Electronic device and method executed by the electronic device

Abstract

An electronic device is provided. The electronic device includes a communication circuit and a processor. The processor can be configured to: obtain information regarding a predicted number of abnormal terminals; allocate different resources to a plurality of terminal groups, wherein the number of the plurality of terminal groups is greater than the predicted number of abnormal terminals; obtain learning data for each of the plurality of terminal groups; and identify a final terminal group from the plurality of terminal groups based on the learning data.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application is based on and claims the benefit of priority from Korean Patent Application No. 10-2021-0062146, filed on May 13, 2021, in the Korean Intellectual Property Office, the disclosure of which is incorporated herein in its entirety by reference. Technical Field

[0003] The present disclosure relates to wireless communication systems and, more particularly, to electronic devices and operating methods thereof. Background Art

[0004] Wireless distributed learning means that the base station updates the system based on the learning results obtained by processing or manipulating learning data sent from wireless communication devices (e.g., terminals). Based on wireless communication connections, the base station can update the entire system by obtaining learning data from each of multiple wireless communication devices.

[0005] The above information is presented as background information only to assist with an understanding of the present disclosure. No determination has been made, and no assertion is made, as to whether any of the above might be applicable as prior art with respect to the present disclosure. Summary of the Invention

[0006] With the increasing use of wireless communication devices in recent years, there is a growing demand for system upgrades using wireless distributed learning. Because a large amount of learning data and high-complexity computations are required to ensure the performance of wireless distributed learning, a base station can perform wireless distributed learning by using multiple wireless communication devices (e.g., terminals). When a wireless communication device is included in the multiple wireless communication devices and is operating abnormally, a method is needed to ensure the performance of machine learning by preventing the learning data caused by the abnormal operation from affecting the overall wireless distributed learning results.

[0007] The present disclosure is proposed to solve problems that occur when some wireless communication devices perform abnormal or disruptive operations in existing wireless distributed learning.

[0008] Aspects of the present disclosure are to at least address the above-mentioned problems and / or disadvantages and provide at least the advantages described below. Therefore, aspects of the present disclosure provide a base station that filters abnormal wireless communication devices (e.g., terminals) based on information about the number of abnormal wireless communication devices, thereby performing wireless distributed learning with high performance.

[0009] Additional aspects will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the presented embodiments.

[0010] According to one aspect of the present disclosure, an electronic device is provided. The electronic device includes a communication circuit and a processor. The processor may be configured to: obtain information regarding a predicted number of abnormal terminals; allocate different resources to a plurality of terminal groups, wherein the number of the plurality of terminal groups is greater than the predicted number of abnormal terminals; obtain learning data for each of the plurality of terminal groups; and identify a final terminal group from the plurality of terminal groups based on the learning data.

[0011] According to another aspect of the present disclosure, a method performed by an electronic device is provided. The method includes: obtaining information about a predicted number of abnormal terminals; allocating different resources to a plurality of terminal groups, wherein the number of the plurality of terminal groups is greater than the predicted number of abnormal terminals; obtaining learning data for each of the plurality of terminal groups; and identifying a final terminal group from the plurality of terminal groups based on the learning data.

[0012] In the apparatus and method according to various embodiments of the present disclosure, a base station filters out abnormal wireless communication devices (eg, terminals) based on the number of abnormal wireless communication devices, thereby reducing the influence caused by learning data of the abnormal wireless communication devices.

[0013] Furthermore, in the apparatus and method according to various embodiments of the present disclosure, the influence caused by the learning data of the abnormal wireless communication device is reduced, thereby performing wireless distributed learning with high performance.

[0014] Other aspects, advantages, and salient features of the disclosure will become apparent to those skilled in the art from the following detailed description, which, taken in conjunction with the annexed drawings, discloses various embodiments of the disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The above and other aspects, features and advantages of certain embodiments of the present disclosure will become more apparent from the following description taken in conjunction with the accompanying drawings, in which:

[0016] Figure 1 An environment for performing wireless distributed learning in a network according to an embodiment of the present disclosure is shown;

[0017] Figure 2 is a flowchart illustrating operations for performing wireless distributed learning according to an embodiment of the present disclosure;

[0018] Figure 3 An example of allocating resources to each terminal group according to an embodiment of the present disclosure is shown;

[0019] Figure 4 is a flowchart illustrating an operation for identifying a final terminal group according to an embodiment of the present disclosure;

[0020] Figure 5is a flowchart illustrating operations for updating a system according to an embodiment of the present disclosure;

[0021] Figure 6 is a flowchart illustrating operations for updating a system based on wireless distributed learning data according to an embodiment of the present disclosure;

[0022] Figure 7A 、 Figure 7B and Figure 7C shows examples of results of wireless distributed learning according to various embodiments of the present disclosure;

[0023] Figure 8 The structure of a base station according to an embodiment of the present disclosure is shown;

[0024] Figure 9 The structure of a terminal according to an embodiment of the present disclosure is shown.

[0025] Throughout the drawings, it should be noted that like reference numbers are used to depict the same or similar elements, features, and structures. DETAILED DESCRIPTION

[0026] The following description, with reference to the accompanying drawings, is provided to assist in a comprehensive understanding of the various embodiments of the present disclosure as defined by the claims and their equivalents. It includes various specific details to assist in that understanding, but these specific details are to be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the various embodiments described herein without departing from the scope and spirit of the present disclosure. Additionally, descriptions of well-known functions and configurations may be omitted for clarity and conciseness.

[0027] The terms and expressions used in the following description and claims are not limited to the bibliographical meanings, but are merely used by the inventor to enable a clear and consistent understanding of the present disclosure. Therefore, it should be apparent to those skilled in the art that the following description of various embodiments of the present disclosure is provided for illustration purposes only and not for the purpose of limiting the present disclosure as defined by the appended claims and their equivalents.

[0028] It should be understood that the singular forms "a," "an," and "the" include plural referents unless the context clearly dictates otherwise. Thus, for example, reference to "a component surface" includes reference to one or more of such surfaces.

[0029] For example, various embodiments of the present disclosure described below are described based on hardware. However, since various embodiments of the present disclosure include technologies using both hardware and software, software-based methods are not excluded in the embodiments of the present disclosure.

[0030] The present disclosure relates to an apparatus and method for updating a wireless communication system based on the results of wireless distributed learning of electronic devices. More specifically, the present disclosure describes a technique for updating a system based on the results of wireless distributed learning, which is more suitable for considering the real-world situation when abnormal terminals exist in a wireless communication system.

[0031] For the sake of convenience of explanation, the terms used in the following description, i.e., terms relating to variables related to location (e.g., distance, length, range, and radius), terms relating to network entities (e.g., electronic devices and external electronic devices), etc., are examples. Therefore, the present disclosure is not limited to the terms described below, and other terms with equivalent technical meanings may also be used. In addition, although the expressions "greater than or equal to" or "less than or equal to" are used in the present disclosure to determine whether a specific condition is met, this is only for the purpose and does not exclude the expressions "greater than" or "less than". A condition described as "greater than or equal to" may be replaced by "greater than". A condition described as "less than or equal to" may be replaced by "less than". A condition described as "greater than or equal to and less than" may be replaced by "greater than and less than or equal to".

[0032] Hereinafter, various embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that this is not intended to limit the technical features set forth herein to specific embodiments, and includes various changes, equivalents, or alternatives to the embodiments of the present disclosure.

[0033] Figure 1 A system for performing wireless distributed learning in a network according to an embodiment of the present disclosure is shown.

[0034] Reference depiction system 100 Figure 1 , a base station may be referred to not only as a base station, but also as an “access point (AP)”, “eNode B (eNB)”, “5th generation (5G) node”, “next generation Node B (gNB)”, “radio point”, “transmit / receive point (TRP)”, or other terms with equivalent technical meanings.

[0035] Figure 1 A terminal is a device that performs machine type communication (MTC) and may be referred to not only as a terminal but also as “user equipment (UE)”, “customer premises equipment (CPE)”, “mobile station”, “subscriber station”, “remote terminal”, “wireless terminal”, “electronic device”, “user device”, or other terms with equivalent technical meanings.

[0036] refer to Figure 1, the system may include a base station 101 and multiple terminals 102-1 to 102-M. Assume that the base station 101 can update the system based on repeated wireless distributed learning, and the base station 101 and each of the M terminal devices 102-1 to 102-M collect B items of learning data and labels in an independent and identically distributed manner. In this case, may represent data collected by base station 101, and It may represent data collected by any terminal_m 102 - m among the M terminal devices 102 - 1 to 102 -M.

[0037] Among the M terminal devices 102-1 to 102-M, the terminal_m 102-m can learn the system based on the learning data of the terminal_m 102-m and the model θ of the entire system in the t-th repeated learning. t The local gradient of terminal_m 102-m is calculated by summing the loss function and the regularization function Therefore, the sum of the loss function and the regularization function can be calculated by the following formula To obtain the local gradient of terminal_m 102-m

[0038]

[0039] In formula 1, (d is an even number) can represent the model of the entire system used for learning, and the result of machine learning with input x can be represented by f(θ,x). In addition, if the label for x is y, the loss function for this can be represented by l(f(θ,x),y). R(θ) can represent the regularization function used to control the overfitting of the model of the entire system.

[0040] Terminal_m 102-m may calculate the local gradient of the t-th learning data based on the sum of the calculated loss function and the regularization function Therefore, the local gradient of terminal_m 102-m It can be calculated by the following formula.

[0041]

[0042] In Formula 2, L(θ) may represent the sum of a loss function and a regularization function, and m∈[1:M] may represent that it is a local gradient of terminal_m102-m among M terminal devices 102-1 to 102-M.

[0043] According to an embodiment, the terminal_m 102-m may send a signal including the calculated local gradient to the base station. According to an embodiment, if the terminal_m 102-m is an abnormally operating terminal, the calculated local gradient is not Instead, any vector may be sent to base station 101. According to an embodiment, an abnormal operation is an operation that degrades the performance of wireless distributed learning in base station 101 and may include at least one of an operation that intentionally interferes with learning, an abnormal operation caused by system aging, and an abnormal operation caused by errors in wireless channel transmission. For example, an abnormally operating terminal may include a Byzantine fault-tolerant terminal, which degrades the performance of wireless distributed learning.

[0044] According to an embodiment, the base station 101 may obtain a local gradient of learning data of each of the M terminal devices 102-1 to 102-M including the terminal_m 102-m based on information received from the M terminal devices 102-1 to 102-M including the terminal_m 102-m.

[0045] According to an embodiment, the base station 101 may determine the estimated gradient by processing the local gradient of the learning data of each of the M terminal devices 102-1 to the terminal device 102-M. For example, the base station 101 may determine the estimated gradient by calculating the average gradient of the local gradient of the learning data of each of the M terminal devices 102-1 to 102-M.

[0046] According to an embodiment, the base station 101 may be based on the estimated gradient According to an embodiment, the base station 101 may broadcast the updated model θ of the entire system to the M terminal devices 102-1 to 102-M. t+1 .

[0047] According to an embodiment, if terminal_m 102-m is an abnormally operating terminal, base station 101 may update the entire system based on an arbitrary vector received from terminal_m 102-m due to the abnormal operation (which may cause performance degradation of wireless distributed learning). In this case, the arbitrary vector may be any vector unrelated to wireless distributed learning.

[0048] However, the base station according to an embodiment of the present disclosure updates the system by identifying a normal terminal group based on information on the predicted number of abnormal terminals, thereby preventing performance degradation of wireless distributed learning caused by abnormal terminals and performing efficient distributed learning.

[0049] Despite Figure 1For example, the base station receives learning data (e.g., local gradients) from multiple terminals to update the system, but the embodiments of the present disclosure are not limited thereto. According to an embodiment, any one of the multiple terminals may receive learning data (e.g., local gradients) from the remaining terminals to update the system.

[0050] Figure 2 is a flowchart illustrating an operation for performing wireless distributed learning according to an embodiment of the present disclosure. Figure 1 The base station 101) is exemplified as the apparatus for performing wireless distributed learning, but a terminal may also be used as described above.

[0051] Reference is made to the flowchart 200. Figure 2 In operation 210, according to an embodiment, the base station may create multiple terminal groups based on information about the predicted number of abnormal terminals. According to an embodiment, the base station may repeatedly perform wireless distributed learning to obtain information about the predicted number of abnormal terminals. According to an embodiment, an abnormal terminal is a terminal that causes performance degradation of wireless distributed learning in the base station and may include a terminal that performs at least one of the following: an operation that intentionally interferes with learning, an abnormal operation caused by system aging, and an abnormal operation caused by errors in transmission on a wireless channel. For example, a terminal that performs an abnormal operation may include a Byzantine fault-tolerant terminal that causes performance degradation of wireless distributed learning.

[0052] According to an embodiment, the base station may create multiple terminal groups so that the number of terminal groups is greater than the predicted number of abnormal terminals. For example, if the predicted number of abnormal terminals is F, the base station may create (F+1) terminal groups. Each terminal may belong to any one of the (F+1) terminal groups. As another example, if the predicted number of abnormal terminals is F, the base station may create (F+2) terminal groups. Each terminal may belong to any one of the (F+2) terminal groups. As another example, if the predicted number of abnormal terminals is F, the base station may create (F+n) terminal groups. Each terminal may belong to any one of the (F+n) terminal groups.

[0053] Because the number of created terminal groups is greater than the number of abnormal terminals, the abnormal terminal does not belong to at least one terminal group. The learning data obtained in the terminal group to which the abnormal terminal does not belong has a specific directionality (i.e., ) and low error rate, thereby improving the performance of wireless distributed learning systems.

[0054] According to an embodiment, the base station may determine the number of terminals belonging to a terminal group based on the total number of terminals of the entire system and the number of terminal groups created. For example, if the total number of terminals of the entire system is M and the number of terminal groups created is E, the base station may determine the number of terminals belonging to a terminal group as {[the total number of terminals of the entire system (M) / the number of groups (E)]}. As another example, the base station may determine the number of terminals belonging to a terminal group as {[the total number of terminals of the entire system (M) / the number of groups (E)]-1}.

[0055] In operation 220, according to an embodiment, the base station may allocate resources to the multiple terminal groups. According to an embodiment, the base station may allocate resources for transmitting the local gradient of the learning data to each of the multiple terminal groups. For example, when each of the multiple terminal groups transmits the local gradient of the learning data by dividing it into a real part and an imaginary part, the base station may allocate resources to each terminal group based on a total of d / 2 resources.

[0056] According to an embodiment, the base station may allocate different resources to multiple terminal groups respectively. For example, the base station may allocate orthogonal resources to multiple terminal groups respectively. According to an embodiment, the base station may allocate different resources to multiple terminal groups respectively based on at least one of an orthogonal frequency division multiplexing (OFDM) scheme, a frequency division multiplexing (FDM) scheme, and a time division multiplexing (TDM) scheme. According to an embodiment, the following will refer to Figure 3 An operation is described in which a base station allocates resources to a plurality of terminal groups by grouping the plurality of terminals so as to allocate different resources to the plurality of terminal groups, respectively.

[0057] In operation 230, according to an embodiment, the base station may obtain learning data for multiple terminal groups. According to an embodiment, the base station may obtain learning data for multiple terminal groups based on resources allocated to the multiple terminal groups. According to an embodiment, the base station may obtain local gradients of learning data for the multiple terminal groups based on resources allocated to the multiple terminal groups. Because different resources are allocated to the multiple terminal groups in the aforementioned operation 202, the base station can identify the terminal group corresponding to the learning data for each of the multiple terminal groups.

[0058] In operation 240, according to an embodiment, the base station may identify the final terminal group. According to an embodiment, the base station may identify the final terminal group based on the received learning data of the multiple terminal groups. According to an embodiment, the base station may identify the final terminal group based on the local gradient of the received learning data of the multiple terminal groups. According to an embodiment, the following will refer to Figure 4 and Figure 5 An operation is described in which a base station compares learning data of a plurality of terminal groups with reference learning data of the base station in order to identify a final terminal group among the plurality of terminal groups.

[0059] Despite Figure 2 For example, the base station receives learning data (e.g., local gradients) from multiple terminals to update the system, but the embodiments of the present disclosure are not limited thereto. According to an embodiment, any one of the multiple terminals may receive learning data (e.g., local gradients) from the remaining terminals to update the system.

[0060] Figure 3 An example of allocating resources to each terminal group according to an embodiment of the present disclosure is shown.

[0061] refer to Figure 3 , depicting example 300, according to an embodiment, a base station may group multiple terminals 301-1 to 301-E into multiple terminal groups 302-1 to 302-E. Multiple terminals 301-1 to 301-E may belong to only any one of the multiple terminal groups 302-1 to 302-E. For example, the base station may perform grouping according to a predetermined scheme so that terminal_1 301-1 belongs to terminal_1 group 302-1. As another example, the base station may perform grouping according to a predetermined scheme so that terminal_2 301-2 belongs to terminal_E group 302-E. As another example, the base station may perform grouping according to a predetermined scheme so that terminal_E 301-E belongs to terminal_2 group 302-2.

[0062] refer to Figure 3 According to an embodiment, the base station may allocate any one of the multiple resources 303-1 to 303-E to each of the multiple terminal groups 302-1 to 302-E. According to an embodiment, the resources 303-1 to 303-E allocated to the multiple terminal groups 302-1 to 302-E may be different resources. According to an embodiment, the resources 303-1 to 303-E allocated to the multiple terminal groups 302-1 to 302-E may be orthogonal resources. For example, the base station may allocate resource_1 303-1 to terminal group_1 302-1 according to a predetermined scheme. As another example, the base station may allocate resource_2 303-2 to terminal group_2 302-2 according to a predetermined scheme. As another example, the base station may allocate resource_E 303-E to terminal group_E 302-E according to a predetermined scheme. In this case, resource_1 303-1, resource_2 303-2, and resource_E 303-E may be different resources that are orthogonal to each other.

[0063] Despite Figure 3For example, the base station groups multiple terminals into groups to create multiple terminal groups and allocates different resources to each of the multiple terminal groups, but the embodiments of the present disclosure are not limited thereto. According to an embodiment, any one of the multiple terminals can group the remaining terminals to create multiple terminal groups and allocate different resources to each of the multiple terminal groups.

[0064] Figure 4 1 is a flowchart illustrating an operation for identifying a final terminal group according to an embodiment of the present disclosure. Figure 1 The base station 101) is exemplified as the apparatus for performing wireless distributed learning, but a terminal may also be used as described above.

[0065] refer to Figure 4 , depicting flow chart 400, in operation 410, according to an embodiment, the base station may estimate reference learning data. According to an embodiment, the reference learning data may include learning data obtained by the base station for the system model. According to an embodiment, the base station may estimate a reference local gradient of the reference learning data.

[0066] According to an embodiment, the base station may perform the t-th iteration of learning based on the learning data obtained from the base station and the model θ of the entire system. t , by summing the loss function and the regularization function, the reference local gradient of the reference learning data of the base station is calculated Therefore, the sum of the loss function and the regularization function can be calculated by the following formula To obtain the reference local gradient of the reference learning data of the base station

[0067]

[0068] In formula 3, (d is an even number) can represent the model of the entire system used for learning, and the result of machine learning with input x can be represented by f(θ,x). In addition, if the label for x is y, the loss function for this can be represented by l(f(θ,x),y). R(θ) can represent the regularization function used to control the overfitting of the model of the entire system.

[0069] According to an embodiment, the base station may calculate the reference local gradient of the t-th reference learning data based on the sum of the calculated loss function and the regularization function. Therefore, the reference local gradient of the base station It can be calculated by the following formula.

[0070]

[0071] In Formula 4, L(θ) can represent the sum of the loss function and the regularization function.

[0072] In operation 420, according to an embodiment, the base station may calculate the difference between the learning data of the plurality of terminal groups and the reference learning data. According to an embodiment, the base station may calculate the local gradient of the learning data of each terminal group in the plurality of terminal groups and the reference local gradient. For example, the base station can calculate the local gradient of the learning data of the terminal group_E among the multiple terminal groups Reference local gradient with reference learning data of base station The Euclidean distance difference between

[0073] In operation 430, according to an embodiment, the base station may identify a terminal group having the smallest difference value among the differences. According to an embodiment, the base station may identify a terminal group having the smallest difference value among the differences between the learning data of the plurality of terminal groups and the reference learning data. According to an embodiment, the base station may identify a reference local gradient between the local gradient of the learning data of each terminal group in the plurality of terminal groups and the reference local gradient of the reference learning data. For example, if the local gradient of the learning data of terminal group _E is Reference local gradient with reference learning data of base station The Euclidean distance difference between minimum, the base station can identify terminal group _E as the final terminal group.

[0074] According to an embodiment, the final terminal group identified by operation 430 may be a group that includes only normal terminals. In the operation of creating terminal groups, the number of terminal groups created is greater than the required number of abnormal terminals, and therefore at least one terminal group may include only normal terminals. This is because when a particular terminal group has more normal terminals than a different terminal group, the difference between the learning data of the terminal group and the reference learning data can be smaller than that of the different terminal group. Therefore, the base station can more accurately perform wireless distributed learning using the learning data of the terminal group in which the ratio of normal terminals is relatively high.

[0075] Despite Figure 4 For example, the base station identifies the final terminal group based on the difference between the learning data of multiple terminal groups and the reference learning data of the base station, but the embodiments of the present disclosure are not limited to this. According to an embodiment, any one of the multiple terminals can identify the final terminal group based on the difference between the learning data of the terminal group to which the remaining terminals belong and the reference learning data of the base station.

[0076] Figure 5is a flowchart illustrating an operation for updating a system according to an embodiment of the present disclosure. Figure 1 The base station 101) is exemplified as the apparatus for performing wireless distributed learning, but a terminal may also be used as described above.

[0077] refer to Figure 5 , depicting flowchart 500, in operation 510, according to an embodiment, the base station may identify whether the minimum difference value of the final terminal group is less than a threshold value. According to an embodiment, if the minimum difference value of the final terminal group is less than the threshold value, operation 520 may be performed. According to an embodiment, if the minimum difference value of the final terminal group is greater than or equal to the threshold value, the base station may not update the system based on the learning data of the final terminal group. According to an embodiment, the base station may preset a threshold value within a range for ensuring minimum performance of wireless distributed learning.

[0078] In operation 520, according to an embodiment, the base station may update the system based on the learning data of the final terminal group. According to an embodiment, the base station may update the system based on the final local gradient. The final optimal local gradient may include the local gradient of the learning data of the final terminal group. Therefore, the base station may calculate the module θ updated based on the final local gradient by the following formula t+1 .

[0079]

[0080] In Formula 5, θ t+1 can represent the updated model of the system in the (t+1)th iterative learning process, and θ t can represent the model of the system during the t-th iteration of learning before being updated. In this case, It can represent the final local gradient.

[0081] Despite Figure 5 For example, the base station updates the model of the entire system based on the learning data of the final terminal group, but the embodiments of the present disclosure are not limited thereto. According to an embodiment, any one of the multiple terminals can update the model of the entire system based on the learning data of the final terminal group among the terminal groups to which the remaining terminals belong.

[0082] Figure 6 is a flowchart illustrating an operation for updating a system based on wireless distributed learning data according to an embodiment of the present disclosure. Figure 6 A base station and a terminal are exemplified in FIG, but a base station can also utilize not only a single terminal but also a terminal such as Figure 1 Each of the multiple terminals shown executes Figure 6 The operations shown in .

[0083] refer to Figure 6 , depicting flowchart 600, assuming that the base station updates the system based on wireless distributed learning, and the base station and each of the M terminals collect B items of learning data and labels in an independent and identically distributed manner. In this case, may represent data collected by base station 101, and It can represent data collected by any terminal_m among the M terminals.

[0084] In operation 611, according to an embodiment, the base station may create a plurality of terminal groups based on information about the predicted number of abnormal terminals.

[0085] According to an embodiment, a base station may repeatedly perform wireless distributed learning to obtain information about the predicted number of abnormal terminals. According to an embodiment, an abnormal terminal is a terminal that causes performance degradation of wireless distributed learning in the base station and may include a terminal that performs at least one of the following operations: an operation that intentionally interferes with learning, an abnormal operation caused by system aging, and an abnormal operation caused by errors in transmission on a wireless channel. For example, a terminal that performs an abnormal operation may include a Byzantine fault-tolerant terminal, which causes performance degradation of wireless distributed learning.

[0086] According to an embodiment, the base station may create multiple terminal groups so that the number of terminal groups is greater than the predicted number of abnormal terminals. For example, if the predicted number of abnormal terminals is F, the base station may create (F+1) terminal groups. Each terminal may belong to any one of the (F+1) terminal groups. As another example, if the predicted number of abnormal terminals is F, the base station may create (F+2) terminal groups. Each terminal may belong to any one of the (F+2) terminal groups. As another example, if the predicted number of abnormal terminals is F, the base station may create (F+n) terminal groups. Each terminal may belong to any one of the (F+n) terminal groups.

[0087] Because the number of created terminal groups is greater than the number of abnormal terminals, the abnormal terminal does not belong to at least one terminal group. The learning data obtained in the terminal group to which the abnormal terminal does not belong has a specific directionality (i.e., ) and low error rate, thereby improving the performance of the wireless distributed learning system described below.

[0088] According to an embodiment, the base station may determine the number of terminals belonging to a terminal group based on the total number of terminals of the entire system and the number of terminal groups created. For example, if the total number of terminals of the entire system is M and the number of terminal groups created is E, the base station may determine the number of terminals belonging to a terminal group as {[the total number of terminals of the entire system (M) / the number of groups (E)]}. As another example, the base station may determine the number of terminals belonging to a terminal group as {[the total number of terminals of the entire system (M) / the number of groups (E)]-1}.

[0089] In operation 613, according to an embodiment, the base station may allocate resources to the multiple terminal groups. According to an embodiment, the base station may allocate resources for transmitting the local gradient of the learning data to each of the multiple terminal groups. For example, when each of the multiple terminal groups transmits the local gradient of the learning data by dividing it into a real part and an imaginary part, the base station may allocate resources to each terminal group based on a total of d / 2 resources.

[0090] According to an embodiment, the base station may allocate different resources to the multiple terminal groups. For example, the base station may allocate orthogonal resources to the multiple terminal groups. According to an embodiment, the base station may allocate different resources to the multiple terminal groups based on at least one of an OFDM scheme, an FDM scheme, and a TDM scheme.

[0091] In operation 615, according to an embodiment, the base station may transmit information about a plurality of terminal groups and resource information for each group. According to an embodiment, the base station may transmit information about resources respectively allocated to the plurality of terminal groups to the corresponding terminal groups.

[0092] In operation 651, according to an embodiment, each terminal belonging to one terminal group among a plurality of terminal groups may calculate a local gradient and a maximum size symbol.

[0093] Among the M terminals, terminal_m can learn the model θ of the entire system based on the learning data of terminal_m in the tth repeated learning. t The local gradient of terminal_m is calculated by summing the loss function and the regularization function Therefore, the sum of the loss function and the regularization function can be calculated by the following formula To obtain the local gradient of terminal_m

[0094]

[0095] In Equation 1, (d is an even number) can represent the overall model used for learning, and the result of machine learning with input x can be represented by f(θ,x). In addition, if the label for x is y, the loss function for this can be represented by l(f(θ,x),y). R(θ) can represent the overfitting regularization function used to control the model of the entire system.

[0096] Terminal_m can calculate the local gradient of the t-th learning data based on the sum of the calculated loss function and the regularization function Therefore, the local gradient of terminal_m It can be calculated by the following formula.

[0097]

[0098] In Equation 2, L(θ) may represent the sum of a loss function and a regularization function, and m∈[1:M] may represent that it is a local gradient of terminal_m among M terminals.

[0099] According to an embodiment, terminal_m can be based on the local gradient To configure the vector In order to transmit the obtained local gradient to the base station by using a specific communication scheme (e.g., OFDM scheme) Therefore, based on the local gradient of terminal_m Vector It can be calculated by the following formula.

[0100]

[0101] In formula 6, It can represent the local gradient of terminal_m The real part of It can represent the local gradient of terminal_m The imaginary part of .

[0102] According to an embodiment, terminal_m can be based on the calculated vector To calculate the maximum size symbol. Send the calculated vector of terminal_m The maximum size of the code element required It can be calculated by the following formula.

[0103]

[0104] In formula 7, d / 2 can represent the vector of sending the calculated terminal_m The number of code elements required.

[0105] In operation 653, according to an embodiment, a terminal belonging to a plurality of terminal groups may transmit a maximum size symbol and a channel estimation signal. According to an embodiment, a terminal may transmit a channel estimation signal to a base station based on a specific numerical data type (e.g., a complex number type). For example, a terminal m belonging to any one of the plurality of terminal groups may transmit a channel estimation signal to the base station in a complex number form.

[0106] In operation 617, according to an embodiment, the base station may calculate a maximum size symbol for the plurality of terminal groups. According to an embodiment, the base station may calculate a maximum size symbol allocated to each of the plurality of terminal groups based on the maximum size of the symbol received from each terminal belonging to the plurality of terminal groups. Thus, the maximum size symbol allocated to each of the plurality of terminal groups is It can be calculated by the following formula.

[0107]

[0108] In formula 8, G e It can represent the e-th terminal group among E terminal groups (e∈[1:E]).

[0109] According to an embodiment, the base station may estimate the channel between the base station and each terminal based on a channel estimation signal (eg, a preamble) received from the terminal.

[0110] In operation 619, according to an embodiment, the base station may transmit channel information and a maximum size symbol of a plurality of terminal groups. According to an embodiment, the base station may transmit channel information and a maximum size symbol of a plurality of terminal groups to the terminal.

[0111] According to an embodiment, the base station may transmit channel information between the base station and the terminal to the terminal based on the channel estimation.

[0112] In operation 655, according to an embodiment, the terminals belonging to the plurality of terminal groups may transmit learning data based on the channel information. According to an embodiment, the terminals belonging to the plurality of terminal groups may transmit local gradients of the learning data based on the channel information. For example, the terminal_m belonging to one of the plurality of terminal groups may calculate the local gradient of the nth subcarrier before transmitting the learning data using the following formula:

[0113]

[0114] In formula 9, is the indicator function and can be expressed by to define.

[0115] In operation 621, according to an embodiment, the base station may obtain learning data of a plurality of terminal groups. According to an embodiment, the base station may obtain learning data of a plurality of terminal groups by receiving learning data from the plurality of terminal groups.

[0116] According to an embodiment, the base station may obtain the learning data of the multiple terminal groups based on the resources allocated to each terminal group in the multiple terminal groups. According to an embodiment, the base station may obtain the local gradient of the learning data of the multiple terminal groups based on the resources allocated to each terminal group in the multiple terminal groups. According to an embodiment, the base station may obtain the local gradient of the learning data received from the terminal by adding the terminal for each group based on a wireless network (e.g., air computing). For example, the terminal group G belonging to the e-th terminal group may be a plurality of terminal groups. e Received signal It can be calculated by the following formula.

[0117]

[0118] In formula 10, is a specific numeric data type (for example, a complex number type), and can represent noise generated when a signal is received.

[0119] According to an embodiment, the base station can calculate the optimal gradient of each terminal group based on the signals received from multiple terminal groups. For example, the optimal gradient of the learning data of the e-th terminal group is It can be calculated by the following formula.

[0120]

[0121] In formula 11, It can represent the optimal gradient of the learning data of the e-th terminal group The real part of .

[0122]

[0123] In formula 12, It can represent the optimal gradient of the learning data of the e-th terminal group The imaginary part of .

[0124] In operation 623, according to an embodiment, the base station may identify a final terminal group.

[0125] According to an embodiment, the base station may identify the final terminal group based on the received learning data of the plurality of terminal groups.According to an embodiment, the base station may identify the final terminal group based on the local gradient of the received learning data of the plurality of terminal groups.

[0126] According to an embodiment, the base station may estimate reference learning data. According to an embodiment, the reference learning data may include learning data obtained by the base station for the system model. According to an embodiment, the base station may estimate a reference local gradient of the reference learning data.

[0127] According to an embodiment, the base station may perform the t-th iteration of learning based on the learning data obtained from the base station and the model θ of the entire system. t , calculate the reference local gradient of the reference learning data of the base station by summing the loss function and the regularization function Therefore, the sum of the loss function and the regularization function can be calculated by the following formula To obtain the reference local gradient of the reference learning data of the base station

[0128]

[0129] In Equation 3, (d is an even number) can represent the model of the entire system used for learning, and the result of machine learning with input x can be represented by f(θ,x). In addition, if the label for x is y, the loss function for this can be represented by l(f(θ,x),y). R(θ) can represent the regularization function used to control the overfitting of the model of the entire system.

[0130] According to an embodiment, the base station may calculate the reference local gradient of the t-th reference learning data based on the sum of the calculated loss function and the regularization function. Therefore, the reference local gradient of the base station It can be calculated by the following formula.

[0131]

[0132] In Equation 4, L(θ) can represent the sum of the loss function and the regularization function.

[0133] According to an embodiment, the base station may identify a terminal group having the smallest difference among the differences. According to an embodiment, the base station may identify a terminal group having the smallest difference among the differences between the learning data of a plurality of terminal groups and the reference learning data. According to an embodiment, the base station may identify a reference local gradient of the learning data of each of the plurality of terminal groups and the reference local gradient of the reference learning data. For example, if the local gradient of the learning data of terminal group _E is Reference local gradient with reference learning data of base station The Euclidean distance difference between minimum, the base station can identify terminal group _E as the final terminal group.

[0134] In operation 625 , according to an embodiment, the base station may update the system based on the learning data of the final terminal group.

[0135] According to an embodiment, the base station may determine whether the minimum difference value of the final terminal group is less than a threshold value. According to an embodiment, if the minimum difference value of the final terminal group is less than the threshold value, the system may be updated based on the learning data of the final terminal group. According to an embodiment, if the minimum difference value of the final terminal group is greater than or equal to the threshold value, the base station may not update the system based on the learning data of the final terminal group. According to an embodiment, the base station may preset a threshold value within a range for ensuring minimum performance of wireless distributed learning.

[0136] According to an embodiment, the base station may update the system based on the learning data of the final terminal group. According to an embodiment, the base station may update the system based on the final local gradient. The final optimal local gradient may include the local gradient of the learning data of the final terminal group. Therefore, the base station may calculate the module θ updated based on the final local gradient by the following formula t+1 .

[0137]

[0138] In Equation 5, θ t+1 can represent the updated model of the system in the (t+1)th iterative learning process, and θ t can represent the model of the system during the t-th iteration of learning before being updated. In this case, is the final local gradient and can represent the final terminal group e * The local gradient of ).

[0139] According to an embodiment, the base station may transmit the updated system model θ to terminals belonging to a plurality of terminal groups through a broadcast scheme. t+1 According to an embodiment, the base station and the terminal may repeatedly perform operations 611 to 625 until the total performance of the wireless distributed learning is greater than or equal to a reference value.

[0140] Despite Figure 6 For example, the base station receives learning data (e.g., local gradients) from multiple terminals to update the system, but the embodiments of the present disclosure are not limited thereto. According to an embodiment, any one of the multiple terminals may receive learning data (e.g., local gradients) from the remaining terminals to update the system.

[0141] As in Figure 6As described in , based on information about the number of abnormal terminals of a base station, a terminal group is created to perform wireless distributed learning. Therefore, it is possible to prevent wireless distributed learning from suffering performance degradation caused by abnormal terminals and to efficiently update the system through wireless distributed learning.

[0142] Figure 7A 、 Figure 7B and Figure 7C Examples of results of wireless distributed learning according to various embodiments of the present disclosure are shown.

[0143] refer to Figure 7A , a graph is shown that compares the test accuracy between the technology in the prior art and the technology proposed according to the present disclosure based on the number F of abnormal terminals when the number of terminals belonging to multiple terminal groups is 20. When the number of terminals belonging to multiple terminal groups is 20 and the number F of abnormal terminals is 4, it can be seen that the test accuracy of the technology in the prior art converges to within a range of approximately 10%, and the test accuracy of the proposed technology converges to within a range of approximately 66%. When the number of terminals belonging to multiple terminal groups is 20 and the number F of abnormal terminals is 9, it can be seen that the test accuracy of the technology in the prior art converges to within a range of approximately 10%, and the test accuracy of the proposed technology converges to within a range of approximately 62%. When the number of terminals belonging to multiple terminal groups is 20 and the number F of abnormal terminals is 0, it can be seen that the test accuracy of the proposed technology converges to within a range of approximately 80%.

[0144] refer to Figure 7B , a graph is shown that compares the test accuracy between the conventional technology and the technology proposed according to the present disclosure based on the number F of abnormal terminals, when the number of terminals belonging to multiple terminal groups is 40. When the number of terminals belonging to multiple terminal groups is 40 and the number F of abnormal terminals is 4, it can be seen that the test accuracy of the technology in the prior art converges to within a range of approximately 10%, and the test accuracy of the proposed technology converges to within a range of approximately 72%. When the number of terminals belonging to multiple terminal groups is 40 and the number F of abnormal terminals is 9, it can be seen that the test accuracy of the technology in the prior art converges to within a range of approximately 10%, and the test accuracy of the proposed technology converges to within a range of approximately 68%. When the number of terminals belonging to multiple terminal groups is 40 and the number F of abnormal terminals is 0, it can be seen that the test accuracy of the proposed technology converges to within a range of approximately 82%.

[0145] refer to Figure 7C, a graph is shown that compares the test accuracy between the prior art technique and the proposed technique based on the number F of abnormal terminals when the number of terminals belonging to multiple terminal groups is 60. When the number of terminals belonging to multiple terminal groups is 60 and the number F of abnormal terminals is 4, it can be seen that the test accuracy of the prior art technique converges within a range of approximately 10%, and the test accuracy of the proposed technique converges within a range of approximately 78%. When the number of terminals belonging to multiple terminal groups is 60 and the number F of abnormal terminals is 9, it can be seen that the test accuracy of the prior art technique converges within a range of approximately 10%, and the test accuracy of the proposed technique converges within a range of approximately 74%. When the number of terminals belonging to multiple terminal groups is 60 and the number F of abnormal terminals is 0, it can be seen that the test accuracy of the proposed technique converges within a range of approximately 85%.

[0146] like Figure 7A 、 Figure 7B and Figure 7C As shown in , if an abnormally operating terminal is included, the base station can update the entire system based on an arbitrary vector received due to the abnormal operation (which may cause performance degradation of wireless distributed learning). In this case, the arbitrary vector may be any vector unrelated to wireless distributed learning. However, according to an embodiment of the present disclosure, the base station updates the system by identifying a normal terminal group based on information about the predicted number of abnormal terminals, thereby preventing performance degradation of wireless distributed learning caused by abnormal terminals and performing efficient distributed learning.

[0147] Although, for example, Figure 7A 、 Figure 7B and Figure 7C The result of wireless distributed learning is shown in FIG, but the embodiments of the present disclosure are not limited thereto. According to the embodiment, the result of wireless distributed learning performed by any one terminal among the multiple terminals may include Figure 7A 、 Figure 7B and Figure 7C The results of wireless distributed learning shown in are the same as or correspond to the results thereof.

[0148] Figure 8 The structure of a base station according to an embodiment of the present disclosure is shown.

[0149] refer to Figure 8 , the base station may include a processor 810 , a memory 820 , and a transceiver 830 .

[0150] The processor 810 can provide comprehensive control of the base station. For example, the processor 810 can create multiple terminal groups for multiple terminals in the system based on information about the number of abnormal terminals. The processor 810 can allocate different resources (e.g., orthogonal resources) to the multiple terminal groups. The processor 810 can identify the final terminal group by comparing the learning data (e.g., local gradient) received from the multiple terminal groups with the reference learning data of the base station. If the difference between the learning data of the identified final terminal group and the reference learning data is less than a threshold, the processor 810 can update the model of the entire system.

[0151] The processor 810 may transmit and receive signals via the transceiver 830. For example, the processor 810 may receive a channel estimation signal and information about a maximum size of a symbol required to transmit learning data of the terminal from the terminal through allocated resources via the transceiver 830.

[0152] In addition, the processor 810 may execute the functions of the protocol stack required in the communication standard. For this purpose, the processor 810 may include at least one processor. The processor 810 may control the base station to perform operations according to the aforementioned embodiments.

[0153] The memory 820 may store data used for the operation of the base station, such as basic programs, application programs, configuration information, etc. The memory 820 may be composed of a volatile memory, a non-volatile memory, or a combination of a volatile memory and a non-volatile memory. The memory 820 may provide stored data at the request of the processor 810.

[0154] The transceiver 830 can perform functions for transmitting and receiving signals through a wired channel or a wireless channel. For example, the transceiver 830 can perform functions for converting between baseband signals and bit streams according to the physical layer standard of the system. For example, in data transmission, the transceiver 830 can generate composite symbols by encoding and modulating the transmitted bit stream. In addition, in data reception, the transceiver 830 can recover the received bit stream by demodulating and decoding the baseband signal. In addition, the transceiver 830 can up-convert the baseband signal into a radio frequency (RF) signal and then transmit it through an antenna, and can down-convert the RF signal received through the antenna into a baseband signal. To this end, the transceiver 830 may include a transmit filter, a receive filter, an amplifier, a mixer, an oscillator, a digital-to-analog converter (DAC), an analog-to-digital converter (ADC), and the like.

[0155] In addition, the transceiver 830 may include multiple transmit paths / receive paths. In addition, the transceiver 830 may include an antenna unit. The transceiver 830 may include at least one antenna array constructed from multiple antenna elements. From a hardware perspective, the transceiver 830 may be composed of digital and analog circuits (e.g., radio frequency integrated circuits (RFICs)). In this article, the digital circuits and analog circuits may be implemented as a single package.

[0156] In addition, the transceiver 830 may further include a backhaul communication interface for performing communication with different nodes in the network. That is, the transceiver 830 converts the bit stream sent from the base station to different nodes (e.g., different access nodes, different base stations, upper nodes, core networks, etc.) into a physical signal, and converts the physical signal received from different nodes into a bit stream.

[0157] In addition, the transceiver 830 may include different communication modules to process signals of different frequency bands. In addition, the transceiver 830 may include multiple communication modules to support multiple different radio access technologies. For example, different radio access technologies may include Bluetooth Low Energy (BLE), Wireless Fidelity (WiFi), cellular networks (e.g., Long Term Evolution (LTE), New Radio (NR)), and the like. In addition, different frequency bands may include ultra-high frequency (SHF) (e.g., 2.5 GHz, 5 GHz) bands and millimeter wave (e.g., 38 GHz, 60 GHz, etc.) bands. In addition, the transceiver 830 may use the same type of radio access technology on different frequency bands (unlicensed bands for license-assisted access (LAA), Citizens Broadband Radio Service (CBRS) (e.g., 3.5 GHz)). Together, the transceiver 830 may be referred to as a communication circuit.

[0158] Figure 9 The structure of a terminal according to an embodiment of the present disclosure is shown.

[0159] refer to Figure 9 , the terminal may include a processor 910 , a memory 920 , and a transceiver 930 .

[0160] The processor 910 may provide comprehensive control of the terminal. For example, the processor 910 may calculate the learning data (e.g., local gradient) of the terminal based on the terminal group information received from the base station. For example, the processor 910 may send and receive signals via the transceiver 930. For example, the processor 910 may provide control to send a channel estimation signal to the base station via the transceiver 930 and the maximum size of the codeword required to send the learning data of the terminal. For example, the processor 910 may provide control to send the learning data of the terminal to the base station based on the resources allocated by the base station. For this, the processor 910 may include at least one processor. The processor 910 may control the terminal to perform the aforementioned operations according to the embodiment.

[0161] The memory 920 may store data used for the operation of the terminal, such as basic programs, application programs, configuration information, etc. The memory 920 may be composed of a volatile memory, a non-volatile memory, or a combination of a volatile memory and a non-volatile memory. The memory 920 may provide stored data at the request of the processor 910.

[0162] The transceiver 930 can perform functions for transmitting and receiving signals through a wired channel or a wireless channel. For example, the transceiver 930 can perform functions for converting between baseband signals and bit streams according to the physical layer standard of the system. For example, in data transmission, the transceiver 930 can generate composite symbols by encoding and modulating the transmitted bit stream. In addition, in data reception, the transceiver 930 can recover the received bit stream by demodulating and decoding the baseband signal. In addition, the transceiver 930 can up-convert the baseband signal into a radio frequency (RF) signal and then transmit it through an antenna, and can down-convert the RF signal received through the antenna into a baseband signal. To this end, the transceiver 930 may include a transmit filter, a receive filter, an amplifier, a mixer, an oscillator, a digital-to-analog converter (DAC), an analog-to-digital converter (ADC), and the like.

[0163] In addition, the transceiver 930 may include multiple transmit paths / receive paths. In addition, the transceiver 930 may include an antenna unit. The transceiver 930 may include at least one antenna array constructed from multiple antenna elements. From a hardware perspective, the transceiver 930 may be composed of digital and analog circuits (e.g., radio frequency integrated circuits (RFICs)). In this article, the digital circuits and analog circuits may be implemented as a single package.

[0164] In addition, the transceiver 930 may include different communication modules to process signals of different frequency bands. In addition, the transceiver 930 may include multiple communication modules to support multiple different radio access technologies. For example, different radio access technologies may include Bluetooth Low Energy (BLE), Wireless Fidelity (WiFi), cellular networks (e.g., Long Term Evolution (LTE), New Radio (NR)), and the like. In addition, different frequency bands may include ultra-high frequency (SHF) (e.g., 2.5 GHz, 5 GHz) bands and millimeter wave (e.g., 38 GHz, 60 GHz, etc.) bands. In addition, the transceiver 930 may use the same type of radio access technology on different frequency bands (unlicensed bands for license-assisted access (LAA), Citizens Broadband Radio Service (CBRS) (e.g., 3.5 GHz)). Together, the transceiver 930 may be referred to as a communication circuit.

[0165] An electronic device according to an embodiment of the present disclosure may include a communication circuit and a processor. The processor may be configured to: obtain information about a predicted number of abnormal terminals; allocate different resources to a plurality of terminal groups, wherein the number of the plurality of terminal groups is greater than the predicted number of abnormal terminals; obtain learning data for each of the plurality of terminal groups; and identify a final terminal group from the plurality of terminal groups based on the learning data.

[0166] According to an embodiment, the processor may be configured to update the system based on the learning data of the final terminal group.

[0167] According to an embodiment, in order to identify a final terminal group among multiple terminal groups, the processor can be configured to: estimate reference learning data, calculate the difference between the reference learning data and the learning data of each terminal group in the multiple terminal groups, and identify the terminal group with the smallest difference among the differences as the final terminal group.

[0168] According to an embodiment, the processor may be further configured to identify whether the minimum difference value of the final terminal group is less than a threshold value.

[0169] According to an embodiment, the processor may be further configured to update the system based on the learning data of the final terminal group if the minimum difference value of the final terminal group is smaller than a threshold value.

[0170] According to an embodiment, the processor may be further configured such that, if the minimum difference value of the final terminal group is greater than or equal to a threshold value, the system is not updated based on the learning data of the final terminal group.

[0171] According to an embodiment, the processor may be further configured to: receive a synchronization signal from at least one terminal, wherein the at least one terminal is included in any one terminal group of a plurality of terminal groups; and estimate a channel between the at least one terminal and the electronic device based on the synchronization signal.

[0172] According to an embodiment, the processor can be further configured to: allow the synchronization signal to include information about resources required in at least one terminal, determine the maximum amount of resources for each terminal group in multiple terminal groups based on the information, and send information about the maximum amount of resources for each terminal in the multiple terminals to at least one terminal.

[0173] According to an embodiment, in order to allocate different resources to each of the multiple terminal groups, the processor can be configured to allocate different resources to each of the multiple terminal groups based on at least one of an orthogonal frequency division multiplexing (OFDM) scheme, a frequency division multiplexing (FDM) scheme, and a time division multiplexing (TDM) scheme.

[0174] According to an embodiment of the present disclosure, a method for operating an electronic device may include: obtaining information about a predicted number of abnormal terminals; allocating different resources to a plurality of terminal groups, respectively, wherein the number of the plurality of terminal groups is greater than the predicted number of abnormal terminals; obtaining learning data for each of the plurality of terminal groups, and identifying a final terminal group among the plurality of terminal groups based on the learning data.

[0175] According to an embodiment, the method may further include updating the system based on the learning data of the final terminal group.

[0176] According to an embodiment, identifying the final terminal group among multiple terminal groups may include: estimating reference learning data, calculating the difference between the reference learning data and the learning data of each terminal group in the multiple terminal groups, and identifying the terminal group with the smallest difference among the differences as the final terminal group.

[0177] According to an embodiment, the method may further include identifying whether the minimum difference value of the final terminal group is less than a threshold value.

[0178] According to an embodiment, the method may further include updating the system based on the learning data of the final terminal group if the minimum difference value of the final terminal group is less than a threshold value.

[0179] According to an embodiment, the method may further include: if the minimum difference value of the final terminal group is greater than or equal to a threshold value, not updating the system based on the learning data of the final terminal group.

[0180] According to an embodiment, the method may further include: receiving a synchronization signal from the at least one terminal, and estimating a channel between the at least one terminal and the electronic device based on the synchronization signal.The at least one terminal may be included in any one of a plurality of terminal groups.

[0181] According to an embodiment, the method may further include: allowing the synchronization signal to include information about resources required in at least one terminal, determining the maximum amount of resources for each of a plurality of terminal groups based on the information, and sending information about the maximum amount of resources for each of the plurality of terminals to the at least one terminal.

[0182] According to an embodiment, respectively allocating different resources to the plurality of terminal groups may include allocating different resources to each of the plurality of terminal groups based on at least one of an OFDM scheme, an FDM scheme, and a TDM scheme.

[0183] The electronic device according to various embodiments disclosed in the present disclosure may be one of various types of electronic devices. The electronic device may include, for example, a portable communication device (e.g., a smartphone), a computer device, a portable multimedia device, a portable medical device, a camera, a wearable device, or a household appliance. According to embodiments of the present disclosure, the electronic device is not limited to those described above.

[0184] It should be understood that the various embodiments disclosed and the terms used therein are not intended to limit the technical features set forth herein to specific embodiments, but rather include various changes, equivalents or alternative forms for the corresponding embodiments. For the description of the accompanying drawings, similar figure numerals may be used to refer to similar or related elements. As used herein, phrases such as "A or B", "at least one of A and B", "at least one of A or B", "A, B, or C", "at least one of A, B, and C", and "at least one of A, B, or C" may include all possible combinations of the items listed together in one of the corresponding ones in the phrase. As used herein, terms such as "1st" and "2nd" or "first" and "second" may be used to simply distinguish a corresponding component from another component, and do not limit the components in other respects (e.g., importance or order). It will be understood that if an element (e.g., a first element) is referred to as being “coupled to another element (e.g., a second element)”, “coupled to another element (e.g., a second element)”, “connected to another element (e.g., a second element)”, or “connected to another element (e.g., a second element)”, whether or not the terms “operably” or “communicatively” are used, it means that the element may be directly (e.g., wired) connected to the other element, wirelessly connected to the other element, or connected to the other element via a third element.

[0185] The term "module" used in various embodiments of the present disclosure may include a unit implemented in hardware, software, or firmware, and may be used interchangeably with other terms such as "logic," "logic block," "portion," or "circuit." A module may be a single integrated component adapted to perform one or more functions or the smallest unit or portion of the single integrated component. For example, depending on the embodiment, a module may be implemented in the form of an application-specific integrated circuit (ASIC).

[0186] Various embodiments of the present disclosure may be implemented as software (e.g., program 140) comprising one or more instructions stored in a storage medium (e.g., internal memory 136 or external memory 138) readable by a machine (e.g., an electronic device). For example, under the control of a processor, a processor (e.g., processor 120) of a machine (e.g., an electronic device) may call at least one of the one or more instructions stored in the storage medium and execute the at least one instruction with or without the use of one or more other components. This enables the machine to operate to perform at least one function according to the called at least one instruction. The one or more instructions may include code generated by a compiler or code that can be executed by an interpreter. The machine-readable storage medium may be provided in the form of a non-transitory storage medium. The term "non-transitory" merely means that the storage medium is a tangible device and does not include signals (e.g., electromagnetic waves), but the term does not distinguish between data being semi-permanently stored in the storage medium and data being temporarily stored in the storage medium.

[0187] According to an embodiment, the method according to various embodiments of the present disclosure may be included and provided in a computer program product. The computer program product may be traded as a product between a seller and a buyer. The computer program product may be released in the form of a machine-readable storage medium (e.g., a compact disc read-only memory (CD-ROM)), or may be downloaded via an application store (e.g., Play Store). TM ) The computer program product may be published online (e.g., downloaded or uploaded) or may be published (e.g., downloaded or uploaded) directly between two user devices (e.g., smartphones). If published online, at least a portion of the computer program product may be temporarily generated or at least temporarily stored in a machine-readable storage medium (e.g., a memory of a manufacturer's server, an application store's server, or a forwarding server).

[0188] According to various embodiments, each component (e.g., module or program) in the aforementioned components may include a single entity or multiple entities, and some of the multiple entities may be arranged separately to different components. According to various embodiments, one or more of the aforementioned components may be omitted, or one or more different components may be added. Alternatively or additionally, multiple components (e.g., module or program) may be integrated into a single component. In this case, the integrated component may perform one or more functions of each component in the multiple components in the same or similar manner as a corresponding component in the multiple components before integration. According to various embodiments, the operations performed by modules, programs or different components may be performed sequentially, in parallel, repeatedly or in a heuristic manner, or one or more operations in the operations may be run in different orders or omitted, or one or more different operations may be added.

[0189] While the present disclosure has been shown and described with reference to various embodiments thereof, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the disclosure as defined by the appended claims and their equivalents.

Claims

1. An electronic device, comprising: Communication circuits; as well as a processor operatively connected to the communication circuitry, wherein the processor is configured to: Obtain information about the number of predicted abnormal terminals, Allocating different resources to a plurality of terminal groups respectively, wherein the number of the plurality of terminal groups is greater than the number of the predicted abnormal terminals, obtaining learning data of each terminal group in the plurality of terminal groups, Estimation of reference learning data, calculating a difference between the reference learning data and the learning data of each of the plurality of terminal groups, and A terminal group having a minimum difference value among the difference values is identified as a final terminal group.

2. The electronic device according to claim 1, wherein The processor is further configured to update a system based on the learning data of the final terminal group.

3. The electronic device according to claim 1, wherein The processor is further configured to identify whether a minimum difference value of the final terminal group is less than a threshold.

4. The electronic device according to claim 3, wherein: The processor is further configured to update the system based on the learning data of the final terminal group if the minimum difference value of the final terminal group is smaller than the threshold.

5. The electronic device according to claim 3, wherein: The processor is further configured to not update the system based on the learning data of the final terminal group if the minimum difference value of the final terminal group is greater than or equal to the threshold. The electronic device according to claim 1 , wherein: The processor is further configured to: receiving a synchronization signal from at least one terminal included in any one of the plurality of terminal groups, and A channel between the at least one terminal and the electronic device is estimated based on the synchronization signal.

7. The electronic device according to claim 6, wherein: The processor is further configured to: allowing the synchronization signal to include information on resources required in the at least one terminal, determining a maximum amount of resources for each of the plurality of terminal groups based on the information, and Information on a maximum amount of resources for each of the plurality of terminal groups is transmitted to the at least one terminal.

8. The electronic device according to claim 1, wherein In order to allocate the different resources to each of the plurality of terminal groups, the processor is further configured to allocate the different resources to each of the plurality of terminal groups based on at least one of an orthogonal frequency division multiplexing scheme, a frequency division multiplexing scheme, and a time division multiplexing scheme.

9. A method performed by an electronic device, the method comprising: Obtaining information about the number of predicted abnormal terminals; Allocating different resources to a plurality of terminal groups respectively, wherein the number of the plurality of terminal groups is greater than the predicted number of abnormal terminals; obtaining learning data of each terminal group in the plurality of terminal groups; Estimating reference learning data; calculating a difference between the reference learning data and learning data of each of the plurality of terminal groups; and A terminal group having a minimum difference value among the difference values is identified as a final terminal group.

10. The method according to claim 9, further comprising: A system is updated based on the learning data of the final terminal group.

11. The method according to claim 9, further comprising: It is identified whether the minimum difference value of the final terminal group is less than a threshold.

12. The method according to claim 11, further comprising: In response to the minimum difference value of the final terminal group being less than the threshold, updating the system based on the learning data of the final terminal group.

13. The method according to claim 11, further comprising: If the minimum difference value of the final terminal group is greater than or equal to the threshold value, the system is not updated based on the learning data of the final terminal group.

14. The method according to claim 9, further comprising: receiving a synchronization signal from at least one terminal; as well as estimating a channel between the at least one terminal and the electronic device based on the synchronization signal, The at least one terminal is included in any one terminal group among the multiple terminal groups.

15. The method according to claim 14, further comprising: allowing the synchronization signal to include information about resources required in the at least one terminal; determining a maximum amount of resources for each of the plurality of terminal groups based on the information; as well as Information on a maximum amount of resources for each of the plurality of terminal groups is transmitted to the at least one terminal.

16. The method according to claim 9, further comprising: A channel estimation signal is received from at least one terminal, the signal including information about a maximum size of a symbol required to transmit learning data for the at least one terminal.

17. The method according to claim 9, wherein The allocating different resources to the plurality of terminal groups respectively includes allocating resources that are orthogonal to each other.

18. The method according to claim 9, further comprising: An optimal gradient for each of the plurality of terminal groups is calculated based on a signal received from each of the plurality of terminal groups.

Images