Learning device, inference device, and communication system

Abstract

To suppress an amount of traffic in a communication system that uses a line network shared by a ground device and a plurality of base station devices.SOLUTION: A learning device includes a data acquisition unit and a model generation unit. The data acquisition unit acquires one or more pieces of base station information that identify each of a plurality of base station devices that receive a wireless signal from a mobile station device, and one or more wireless link qualities that represent the quality of the wireless signal received by each of the base station devices. The model generation unit generates a learned model for inferring, from the base station information and the wireless link qualities, transmission control information that controls whether or not each of the base station devices is allowed to transmit to the same ground device, by learning using learning data generated by associating the base station information and the wireless link qualities both acquired from the data acquisition unit.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present disclosure relates to a learning device, an inference device, and a communication system. [Background technology]

[0002] BACKGROUND ART A known communication system has a configuration in which a plurality of base station devices receive a radio signal from the same mobile station device, and a central device receives signals output from each base station (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2002-325270 Summary of the Invention [Problem to be solved by the invention]

[0004] A configuration is assumed in which one central device and multiple base station devices can communicate by sharing a line network (e.g., Ethernet (registered trademark)). In such a configuration, it is assumed that multiple base station devices receive radio signals from the same mobile station device with a short time difference from each other and output them to the central device. Outputting signals with such a short time difference will increase the amount of traffic.

[0005] The central device does not need to obtain information contained in radio signals from the same mobile station device from multiple base station devices, and the information transmitted to the central device may be obtained only from the signal output from the base station device that received a specific radio signal from the same mobile station device.

[0006] The present disclosure aims to reduce the amount of traffic in a communication system that uses a line network shared by a ground device as a central device and a plurality of base station devices. [Means for solving the problem]

[0007] A learning device according to the present disclosure includes a data acquisition unit and a model generation unit. The data acquisition unit acquires one or more pieces of base station information identifying each of a plurality of base station devices that receive a wireless signal from a mobile station device, and one or more pieces of wireless link quality indicating the quality of the wireless signal received by each of the base station devices. The model generation unit generates a trained model for inferring, from the base station information and the wireless link quality, transmission control information that controls whether or not each of the base station devices transmits to the same ground device, by learning using training data generated by associating the base station information and the wireless link quality acquired from the data acquisition unit.

[0008] The inference device according to the present disclosure includes a data acquisition unit and an inference unit. The data acquisition unit acquires and outputs base station information identifying each of a plurality of base station devices that receive a wireless signal from a mobile station device and wireless channel quality indicating the quality of the wireless signal received by each of the base station devices. The inference unit infers the transmission control information for the base station device corresponding to the base station information output from the data acquisition unit, based on a trained model in which transmission control information is trained using training data, and the base station information and the wireless channel quality output from the data acquisition unit.

[0009] The transmission control information controls whether or not transmission from the base station device to the same ground device is permitted. The learning data includes the base station information and the wireless channel quality, which correspond to each other. [Effects of the Invention]

[0010] According to the present disclosure, the amount of traffic is reduced in a communication system using a line network shared by a ground device and a plurality of base station devices. [Brief explanation of the drawings]

[0011] [Figure 1] 1 is a block diagram illustrating a configuration for performing learning in a learning device according to a first embodiment of the present disclosure. [Figure 2] 1 is a flowchart illustrating a process for obtaining the best wireless link quality and corresponding base station information. [Figure 3] 1 is a block diagram illustrating a configuration and input / output of a learning device according to a first embodiment. [Figure 4] 1 is a flowchart illustrating a model generation process according to the first embodiment. [Figure 5] 2 is a block diagram illustrating an example of input and output of the inference device in the first embodiment. FIG. [Figure 6] 1 is a block diagram illustrating the configuration and input / output of an inference device according to a first embodiment. [Figure 7] 4 is a flowchart illustrating a process of generating transmission control information in the first embodiment. [Figure 8] 5 is a flowchart illustrating an operation of the base station device in the first embodiment. [Figure 9] FIG. 10 is a block diagram illustrating the configuration and input / output of a learning device according to a second embodiment of the present disclosure. [Figure 10] 10 is a flowchart illustrating a model generation process according to the second embodiment. [Figure 11] FIG. 10 is a block diagram illustrating the configuration and input / output of an inference device according to a second embodiment. [Figure 12] 10 is a flowchart illustrating a process of generating transmission control information in the second embodiment. [Figure 13] FIG. 11 is a block diagram illustrating a configuration for performing learning in a learning device according to a third embodiment of the present disclosure. [Figure 14] FIG. 11 is a block diagram illustrating the configuration and input / output of a learning device according to a third embodiment. [Figure 15] 11 is a flowchart illustrating an operation of a ground device in a learning phase according to the third embodiment. [Figure 16] 11 is a flowchart illustrating a model generation process according to the third embodiment. [Figure 17]FIG. 11 is a block diagram illustrating a neural network employed in the third embodiment. [Figure 18] FIG. 11 is a block diagram illustrating the configuration and input / output of an inference device according to a third embodiment. [Figure 19] 11 is a flowchart illustrating a process of generating transmission control information in the third embodiment. [Figure 20] FIG. 10 is a block diagram illustrating a configuration for performing learning in a learning device according to a fourth embodiment of the present disclosure. [Figure 21] FIG. 10 is a block diagram illustrating the configuration and input / output of a learning device according to a fourth embodiment. [Figure 22] 13 is a flowchart illustrating a model generation process according to the fourth embodiment. [Figure 23] FIG. 13 is a block diagram illustrating a configuration for performing learning and inference in a fifth embodiment. [Figure 24] FIG. 13 is a block diagram illustrating the configuration and input / output of a learning / inference device according to a fifth embodiment. [Figure 25] 13 is a flowchart illustrating a part of the operation of the base station device according to the fifth embodiment. [Figure 26] 13 is a flowchart illustrating a model generation process according to the fifth embodiment. [Figure 27] FIG. 1 is a block diagram illustrating a hardware configuration applicable to the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0012] The communication system in the present disclosure utilizes machine learning. In machine learning, for example, a trained model is generated by a learning device. In machine learning, for example, an inference device utilizes the trained model to output an inference result. Trained parameters may be generated by the learning device, and the trained parameters may be utilized in the inference device to obtain an inference result.

[0013] In the present disclosure, the inference result is exemplified as transmission control information that controls whether or not transmission from the base station device to the central device is permitted.

[0014] In the embodiments described below, the form of communication in the communication system is described as being divided into a learning phase in which a process of generating a trained model is performed, and an utilization phase in which a process of obtaining an inference result using the generated trained model is performed.

[0015] When the learning phase is executed and when the utilization phase is executed, a configuration in which the connection relationships between base station devices or between base station devices and a central device are different may be adopted. Such differences are not essential to the present disclosure, and both the learning phase and the utilization phase may be executed with the same connection relationships.

[0016] In the connection relationship, when the learning phase is executed, connections unnecessary for the execution of the utilization phase may be redundant, and vice versa. Such redundancy does not necessarily correspond to the configuration for preparing for increased traffic volume, commonly known as redundancy in communication systems.

[0017] As will be mentioned in the individual embodiments below, the machine learning may be supervised learning, unsupervised learning, or reinforcement learning.

[0018] <1. First Embodiment> <1-1. Learning Phase> 1 is a block diagram illustrating a communication system 91A in which a learning phase is executed according to the first embodiment of the present disclosure. The block diagram can also be said to be a block diagram illustrating a configuration for performing learning in a learning device 10A according to the first embodiment of the present disclosure.

[0019] The communication system 91A includes a plurality of base station devices 21, 22, ... 2N (where N is an integer of 2 or more), a ground device 1 functioning as a central device, and a mobile station device 3. It can also be said that there are a plurality of base station devices 21 to 2N.

[0020] Base station devices 21, 22, ..., 2N are connected in this order by wired lines 71. For convenience, the direction from base station device 21 to base station device 2N is referred to as downstream, and the direction from base station device 2N to base station device 21 is referred to as upstream. For example, base station device 2N is downstream of base station device 22, base station devices 22 and 2N are downstream of base station device 21, base station devices 21 and 22 are upstream of base station device 2N, and base station device 21 is upstream of base station device 22.

[0021] Base station device 21 is also connected to ground equipment 1 by a line 71. Base station device 2N is also connected to ground equipment 1 by a line 71. It can be said that line 71 connects ground equipment 1 and base stations 21 to 2N in a ring shape.

[0022] This disclosure describes suppression of traffic volume when multiple base station devices 21, 22, ... 2N each receive a radio signal (hereinafter also referred to as a "mobile station signal") 31, 32, ... 3N transmitted from the same mobile station device 3. There may be multiple mobile station devices 3, and the technology according to this disclosure may be applied to each of them.

[0023] <1-1-1. Wireless line quality> 2 is a flowchart illustrating a process for obtaining the best wireless channel quality and corresponding base station information (hereinafter tentatively referred to as the "best quality selection process"). The best quality selection process determines the best quality among mobile station signals 31-3N, which are wireless signals, and selects the base station device 21-2N that has received the mobile station signal with the best quality.

[0024] The quality of each of the mobile station signals 31 to 3N is represented by wireless channel quality B11 to B1N. The quality is determined based on one or more of received signal strength commonly known as RSSI (Received Signal Strength Indicator), frequency deviation from a reference frequency, and error correction.

[0025] The best quality selection step includes steps S201, S202, S203, S204, S205, and S206. The best quality selection step is executed in each of the base station devices 21 to 2N.

[0026] 2 illustrates the best quality selection process executed in base station device 2q (q represents all integers greater than or equal to 1 and less than or equal to N). However, when q=1, that is, when the best quality selection process is executed in base station device 21, step S206 is partially reinterpreted as described below. When q=N, the best quality selection process is executed in base station device 2N by omitting steps S201 and S204.

[0027] In step S201, it is determined whether wireless channel quality B10 and base station information B20 have been acquired from downstream base station device 2d (d is an integer greater than q by 1). Specifically, this acquisition is reception via line 71.

[0028] Wireless channel quality B10 is the best wireless channel quality for all downstream base station devices 2d to 2N. Base station devices 21 to 2N are identified by base station information B21 to B2N, respectively. An example of base station information is a base station number. Base station information B20 is information that identifies the base station device 2d to 2N that has received a mobile station signal having wireless channel quality B10.

[0029] Step S201 is repeatedly executed, making a negative determination until the wireless channel quality B10 and the base station information B20 are acquired. Step S201 can be considered a process of waiting until the base station device 2q acquires the wireless channel quality B10 and the base station information B20 from the downstream side.

[0030] When the wireless channel quality B10 and the base station information B20 are acquired, an affirmative determination is made in step S201, and step S202 is executed.

[0031] In step S202, it is determined whether the base station device 2q has received the mobile station signal 3q. Until the mobile station signal 3q is received, a negative determination is made in step S202, and step S202 is repeatedly executed. Step S202 can be considered a process of waiting until the base station device 2q receives the mobile station signal 3q.

[0032] When the mobile station signal 3q is received, the determination in step S202 is affirmative, and step S203 is executed. In step S203, the wireless channel quality B1q of the mobile station signal 3q is calculated.

[0033] After a positive determination is made in step S202, step S203 can be executed. The order of execution of steps S201 and S202 may be reversed. Step S201 may be executed after step S203, without preceding step S202.

[0034] After step S203 is executed and a positive determination is made in step S201, step S204 is executed.

[0035] In step S204, it is determined whether the wireless channel quality B1q is better than the wireless channel quality B10. If the result of the determination is positive, step S205 is executed.

[0036] If the result of the determination in step S204 is affirmative, wireless channel quality B1q is the best wireless channel quality for all of base station devices 2q to 2N, and so wireless channel quality B1q is adopted as wireless channel quality B10 in step S205. Correspondingly, base station information B2q is adopted as base station information B20. Execution of step S205 can be said to be updating wireless channel quality B10 and base station information B20.

[0037] If the result of the determination in step S204 is negative, step S205 is not executed, and the wireless channel quality B10 and the base station information B20 are maintained without being updated. Even when the wireless channel quality B10 is maintained, the wireless channel quality B10 is the best wireless channel quality for all of the base station devices 2q to 2N, just as when the wireless channel quality B10 is updated. The same applies to the base station information B20.

[0038] If the result of the determination in step S204 is negative, or if step S205 has been executed, step S206 is executed. In step S206, the wireless channel quality B10 and the base station information B20 are output to an upstream base station device 2u (u is an integer smaller than q by 1).

[0039] Step S206 is executed, and the best quality selection process is completed. Execution of step S206 in base station device 2q makes the determination result of step S201 in base station device 2u positive.

[0040] In the best quality selection process performed in the base station device 21, the "upstream base station device 2u" in step S206 is replaced with the "terrestrial device 1".

[0041] In the best quality selection process performed in base station device 2N, steps S201 and S204 are omitted, and steps S202, S203, S205, and S206 are performed in this order, because there is no base station device downstream of base station device 2N.

[0042] As a result of the best quality selection process being executed in the base station devices 21 to 2N, the best wireless link quality B10 among the base station devices 21 to 2N is output to the ground equipment 1 together with base station information B20 that identifies the base station device that received the mobile station signal, which is a wireless signal having this quality.

[0043] In the above, "upstream" and "downstream" are used for convenience in terms of their directions, and the direction from base station device 21 to base station device 2N may be considered upstream, and the direction from base station device 2N to base station device 21 may be considered downstream. Alternatively, for example, N = n + n and b = n + 1 (n is an integer greater than or equal to 1), and both the direction from base station device 21 to base station device 2n and the direction from base station device 2N to base station device 2b may be set downstream, with base station devices 2n and 2b being considered to be furthest downstream. In this case, "downstream" can be said to be the direction from near to ground device 1 to far away.

[0044] <1-1-2. Learning using good wireless link quality> 3 is a block diagram illustrating the configuration and input / output of a learning device 10A according to the first embodiment of the present disclosure. The learning device 10A is connected to a ground device 1. The ground device 1 receives wireless channel quality B10 and base station information B20 output from, for example, a base station device 21 or a base station device 2N. The learning device 10A includes a data acquisition unit 101 and a model generation unit 102.

[0045] 4 is a flowchart illustrating a process for generating a trained model M01 (hereinafter also tentatively referred to as a "model generation process"). The model generation process is executed by a learning device 10A. The model generation process includes steps S101, S102, S103, and S104.

[0046] In step S101, the data acquisition unit 101 acquires the wireless channel quality B10 and the base station information B20 from the ground equipment 1. After step S101 is executed, step S102 is executed. In step S102, learning data D1 is generated. The learning data D1 is data in which the wireless channel quality B10 and the base station information B20 are associated with each other.

[0047] For example, the data acquisition unit 101 provides the wireless channel quality B10 and the base station information B20 to the model generation unit 102 (FIG. 3). In this case, step S102 is executed by the model generation unit 102.

[0048] In this embodiment, the wireless channel quality B10 and base station information B20 input to the data acquisition unit 101 are associated with each other. The data acquisition unit 101 may generate learning data D1 and output it to the model generation unit 102. In this case, step S102 is executed by the data acquisition unit 101.

[0049] After step S102 is executed, step S103 is executed. In step S103, the model generation unit 102 performs learning based on the learning data D1, specifically, learns the transmission control information C0. The transmission control information C0 is information used for inference to control whether or not each of the base station devices 21 to 2N is permitted to transmit to the same ground device 1 (i.e., whether or not transmission is possible).

[0050] Through the learning in step S103, the model generation unit 102 generates a trained model M01. After step S103 is executed, step S104 is executed.

[0051] In step S104, the generated trained model M01 is output from the learning device 10A, more specifically, from the model generation unit 102 to the trained model storage unit 13. After step S104 is executed, the model generation process ends.

[0052] The trained model storage unit 13 stores the trained model M01. The trained model storage unit 13 may be considered to be included in the communication system 91A, or may be considered to be separate from the communication system 91A.

[0053] For example, the best quality selection step is performed on a plurality of mobile station signals 3q transmitted from different positions by the mobile station device 3. The learning device 10A acquires wireless channel qualities B10 and base station information B20 for a plurality of positional relationships between the mobile station device 3 and the base station devices 21 to 2N. The learning data D1 acquired in this manner for the plurality of different positional relationships is stored in the model generation unit 102. The trained model M01 may be updated each time the learning data D1 is stored.

[0054] The trained model M01 functions as a model for classifying wireless link quality and base station information into one or more cluster groups each consisting of good, e.g., the best, wireless link quality and base station information from which a mobile station signal having the good wireless link quality has been received. This function contributes to inference in the utilization phase, which will be described later.

[0055] <1-1-3. Example of unsupervised learning> In step S103, for example, unsupervised learning is adopted as the learning algorithm. Unsupervised learning is a method of learning features in training data by providing the training data, which does not include results (labels), to a learning device, particularly the model generation unit 102 in this disclosure.

[0056] For example, a grouping method using k-means clustering can be applied as unsupervised learning. The k-means clustering is a non-hierarchical clustering algorithm that uses the cluster mean to classify a given number of clusters into k.

[0057] Specifically, the k-means algorithm works as follows: First, each data item xi is randomly assigned to a cluster. Next, the center Vj of each cluster is calculated based on the assigned data. Next, the distance between each data item xi and the center Vj is calculated, and the data item xi is reassigned to the cluster with the closest center. If the cluster assignment for all data items xi remains unchanged through the above process, or if the amount of change falls below a preset threshold, the algorithm is deemed to have converged and the process ends.

[0058] <1-2. Utilization Phase> 5 is a block diagram illustrating a communication system 91B in which the utilization phase according to the first embodiment of the present disclosure is executed. This block diagram can also be said to be a block diagram illustrating inputs and outputs of inference devices 41, 42, ..., 4N according to the first embodiment of the present disclosure.

[0059] Similar to the communication system 91A, the communication system 91B includes a plurality of base station devices 21, 22, ... 2N, a ground device 1, and a mobile station device 3. However, the ground device 1 shares a line network (illustrated as a LAN (Local Area Network) in the figure) 8 with all the base station devices 21 to 2N. Unlike the communication system 91A, the communication system 91B allows transmission from any of the base station devices 21 to 2N to the same ground device 1 via the line network 8. The line network 8 can be considered to be included in the communication system 91B.

[0060] In the communication system 91A, it is not necessary that transmission from all of the base station devices 21 to 2N to the ground equipment 1 is impossible. In the communication system 91A, it is necessary that transmission from either one or both of the base station devices 21 and 2N to the ground equipment 1 is possible.

[0061] The ground equipment 1 and the base station equipment 21 to 2N are connected to the network 8 by a line 72. The line 72 may be a wired line or a wireless line. The network 8 may be a wired line or a wireless line.

[0062] The communication system 91B includes inference devices 41, 42, ..., 4N, which are connected to base station devices 21, 22, ..., 2N, respectively.

[0063] All of the inference devices 41 to 4N are connected to the trained model storage unit 13. In the utilization phase, the trained model M01 generated in the learning phase and stored in the trained model storage unit 13 is utilized for inference in machine learning.

[0064] The trained model storage unit 13 may be considered to be included in the communication system 91B, or may be considered to be separate from the communication system 91B.

[0065] 5 illustrates a configuration in which communication system 91B does not include line 71 and learning device 10A (see FIG. 1 for both), but these may be included in communication system 91B. However, line 71 and learning device 10A are not essential in the utilization phase of embodiment 1.

[0066] 1 illustrates a configuration in which the communication system 91A does not include the line 72, the line network 8, and the inference devices 41 to 4N, but these may be included in the communication system 91A. However, the line 72, the line network 8, and the inference devices 41 to 4N are not essential for the learning phase of the first embodiment.

[0067] <1-2-1. Reasoning> 6 is a block diagram illustrating the configuration and input / output of inference devices 41 to 4N. Inference device 41 includes a data acquisition unit 401 and an inference unit 402. Inference devices 42 to 4N are configured in the same way as inference device 41, and are therefore illustrated in a simplified manner.

[0068] The data acquisition unit 401 of the inference device 41 receives as input wireless channel quality B11 and base station information B21 from the base station device 21. The inference units 402 of all of the inference devices 41 to 4N receive as input trained model M01 from the trained model storage unit 13. The inference unit 402 of the inference device 41 receives as input wireless channel quality B11 and base station information B21 from the data acquisition unit 401.

[0069] The inference unit 402 of the inference device 41 outputs the transmission control information C1 to the base station device 21. Whether or not the base station device 21 is allowed to transmit the base station signal T1 to the network 8 is controlled by the transmission control information C1.

[0070] Base station device 21 includes a transmission control unit 211, a transmission / reception circuit 212, a memory 213, and a control unit 210. The transmission control unit 211, the transmission / reception circuit 212, the memory 213, and the control unit 210 are connected by a bus 218. The bus 218 is connected to a line 72. Base station devices 22 to 2N are configured in the same manner as base station device 21, and are therefore illustrated in a simplified manner.

[0071] The transmitting / receiving circuit 212 receives the mobile station signal 31. The transmitting / receiving circuit 212 transmits the wireless channel quality B11 and the base station information B21 to the inference device 41, more specifically to the data acquisition unit 401 thereof.

[0072] The transmission / reception circuit 212 obtains the wireless channel quality B11 from the mobile station signal 31. The mobile station signal 31 received by the transmission / reception circuit 212 may be provided to the control unit 210 via the bus 218, the control unit 210 may obtain the wireless channel quality B11, and the wireless channel quality B11 may be provided to the transmission / reception circuit 212 via the bus 218.

[0073] Transmission of the base station signal T1 to the network 8 via the bus 218 and the line 72, and ultimately to the ground equipment 1, is performed by the transmission / reception circuit 212 under the control of the transmission control unit 211. When the transmission control information C1 input from the inference unit 402 of the inference device 41 indicates permission for transmission, the transmission control unit 211 permits the transmission / reception circuit 212 to transmit the base station signal T1. When the input transmission control information C1 does not indicate permission for transmission, the transmission control unit 211 prohibits the transmission / reception circuit 212 from transmitting the base station signal T1.

[0074] Similarly, base station devices 22 to 2N each receive mobile station signals 32 to 3N, transmit wireless line qualities B12 to B1N to inference devices 42 to 4N, transmit base station information B22 to B2N to inference devices 42 to 4N, receive transmission control information C2 to CN, and transmit base station signals T2 to TN if permitted by the transmission control information C2 to CN.

[0075] 7 is a flowchart illustrating a process for generating transmission control information Cq. This process is executed by, for example, the inference unit 402 of the inference device 4q. This process includes steps S41, S42, S43, and S44.

[0076] In step S41, the inference unit 402 acquires the wireless channel quality B1q and the base station information B2q from the data acquisition unit 401.

[0077] Thereafter, in step S42, the inference unit 402 acquires the trained model M01 from the trained model storage unit 13.

[0078] In step S43, the inference unit 402 inputs the wireless channel quality B1q and the base station information B2q into the trained model M01, and generates transmission control information Cq as an inference result using a known method in machine learning.

[0079] Only when it is inferred that the wireless channel quality B1q is good, the transmission control information Cq indicates permission for the transmission of the base station signal Tq from the base station device 2q.

[0080] Specifically, for example, only when the wireless line quality B1q and the base station information B2q are classified into one or more cluster groups consisting of the wireless line quality B10 and the base station information B20 using the trained model M01, the transmission control information Cq indicates permission to transmit the base station signal Tq from the base station device 2q.

[0081] In step S44, the inference unit 402 outputs the transmission control information Cq to the base station device 2q, more specifically to the transmission control unit 211 thereof.

[0082] <1-2-2. Transmission to Ground Device 1> 8 is a flowchart illustrating an operation of the base station device 2q in the utilization phase according to the first embodiment. This process includes steps S207, S208, S209, S210, S211, and S212.

[0083] In step S207, it is determined whether the base station device 2q has received the mobile station signal 3q. If the determination is negative, step S207 is repeatedly executed until a positive determination is obtained. Step S207 can be considered a step of waiting until the mobile station signal 3q is received.

[0084] If the determination result in step S207 is affirmative, step S208 is executed. In step S208, the wireless channel quality B1q of the mobile station signal 3q is calculated. This calculation is executed by the transmission / reception circuit 212, for example.

[0085] When step S208 is completed, step S209 is executed. In step S209, for example, the transmission / reception circuit 212 outputs the wireless line quality B1q and the base station information B2q to the inference device 4q, more specifically to its data acquisition unit 401. The base station information B2q is stored in, for example, the memory 213, and is provided from the memory 213 to the transmission / reception circuit 212 via the bus 218. The exchange of the base station information B2q is simply represented in FIG. 6 by an arrow pointing from the memory 213 to the transmission / reception circuit 212.

[0086] 7, the inference device 4q generates transmission control information Cq as an inference result using the trained model M01, the wireless channel quality B1q, and the base station information B2q. In step S210, it is determined whether the transmission control information Cq has been input to the base station device 2q, more specifically, to its transmission control unit 211.

[0087] If the determination is negative, step S210 is repeatedly executed until a positive determination is obtained. Step S210 can be said to be a step of waiting for the input of transmission control information Cq.

[0088] If the determination result in step S210 is affirmative, step S211 is executed. In step S211, the transmission control unit 211 determines whether the transmission control information Cq permits transmission of the base station signal Tq.

[0089] When the wireless channel quality B1q and the base station information B2q are classified into one or more cluster groups each consisting of the wireless channel quality B10 and the base station information B20, the determination is affirmative. If the determination is affirmative, the base station device 2q is permitted to transmit the base station signal Tq by the transmission control unit 211. The base station device 2q transmits the mobile station signal 3q as the base station signal Tq to the ground device 1 via the line network (for example, the "LAN" as exemplified in FIG. 5) 8.

[0090] If the determination is negative, the operation of the base station device 2q ends without transmitting the base station signal Tq. The base station device 2q operates again in accordance with the flowchart starting from step S207.

[0091] According to the first embodiment of the present disclosure, by using the above-described learning and inference, it is possible to determine whether or not to transmit base station signals T1-TN to the network 8, and ultimately to the ground equipment 1. Such a determination contributes to reducing the amount of traffic in the communication system 91B that uses the network 8 shared by the ground equipment 1 and the base station devices 21-2N.

[0092] Determining whether to transmit base station signals T1 to TN to the ground equipment 1 based on the level of wireless line qualities B11 to B1N contributes to reducing traffic volume. Inference using the trained model M01 described above contributes to either avoiding transmission of poor-quality base station signals T1 to TN or transmitting good-quality signals, or both.

[0093] For example, the trained model M01 may be generated not only when the wireless channel quality B10 is the best, but also when multiple good wireless channel qualities B10 are adopted. The use of such a trained model M01 contributes to suppressing the occurrence of an event in which all of the base station signals T1 to TN are not transmitted.

[0094] The trained model M01 is generated, for example, in the model generation unit 102 by training using a quality lower than the wireless channel quality B10.

[0095] Alternatively, one or more poor wireless channel qualities B1q may be adopted to generate the trained model M01. For example, in this case, only when it is inferred that the wireless channel quality B1q is poor, the transmission control information Cq indicates prohibition of transmission of the base station signal Tq from the base station device 2q.

[0096] Specifically, for example, only when the wireless channel quality B1q and the base station information B2q are classified into one or more cluster groups consisting of the wireless channel quality B10 and the base station information B20 using the trained model M01, the transmission control information Cq indicates prohibition of transmission of the base station signal Tq from the base station device 2q. In this case, too, it is possible to determine whether or not to transmit the base station signal Tq to the ground device 1 based on the level of the wireless channel quality B1q, which contributes to transmitting the base station signal Tq with good quality while suppressing the traffic volume.

[0097] <2. Second Embodiment> <2-1. Learning Phase> 9 is a block diagram illustrating the configuration and input / output of a learning device 10B according to the second embodiment of the present disclosure. The learning device 10B includes a data acquisition unit 101 and a model generation unit 102. The learning device 10B is connected to a ground device 1. The ground device 1 receives wireless channel qualities B11 to B1N and base station information B21 to B2N output from base station devices 21 to 2N.

[0098] To implement the learning phase using learning device 10B, the communication system 91A illustrated in Fig. 1 may be employed. In this case, for example, base station device 2q receives wireless channel qualities B1d-B1N and base station information B2d-B2N from downstream base station devices 2d-2N, and outputs wireless channel qualities B1q-B1N and base station information B2q-B2N to base station device 2u.

[0099] Such operations are, for example, as shown in the flowchart of FIG. 2: In step S201, "input wireless line quality B10 and base station information B20" is changed to "input wireless line quality B1d to B1N and base station information B2d to B2N"; Omission of execution of steps S204 and S205; In step S207, "wireless line quality B10 and base station information B20" is replaced with "wireless line quality B1q to B1N and base station information B2q to B2N." This can be achieved by performing the following.

[0100] However, base station device 2N calculates wireless channel quality B1N and outputs wireless channel quality B1N and base station information B2N to base station device 2t (t is an integer smaller than N by 1). Base station device 21 receives wireless channel qualities B12-B1N and base station information B22-B2N from downstream base station devices 22-2N, calculates wireless channel quality B11, and outputs wireless channel qualities B11-B1N and base station information B21-B2N to ground device 1.

[0101] To realize the learning phase using learning device 10B, communication system 91B illustrated in Fig. 5 may be employed. In this case, in communication system 91B, learning device 10B is connected to ground equipment 1. In this case, for example, base station devices 21 to 2N transmit wireless link qualities B11 to B1N, respectively, to ground equipment 1, and transmit base station information B21 to B2N, respectively, to ground equipment 1 via network 8. When the learning phase is realized in communication system 91B, there is a possibility that the amount of traffic will increase due to the transmission of wireless link qualities B11 to B1N and base station information B21 to B2N, which are used for learning, to ground equipment 1, but the amount of traffic in the utilization phase is suppressed as will be described later.

[0102] 10 is a flowchart illustrating a model generation process according to the second embodiment of the present disclosure. Trained models M1 to MN are generated by the model generation process. The model generation process includes steps S105, S106, S107, and S108.

[0103] In step S105, the data acquisition unit 101 acquires wireless channel qualities B11 to B1N and base station information B21 to B2N from the ground equipment 1. After step S105 is executed, step S106 is executed. In step S106, learning data D11 to D1N are generated. The learning data D1q is data in which the wireless channel quality B1q and the base station information B2q are associated with each other for each base station device 2q.

[0104] For example, the data acquisition unit 101 provides the wireless channel qualities B11 to B1N and the base station information B21 to B2N to the model generation unit 102. In this case, step S106 is executed by the model generation unit 102.

[0105] In this embodiment, the wireless channel quality B1q and the base station information B2q input to the data acquisition unit 101 are associated with each other. The data acquisition unit 101 may generate learning data D11 to D1N and output them to the model generation unit 102. In this case, step S106 is executed by the data acquisition unit 101.

[0106] The set of training data D11 to D1N can be seen as training data D1 used to generate trained models M1 to MN (see FIG. 9).

[0107] After step S106 is executed, step S107 is executed. In step S107, the model generation unit 102 performs learning based on the learning data D1, specifically, learns the transmission control information C01 to C0N. The transmission control information C01 to C0N is information used for inference to control whether or not to permit transmission from the base station devices 21 to 2N to the ground device 1. The set of the transmission control information C01 to C0N can be seen as the transmission control information C0 used for learning the learned models M1 to MN (see FIG. 9).

[0108] Through the learning in step S107, the model generation unit 102 generates trained models M1 to MN. In other words, a trained model Mq is generated for each base station device 2q to infer whether or not a base station signal Tq can be transmitted from the base station device 2q to the ground device 1.

[0109] After step S107 is executed, step S108 is executed. In step S108, the generated trained models M1 to MN are output from the learning device 10B, more specifically, from the model generation unit 102 to the trained model storage unit 13. After step S108 is executed, the model generation process ends.

[0110] As with the learning device 10A described in the first embodiment, the learning device 10B also generates trained models M1 to MN by, for example, unsupervised learning.

[0111] Trained models M1 to MN are stored in trained model storage unit 13. Trained model storage unit 13 may be considered to be included in communication system 91A or communication system 91B, or may be considered to be separate from communication system 91A or communication system 91B.

[0112] For example, the learning device 10B acquires wireless channel qualities B11 to B1N and base station information B21 to B2N for a plurality of positional relationships between the mobile station device 3 and the base station devices 21 to 2N. The learning data D1 acquired in this manner for a plurality of different positional relationships is stored in the model generation unit 102. The trained models M1 to MN may be updated each time the learning data D1 is stored.

[0113] <2-2. Utilization Phase> The utilization phase in the second embodiment of the present disclosure is also executed in, for example, the communication system 91B, similarly to the utilization phase in the first embodiment.

[0114] <2-2-1. Reasoning> FIG. 11 is a block diagram illustrating the configuration and input / output of an inference device 4q according to the second embodiment.

[0115] The inference devices 41 to 4N in the first embodiment can be used as the inference devices 41 to 4N in the second embodiment.

[0116] However, unlike the first embodiment, in the second embodiment, the inference unit 402 of the inference device 4q acquires the trained model Mq from the trained model storage unit 13. In other words, the inference in the inference devices 41 to 4N, specifically, the individual trained models M1 to MN corresponding to the base station devices 21 to 2N, are used for inference, specifically for determining whether or not to transmit the base station signals T1 to TN from the base station devices 21 to 2N.

[0117] 12 is a flowchart illustrating a process of generating transmission control information Cq in embodiment 2. This process is executed by, for example, the inference unit 402 of the inference device 4q. This process includes steps S41, S45, S49, and S44.

[0118] Steps S41 and S44 in the second embodiment are the same as steps S41 and S44 in the first embodiment, respectively, and therefore will not be described further.

[0119] After step S41 is executed, in step S45, the inference unit 402 acquires the trained model Mq from the trained model storage unit 13. This is different from step S42 shown in the first embodiment because inference is performed using an individual trained model Mq.

[0120] After step S45 is executed, in step S49, the wireless channel quality B1q and the base station information B2q are input to the trained model Mq to generate transmission control information Cq. After step S49 is executed, step S44 is executed.

[0121] <2-2-2. Transmission to Ground Device 1> The operations of the base station devices 21 to 2N in the utilization phase of the second embodiment can also be performed according to the flowchart illustrated in FIG.

[0122] Even if individual trained models M1 to MN corresponding to each of the base station devices 21 to 2N are used to generate the transmission control information C01 to C0N for each of the base station devices 21 to 2N in this way, it is possible to determine whether or not to transmit the base station signals T1 to TN to the network 8, and ultimately to the ground equipment 1, in the same manner as in embodiment 1. Such a determination contributes to reducing the amount of traffic in the communication system 91B that uses the network 8 shared by the ground equipment 1 and the base station devices 21 to 2N.

[0123] Determining whether to transmit the base station signals T1 to TN to the ground equipment 1 based on the level of the wireless line qualities B11 to B1N contributes to reducing traffic volume. Inference using the trained models M1 to MN described above contributes to either avoiding the transmission of poor-quality base station signals T1 to TN or transmitting good-quality signals, or both.

[0124] <3. Embodiment 3> <3-1. Learning Phase> 13 is a block diagram illustrating a communication system 91C in which the learning phase is executed according to the third embodiment of the present disclosure. This block diagram can also be said to be a block diagram illustrating a configuration for performing learning in a learning device 10C according to the third embodiment of the present disclosure.

[0125] The communication system 91C includes base station devices 21 to 2N, a ground device 1, and a mobile station device 3, similar to the communication systems 91A (see FIG. 1) and 91B (see FIG. 5) described in the first embodiment.

[0126] Unlike communication system 91A, communication system 91C is provided with learning device 10C instead of learning device 10A.

[0127] Unlike the communication system 91A, the communication system 91C does not require the lines 71 that connect the base station devices 21 to 2N with each other. The communication system 91C is provided with lines 73 that individually connect each of the base station devices 21 to 2N with the ground equipment 1. It can be said that the lines 71 that connect the base station devices 21 and 2N with the ground equipment 1 respectively are the lines 73 that connect the base station devices 21 and 2N with the ground equipment 1 respectively in the communication system 91C.

[0128] Base station devices 21 to 2N receive mobile station signals 31 to 3N, respectively, from mobile station device 3. Base station device 2q calculates wireless channel quality B1q from mobile station signal 3q. Base station device 2q outputs wireless channel quality B1q together with base station information B2q to ground device 1 via line 73.

[0129] The learning device 10C acquires wireless channel qualities B11 to B1N, base station information B21 to B2N, and evaluation criterion information B31 to B3N (described later) from the ground device 1, and generates a trained model M02. The trained model M02 is stored in the trained model storage unit 13.

[0130] In view of such operations, the learning phase in the second embodiment may be executed in the communication system 91C. In that case, the learning device 10C in the communication system 91C is replaced with the learning device 10B, and trained models M1 to MN are generated in place of the trained model M02. The evaluation criterion information B31 to B3N is not necessarily required for the trained models M1 to MN.

[0131] 14 is a block diagram illustrating the configuration and input / output of a learning device 10C according to the third embodiment. The learning device 10C includes a data acquisition unit 101 and a model generation unit 102. The learning device 10C is connected to the ground device 1.

[0132] The ground equipment 1 receives the wireless channel qualities B11 to B1N and the base station information B21 to B2N output from the base station devices 21 to 2N. The ground equipment 1 generates evaluation reference information B31 to B3N from the wireless channel qualities B11 to B1N. The evaluation reference information B31 to B3N functions as correct answer information corresponding to each of the transmission control information C01 to C0N in the learning phase.

[0133] The ground device 1 provides the evaluation reference information B31 to B3N to the learning device 10C, more specifically to the data acquisition unit 101, together with the wireless channel qualities B11 to B1N and the base station information B21 to B2N.

[0134] <3-1-1. Generation of correct answer information> 15 is a flowchart illustrating a process of generating evaluation reference information B31 to B3N (hereinafter also referred to as a "correct answer information generating process"). The correct answer information generating process is executed by, for example, the ground equipment 1 in the learning phase. FIG. 15 can also be said to be a flowchart illustrating the operation of the ground equipment 1 in the learning phase of the third embodiment.

[0135] The correct answer information generation process includes steps S11, S12, S13, and S14. In step S11, the ground equipment 1 determines whether it has received the wireless channel qualities B11 to B1N and the base station information B21 to B2N. Until the wireless channel qualities B11 to B1N and the base station information B21 to B2N are acquired, a negative determination is made in step S11, and step S11 is repeatedly executed. Step S11 can be considered a process in which the ground equipment 1 waits until it acquires the wireless channel qualities B11 to B1N and the base station information B21 to B2N.

[0136] When the wireless channel qualities B11 to B1N and the base station information B21 to B2N are acquired, an affirmative determination is made in step S11, and step S12 is executed.

[0137] In step S12, the ground device 1 determines one or more wireless channel qualities B1j (j is one or more of (N-1) integers between 1 and N) from among the wireless channel qualities B11 to B1N in order of best. The best wireless channel quality B1j corresponds to the wireless channel quality B10 in the first embodiment.

[0138] After step S12 is executed, step S13 is executed. In step S13, different information is assigned to evaluation criterion information B3j corresponding to wireless channel quality B1j and all other evaluation criterion information B3i (where i is an integer between 1 and N, and represents all integers different from integer j). For example, a value of 1 is assigned to evaluation criterion information B3j, and a value of 0 is assigned to other evaluation criterion information B3i.

[0139] After step S13 is executed, step S14 is executed. In step S14, the ground equipment 1 outputs the wireless channel qualities B11 to B1N, the base station information B21 to B2N, and the evaluation criterion information B31 to B3N to the learning device 10C.

[0140] Alternatively, the evaluation criterion information B31 to B3N may be generated by the data acquisition unit 101. In this case, step S14 is not necessary.

[0141] After obtaining the evaluation criterion information B31 to B3N, the data acquisition unit 101 outputs the associated data group D2 and the evaluation criterion information group B3 to the model generation unit 102. The model generation unit 102 collects the associated data group D2 and the evaluation criterion information group B3 to generate learning data D3, which is used for learning the transmission control information C01 to C0N.

[0142] The associated data group D2 is a collection of associated data D21 to D2N. The associated data D2q is data in which the wireless channel quality B1q and the base station information B2q are associated with each other. The learning data D3 includes the associated data group D2.

[0143] The data acquisition unit 101 may not generate the associated data group D2, and the model generation unit 102 may obtain the associated data D21 to D2N from the data acquisition unit 101 and generate the associated data group D2.

[0144] The evaluation criterion information group B3 is a collection of evaluation criterion information B31 to B3N. The data acquisition unit 101 may not generate the evaluation criterion information group B3, but the model generation unit 102 may obtain the evaluation criterion information B31 to B3N from the data acquisition unit 101 and generate the evaluation criterion information group B3.

[0145] Either or both of the associated data group D2 and the evaluation criterion information group B3 may not be generated by the data acquisition unit 101, and the associated data D21 to D2N and the evaluation criterion information B31 to B3N may be collectively treated as learning data D3 in the model generation unit 102.

[0146] 16 is a flowchart illustrating a model generation process according to the third embodiment. A trained model M02 is generated by the model generation process. The model generation process includes steps S109, S110, S111, and S112.

[0147] In step S109, the data acquisition unit 101 acquires wireless channel qualities B11 to B1N, base station information B21 to B2N, and evaluation reference information B31 to B3N from the ground equipment 1. After step S109 is executed, step S110 is executed. In step S110, learning data D3 is generated.

[0148] For example, the data acquisition unit 101 acquires wireless channel qualities B11 to B1N, base station information B21 to B2N, and evaluation criterion information B31 to B3N from the ground equipment 1 at different times.

[0149] In this embodiment, the wireless channel quality B1q and base station information B2q input to the data acquisition unit 101 are associated with each other. For example, the data acquisition unit 101 generates an associated data group D2 (or associated data D21 to D2N) and provides this and an evaluation criterion information group B3 (or evaluation criterion information B31 to B3N) to the model generation unit 102 (see FIG. 14). In this case, step S110 is executed by the model generation unit 102.

[0150] For example, the data acquiring unit 101 acquires evaluation reference information B31 to B3N at the same time that it acquires wireless channel qualities B11 to B1N and base station information B21 to B2N from the ground equipment 1. For example, the data acquiring unit 101 provides the wireless channel qualities B11 to B1N, the base station information B21 to B2N, and the evaluation reference information B31 to B3N to the model generating unit 102. In this case, too, the associated data D21 to D2N may be generated by the data acquiring unit 101, and step S110 may be executed by the model generating unit 102. On the other hand, the wireless channel quality B1q, the base station information B2q, and the evaluation reference information B3q input to the data acquiring unit 101 are associated with each other, so step S110 may be executed by the data acquiring unit 101.

[0151] After step S110 is executed, step S111 is executed. In step S111, the model generation unit 102 performs learning based on the learning data D3, specifically, learning the transmission control information C01 to C0N. The set of the transmission control information C01 to C0N can be regarded as the transmission control information C0 used for learning the learned model M02 (see FIG. 14).

[0152] <3-1-2. Example of supervised learning> In the above step S111, for example, supervised learning is adopted as the learning algorithm. Supervised learning refers to a technique in which a set of input and result (label) data is provided to a learning device, particularly the model generation unit 102 in this disclosure, to learn features in the learning data and infer a result from the input.

[0153] FIG. 17 is a block diagram illustrating a neural network 93 that can be employed by the model generating unit 102 in the third embodiment for supervised learning, for example.

[0154] The neural network 93 includes an input layer X0, an intermediate layer (hidden layer) Y0, and an output layer Z0. The input layer X0 is made up of a plurality of neurons, N in this case, X1, X2, ..., XN. The intermediate layer Y0 is made up of a plurality of neurons, two in this case, Y1, Y2. The output layer Z0 is made up of a plurality of neurons, N in this case, Z1, Z2, ..., ZN. The intermediate layer Y0 may be a single layer as illustrated, or may be two or more layers.

[0155] For example, neuron Xq receives associated data D2q as input. Neuron Xq multiplies associated data D2q by a weight vq1 and outputs the result to neuron Y1. Neuron Xq multiplies associated data D2q by a weight vq2 and outputs the result to neuron Y2.

[0156] Neuron Y1 receives N outputs from neurons X1 to XN and performs an operation (for example, addition) on them. Neuron Y1 multiplies the result of the operation by a weight w1q and outputs the result to neuron Zq.

[0157] Neuron Y2 receives N outputs from neurons X1 to XN and performs an operation (for example, addition) on them. Neuron Y2 multiplies the result of the operation by a weight w2q and outputs the result to neuron Zq.

[0158] Neuron Zq receives two outputs from neurons Y1 and Y2, performs an operation (for example, addition) on these outputs, and outputs the result of the operation as transmission control information C0q.

[0159] The weights v11 to vN1, v21 to vN2, w11 to v1N, and w21 to w2N are adjusted by using the evaluation reference information B31 to B3N as correct answer information for each of the transmission control information C01 to C0N. This adjustment can be considered as generation of trained parameters or generation of a trained model from the neural network 93.

[0160] In the present disclosure, the neural network 93 can be said to learn the transmission control information C0 by so-called supervised learning in accordance with the learning data D3 (step S110) created based on the corresponding data D21 to D2N acquired by the data acquisition unit 101 and the evaluation criteria information B31 to B3N (step S109) as correct answer information.

[0161] The neural network 93 receives, as input to the input layer X0, associated data D21-D2N that associates wireless channel qualities B11-B1N with base station information B21-B2N. The neural network 93 performs learning by adjusting the weights v11-vN1, v21-vN2, w11-v1N, and w21-w2N so that the resulting transmission control information C01-C0N output from the output layer Z0 approaches the evaluation reference information B31-B3N. The model generation unit 102 performs the above learning to generate a trained model M02 (step S111) and outputs the trained model M02 to the trained model storage unit 13 (step S112).

[0162] <3-2. Utilization Phase> The utilization phase in the third embodiment of the present disclosure is also executed in, for example, the communication system 91B, similarly to the utilization phase in the first embodiment.

[0163] <3-2-1. Reasoning> FIG. 18 is a block diagram illustrating the configuration and input / output of an inference device 4q according to the third embodiment.

[0164] The inference devices 41 to 4N in the third embodiment can be the inference devices 41 to 4N in the first embodiment.

[0165] However, unlike the first embodiment, in the second embodiment, the inference unit 402 of the inference device 4q acquires the learned model M02 from the learned model storage unit 13 and acquires the associated data D2q from the data acquisition unit 401. The data acquisition unit 401 of the inference device 4q receives the wireless channel quality B1q and the base station information B2q from the base station device 2q as input.

[0166] 19 is a flowchart illustrating a process for generating transmission control information Cq in the third embodiment. This process is executed by, for example, the inference unit 402 of the inference device 4q. This process includes steps S46, S47, S48, and S44.

[0167] Step S44 in the third embodiment is the same as step S44 in the first embodiment, and therefore a description thereof will be omitted.

[0168] In step S46, the inference unit 402 acquires the association data D2q. For example, the inference unit 402 acquires the association data D2q from the data acquisition unit 401. Alternatively, similar to step S41 (see FIG. 7), the inference unit 402 acquires the wireless channel quality B1q and the base station information B2q from the data acquisition unit 401, and generates the association data D2q based on these.

[0169] After step S46 is executed, in step S47, the inference unit 402 acquires the trained model M02 from the trained model storage unit 13.

[0170] After step S47 is executed, in step S48, the correspondence data D2q is input to the trained model M02 to generate transmission control information Cq. After step S48 is executed, step S44 is executed.

[0171] <3-2-2. Transmission to Ground Device 1> The operations of the base station devices 21 to 2N in the utilization phase of the third embodiment can also be performed according to the flowchart illustrated in FIG.

[0172] Even if the trained model M02 is used to generate the transmission control information C01 to C0N for each of the base station devices 21 to 2N in this way, it is possible to determine whether or not to transmit the base station signals T1 to TN to the network 8, and ultimately to the ground equipment 1, in the same manner as in embodiment 1. Such a determination contributes to reducing the amount of traffic in the communication system 91B that uses the network 8 shared by the ground equipment 1 and the base station devices 21 to 2N.

[0173] <3-3. Combination with embodiment 2> In the third embodiment, as in the second embodiment, an individual trained model Mq may be generated for the base station device 2q, and the transmission control information Cq may be obtained by inference using the trained model Mq.

[0174] For example, the learning device 10C associates the wireless channel quality B1q and base station information B2q output from the ground equipment 1 to obtain associated data D2q. The learning device 10C generates a trained model Mq by learning using the evaluation reference information B3q and the associated data D2q output from the ground equipment 1. The inference device 4q performs inference using the wireless channel quality B1q and base station information B2q obtained from the base station device 2q and the trained model Mq to generate transmission control information Cq.

[0175] <4. Embodiment 4> <4-1. Learning Phase> 20 is a block diagram illustrating a communication system 91D in which the learning phase is executed according to the fourth embodiment of the present disclosure. This block diagram can also be said to be a block diagram illustrating a configuration for performing learning in a learning device 10D according to the fourth embodiment of the present disclosure.

[0176] The communication system 91D includes base station devices 21 to 2N, a ground device 1, a mobile station device 3, and a line 73, similar to the communication system 91C (see FIG. 13) described in the third embodiment.

[0177] Unlike the communication system 91C, the communication system 91D is provided with a learning device 10D instead of the learning device 10C. The learning device 10D is connected to the ground equipment 1.

[0178] Unlike in the third embodiment, the ground equipment 1 does not need to generate evaluation reference information B31 to B3N. In the fourth embodiment, the ground equipment 1 receives, via a line 73, wireless channel qualities B11 to B1N and base station information B21 to B2N output from the base station devices 21 to 2N, respectively.

[0179] The learning device 10D acquires the wireless channel qualities B11 to B1N and the base station information B21 to 2N from the ground device 1, and generates a trained model M03. The trained model M03 is stored in the trained model storage unit 13.

[0180] 21 is a block diagram illustrating the configuration and input / output of a learning device 10D according to Embodiment 4. The learning device 10D includes a data acquiring unit 101 and a model generating unit .

[0181] The ground equipment 1 receives the wireless channel qualities B11 to B1N and the base station information B21 to B2N output from the base station devices 21 to 2N. The ground equipment 1 provides the wireless channel qualities B11 to B1N and the base station information B21 to B2N to the learning device 10D, more specifically, to the data acquisition unit 101.

[0182] In the fourth embodiment, neither the best wireless channel quality B10 as in the first embodiment nor the evaluation reference information B31 to B3N as correct answer information as in the third embodiment is required.

[0183] The model generation unit 102 obtains training data D4 using the wireless channel qualities B11 to B1N and the base station information B21 to B2N. For example, the training data D1 (see FIG. 9) as a set of the training data D11 to D1N exemplified in the second embodiment is used as the training data D4.

[0184] Alternatively, for example, the associated data group D2 exemplified in the third embodiment is employed, and in the associated data D2q, the wireless channel quality B1q and the base station information B2q are associated with each other (see FIG. 14). In the learning data D4, for example, the associated data D21 to D2N are ranked based on the quality of the wireless channel qualities B11 to B1N, respectively.

[0185] In the learning device 10D, the model generation unit 102 selects one or more of the wireless link qualities B11 to B1N in descending order of rank, for example, in descending order of quality. The selected one of the wireless link qualities B11 to B1N corresponds to the wireless link quality B1j in the third embodiment. In other words, the learning device 10D executes steps S11, S12, and S13 (see FIG. 15) that were performed by the ground equipment 1 in the third embodiment.

[0186] In the learning device 10D, the model generation unit 102 learns the transmission control information C0 so as to permit only transmission from the base station device 2j corresponding to the wireless channel quality B1j. As a result of this learning, a trained model M03 is generated.

[0187] Alternatively, the model generation unit 102 in the learning device 10D selects one or more of the wireless channel qualities B11 to B1N, for example, from the lowest quality, as the wireless channel quality B1i. The model generation unit 102 in the learning device 10D learns the transmission control information C0 so as to prohibit only transmission from the base station device 2j corresponding to the wireless channel quality B1i. As a result of this learning, a trained model M03 is generated.

[0188] 22 is a flowchart illustrating a model generation process according to the fourth embodiment. A trained model M03 is generated by the model generation process. The model generation process includes steps S105, S196, S197, and S198.

[0189] Step S105 has been described in the second embodiment (see FIG. 10), and therefore its description will be omitted in the fourth embodiment.

[0190] After step S105 is executed, step S196 is executed. In step S196, training data D4 is generated. The training data D4 is, for example, the training data D1 exemplified in the second embodiment (see FIG. 9), or the associated data group D2 exemplified in the third embodiment (see FIG. 14).

[0191] For example, the data acquisition unit 101 provides the wireless channel qualities B11 to B1N and the base station information B21 to B2N to the model generation unit 102 (see FIG. 21). In this case, step S196 is executed by the model generation unit 102.

[0192] In this embodiment, the wireless channel quality B1q and the base station information B2q input to the data acquisition unit 101 are associated with each other. The data acquisition unit 101 may generate learning data D4 and output it to the model generation unit 102. In this case, step S196 is executed by the data acquisition unit 101.

[0193] After step S196 is executed, step S197 is executed. In step S197, the model generation unit 102 performs learning based on the learning data D4, specifically, learning the transmission control information C0. In this learning, the above-mentioned ranking is adopted.

[0194] Through the learning in step S197, the model generation unit 102 generates a trained model M03. After step S197 is executed, step S198 is executed.

[0195] In step S198, the generated trained model M03 is output from the learning device 10D, more specifically, from the model generation unit 102 to the trained model storage unit 13. After step S198 is executed, the model generation process ends.

[0196] As with the learning device 10A described in embodiment 1, the learning device 10D also generates a trained model M03 by, for example, unsupervised learning. As with the learning device 10B described in embodiment 2, the learning device 10D may also generate trained models M1 to MN.

[0197] The trained model storage unit 13 stores the trained model M03. The trained model storage unit 13 may be considered to be included in the communication system 91D, or may be considered to be separate from the communication system 91D.

[0198] For example, the learning device 10D acquires wireless channel qualities B11 to B1N and base station information B21 to B2N for a plurality of positional relationships between the mobile station device 3 and the base station devices 21 to 2N. The learning data D4 acquired in this manner for a plurality of different positional relationships is stored in the model generation unit 102. The trained model M03 may be updated each time the learning data D4 is stored.

[0199] <4-2. Utilization Phase> The utilization phase in the fourth embodiment is also executed in, for example, the communication system 91B, similarly to the utilization phase in the first embodiment.

[0200] For example, in FIG. 6, the trained model M01 is replaced with the trained model M03, and the explanation of the inference in the first embodiment is applicable to the explanation of the inference in the fourth embodiment.

[0201] In steps S42 and S43 included in the flowchart of Figure 7 illustrated in embodiment 1, the learned model M01 is replaced with the learned model M03, and the explanation of the transmission control information Cq in embodiment 1 applies to the explanation of the transmission control information Cq in embodiment 4.

[0202] The operations of the base station devices 21 to 2N in the utilization phase of the fourth embodiment can also be performed according to the flowchart illustrated in FIG.

[0203] In this way, in the fourth embodiment as well, it is possible to determine whether or not to transmit the base station signals T1 to TN to the network 8, and ultimately to the ground equipment 1, in the same manner as in the first embodiment. Such a determination contributes to suppressing the amount of traffic in the communication system 91B that uses the network 8 shared by the ground equipment 1 and the base station devices 21 to 2N.

[0204] Furthermore, determining whether to transmit the base station signals T1 to TN to the ground equipment 1 based on the level of the wireless line qualities B11 to B1N contributes to reducing traffic volume. Inference using the trained model M03 described above contributes to either avoiding transmission of poor-quality base station signals T1 to TN or transmitting good-quality signals, or both.

[0205] In the fourth embodiment, the wireless channel quality B10 is not determined by the base station devices 21 to 2N (see FIG. 2) as in the first embodiment, nor is the evaluation reference information B31 to B3N generated by the ground device 1 (see FIG. 15) as in the third embodiment. However, using the trained model M03 contributes to suppressing the occurrence of an event in which all of the base station signals T1 to TN are not transmitted.

[0206] Alternatively, one or more poor wireless channel qualities B1q may be adopted to generate the trained model M03. For example, in this case, the transmission control information Cq indicates prohibition of transmission of the base station signal Tq from the base station device 2q.

[0207] <4-3. Combination with embodiment 2> Similar to the learning device 10B described in embodiment 2, the learning device 10D may also generate trained models M1 to MN. In this case, the trained model M03 in the above description is replaced with the trained model Mq.

[0208] <5. Embodiment 5> FIG. 23 is a block diagram schematically illustrating a communication system 91E in which the learning phase and the utilization phase are executed according to the fifth embodiment of the present disclosure.

[0209] Similar to the communication system 91B, the communication system 91E includes a ground equipment 1, base station devices 21 to 2N, a mobile station device 3, a network 8, and a line 72. The ground equipment 1 shares the network 8 with all of the base station devices 21 to 2N. The line 72 individually connects the ground equipment 1 and the base station devices 21 to 2N to the network 8. The base station device 2q receives a mobile station signal 3q transmitted from the mobile station device 3.

[0210] Similar to communication system 91A, communication system 91E includes line 71. Line 71 in communication system 91E connects base station devices 21, 22, ..., 2N in this order. However, unlike communication system 91A, communication system 91E does not employ line 71 connecting base station device 21 and ground equipment 1, nor line 71 connecting base station device 2N and ground equipment 1. Line 71 in communication system 91E connects base station devices 21 to 2N in series.

[0211] The communication system 91E includes learning / inference devices 51, 52, ..., 5N. Here, the symbol " / " in the name "learning / inference device" indicates that it performs both the functions before and after it, in this case, both the "learning" and "inference" functions. The block diagram can also be said to be a block diagram illustrating an example of a configuration for performing learning and inference in the learning / inference devices 51 to 5N according to the fifth embodiment of the present disclosure.

[0212] The learning / inference devices 51 to 5N are connected to the base station devices 21 to 2N, respectively. The learning / inference devices 51 to 5N may be included in the base station devices 21 to 2N, respectively.

[0213] 24 is a block diagram illustrating the configuration and input / output of a learning / inference device 5q according to Embodiment 5. The learning / inference device 5q includes an inference device 4q, a learned model storage unit 13q, and a learning device 10Eq.

[0214] <5-1. Learning Phase> Similar to the learning device 10B described in the second embodiment, the learning device 10Eq includes a data acquisition unit 101 and a model generation unit 102 (see FIG. 9). Wireless channel qualities B11 to B1N and base station information B21 to B2N are input to the data acquisition unit 101 of each learning device 10Eq. Learning data D11 to D1N and transmission control information C01 to C0N are used in the model generation unit 102 of each learning device 10Eq.

[0215] Unlike the model generation unit 102 of the learning device 10B, the model generation unit 102 of the learning device 10Eq only needs to generate a trained model Mq. The trained model Mq controls whether or not a base station signal Tq is transmitted from the base station device 2q to the ground device 1. The trained model Mq is generated by learning similar to that in the learning device 10B (see embodiment 2). The trained model Mq is stored in a trained model storage unit 13q.

[0216] <5-1-1. Broadcasting> The wireless channel qualities B11 to B1N and the base station information B21 to B2N are input to the data acquisition unit 101 of each learning device 10Eq, and the learning device 10Eq is individually connected to or included in a base station device 2q corresponding to the learning device 10Eq. All the base station devices 2q broadcast the wireless channel qualities B1q and the base station information B2q to each other.

[0217] in particular: The transmitting / receiving circuit 212 of the base station device 2q transmits the wireless line quality B1q and the base station information B2q to the line 71; The transmission / reception circuit 212 of the base station device 2q (where q≠1 or q≠N) receives wireless channel quality B1m (where the integer m is 1 or more and N or less, and represents all values ​​other than the integer q) and base station information B2m from an adjacent base station device 2s (where the integer s represents both the case where the integer s is 1 less than the integer q and the case where the integer s is 1 greater than the integer q) via a line 71; The transmission / reception circuit 212 of the base station device 21 receives wireless channel qualities B12 to B1N and base station information B22 to B2N from the adjacent base station device 22 via the line 71; The transmission / reception circuit 212 of the base station device 2N receives, via the line 71, wireless channel qualities B11 to B1t and base station information B21 to B2t from the adjacent base station device 2t.

[0218] Such broadcasting via the line 71 contributes to outputting the wireless line qualities B11 to B1N and the base station information B21 to B2N from any of the base station devices 2q to the data acquisition unit 101 of the learning device 10Eq.

[0219] 25 is a flowchart illustrating part of the operation of base station device 2q in embodiment 5, which part contributes to learning in learning device 10Eq. This flowchart is executed, for example, between step S208 and step S209 of the flowchart illustrated in FIG.

[0220] After the wireless channel quality B1q is calculated in step S208, broadcasting is performed in step S221. Specifically, in step S221, the base station device 2q broadcasts the wireless channel quality B1q and the base station information B2q to all the base stations 2m other than the base station device 2q via the line 71 (or further via the base station device 2m) using the transmission / reception circuit 212.

[0221] After step S221 is executed, step S222 is executed, in which it is determined whether all of the wireless channel qualities B1m and all of the base station information B2m have been received by the transmission / reception circuit 212.

[0222] Step S222 is repeatedly executed and a negative determination is made until all of the wireless channel quality B1m and all of the base station information B2m are received. Step S222 can be considered a step of waiting until the base station device 2q receives all of the wireless channel quality B1m and all of the base station information B2m.

[0223] When all of the wireless channel qualities B1m and all of the base station information B2m have been received, a positive determination is made in step S222. After step S208 is executed, the base station device 2q is able to transmit its own wireless channel qualities B1q and base station information B2q. Therefore, at the time when a positive determination is made in step S222, the base station device 2q is in a state where it can transmit wireless channel qualities B11 to B1N and base station information B21 to B2N.

[0224] If the determination result of step S222 is affirmative, step S223 is executed. In step S223, the wireless channel qualities B11 to B1N and the base station information B21 to B2N are output from the base station device 2q to the learning device 10Eq by the transmission / reception circuit 212. After step S223 is executed, step S209 is executed.

[0225] <5-1-2. Generating a trained model> 26 is a flowchart illustrating a model generation process according to the fifth embodiment. A trained model Mq is generated by the model generation process. The model generation process includes steps S105, S106, S197, and S198.

[0226] Steps S105 and S106 in the model generation process are the same as steps S105 and S106 illustrated in the flowchart of Fig. 10 in embodiment 2. In this embodiment, a description of steps S105 and S106 will be omitted.

[0227] By executing step S223, the data acquisition unit 101 of the learning device 10Eq obtains the wireless channel qualities B11 to B1N and the base station information B21 to B2N. These are provided to the model generation unit 102, and step S105 in the model generation process is executed.

[0228] In the model generation process, step S106 is executed, followed by step S197. Step S197 can be obtained by replacing "trained models M1 to MN" with "trained model Mq" in step S107 illustrated in the flowchart of FIG. 10 in the second embodiment.

[0229] As will be described later in the fifth embodiment, in the utilization phase, the transmission control information Cq is generated by an inference device 4q provided separately for the base station device 2q. Therefore, the model generation unit 102 of the learning device 10Eq does not need to generate trained models other than the trained model Mq.

[0230] After step S197 is executed, step S198 is executed. Step S198 can be obtained by replacing "trained models M1 to MN" with "trained model Mq" and "trained model storage unit 13" with "trained model storage unit 13q" in step S108 illustrated in the flowchart of Fig. 10 in the second embodiment.

[0231] <5-2. Utilization Phase> <5-2-1. Reasoning> The inference device 4q in the fifth embodiment, like the inference device 4q in the second embodiment (see FIG. 11), has a data acquisition unit 401 and an inference unit 402. Wireless channel quality B1q and base station information B2q are output from the base station device 2q to the inference device 4q, specifically from the transmission / reception circuit 212 to the data acquisition unit 401. Transmission control information Cq is output from the inference device 4q to the base station device 2q, specifically from the inference unit 402 to the transmission control unit 211.

[0232] The operation of the inference device 4q in the fifth embodiment is similar to that in the second embodiment, and is explained by the flowchart shown in FIG.

[0233] <5-2-2. Transmission to Ground Device 1> The operations of the base station devices 21 to 2N in the utilization phase of the fifth embodiment can also be executed according to the flowchart illustrated in Fig. 8. For example, by inserting steps S221, S222, and S223 illustrated in Fig. 25 between steps S208 and S209 of the flowchart illustrated in Fig. 8, the operation of the base station device 2q when determining, by machine learning, whether or not to transmit the base station signal Tq to the ground device 1 can be explained.

[0234] As in the first embodiment, the fifth embodiment also contributes to suppressing the amount of traffic in the communication system 91B using the line network 8 shared by the ground equipment 1 and the base station equipment 21 to 2N.

[0235] <6. Hardware configuration example> 27 is a block diagram illustrating a configuration of hardware 60 applicable to the present disclosure. The hardware 60 includes a transmitter / receiver (in the figure, the symbol "I / O" indicating input / output) 66, a processor (CPU: Central Processing Unit) 67, a memory (ROM: Read Only Memory) 68, and a memory (RAM: Random Access Memory) 69.

[0236] <6-1. Application to learning devices> The hardware 60 can be applied to, for example, either or both of the data acquisition unit 101 and the model generation unit 102 in each of the learning devices 10A to 10D and 10E1 to 10EN.

[0237] <6-1-1. Application to the data acquisition unit 101> For example, when the hardware 60 is applied to the data acquisition unit 101, the processor 67 executes a predetermined program stored in the memory 68, for example, using the memory 69 as a working memory, and in the learning device 10A, the transmitting / receiving device 66 receives the wireless line quality B10 and the base station information B20 from the ground device 1 and transmits them to the model generation unit 102 (see Figure 3).

[0238] Similarly, in the learning devices 10B and 10D, the transmitting / receiving device 66 receives the wireless channel qualities B11 to B1N and the base station information B21 to B2N from the ground device 1 and transmits them to the model generating unit 102 (see FIGS. 9 and 21).

[0239] Similarly, in the learning device 10C, the transceiver 66 receives wireless line qualities B11 to B1N, base station information B21 to B2N, and evaluation criterion information B31 to B3N from the ground equipment 1, the processor 67 generates a corresponding data group D2 and an evaluation criterion information group B3, and the transceiver 66 transmits the corresponding data group D2 and the evaluation criterion information group B3 to the model generation unit 102 (see Figure 14).

[0240] Similarly, in learning device 10Eq, transmitting / receiving device 66 receives wireless channel qualities B11 to B1N and base station information B21 to B2N from base station device 2q and transmits them to model generating unit 102 (see FIG. 24).

[0241] <6-1-2. Application to the model generation unit 102> For example, when the hardware 60 is applied to the model generation unit 102, the processor 67 executes a predetermined program stored in the memory 68, for example, using the memory 69 as a working memory, so that in the learning device 10A, the transmission / reception device 66 receives the wireless line quality B10 and the base station information B20 from the data acquisition unit 101, the processor 67 generates the learning data D1, learns the transmission control information C0, and generates the learned model M01, and the transmission / reception device 66 outputs the learned model M01 to the learned model memory unit 13 (see Figure 4).

[0242] Similarly, in the learning device 10B, the transmission / reception device 66 receives wireless line qualities B11 to B1N and base station information B21 to B2N from the data acquisition unit 101, the processor 67 generates learning data D1, learns the transmission control information C0, and generates learned models M1 to MN, and the transmission / reception device 66 outputs the learned models M1 to MN to the learned model memory unit 13 (see Figure 10).

[0243] Similarly, in the learning device 10C, the transmission / reception device 66 receives the corresponding data group D2 and the evaluation criterion information group B3 from the data acquisition unit 101, and the processor 67 uses the learning data D3 that compiles the corresponding data group D2 and the evaluation criterion information group B3 to learn the transmission control information C0 and generate the learned model M02, and the transmission / reception device 66 outputs the learned model M02 to the learned model memory unit 13 (see Figure 16).

[0244] Similarly, in the learning device 10D, the transceiver 66 receives wireless line qualities B11 to B1N and base station information B21 to B2N from the data acquisition unit 101, the processor 67 generates learning data D4, learns the transmission control information C0, and generates a learned model M03, and the transceiver 66 outputs the learned model M03 to the learned model memory unit 13 (see Figure 22).

[0245] Similarly, in the learning device 10Eq, the transceiver 66 receives wireless line qualities B11 to B1N and base station information B21 to B2N from the data acquisition unit 101, the processor 67 generates learning data D11 to D1N, learns the transmission control information C01 to C0N, and generates a learned model Mq, and the transceiver 66 outputs the learned model Mq to the learned model memory unit 13q (see Figure 26).

[0246] For example, in the learning devices 10A, 10B, 10D, and 10Eq, unsupervised learning is realized by the operation of the processor 67 using the memory 69 as a working memory.

[0247] For example, in learning device 10C, configuring neural network 93 (see FIG. 17), including adjusting weights v11 to vN1, v21 to vN2, w11 to v1N, and w21 to w2N, is realized by the operation of processor 67 using memory 69 as a working memory, for example.

[0248] <6-2. Application to inference devices> The hardware 60 may be applied to either or both of the data acquisition unit 401 and the inference unit 402 in the inference device 4q, for example.

[0249] <6-2-1. Application to the data acquisition unit 401> For example, when the hardware 60 is applied to the data acquisition unit 401, the processor 67 executes a predetermined program stored in the memory 68, for example, using the memory 69 as a working memory, to receive the wireless channel quality B1q and the base station information B2q from the base station device 2q and output the wireless channel quality B1q and the base station information B2q to the inference unit 402 (see FIGS. 6 and 11). The data acquisition unit 401 in the third embodiment may output the associated data D2q to the inference unit 402 (see FIG. 18).

[0250] <6-2-2. Application to the inference unit 402> For example, when the hardware 60 is applied to the inference unit 402, the processor 67 executes a predetermined program stored in the memory 68, for example, using the memory 69 as a working memory, so that in embodiment 1 the transceiver device 66 receives the wireless line quality B1q, the base station information B2q, and the trained model M01 (see steps S41 and S42 in Figure 7), and the processor 67 inputs the wireless line quality B1q and the base station information B2q into the trained model M01 to generate transmission control information Cq (see step S43 in Figure 7).

[0251] In embodiments 2 and 5, the transceiver 66 receives the wireless line quality B1q, the base station information B2q, and the trained model Mq (see FIG. 12, steps S41 and S45), and the processor 67 inputs the wireless line quality B1q and the base station information B2q into the trained model Mq to generate the transmission control information Cq (see FIG. 12, step S49).

[0252] In embodiment 3, the transmission / reception device 66 receives the correspondence data D2q and the trained model M02 (see Figure 19, steps S46 and S47), and the processor 67 inputs the correspondence data D2q into the trained model M02 to generate transmission control information Cq (see Figure 19, step S48).

[0253] In embodiment 4, the transmission / reception device 66 receives the wireless line quality B1q, the base station information B2q, and the trained model M03 (see Figure 7, step S41, reinterpretation of step S42), and the processor 67 inputs the wireless line quality B1q and the base station information B2q into the trained model M03 to generate the transmission control information Cq (Figure 7, reinterpretation of step S43).

[0254] In any of the first to fifth embodiments, the transmitting / receiving device 66 outputs the transmission control information Cq to the base station device 2q (see step S44).

[0255] <6-3. Application to base station equipment> The hardware 60 may be applied to, for example, a base station device 2q. For example, the transceiver circuit 212 may be realized by a transceiver device 66, the memory 213 may be realized by memories 68 and 69, and the control unit 210 and the transmission control unit 211 may be realized by a processor 67.

[0256] Either or both of the control unit 210 and the transmission control unit 211 may be realized by hardware 60.

[0257] For example, when hardware 60 is adopted as base station device 2q in the first embodiment: The transceiver 66 functioning as the transceiver circuit 212 acquires the wireless channel quality B10 and the base station information B20 from the downstream base station device 2d (where q≠N) (see step S201), and receives the mobile station signal 3q (see step S202); The processor 67 functioning as the control unit 210 calculates the wireless line quality B1q and compares it with the calculated wireless line quality B10 to maintain or update the wireless line quality B10 and the base station information B20 (see steps S204 and S205); The transceiver 66 transmits wireless line quality B10 and base station information B20 to the upstream base station device 2u (or to the ground device 1 when q=1) (see step S206), and transmits wireless line quality B1q and base station information B2q to the inference device 4q (see step S209).

[0258] For example, when hardware 60 is adopted as the base station device 2q in the second to fourth embodiments: The transceiver 66 functioning as the transceiver circuit 212 receives the mobile station signal 3q (see step S207); The processor 67 functioning as the control unit 210 calculates the wireless channel quality B1q (see step S208); The transmitting / receiving device 66 transmits the wireless line quality B1q and the base station information B2q to the ground device 1 and the inference device 4q (see steps S105 and S209).

[0259] For example, when hardware 60 is adopted as the base station device 2q in the fifth embodiment: The transceiver 66 functioning as the transceiver circuit 212 receives the mobile station signal 3q (see step S207); The processor 67 functioning as the control unit 210 calculates the wireless channel quality B1q (see step S208); The transmitting / receiving device 66 broadcasts the wireless line quality B1q and the base station information B2q (step S221), receives the wireless line quality B1m and the base station information B2m (see step S222), and transmits the wireless line qualities B11 to B1N and the base station information B21 to B2N to the learning device 10Eq (see step S223).

[0260] When the hardware 60 is employed as the base station device 2q in any of the first to fifth embodiments: The processor 67 functioning as the transmission control unit 211 determines whether the input transmission control information Cq permits the transmission of the base station signal Tq (see steps S210 and S211); When the transmission of the base station signal Tq is permitted, the transmitting / receiving device 66 transmits the base station signal Tq to the network 8 (see step S212).

[0261] <6-4. Application to ground equipment> The hardware 60 may be applied, for example, in the ground equipment 1. For example, a transceiver 66 receives base station signals T1 to TN.

[0262] For example, when the hardware 60 is applied to the ground device 1 in embodiment 1, the transceiver device 66 receives the wireless line quality B10 and the base station information B20 from the base station device 21 and transmits the wireless line quality B10 and the base station information B20 to the learning device 10A.

[0263] For example, when the hardware 60 is applied to the ground device 1 in embodiment 2, the hardware 60 receives wireless channel qualities B11 to B1N and base station information B21 to B2N from the base station device 21, and transmits the wireless channel qualities B11 to B1N and base station information B21 to B2N to the learning device 10B.

[0264] For example, when the hardware 60 is applied to the ground device 1 in embodiment 3, the hardware 60 receives the wireless line quality B1q and the base station information B2q from the base station device 2q (see step S11), and transmits the wireless line qualities B11 to B1N, the base station information B21 to B2N, and the evaluation criterion information B31 to B3N to the learning device 10C (see step S14).

[0265] For example, when the hardware 60 is applied to the ground device 1 in the fourth embodiment, it receives wireless channel quality B1q and base station information B2q from the base station device 2q, and transmits wireless channel qualities B11 to B1N and base station information B21 to B2N to the learning device 10D.

[0266] When the hardware 60 is applied to the ground device 1 in embodiment 3, the processor 67 executes a predetermined program stored in the memory 68, for example, using the memory 69 as a working memory, thereby generating evaluation criteria information B31 to B3N (see steps S12 and S13).

[0267] Each functional module of the hardware 60 may be realized by the processor 67 executing software processing in accordance with a pre-set program as described above, or at least some of the functional modules may be configured to execute predetermined numerical and logical operation processing using hardware such as electronic circuits having each function.

[0268] The various functions may be realized by a single piece of hardware 60, or may be realized by the cooperative operation of multiple control devices.

[0269] <7. Transformation> <7-1. Modifications of base station information B2q> <7-1-1. Location Information of Mobile Station Device 3> The base station information B2q may include location information indicating the location of the mobile station device 3. The location information is included in, for example, the mobile station signal 3q. By learning the location information of the mobile station device 3 in the learning phase, the transmission control information Cq can be more appropriately inferred.

[0270] The location information may be handled and learned independently of the base station information B2q. In this case, the learning data D1q in the second embodiment is data in which the wireless channel quality B1q, the base station information B2q, and the location information of the mobile station device 3 obtained from the mobile station signal 3q are associated with each other. The same applies to the association data D2q in the third embodiment.

[0271] In determining the wireless channel quality B1j in the third embodiment, the location information of the mobile station device 3 may be taken into consideration. For example, if the wireless channel qualities B11 to B1N are the same, the wireless channel quality B1j is determined to be the one corresponding to the mobile station signal 3q indicating that the location of the mobile station device 3 is closest to any of the base station devices 21 to 2N.

[0272] <7-1-2. Storage of base station information B2q> Since an inference device 4q is provided for each base station device 2q, base station information B2q from the base station device 2q may be stored in the inference device 4q. In this case, transmission and reception of the base station information B2q between the base station device 2q and the inference device 4q is not necessarily required.

[0273] The memory 403 for storing the base station information B2q in the inference device 4q is shown by a dashed line to indicate that it is not necessarily required in the present disclosure (FIGS. 6, 11, 18, and 24).

[0274] <7-2. Variations on the learning device> <7-2-1. Memory of trained models> The present disclosure does not assume that the learning phase is performed. For example, the trained model used in the operation of the inference device 4q in the utilization phase may be obtained from a source other than the base station device 2q, such as a server. Even in this case, transmission control information Cq is obtained as the inference result.

[0275] <7-2-2. Connecting the learning device> The learning devices 10A, 10B, 10C, and 10D may be included in the ground device 1. The connection between the learning devices 10A, 10B, 10C, and 10D and the ground device 1 may be realized via a line network. The learning devices 10A, 10B, 10C, and 10D may exist on a cloud server, and the connection may be realized by the ground device 1 communicating with the cloud server.

[0276] In the utilization phase, the learning devices 10A, 10B, 10C, and 10D are not necessarily required, and they do not need to be connected to either or both of the ground equipment 1 and the trained model storage units 13 and 13q. While inference is being performed in the utilization phase, the learning devices 10A, 10B, 10C, and 10D may be connected to either or both of the ground equipment 1 and the trained model storage units 13 and 13q.

[0277] <7-2-3. Variations in the learning phase> As the unsupervised learning, a known method other than non-hierarchical clustering, such as hierarchical clustering exemplified by the shortest distance method, may be adopted.

[0278] <7-3. Variations on the inference device> The inference devices 41 to 4N may be included in the base station devices 21 to 2N, respectively. The connection between the inference device 4q and the base station device 2q may be realized via a line network. The inference device 4q may exist on a cloud server, and the ground device 1 may communicate with the cloud server, thereby realizing the connection.

[0279] 8. General Description of the Disclosure The base station information B2q identifies each base station device 2q that receives a wireless signal 3q from the mobile station device 3. The wireless channel quality B1q represents the quality of the wireless signal 3q received by the base station device 2q.

[0280] <8-1. General explanation of the learning device> Each of the learning devices 10A, 10B, 10C, 10D, and 10Eq includes a data acquisition unit 101 and a model generation unit .

[0281] The wireless link quality B10 acquired by the data acquisition unit 101 in the learning device 10A is one of the wireless link qualities B11 to B1N. The base station information B20 acquired by the data acquisition unit 101 in the learning device 10A is one of the base station information B21 to B2N (see the first embodiment).

[0282] The learning devices 10B, 10C, 10D, and 10Eq acquire wireless channel qualities B11 to B1N and base station information B21 to B2N (see the second to fifth embodiments).

[0283] From the above, it can be said that any data acquisition unit 101 acquires one or more pieces of base station information B21 to B2N and one or more pieces of wireless channel quality B11 to B1N.

[0284] A data acquisition unit 101 of the learning device 10A acquires wireless link quality B10 and base station information B20. The data acquisition unit 101 acquires wireless link quality B10, which is the best among wireless link qualities B11 to B1N, and base station information B20 that identifies the base station device that received one of wireless signals 31 to 3N having wireless link quality B10 (see FIGS. 1, 2, and 3).

[0285] The learning data D1q in the learning devices 10B and 10Eq (see FIGS. 9 and 24) and the associated data D2q in the learning device 10C (see FIG. 14) are generated by associating base station information B2q with wireless channel quality B1q for each base station device 2q. The same applies to the learning data D4 in the learning device 10D (see FIG. 21).

[0286] The data acquisition unit 101 of the learning device 10C further acquires evaluation reference information B31-B3N that functions as correct answer information for the transmission control information C01-C0N in learning. The learning data D3 is generated by collectively generating the evaluation reference information B31-B3N in addition to the wireless channel qualities B11-B1N and base station information B21-B2N as the associated data D21-D2N (see FIG. 14).

[0287] The trained model Mq is generated for each base station device 2q (Embodiments 2 and 5: see FIGS. 9 and 24). The trained model Mq is stored in a trained model storage unit 13q for each base station device 2q (Embodiment 5: see FIG. 24).

[0288] Each base station device 2q broadcasts its own wireless channel quality B1q to other base station devices 2m (Fifth Embodiment: see FIGS. 24 and 25). At this time, a learning device 10Eq is provided in association with each base station device 2q.

[0289] <8-2. General explanation of the inference device> The inference device 4q includes a data acquisition unit 401 and an inference unit 402. The data acquisition unit 401 acquires and outputs base station information B21 to B2N and wireless channel qualities B11 to B1N.

[0290] The inference unit 402 in embodiment 1 infers transmission control information Cq for the base station device 2q corresponding to the base station information B2q from the learned model M01 in which the transmission control information C0 has been learned using the learning data D1, and the base station information B2q and wireless line quality B1q output from the data acquisition unit 401.

[0291] The inference unit 402 in embodiments 2 and 5 infers transmission control information Cq for the base station device 2q corresponding to the base station information B2q from the learned model Mq in which the transmission control information C0q is learned using the learning data D1q, and the base station information B2q and wireless line quality B1q output from the data acquisition unit 401.

[0292] The inference unit 402 in embodiments 3 and 4 infers transmission control information Cq for the base station device 2q corresponding to the base station information B2q from the trained models M02 and M03 in which the transmission control information C0 has been trained using the training data D3 and D4, respectively, and the base station information B2q and wireless line quality B1q output from the data acquisition unit 401.

[0293] The learning data D1q, D3, and D4 are generated by associating base station information B2q and wireless channel quality B1q for each base station device 2q (see embodiments 2, 3, 4, and 5). The learning data D3 further includes evaluation criterion information B3q that functions as correct answer information for transmission control information Cq in learning (see embodiment 3).

[0294] <8-3. General explanation of communication systems> The communication systems 91A, 91B, 91C, 91D, and 91E each include a plurality of base station devices 21 to 2N, a ground device 1, and a mobile station device 3.

[0295] Communication systems 91A, 91C, and 91D include learning devices 10A, 10C, and 10D, respectively (see FIGS. 1, 13, and 20). Communication system 91E includes a learning device 10Eq for each base station device 2q (see FIG. 24).

[0296] The communication systems 91C and 91D further include lines 73 that individually connect the base station devices 21 to 2N and the ground equipment 1 (see FIGS. 13 and 20).

[0297] The communication system 91B further includes a network 8 and lines 72 that individually connect the network 8 to the base station devices 21 to 2N and the ground device 1 (see FIG. 5).

[0298] The communication system 91E further includes a line 71 that connects the base station devices 21 to 2N in series, a network 8, and lines 72 that individually connect the network 8 with the base station devices 21 to 2N and the ground equipment 1 (see FIG. 23).

[0299] The communication systems 91B and 91E include an inference device 4q for each base station device 2q (see FIGS. 5 and 23).

[0300] It should be noted that the embodiments can be freely combined, and each embodiment can be modified or omitted as appropriate.

[0301] Various aspects of the present disclosure are summarized below as appendices.

[0302] (Appendix 1) a data acquisition unit that acquires one or more pieces of base station information that identify each of a plurality of base station devices that receive a wireless signal from a mobile station device, and one or more pieces of wireless link quality that represent the quality of the wireless signal received by each of the base station devices; a model generation unit that generates a trained model for inferring transmission control information for controlling whether or not transmission from each of the base station devices to the same ground device is permitted from the base station information and the wireless line quality by learning using learning data that is generated in association with the base station information and the wireless line quality acquired from the data acquisition unit; and A learning device comprising:

[0303] (Appendix 2) The learning device described in Appendix 1, wherein the data acquisition unit acquires the best wireless line quality among the wireless line qualities and the base station information that identifies the base station device that received the wireless signal having the best wireless line quality.

[0304] (Appendix 3) 2. The learning device according to claim 1, wherein the learning data is generated for each base station device in such a way that the base station information and the wireless line quality are associated with each other.

[0305] (Appendix 4) the data acquisition unit further acquires evaluation criterion information that functions as correct answer information for the transmission control information in the learning; The learning device according to claim 3, wherein the learning data is generated together with the evaluation criterion information.

[0306] (Appendix 5) 4. The learning device according to claim 3, wherein the trained model is generated for each of the base station devices.

[0307] (Appendix 6) The learning device according to claim 5, wherein the trained model is stored for each base station device.

[0308] (Appendix 7) each of the base station devices broadcasts its own wireless channel quality to the other base station devices; 7. A learning device according to claim 6, provided in association with each of the base station devices.

[0309] (Appendix 8) a data acquisition unit that acquires and outputs base station information that identifies each of a plurality of base station devices that receive wireless signals from a mobile station device and wireless channel quality that indicates the quality of the wireless signals received by each of the base station devices; an inference unit that infers the transmission control information for the base station device corresponding to the base station information output from the data acquisition unit, from a trained model in which transmission control information is trained using training data and the base station information and the wireless line quality output from the data acquisition unit; Equipped with the transmission control information controls whether or not transmission from the base station device to the same ground device is permitted; An inference device, wherein the learning data includes the base station information and the wireless line quality that correspond to each other.

[0310] (Appendix 9) 9. The inference device according to claim 8, wherein the learning data is generated for each base station device in association with the base station information and the wireless line quality.

[0311] (Appendix 10) An inference device as described in Appendix 9, wherein the learning data further includes evaluation criteria information that functions as correct answer information for the transmission control information in the learning.

[0312] (Appendix 11) 10. The inference device of claim 9, wherein the trained model is generated for each base station device.

[0313] (Appendix 12) An inference device as described in Appendix 11, wherein the trained model is stored for each base station device.

[0314] (Appendix 13) A learning device according to any one of Supplementary Note 1 to Supplementary Note 7; A plurality of the base station devices; The ground device; the mobile station device; A communication system comprising:

[0315] (Appendix 14) A wired line that connects the plurality of base station devices and the ground device in a ring shape 14. The communication system of claim 13, further comprising:

[0316] (Appendix 15) Lines individually connecting the plurality of base station devices and the ground equipment 14. The communication system of claim 13, further comprising:

[0317] (Appendix 16) a line connecting the plurality of base station devices in series; The network and lines individually connecting the line network to the plurality of base station devices and the ground equipment; 14. The communication system of claim 13, further comprising:

[0318] (Appendix 17) An inference device according to any one of Supplementary Note 8 to Supplementary Note 12; A plurality of the base station devices; The ground device; the mobile station device; A communication system comprising:

[0319] (Appendix 18) The network and lines individually connecting the line network to the plurality of base station devices and the ground equipment; 18. The communication system of claim 17, further comprising:

[0320] (Appendix 19) 18. The communication system of claim 17, wherein the inference device is provided for each of the plurality of base station devices. [Explanation of symbols]

[0321] 1 ground equipment, 21 to 2N, 2b, 2d, 2j, 2m, 2n, 2q, 2s, 2t, 2u base station equipment, 3 mobile station equipment, 31 to 3N radio signal (mobile station signal), 41 to 4N, 4q inference device, 8 line network, 10A to 10D, 10Eq learning device, 13, 13q trained model memory unit, 71 to 73 line, 91A to 91E communication system, 101 data acquisition unit, 102 model generation unit, 401 data acquisition unit, 402 inference unit, B10 to B1N, B1d, B1i, B1j, B1m, B1q, B1t radio line quality, B20 to B2N, B2d, B2m, B2q, B2t base station information, B31 to B3N, B3i, B3j, B3q Evaluation criteria information, C0, C01 to C0N, C0q, C1 to CN, Cq transmission control information, D1, D11 to D1N, D1q, D3, D4 learning data, D2 data group with correspondence, D21 to D2N, D2q data with correspondence, M01, M02, M03, M1 to MN, Mq trained model, T1 to TN, Tq base station signal.

Claims

1. a data acquisition unit that acquires one or more pieces of base station information that identify each of a plurality of base station devices that receive a wireless signal from a mobile station device, and one or more pieces of wireless link quality that represent the quality of the wireless signal received by each of the base station devices; a model generation unit that generates a trained model for inferring transmission control information for controlling whether or not transmission from each of the base station devices to the same ground device is permitted from the base station information and the wireless line quality by learning using learning data that is generated in association with the base station information and the wireless line quality acquired from the data acquisition unit; and A learning device comprising:

2. The learning device according to claim 1, wherein the data acquisition unit acquires the best wireless line quality among the wireless line qualities and the base station information that identifies the base station device that received the wireless signal having the best wireless line quality.

3. The learning device according to claim 1 , wherein the learning data is generated for each of the base station devices in such a way that the base station information and the wireless channel quality are associated with each other.

4. the data acquisition unit further acquires evaluation criterion information that functions as correct answer information for the transmission control information in the learning; The learning device according to claim 3 , wherein the learning data is generated together with the evaluation criterion information.

5. The learning device according to claim 3 , wherein the trained model is generated for each of the base station devices.

6. The learning device according to claim 5 , wherein the trained model is stored for each of the base station devices.

7. each of the base station devices broadcasts its own wireless channel quality to the other base station devices; The learning device according to claim 6 , provided in association with each of the base station devices.

8. a data acquisition unit that acquires and outputs base station information that identifies each of a plurality of base station devices that receive wireless signals from a mobile station device and wireless channel quality that indicates the quality of the wireless signals received by each of the base station devices; an inference unit that infers the transmission control information for the base station device corresponding to the base station information output from the data acquisition unit, from a trained model in which transmission control information is trained using training data and the base station information and the wireless line quality output from the data acquisition unit; Equipped with the transmission control information controls whether or not transmission from the base station device to the same ground device is permitted; An inference device, wherein the learning data includes the base station information and the wireless line quality that correspond to each other.

9. The inference device according to claim 8 , wherein the learning data is generated for each of the base station devices in such a way that the base station information and the wireless line quality are associated with each other.

10. The inference device according to claim 9 , wherein the learning data further includes evaluation criterion information that functions as correct answer information for the transmission control information in the learning.

11. The inference device according to claim 9 , wherein the trained model is generated for each of the base station devices.

12. The inference device according to claim 11 , wherein the trained model is stored for each of the base station devices.

13. A learning device according to any one of claims 1 to 7; A plurality of the base station devices; The ground device; the mobile station device; A communication system comprising:

14. A wired line that connects the plurality of base station devices and the ground device in a ring shape The communication system of claim 13 further comprising:

15. Lines individually connecting the plurality of base station devices and the ground equipment The communication system of claim 13 further comprising:

16. a line connecting the plurality of base station devices in series; The network and lines individually connecting the line network to the plurality of base station devices and the ground equipment; The communication system of claim 13 further comprising:

17. An inference device according to any one of claims 8 to 12; A plurality of the base station devices; The ground device; the mobile station device; A communication system comprising:

18. The network and lines individually connecting the line network to the plurality of base station devices and the ground equipment; The communication system of claim 17 further comprising:

19. The communication system according to claim 17 , wherein the inference device is provided for each of the plurality of base station devices.

Images