Method and apparatus for generating communication interference extension data, program product, and training method

Abstract

To improve accuracy of classification of an artificial intelligence model.SOLUTION: A method for generating communication interference extension data includes: acquiring communication interference detection data including distributions of a plurality of detection data, the communication interference detection data being used for training an artificial intelligence model for communication interference classification together with a corresponding interference type, a distribution of each of the detection data representing a change of communication interference power in a plurality of frequency ranges in one sampling period; and generating one or more distributions of extension data based on a change trend of distributions of the plurality of detection data, the one or more distributions of the extension data being used for training the artificial intelligence model for communication interference classification together with the interference type.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to a method and apparatus for generating communication interference augmented data, a computer program product, and a method for training an artificial intelligence model for communication interference classification. [Background technology]

[0002] In mobile network communications, with the increasing number of users and connected devices, network interference is becoming a more prominent issue, seriously affecting network performance and user experience. However, there are already technologies that use artificial intelligence models to automatically classify interference types. Summary of the Invention [Problem to be solved by the invention]

[0003] However, artificial intelligence models often require a large amount of training data to improve classification accuracy, and collecting and marking the training data requires enormous human and material resources. [Means for solving the problem]

[0004] The present disclosure has been made in view of the above-mentioned problems. An object of the present disclosure is to provide a method and apparatus for generating communication interference extended data, a computer program product, and a training method using an artificial intelligence model for communication interference classification, which automatically generate multiple pieces of communication interference extended data as training data for an artificial intelligence model for communication interference classification based on existing communication interference detection data, thereby improving the classification accuracy of the artificial intelligence model for communication interference classification.

[0005] According to one aspect of the present disclosure, there is provided a method for generating communication interference extended data, the method including: acquiring communication interference detection data including a distribution of a plurality of pieces of detection data, the communication interference detection data being used together with corresponding interference types to train an artificial intelligence model for classifying communication interference, wherein each of the detection data distributions represents a change in communication interference power across a plurality of frequency ranges during one sampling period; and generating one or more extended data distributions as communication interference extended data based on the tendency of change among the distributions of the plurality of detection data, the communication interference extended data being used together with interference types to train an artificial intelligence model for classifying communication interference.

[0006] According to one aspect of the present disclosure, there is provided a method for training an artificial intelligence model for classifying communication interference, the method including: generating communication interference extension data using the above-mentioned method for generating communication interference extension data; and training an artificial intelligence model for classifying communication interference using the communication interference detection data, the communication interference extension data, and the corresponding interference type as training data.

[0007] According to one aspect of the present disclosure, there is provided a communication interference extension data generation device including a processor and a memory in which one or more computer programs are stored, and when the one or more computer programs are executed by the processor, the above-mentioned communication interference extension data generation method is executed by the processor.

[0008] According to one aspect of the present disclosure, there is provided a computer program product including a computer program that, when executed by a processor, performs the method for generating communication interference extension data. [Effects of the Invention]

[0009] According to the method and device for generating communication interference extension data, computer program product, and method for training an artificial intelligence model for classifying communication interference provided by the present disclosure, communication interference extension data is automatically generated by applying appropriate processing according to different change trends between the distributions of multiple detection data corresponding to different interference types, and the generated communication interference extension data is used together with the interference type to train an artificial intelligence model for classifying communication interference, thereby improving the classification accuracy of the artificial intelligence model for classifying communication interference. [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 1 is a schematic diagram illustrating a method for generating communication interference enhancement data according to an embodiment of the present disclosure. [Figure 2] FIG. 2 is a diagram illustrating an example of the configuration of communication interference detection data according to an embodiment of the present disclosure. [Figure 3] FIG. 3 is a diagram illustrating an example of distribution of detection data according to an embodiment of the present disclosure. [Figure 4] FIG. 4 is a diagram illustrating another example of the distribution of detection data according to an embodiment of the present disclosure. [Figure 5] FIG. 5 is a diagram illustrating an example of a distribution of communication interference extension data generated when the distribution of detection data according to an embodiment of the present disclosure has a first tendency. [Figure 6] FIG. 6 is a diagram illustrating an example of a distribution of extended data generated when the distribution of detection data according to an embodiment of the present disclosure has a first trend and has an intersection. [Figure 7] FIG. 7 is a diagram illustrating an example of the distribution of extended data generated when the distribution of detection data according to an embodiment of the present disclosure has the second tendency. [Figure 8] FIG. 8 is a diagram illustrating another example of the distribution of extended data generated when the distribution of detection data has the second tendency according to an embodiment of the present disclosure. [Figure 9] FIG. 9 is a diagram illustrating the evaluation results of the classification accuracy of a classifier trained using communication interference extended data generated by a method according to an embodiment of the present disclosure. [Figure 10] FIG. 10 is a schematic diagram illustrating the configuration of an apparatus for generating communication interference extension data according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, embodiments of the present disclosure will be described in more detail with reference to the drawings. Although the drawings illustrate several embodiments of the present disclosure, the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are merely illustrative and are not intended to limit the scope of protection of the present disclosure.

[0012] As used herein, the term "comprises" and variations thereof are intended to be open inclusive, meaning "including, but not limited to." The term "based on" means "based at least in part on." The term "in one embodiment" means "at least one embodiment," the term "in another embodiment" means "at least one other embodiment," and the term "in some embodiments" means "at least some embodiments." Definitions relating to other terms are provided below.

[0013] It should be understood that the concepts of "first," "second," etc. referred to in this disclosure are used only to distinguish between different devices, modules, units, models, data, etc., and are not used to limit the order in which functions are performed by these devices, modules, units, models, data, or the order in which they are produced, or the interdependencies between them.

[0014] It will be appreciated by those skilled in the art that the modifiers "a" and "plurality" referred to in this disclosure are intended to be illustrative rather than limiting, and should be understood as "one or more" unless the context clearly indicates otherwise.

[0015] In a wireless communication system, a physical channel for transmitting signals from a mobile station (terminal) to a base station is called an uplink, and a physical channel for transmitting signals from a base station to a mobile station (terminal) is called a downlink. Here, various types of interference may exist in the uplink or downlink transmission of signals. In the following description, uplink interference is used as an example, but it should be noted that the embodiments of the present disclosure are not limited to uplink interference. In this disclosure, uplink interference is classified into three types: internal interference, external interference, and mixed internal and external interference. Internal interference refers to interference caused by factors internal to a wireless communication system, and external interference refers to interference caused by factors external to a wireless communication system. Internal interference and external interference may each be further subdivided into multiple interference types. For example, internal interference includes interference types such as Telecom 800M interference (interference in the 800M operating frequency specific to China Telecom), false base station interference, NR system interference, LTE D-band interference, and 700M interference. Additionally, external interference includes interference types such as intelligent street lamp interference, wireless bridge / video surveillance interference, external interference, jammers, routers, clock source failures, radio / television signal interference, and other interference. Interference that combines multiple interference types, including internal interference and external interference, is referred to as mixed internal / external interference. The list of interference categories and interference types related to interference data is an example, and the present disclosure is not limited to the above-mentioned interference types and interference types.

[0016] When using a supervised artificial intelligence model to automatically determine the category and / or type of uplink interference, the artificial intelligence model must be trained in advance using communication interference detection data. For example, the communication interference detection data may be Northbound Interface (NBI) interference data for a certain region over a certain period of time, and the corresponding labels (i.e., interference categories and / or types) may be generated by a frontline work order and manually verified, or may be fully labeled manually. Collecting and labeling communication interference detection data requires the consumption of human and material resources. A lack of communication interference detection data may affect the classification accuracy of the artificial intelligence model.

[0017] FIG. 1 is a schematic diagram illustrating a method for generating communication interference enhancement data according to an embodiment of the present disclosure.

[0018] A method 100 for generating communication interference extended data according to an embodiment of the present disclosure includes steps S110 and S120. In step S110, communication interference detection data can be obtained. The communication interference detection data includes a plurality of detection data distributions, which, together with corresponding interference types, can be used to train an artificial intelligence model for communication interference classification. Each detection data distribution can represent the change in communication interference power across multiple frequency ranges during a sampling period. In step S120, one or more extended data distributions can be generated as communication interference extended data based on the change trends among the plurality of detection data distributions. The communication interference extended data, together with interference types, can train an artificial intelligence model for communication interference classification.

[0019] The communication interference detection data is obtained by collecting actual communication interference data, and the extended communication interference data is automatically generated based on the communication interference detection data using the method 100 according to the embodiment of the present disclosure. The applicant discovered that for a specific interference type, there is a specific change trend between the distributions of the detection data. Therefore, the applicant proposes generating an extended data distribution by performing processing corresponding to the change trend between the distributions of multiple detection data for the specific interference type. In this way, the generated extended data distribution can satisfy the change trend between the distributions of existing detection data for the specific interference type and, therefore, the characteristics of actual communication interference detection data for the specific interference type, thereby improving the likelihood that the communication interference extended data corresponding to the distribution of the extended data will be used. Furthermore, mixing the extended communication interference data with existing communication interference detection data increases the amount of training data in the training dataset used to train an artificial intelligence model for communication interference classification, thereby improving the accuracy of the artificial intelligence model's communication interference classification.

[0020] As an example, the communication interference detection data in the present disclosure is data obtained by measuring the total power of the noise floor and interference from adjacent cells for each uplink PRB, using a frequency range of 273 bands of physical resource blocks (PRBs) in the 2515 MHz to 2675 MHz frequency band, a sampling period of one second (an example of a sampling period), and defining the average value of the interference power of each PRB in the sampling period as the communication interference power corresponding to that PRB. In the present disclosure, the communication interference detection data includes multiple data obtained by sampling for multiple sampling periods. FIG. 2 is a diagram illustrating an example of the configuration of communication interference detection data according to an embodiment of the present disclosure. In FIG. 2, the leftmost column indicates the index (number) of the communication interference detection data, and each index of the communication interference detection data corresponds to one sampling period. The top row indicates the index of the PRB (frequency range), and the number at the intersection of the row and column indicates the average value of the interference power detected in the PRB (frequency range) corresponding to the index of the PRB in the sampling period corresponding to the index of the communication interference detection data. In the following, unless otherwise specified, the terms frequency range and PRB may be interpreted interchangeably. The above-described communication interference detection data is merely an example, and it is clear that the sampling frequency band, the number of frequency ranges, the sampling period (cycle), the interference category / type, and the like of the communication interference detection data in the present disclosure are not limited to these. Furthermore, communication interference power is not limited to being expressed by the average value of interference power, but can be expressed in various possible ways, such as the maximum value, minimum value, and median value of interference power. An example of expressing communication interference power by the average value of interference power will be described below.

[0021] Each piece of communication interference detection data may correspond to a distribution of one piece of detection data. For one piece of interference detection data, the corresponding distribution of detection data represents the change in the average value of interference power within one sampling period across multiple frequency ranges. In the present disclosure, the distribution of detection data can be represented in a coordinate system in which each frequency range of communication interference detection data is represented on the horizontal axis and the average value of interference power corresponding to the frequency range is represented on the vertical axis. In the following description, "the distribution of detection data corresponding to communication interference detection data" has the same meaning as "the distribution of detection data of communication interference detection data."

[0022] 3 and 4 are diagrams illustrating an example of the distribution of detection data according to an embodiment of the present disclosure. In Fig. 3 and Fig. 4, the horizontal axis represents the frequency range, and the value range is an integer from 0 to 272 (i.e., an index of 273 frequency ranges), and the vertical axis represents the average value of interference power (unit: decibel milliwatts (dBm)).

[0023] FIG. 3 shows detection data distributions a and b for two pieces of communication interference detection data in which the type of interference is telecommunication 800M interference. In FIG. 3, detection data distribution a is shifted upward relative to detection data distribution b. That is, an approximation of detection data distribution a can be obtained by shifting detection data distribution b upward. FIG. 4 shows detection data distributions c and d for two pieces of communication interference detection data in which the type of interference is router. In FIG. 4, detection data distribution c is shifted left relative to detection data distribution d. That is, an approximation of detection data distribution c can be obtained by shifting detection data distribution d left. The detection data distributions make it possible to intuitively represent the change in communication interference power (e.g., the average value of interference power) of each piece of communication interference detection data over one sampling period depending on the frequency range, as well as the tendency of change between multiple detection data distributions.

[0024] In the present disclosure, a change trend can represent the relative positional relationship in a coordinate system of multiple detection data distributions belonging to the same interference category. In the present disclosure, a change trend in which detection data distribution a as shown in FIG. 3 shifts upward relative to another detection data distribution b is referred to as a first trend, and a change trend in which detection data distribution c as shown in FIG. 4 shifts leftward relative to another detection data distribution d is referred to as a second trend. Furthermore, the first trend may be a downward shift of detection data distribution b relative to detection data distribution a, and the second trend may be a rightward shift of detection data distribution d relative to detection data distribution c. In other words, the first trend may include an upward or downward shift of different detection data distributions, and the second trend may include a leftward or rightward shift of different detection data distributions.

[0025] Note that the first tendency does not mean that the communication interference power (e.g., the average value of the interference power) corresponding to the distribution of one piece of detection data is strictly greater (or smaller) than the communication interference power corresponding to the distribution of other piece of detection data across the entire frequency range, but means that there is a possibility that one or more intersections exist between the distributions of the detection data.Furthermore, the second tendency does not mean that the distribution of detection data of one piece of communication interference detection data, after shifting to the left or right, strictly overlaps with the distribution of detection data of other piece of communication interference detection data.

[0026] As described above, in step S120, one or more distributions of extended data may be generated as communication interference extended data based on the change tendency between the distributions of a plurality of pieces of detection data. Hereinafter, how to generate the distribution of extended data when the change tendency between the distributions of detection data is the first tendency will be described.

[0027] For a certain type of interference, if the tendency of change between the distributions of detection data is a first tendency, two distributions of detection data can be randomly selected from the plurality of distributions of detection data, and an extended data distribution can be generated based on the interpolated value of the selected two distributions of detection data.

[0028] That is, in step S120, if the trend of change between the detection data distributions is a first trend, generating one or more extended data distributions may include randomly selecting two detection data distributions from the multiple detection data distributions and calculating an interpolated value of the selected two detection data distributions.

[0029] Here, generating a distribution of extended data based on an interpolated value of the distribution of the two selected detection data may involve performing an interpolation operation on the average value of the interference power corresponding to the same frequency range of the distribution of the two selected detection data, and using the result of the interpolation operation as the average value of the corresponding interference power in that frequency range of the distribution of the extended data.

[0030] If it is desired to generate only one extended data distribution, it is sufficient to perform the interpolation calculation process only once for the two selected detection data distributions. If it is desired to generate multiple extended data distributions, it is possible to perform the interpolation calculation multiple times for the two selected detection data distributions (using different weights in the following formula (1)), or it is possible to select two different detection data distributions multiple times and perform the interpolation calculation once or multiple times for the two selected detection data distributions each time.

[0031] Specifically, the interpolated value of the distribution of the two pieces of detection data may be calculated using the following formula (1).

number

[0032] where y represents the distribution of the generated extended data, i represents the index of the frequency range (i.e., the index of the PRB), a and b represent the distribution of the selected two detected data, and weight is a predefined weight. For example, weight may be generated randomly. Weight can be a positive or negative number according to needs.

[0033] FIG. 5 is a diagram illustrating an example of an extended data distribution generated when the detection data distribution according to an embodiment of the present disclosure has a first trend. In FIG. 5, detection data distributions a and b of two randomly selected communication interference detection data, the type of interference being telecommunication 800M interference, are indicated by solid lines, and three extended data distributions y1, y2, and y3 generated based on the detection data distributions a and b and the above-described formula (1) are indicated by dashed-dotted lines. In FIG. 5, the extended data distributions y1, y2, and y3 are generated based on the same detection data distributions a and b, but differ depending on the weight values ​​in the above-described formula (1). For example, since the detection data distribution a shown in FIG. 5 is shifted upward relative to b, the top-most extended data distribution y1 is obtained by assigning a negative weight to the extended data distributions y2 and y3. The extended data distribution y2 is obtained by assigning a positive and small weight to the extended data distributions y2 and y3, and the extended data distribution y3 is obtained by assigning a positive and large weight to the extended data distributions y2 and y3.

[0034] As shown in Figure 5, by generating an extended data distribution using interpolated values ​​for the distribution of detection data with a first trend, the generated extended data distribution can meet the characteristics of the actual communication interference detection data for that interference type. For example, in Figure 5, the generated extended data distributions y1, y2, and y3 have characteristics similar to the existing detection data distributions a and b, i.e., the average value of the interference power on the low frequency side changes less as the frequency range increases, and the average value of the interference power on the high frequency side changes more as the frequency range increases. This improves the likelihood that the interference extended data will be used when training an artificial intelligence model for communication interference classification. Specific evaluation results will be described later.

[0035] As mentioned above, the first tendency does not mean that one communication interference detection data is strictly larger or smaller than the other communication interference detection data across the entire frequency range, but means that there may be an intersection point in the distribution of the detection data of the two communication interference detection data. In this case, b in the above formula (1) i -a iThe sign of changes at the intersection point. i -a i When the distribution of extended data is generated using Equation (1) for positive and negative changes in , the distribution of the generated extended data may not meet the characteristics of the actual communication interference detection data of that interference type, which will ultimately affect the accuracy of the artificial intelligence model in communication interference classification.

[0036] Therefore, if there is an intersection between two detection data distributions a and b randomly selected from multiple detection data distributions, b in the above formula (1) i -a i Different processing can be applied when is positive or negative. For example, the interpolated value may be calculated based on the following formula (2):

number

[0037] The meaning of each parameter is the same as in equation (1), and weight takes a negative value.

[0038] 6 is a diagram illustrating an example of the distribution of extended data generated when the distribution of detected data according to an embodiment of the present disclosure has a first trend and has an intersection. In FIG. 6, the distributions a and b of detected data of two randomly selected communication interference detection data, whose interference type is telecommunication 800M interference, are shown by solid lines, and the detected data distributions a and b intersect at frequency range indexes i1 and i2. y is the distribution of extended data generated based on the detected data distributions a and b and the above formula (2), and is shown by a dashed line. In FIG. 6, in the frequency range less than i1 and greater than i2, a i Ga b i Since it is less than or equal to y i is a i (In Figure 6, y i is a i and is represented by a(y), in the frequency range greater than or equal to i1 and less than or equal to i2, i A i Since it is smaller than y, we take a negative number as weight and use the y generated by the above equation (2). iis a i It becomes bigger.

[0039] When an intersection exists between the distributions of detection data of two pieces of communication interference detection data, by generating a distribution of extended data using the above method, the generated distribution of extended data can satisfy the characteristics of the actual communication interference detection data of that interference type.In the following description, the mode of generating a distribution of extended data by calculating an interpolated value of the distribution of detection data using equation (1) is called the basic interpolation mode, and the mode of generating a distribution of extended data by calculating an interpolated value of the distribution of detection data where an intersection exists using equation (2) is called the optimized interpolation mode.

[0040] The above has described a method for generating communication interference extension data when the trend of changes between the distributions of detection data is the first trend. Below, a method for generating communication interference extension data when the trend of changes between the distributions of detection data is the second trend will be described.

[0041] For a certain interference type, if the change tendency between the detection data distributions is the second tendency, one detection data distribution can be randomly selected from the multiple detection data distributions, and the selected detection data distribution can be rolled or shifted to the left or right to generate an extended data distribution.

[0042] That is, in step S120, if the trend of change between the detection data distributions is the second trend, generating one or more extended data distributions may include randomly selecting one detection data distribution from the multiple detection data distributions, and rolling-shifting the selected one detection data distribution to the left or right.

[0043] Here, by rolling-shifting the distribution of one selected detection data to the left or right to generate the distribution of extended data, the average value of interference power corresponding to a certain frequency range of the distribution of one selected detection data can be set to the average value of interference power corresponding to another frequency range to the left or right of the frequency range of the distribution of extended data.

[0044] A rolling shift to the left or right both refers to a cyclic rolling shift. For example, when the distribution of detected data undergoes a rolling shift to the left, its low-frequency side may exceed the minimum value of the frequency range, and in such cases, the exceeded part becomes the high-frequency side after the rolling shift. Similarly, when the distribution of detected data undergoes a rolling shift to the right, its high-frequency side may exceed the maximum value of the frequency range, and in such cases, the exceeded part becomes the low-frequency side after the rolling shift.

[0045] If you want to generate only one augmented data distribution, you only need to perform the rolling shift process once for the selected detection data distribution. If you want to generate multiple augmented data distributions, you can perform the rolling shift process multiple times for the selected detection data distribution, or you can select different detection data distributions multiple times and perform the rolling shift process once or multiple times for each selected detection data distribution. The amplitude of the rolling shift can be the same or different each time the rolling shift process is performed.

[0046] Specifically, the distribution of the extended data after rolling shift to the left or right can be calculated using the following equation (3).

number

[0047] 7 is a diagram illustrating an example of the distribution of extended data generated when the distribution of detection data according to an embodiment of the present disclosure has a second trend. In FIG. 7, the distribution a of detection data of one piece of randomly selected communication interference detection data in which the interference type is router is indicated by a solid line, and the distribution y of extended data generated by rolling-shifting the distribution a of detection data to the left is indicated by a dashed dotted line.

[0048] As shown in Figure 7, by generating an extended data distribution by rolling or shifting the existing detection data distribution to the left or right for the second trend detection data distribution, the generated extended data distribution can meet the characteristics of the actual communication interference detection data for that interference type. For example, in Figure 7, detection data distribution a has the characteristic of showing two maximum values ​​of the average values ​​of interference power in different frequency ranges, and extended data distribution y obtained by rolling or shifting detection data distribution a to the left retains this characteristic. Therefore, it is possible to improve the likelihood that interference extended data will be used when training an artificial intelligence model for communication interference classification. Specific evaluation results will be described later.

[0049] In Figure 7, the distribution a of the detected data shows some fluctuations in the low frequency range, while the distribution y of the extended data obtained by rolling-shifting it to the left shows the same fluctuations in the high frequency range. This distribution y of the extended data does not satisfy the characteristics of the communication interference detection data of the interference type with the router, and the classification accuracy of the artificial intelligence model trained using the distribution y of the extended data may be reduced.

[0050] To solve the above problem, when the change tendency between the detection data distributions is the second tendency, instead of rolling-shifting the entire existing detection data distribution to the left or right, at least a part at one or both ends of the detection data distribution can be kept without rolling-shifting and used as a corresponding part at one or both ends of the generated extended data distribution. That is, the following processing can be further performed on the extended data distribution generated by the rolling-shift defined in the above formula (3). y for i∈[p, q] i About y i =a i where p and q are predefined integers, 0≦p≦q≦n. In other words, the frequency range [p, q] where rolling shift is not desired is defined in advance, and after performing rolling shift to obtain the distribution of extended data, the corresponding communication interference power (average value of interference power) in the frequency range [p, q] of the distribution of extended data is set to the communication interference power (average value of interference power) in the frequency range [p, q] of the distribution of detected data.

[0051] 8 is a diagram illustrating another example of the distribution of extended data generated when the distribution of detection data according to an embodiment of the present disclosure has a second trend. In FIG. 8, similar to FIG. 7, the solid line indicates the distribution a of detection data of randomly selected communication interference detection data in which the interference type is router, and the dashed-dotted line indicates the distribution y of extended data generated after rolling-shifting the distribution a of detection data to the left. Here, since the frequency range [p, q] is predefined, the distribution y of extended data overlaps with the distribution a of detection data in the frequency range [p, q].

[0052] In this example, the value at the end point q of the frequency range is the maximum value of the frequency range. That is, the extended data distribution y shown in FIG. 8 is obtained by retaining the communication interference power (average value of interference power) on the high frequency side of the detection data distribution a after a rolling shift. However, the present disclosure is not limited to this, and the left end, both ends, or a part of the middle of the detection data distribution may be set as the retention target. The position of the detection data distribution and the size of the frequency range to be retained may be determined based on the actual characteristics of the communication interference detection data of the interference type.

[0053] As shown in Figure 8, by maintaining the communication interference power (average value of interference power) corresponding to the high frequency side of the detection data distribution a without rolling or shifting, the extended data distribution y more closely matches the characteristics of actual communication interference detection data for the type of interference with the router compared to the case of Figure 7, which improves the possibility of using the interference extended data when training an artificial intelligence model for communication interference classification.

[0054] In the following description, the mode of generating the distribution of extended data by rolling / shifting the distribution of detected data to the left or right using equation (3) is called the basic rolling mode, and the mode of generating the distribution of extended data so as to retain at least a part of the distribution of detected data without rolling / shifting is called the optimized rolling mode.

[0055] Note that for some interference types, the communication interference detection data may not only exhibit either the first or second trend described above, but may also exhibit a mixture of the first and second trends. For example, for a certain interference type, the distributions of some of the detection data may follow the first trend, while the distributions of other parts of the detection data may follow the second trend. For such interference types, both the interpolation and rolling-shift methods described above may be used to generate the extended data distribution. For example, multiple extended data distributions may be generated using the basic interpolation method, the generated extended data distributions may be mixed with existing detection data distributions, and then an optimized rolling method may be used to generate additional extended data distributions for the mixed data. This is merely an example, and the method of mixing the interpolation and rolling-shift method is not limited to this.

[0056] Below, the effectiveness of the basic interpolation mode, the optimized interpolation mode, the basic rolling mode, and the optimized rolling mode of the present disclosure will be evaluated and explained.

[0057] In this disclosure, the effectiveness of each of the above modes is evaluated using 13 classification tasks for 13 interference types. A known simple classification model and ResNet are used as classifiers (examples of communication interference artificial intelligence models). For each interference type, a distribution of extended data is generated using each of the four modes described above: a mode combining the basic interpolation mode and the basic rolling mode (hereinafter referred to as the basic mixed mode), and a mode combining the optimized interpolation mode and the optimized rolling mode (hereinafter referred to as the optimized mixed mode), to obtain communication interference extended data. The communication interference extended data is mixed with existing communication interference detection data, and the mixed data is divided into training data and test data in a ratio of 8:2. All training data, 50 training data for each interference type, 20 training data for each interference type, and 10 training data for each interference type are extracted from the training data to train a simple classification model. Then, an interference type classification test is performed on the test data using the trained model. Accuracy is used as an evaluation index for the classification test. Here, the type of classifier, the ratio of the number of training data to the number of test data, the amount of training data used when training the communication interference artificial intelligence model, etc. are merely examples, and the present disclosure is not limited to these.

[0058] FIG. 9 is a diagram illustrating the evaluation results of the classification accuracy of a classifier trained using communication interference extended data generated by the above-described embodiment of the present disclosure. In FIG. 9, the values ​​corresponding to "conventional embodiment" indicate classification results when a simple classifier and a ResNet are trained using only existing communication interference detection data, when the training data does not include communication interference extended data generated by the method for extending communication interference data according to an embodiment of the present disclosure. The values ​​corresponding to "present application" indicate classification results when the training data includes communication interference extended data generated by the method for extending communication interference data according to an embodiment of the present disclosure and is mixed with existing communication interference detection data to train a simple classifier. As shown in FIG. 9, when a simple classification model is used as a classifier without using the method for extending communication interference data according to an embodiment of the present disclosure, the classification test accuracies obtained after training the classifier with all training data, 50 training data per type (i.e., per interference type), 20 interference data per type, and 10 interference data per type are 90.92%, 79.10%, 71.57%, and 63.31%, respectively. In contrast, when the communication interference extended data generated by the optimization mixing aspect of the present disclosure is mixed with existing communication interference detection data and used as training data to train a base classifier, the classification accuracy is improved by 1.8%, 7.11%, 12.82%, and 13.42%, respectively. Therefore, by generating communication interference extended data using the above aspects of the embodiments of the present disclosure and training an artificial intelligence model for communication interference classification, the classification accuracy of the artificial intelligence model for communication interference classification can be improved.

[0059] The method for generating communication interference enhancement data according to the present disclosure has been described in detail above.

[0060] According to an aspect of the present disclosure, there is further provided an apparatus for generating communication interference enhancement data, Figure 10 is a schematic diagram illustrating the configuration of an apparatus for generating communication interference enhancement data according to an embodiment of the present disclosure.

[0061] 10, a generating apparatus 1000 for generating communication interference extension data according to an embodiment of the present invention includes a processor 1100 and a memory 1200. The memory 1200 stores one or more computer programs.

[0062] The processor 1100 is a program-controlled device such as a microprocessor, and operates according to a program installed in the memory 1200. The memory 1200 is a storage element such as a ROM or RAM. The program executed by the processor 1100 is stored in the memory 1200. The apparatus 1000 shown in FIG. 10 is used to perform the method for generating communication interference extension data according to an embodiment of the present disclosure.

[0063] According to one aspect of the present disclosure, a computer program product is provided, the computer program product including a computer program that, when executed by a processor, performs a method for generating communication interference enhancement data according to an embodiment of the present disclosure.

[0064] According to one aspect of the present disclosure, there is provided a method for training an artificial intelligence model for classifying communication interference, the training method including: generating extended communication interference data by the method for generating extended communication interference data described above; and training an artificial intelligence model for classifying communication interference using the communication interference detection data, the extended communication interference data, and corresponding interference types as training data.

[0065] Block diagrams of apparatuses, devices, methods, systems, etc. included in this disclosure are merely examples and do not require or suggest that the circuits, elements, apparatuses, devices, systems, etc. be connected, arranged, or configured in the manner shown in the block diagrams. Those skilled in the art will understand that these circuits, elements, apparatuses, devices, systems, etc. may be connected, arranged, or configured in any manner so long as the desired purpose is achieved.

[0066] In the above description, the present invention has been described based on an embodiment, but the embodiment of the present disclosure is merely illustrative, and a person skilled in the art will understand that various modifications can be made to the combination of components and processes in the embodiment, and that such modifications also fall within the scope of protection of the present invention.

Claims

1. obtaining communication interference detection data including a plurality of detection data distributions, the communication interference detection data and corresponding interference types being used to train an artificial intelligence model for classifying communication interference, each of the detection data distributions representing variations in communication interference power across a plurality of frequency ranges during one sampling period; A method for generating communication interference extended data, comprising: a step of generating one or more extended data distributions based on the tendency of changes between the distributions of the plurality of detection data to use as the communication interference extended data, wherein the communication interference extended data is used together with the interference type to train an artificial intelligence model for classifying the communication interference.

2. The trend of the change is as follows: a first tendency that one of the distributions of the plurality of detection data shifts upward or downward relative to the distributions of other different detection data; and a second tendency for any one of the plurality of detection data distributions to shift left or right relative to a different other detection data distribution.

3. The step of generating one or more distributions of extended data based on the tendency of change among the distributions of the plurality of detection data and setting the distributions as the extended communication interference data includes: randomly selecting two detection data distributions from the plurality of detection data distributions in response to the determination that the change trend is the first trend; and generating a distribution of the extended data based on an interpolated value of the distributions of the two selected detection data.

4. generating the distribution of the extended data based on an interpolated value of the distributions of the two selected detection data, The interpolated value of the distribution of the two selected detection data is expressed by the following equation (1): [Equation 1] and calculating the value based on the 4. The method for generating communication interference extension data according to claim 3, wherein y indicates the distribution of the generated extension data, i is an integer representing the index of the frequency range, a and b indicate the distribution of the selected two detection data, and weight is a predefined weight.

5. If there is an intersection point between the distributions of the two selected detection data, the interpolated value of the distributions of the two selected detection data is expressed by the following equation (2): [Equation 2] Further comprising the step of calculating based on The method for generating communication interference extension data according to claim 4 , wherein weight is a negative number.

6. The step of generating one or more distributions of extended data based on the tendency of change among the distributions of the plurality of detection data and setting the distributions as the extended communication interference data includes: randomly selecting one detection data distribution from the plurality of detection data distributions in response to the determination that the change trend is the second trend; 3. The method for generating communication interference extended data according to claim 2, further comprising: a step of generating a distribution of the extended data by rolling shifting the distribution of one selected piece of detection data to the left or right.

7. The step of generating the augmented data distribution by rolling shifting the selected one of the detected data distributions to the left or right includes: The distribution of the extended data after rolling shift to the left or right is given by Equation (3): [Equation 3] and calculating the value based on the 7. The communication interference extension data generating method of claim 6, wherein y indicates the distribution of the generated extension data, a indicates the distribution of one selected detection data, n indicates the number of good frequency ranges, i is an integer indicating the index of the frequency range, 0≦i≦n−1, j is an integer indicating the amplitude of a predefined rolling shift, 0≦j≦n, and mod indicates modulo operation.

8. y for i∈[p,q] ｉ Regarding y ｉ = a ｉ The method further includes the step of: The method for generating communication interference extension data according to claim 7, wherein p and q are predefined integers, and 0≦p≦q≦n.

9. A step of generating communication interference extension data by the communication interference extension data generation method according to any one of claims 1 to 8; A method for training an artificial intelligence model for classifying communication interference, comprising: a step of training an artificial intelligence model for classifying communication interference using the communication interference detection data, the communication interference extended data, and the interference type as training data.

10. a processor and a memory having one or more computer programs stored therein; A device for generating communication interference extension data, wherein when the one or more computer programs are executed by the processor, the processor performs the method for generating communication interference extension data described in any one of claims 1 to 8.

11. A computer program product comprising a computer program, A computer program product, which, when executed by a processor, performs the method for generating communication interference extension data according to any one of claims 1 to 8.

Images