the model learning device and the spectrum monitoring devices installed at multiple sites are connected to a network, and an operator uses an operator console to remotely control the model learning device and the spectrum monitoring devices.

[0038] 9 is a block diagram showing a configuration of a spectrum monitoring system 3 according to a third embodiment. The spectrum monitoring system 3 includes a plurality of spectrum monitoring devices 100 installed at a plurality of sites, and a model learning device 200 that is communicatively connected to the plurality of spectrum monitoring devices 100 via a network. The spectrum monitoring system 3 is obtained by dividing the configuration of the spectrum monitoring device 2 according to the second embodiment and communicatively connecting the components. The spectrum monitoring device 100 includes a receiving unit 11, a spectrum image generating unit 12, a spectrum reconstruction unit 13, a determining unit 14, and a visualizing unit 15, and the model learning device 200 includes a storage unit 21 and a learning unit 22. In the model learning device 200, the storage unit 21 receives and stores a received spectrum image generated by the spectrum image generating unit 12, and the learning unit 22 transmits a learning model generated by learning using the learning data to the spectrum reconstruction unit 13.

[0039] Fig. 10 is a sequence diagram (steps S301 to S303) showing the processing procedure of the spectrum monitoring system 3 according to the third embodiment, and shows interactions between an operator, the spectrum monitoring device 100, and the model learning device 200. The sequence diagram in Fig. 10 is composed of two areas corresponding to the flowchart in Fig. 7, where area A corresponds to model learning (S21) and area B corresponds to inference using a trained model (anomaly detection processing) (S22).

[0040] In order to carry out processing equivalent to model learning in area A of FIG. 10, the operator instructs the model learning device 200 to collect learning data, and in response to the instruction, the model learning device 200 transmits a data collection instruction to the spectrum monitoring device 100. The spectrum monitoring device 100 transmits a received spectrum image generated by the spectrum image generating unit 12 to the model learning device 200, and the model learning device 200 stores the received spectrum image in the storage unit 21 (S301). The storage unit 21 of the model learning device 200 sequentially stores a plurality of received spectrum images. Then, when the spectrum monitoring device 100 completes collection of the number of received spectrum images designated by the operator, it sends a report of data collection completion to the model learning device 200. After that, the model learning device 200 notifies the operator of a report of completion of the learning data. The operator sends an instruction to execute learning to model learning device 200, and learning unit 22 of model learning device 200 executes learning processing using the learning data to generate a trained model (S302). Model learning device 200 notifies the operator of the completion of learning and transfers the trained model to spectrum monitoring device 200.

[0041] In order to perform processing equivalent to inference (anomaly detection processing) using the trained model in area B of Fig. 10, the operator sends a spectrum monitoring instruction to the spectrum monitoring device 100, and the spectrum monitoring device 100 executes the anomaly detection processing (S303). Specifically, the spectrum reconstruction unit 13 applies the received spectrum image to the trained model to generate a reconstructed spectrum image, the determination unit 14 performs anomaly determination using the received spectrum image and the reconstructed spectrum image, and when an anomaly is detected, the visualization unit 15 visualizes the anomaly occurrence location in the spectrum image by coloring it. The spectrum monitoring device 100 notifies the operator of the visualization result of the anomaly occurrence location.

[0042] 10 shows the spectrum monitoring device 100 and the model learning device 200 for one site, but when spectrum monitoring devices 100 for multiple sites are provided, the model learning device 200 may collect received spectrum images for the multiple sites from the multiple spectrum monitoring devices 100, execute learning processing, and generate multiple trained models to be applied individually to the multiple spectrum monitoring devices 100. Also, multiple model learning devices 200 may be provided to generate multiple trained models for the multiple spectrum monitoring devices 100. By performing the sequence shown in FIG. 10 in the spectrum monitoring system 3 as shown in FIG. 9, it is possible to generate a trained model using the trained data and detect anomalies using the spectrum images as described above, even when the model learning device 200 and the spectrum monitoring device 100 are installed remotely.

[0043] In the above-mentioned embodiment, the spectrum monitoring device implements the above-mentioned functions by executing computer programs such as a spectrum monitoring program and an anomaly determination program (or anomaly detection program) using a processor (CPU, GPU, etc.), but the computer programs may be stored in a storage medium (semiconductor memory such as ROM or RAM), or may be distributed from another device. Also, the computer program may be for realizing part of the above-mentioned functions, or may be a so-called difference file (or difference program) that can realize the above-mentioned functions in combination with a program already recorded in the computer system.

[0044] Next, the features and effects of the present invention will be described. In the present invention, in a spectrum monitoring device (or an information processing device implementing a spectrum monitoring method) that monitors received data obtained by receiving radio waves from a radio wave source such as a transmitting device, in addition to determining that there is a radio wave abnormality (or radio wave interference) when the sum of the difference value (first difference value) for each pixel between the received spectrum image and the reconstructed spectrum image exceeds a first threshold, the difference value (second difference value) in the level direction (amplitude direction) between the received spectrum image and the reconstructed spectrum image is calculated for each frequency bin, and a frequency bin where the second difference value is lower than the second threshold is determined not to be a fluctuation due to noise, thereby making it possible to visualize the location where an interference signal occurs while reducing the influence of noise fluctuation.

[0045] In the conventional method, even if it is possible to detect the occurrence of a radio wave abnormality (radio wave interference) at a certain time, it is not possible to identify which frequency bin the radio wave abnormality occurred in. Even if it is possible to visualize the frequency band in which the radio wave abnormality occurred, the radio wave fluctuation due to noise is also visualized, making it impossible to specifically identify the frequency bin in which the radio wave abnormality occurred. With the configuration and operation of this embodiment, it is possible to visualize the location of the interference signal after reducing the influence of the noise fluctuation, in other words, it is possible to eliminate the difficulty in distinguishing between the location of the noise and the location of the interference signal due to the visualization of the noise fluctuation.

[0046] The present invention is not limited to the above-described embodiments, and the configuration and processing procedure of the spectrum monitoring device are not limited to the first to third embodiments (FIGS. 1 to 3, 6 to 10), and the method of setting the threshold value is not limited to the first embodiment (FIGS. 4 to 5). In FIG. 1, it is not necessary to mount the receiving unit 11 equipped with an antenna on the spectrum monitoring device 1, and received data may be input from an external receiving device, and the spectrum image generating unit 12, spectrum reconstructing unit 13, and determining unit 14 may be integrated into one unit.

[0047] 11 is a block diagram of the spectrum monitoring device 4 having the minimum configuration of the above-mentioned embodiment, and includes an abnormality determination unit 41 and a visualization unit 42. The abnormality determination unit 41 receives reception data corresponding to radio waves from an external radio wave generation source (such as a wireless transmission device), performs a Discrete Fourier Transform (DFT) to calculate spectrum information, generates a spectrum image from the spectrum information, and generates a reconstructed spectrum image by reconstruction using deep learning reconstruction or a learning model such as a Generative Adversarial Network (GAN) or a Convolutional Neural Network (CNN). Thereafter, the abnormality determination unit 41 calculates the sum of the difference values ​​(first difference values) for each pixel between the received spectrum image and the reconstructed spectrum image, and determines an abnormality when the sum of the first difference values ​​is greater than a first threshold value. Furthermore, the abnormality determination unit 41 calculates the difference in the level direction (second difference value) between the received spectrum image and the reconstructed spectrum image for each frequency bin, and determines that the frequency bins in which the second difference value is smaller than the second threshold value are fluctuations due to noise. As a result, the frequency bin region in which the sum of the first difference values ​​is larger than the first threshold value and the second difference value is larger than the second threshold value is determined to be an abnormal region. The visualization unit 42 visualizes the abnormal region of the frequency bins of the received spectrum image by coloring them. As a coloring method, a color (e.g., red) that can be distinguished from the image of the frequency bin may be applied, or a different display mode (dashed line, shading, etc.) may be applied.

[0048] 12 is a flowchart showing the processing procedure of the spectrum monitoring device 4 (steps S401 to S402). The spectrum monitoring device 4 executes two processing steps: abnormality determination (S401) and visualization (S402). First, a reception spectrum image generated from reception data and a reconstructed spectrum image obtained by reconstructing the reception spectrum image are prepared, and a frequency bin region in which a first difference value between the reception spectrum image and the reconstructed spectrum image is greater than a first threshold value and a second difference value is greater than a second threshold value is determined to be an abnormal region. Then, the abnormal region of the frequency bin is visualized by coloring it in the reception spectrum image.

[0049] Finally, in the above-mentioned embodiment, radio wave monitoring is performed by a spectrum diagram to detect radio wave abnormalities (radio wave interference) in wireless communication, but various wireless communication methods can be applied. For example, the present invention can be applied to software radio in which communication parameters can be changed by software, cognitive radio compatible with multiple wireless methods (Wi-Fi, LTE, etc.), and wireless communication using dynamic spectrum sharing (DSS: Dynamic Spectrum Sharing) that realizes dynamic frequency allocation in milliseconds between LTE and 5G services. The present invention also includes various modified examples, design changes, and modifications within the technical scope defined in the appended claims. [Industrial Applicability]

[0050] The present invention detects radio wave anomalies (radio wave interference) in important wireless communications such as police radio, fire radio, aviation radio, and railway radio, and visualizes them in a spectrum image. However, the present invention is not limited to important wireless communications and can also be applied to telecommunications carriers' communication systems (LTE, 5G, etc.), broadcasting carriers' terrestrial communication equipment, satellite communication equipment, etc. [Explanation of symbols]

[0051] 1, 2, 4, 100 Spectrum Monitoring Equipment 3 Spectrum Monitoring System 11 Receiving section 12 Spectral image generator 13 Spectrum reconstruction unit 14 Judgment section 15 Visualization section 21 Storage Department 22 Learning Department 41 Abnormality determination section 42 Visualization section 200 Model Learning Device

Claims

1. a receiving unit that receives radio waves transmitted wirelessly and generates a received signal; a spectrum image generating unit that calculates spectrum information from the received signal to generate a received spectrum image; a spectrum reconstruction unit that reconstructs the received spectrum image through learning to generate a reconstructed spectrum image; a determination unit that determines the presence or absence of an interference signal based on a difference between the received spectrum image and the reconstructed spectrum image; a visualization unit that visualizes a frequency band in which the interference signal exists when the determination unit determines that the interference signal exists; A spectrum monitoring device comprising:

2. 2. The spectrum monitoring device according to claim 1, wherein the determination unit determines an abnormality when a sum of first difference values ​​for each pixel between the reception spectrum image and the reconstructed spectrum image is greater than a first threshold value, calculates a second difference value in a level direction between the reception spectrum image and the reconstructed spectrum image for each frequency band, and determines that the frequency band in which the second difference value is greater than a second threshold value is not due to noise fluctuation, and the visualization unit visualizes the frequency band that is determined to be abnormal by the determination unit and is also determined to be not due to noise fluctuation.

3. 3. The spectrum monitoring device according to claim 2, wherein the first threshold is calculated from the first difference value when a first frequency distribution associated with first spectrum data determined to be normal when generating the learning model for learning and a second frequency distribution associated with second spectrum data determined to be abnormal overlap, and the second threshold is calculated from the second difference value when a third frequency distribution associated with third spectrum data due only to noise fluctuations and a fourth frequency distribution associated with fourth spectrum data including the interference signal overlap.

4. Receives the wirelessly transmitted radio waves and generates a received signal; Calculating spectrum information from the received signal to generate a received spectrum image; Reconstructing the received spectrum image through learning to generate a reconstructed spectrum image; determining whether or not an interference signal exists based on a difference between the received spectrum image and the reconstructed spectrum image; When the presence of the interfering signal is determined, the frequency band in which the interfering signal exists is visualized.

5. A program implemented in a computer of a spectrum monitoring device, Receives the wirelessly transmitted radio waves and generates a received signal; Calculating spectrum information from the received signal to generate a received spectrum image; Reconstructing the received spectrum image through learning to generate a reconstructed spectrum image; determining whether or not an interference signal is present based on a difference between the received spectrum image and the reconstructed spectrum image; When the presence of the interference signal is determined, a frequency band in which the interference signal exists is visualized.

6. an abnormality determination unit that determines an abnormal region of a frequency band based on a difference between a reception spectrum image generated from a reception signal related to a wirelessly transmitted radio wave and a reconstructed spectrum image generated by reconstructing the reception spectrum image through learning; a visualization unit that visualizes the abnormal region in the frequency band; A spectrum monitoring device comprising:

7. determining an abnormal region in a frequency band based on a difference between a reception spectrum image generated from a reception signal of the wirelessly transmitted radio wave and a reconstructed spectrum image generated by reconstructing the reception spectrum image through learning; A spectrum monitoring method for visualizing the anomalous regions in the frequency band.

8. A program implemented in a computer of a spectrum monitoring device, the program determining an abnormal region of a frequency band based on a difference between a reception spectrum image generated from a reception signal of a wirelessly transmitted radio wave and a reconstructed spectrum image generated by reconstructing the reception spectrum image through learning, A program that visualizes the abnormal region in the frequency band.

Citation Information

Patent Citations

  • Transmission device collation device, transmission device collation method, and program

    JP2021180388A

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

    JP2022182844A

  • Abnormality detection device, abnormality detection method, and non-transitory computer-readable medium

    WO2021176805A1

Cited By

  • Identification method based on wireless spectrum feature position statistics

    CN121078441A

  • Electromagnetic spectrum abnormal signal detection method and system

    CN122119802A

  • Abnormality management device, abnormality management method, and abnormality management system

    JP7714828B1