Learning device

Abstract

This learning device comprises: a spatial information preprocessing unit that acquires spatial information between a first communication device and a second communication device that perform wireless communication, and performs preprocessing on the spatial information; a data generation unit that acquires wireless communication quality information, which is information relating to the quality of the wireless communication, and generates training data in which the preprocessed spatial information and the wireless communication quality information are associated with each other; and a learning model generation unit that performs machine learning in advance using the training data, and generates a learning model for predicting the quality of the wireless communication between the first communication device and the second communication device.

Description

Learning device

[0001] The present invention relates to a learning device.

[0002] In wireless communication between a base station device and a terminal device, for example, when the position and orientation of the terminal device change, the wireless communication quality such as communication capacity and throughput may change. Such a change in wireless communication quality may be a factor that causes a situation where the level of wireless communication quality required in the service and system used by the terminal device is not satisfied.

[0003] In particular, in IEEE 802.11ad, which is a wireless LAN (Local Area Network) standard, and the cellular-based fifth-generation mobile communication system (5G), radio waves in a high-frequency band (millimeter-wave band) with strong directivity are used, so radio wave attenuation caused by shielding between the base station device and the terminal device becomes a major problem. In addition to this, it is also known that Doppler shift caused by the movement of the terminal device and the movement of reflectors that exist between the base station device and the terminal device and reflect radio waves also affects the wireless communication quality.

[0004] If it is possible to predict the future wireless communication quality for such a communication environment where the wireless communication quality changes, it becomes possible to take some countermeasures before the service and system are affected by the deterioration of the wireless communication quality. For example, in a high-capacity wireless communication service such as XR (registered trademark), which is a long-distance technology for wireless LAN, it becomes possible to take countermeasures such as performing transmission bit rate control and buffer control in advance according to the prediction result of the wireless communication quality. In this way, more stable service provision is realized.

[0005] Technologies for predicting future wireless communication quality have been proposed for some time. For example, Non-Patent Document 1 describes a technology for predicting future received signal strength (RSSI) using images obtained from a depth camera in a situation where the positions of the terminal device and the access point are fixed, and two people walk in a way that obstructs the line of sight between them. Also, for example, Non-Patent Document 2 describes a technology for predicting future throughput using the position of the terminal device and past throughput values ​​in a situation where a person carrying the terminal device is walking.

[0006] Arvind Narayanan, et al., "Lumos5G: Mapping and Predicting Commercial mmWave 5G Throughput," Proceedings of the ACM on Internet Measurement Conference, pp. 176-193, October 2020.

[0007] Non-patent documents 1 and 2 demonstrate that wireless communication quality can be predicted with high accuracy by preparing a vast amount of training data in advance and performing machine learning. However, while the deep learning techniques described in non-patent documents 1 and 2 can predict with high accuracy, they predict wireless communication quality for a predetermined, specific communication environment. Therefore, with these conventional techniques, if the communication environment changes, for example, due to a change in the movement path of a terminal device or a change in the location where the terminal device is placed, it becomes necessary to prepare a vast amount of new training data and redo the machine learning. Furthermore, it is often difficult in practice to prepare a vast amount of training data for various potentially changing communication environments and to construct a separate predictive model for each.

[0008] Thus, conventionally, there has been a challenge in accurately predicting future wireless communication quality using only a small amount of training data, or without performing machine learning again at all, when the communication environment changes over time or when machine learning is performed in a different environment, such as a location different from where it was performed before.

[0009] In view of the above circumstances, the present invention aims to provide a technology that can accurately predict wireless communication quality by performing machine learning again using only a small amount of training data, or without performing machine learning again at all, even in situations where the communication environment changes or in an environment different from the environment in which machine learning was performed.

[0010] One aspect of the present invention is a learning device comprising: a spatial information preprocessing unit that acquires spatial information between a first communication device and a second communication device that perform wireless communication and performs preprocessing on the spatial information; a data generation unit that acquires wireless communication quality information, which is information relating to the quality of the wireless communication, and generates training data to which the preprocessed spatial information and the wireless communication quality information are associated; and a learning model generation unit that performs machine learning in advance using the training data and generates a learning model to predict the quality of the wireless communication between the first communication device and the second communication device.

[0011] This invention makes it possible to predict wireless communication quality with high accuracy, even in situations where the communication environment changes or in an environment different from the one in which machine learning was performed, by performing machine learning again using only a small amount of training data, or even without performing machine learning again at all.

[0012] This is an overall configuration diagram of the wireless communication system 1 in one embodiment of the present invention. This is a block diagram showing the functional configuration of the learning device 10 in one embodiment of the present invention. This is a block diagram showing the functional configuration of the estimation device 20 in one embodiment of the present invention. This is a flowchart showing the operation of the learning device 10 when generating a learning model in one embodiment of the present invention. This is a flowchart showing the operation of the learning device 10 when updating the learning model in one embodiment of the present invention. This is a flowchart showing the operation of the estimation device 20 in one embodiment of the present invention. This is a diagram illustrating an example of how the communication control system 4 predicts wireless communication quality in one embodiment of the present invention. This is a diagram illustrating the spatial information of the terminal device 3 before preprocessing is performed. This is a diagram illustrating the spatial information of the terminal device 3 after preprocessing is performed. This is a diagram illustrating an example of the implementation effect of the communication control system 4 predicting wireless communication quality in one embodiment of the present invention. This is a diagram illustrating an example of how the communication control system 4 predicts wireless communication quality in one embodiment of the present invention. This is a diagram illustrating another example of how the communication control system 4 predicts wireless communication quality in one embodiment of the present invention. This is a diagram illustrating another example of how the communication control system 4 predicts wireless communication quality in one embodiment of the present invention.

[0013] The following describes a wireless communication system 1 in one embodiment of the present invention with reference to the drawings.

[0014] [Configuration of the Wireless Communication System] Figure 1 is an overall configuration diagram of a wireless communication system 1 in one embodiment of the present invention. As shown in Figure 1, the wireless communication system 1 is composed of a base station device 2, a terminal device 3, and a communication control system 4. For the sake of clarity, only one terminal device 3 is shown in Figure 1, but in reality, it is assumed that multiple terminal devices 3 are connected to one base station device 2. It is also possible to have a configuration in which there are multiple base station devices 2. In this case, it is also possible to configure the system so that data generated by one base station device 2 (for example, a learning model obtained by machine learning, etc.) can be used by multiple base station devices 2.

[0015] The wireless communication system 1 is a communication system that realizes wireless communication between a base station device 2 and a terminal device 3. In this embodiment, the communication environment between the base station device 2 and the terminal device 3 changes over time. For example, the communication environment changes due to Doppler shift caused by the movement of the terminal device 3 or the movement of a reflective object (not shown) that reflects radio waves and exists between the base station device 2 and the terminal device 3. Alternatively, in this embodiment, the communication environment between the base station device 2 and the terminal device 3 may differ from the communication environment at the location where machine learning, described later, is performed. As the communication environment changes, the quality of wireless communication between the base station device 2 and the terminal device 3 also changes.

[0016] The communication control system 4 predicts future wireless communication quality based on changes in the communication environment between the base station device 2 and the terminal device 3. Then, according to the predicted wireless communication quality, the communication control system 4 performs communication control, such as transmission bitrate control and buffer control. In this way, the communication control system 4 can prevent situations from occurring where the wireless communication quality falls below the level required for the services and systems used by the terminal device 3.

[0017] Base station device 2 is, for example, a communication device such as a wireless LAN access point. Terminal device 3 is, for example, a mobile terminal such as a smartphone. However, base station device 2 and terminal device 3 are not limited to these devices, and other devices may be used as communication devices that communicate wirelessly with each other in situations where the communication environment with opposing communication devices may change.

[0018] As shown in Figure 1, the communication control system 4 is comprised of a learning device 10, an estimation device 20, a communication control device 30, and a measurement device 40.

[0019] The learning device 10 is, for example, an information processing device such as a general-purpose computer. The learning device 10 acquires information regarding the communication environment between the base station device 2 and the terminal device 3, and information regarding the wireless communication quality between the base station device 2 and the terminal device 3. The learning device 10 generates training data using the acquired data and performs machine learning in advance. As a result, the learning device 10 generates a learning model that takes information regarding the communication environment between the base station device 2 and the terminal device 3 as input and outputs information regarding future wireless communication quality.

[0020] Furthermore, the learning device 10 determines the need to update the learning model (i.e., perform further machine learning) based on the difference between the predicted result (predicted value) of wireless communication quality output from the generated learning model and the measured value. If the learning device 10 determines that an update to the learning model is necessary, it generates additional training data and performs further machine learning.

[0021] The estimation device 20 is, for example, an information processing device such as a general-purpose computer. The estimation device 20 acquires information about the communication environment between the base station device 2 and the terminal device 3. Using the acquired information about the communication environment and a learning model pre-generated by the learning device 10, the estimation device 20 predicts the future wireless communication quality between the base station device 2 and the terminal device 3.

[0022] The communication control device 30 acquires information indicating the prediction result of the wireless communication quality predicted by the estimation device 20. Based on the acquired information, the communication control device 30 performs communication control, such as transmission bitrate control and buffer control, according to the prediction result of the wireless communication quality.

[0023] Furthermore, any two or all of the learning device 10, estimation device 20, and communication control device 30 may be an integrated device. Also, the learning device 10 and estimation device 20 may be devices that share the same functional parts. In addition, at least one of the learning device 10, estimation device 20, and communication control device 30 may be a device located on the cloud.

[0024] The measuring device 40 is, for example, a camera and a LiDAR (Light Detection and Ranging) device. The measuring device 40 is a device for externally measuring the communication environment between the base station device 2 and the terminal device 3. The measuring device 40 measures, for example, the position, speed and direction of movement of the terminal device 3, and the position, speed and direction of movement of reflective objects (not shown) that exist between the base station device 2 and the terminal device 3 and can reflect radio waves. The measuring device 40 outputs spatial information showing these measurement results to the learning device 10 and the estimation device 20.

[0025] The functional configurations of the learning device 10 and the estimation device 20 will be described in more detail below.

[0026] [Functional Configuration of the Learning Device] Figure 2 is a block diagram showing the functional configuration of the learning device 10 in one embodiment of the present invention. As shown in Figure 2, the learning device 10 is composed of a spatial information acquisition unit 11, a spatial information preprocessing unit 12, a quality information acquisition unit 13, a training data generation unit 14, a learning model generation unit 15, a learning model storage unit 16, an environmental change determination unit 17, and a learning model update unit 18.

[0027] The spatial information acquisition unit 11 acquires spatial information of the base station device 2 and the terminal device 3, which are the targets for prediction of wireless communication quality, and spatial information of reflective objects (not shown) that exist between the base station device 2 and the terminal device 3, which are the targets for prediction of wireless communication quality. As mentioned above, the spatial information includes information indicating the position, speed of movement, and direction of movement of the base station device 2, the terminal device 3, and the reflective objects (not shown). Furthermore, as mentioned above, reflective objects are objects that exist between the base station device 2 and the terminal device 3 and can reflect radio waves.

[0028] The spatial information acquisition unit 11 acquires spatial information indicating the position measured by a receiver of a positioning system such as a GPS (Global Positioning System) mounted on the terminal device 3 from the terminal device 3. The spatial information acquisition unit 11 also acquires spatial information indicating the speed and direction of movement measured by various sensors such as an acceleration sensor mounted on the terminal device 3 from the terminal device 3. The spatial information acquisition unit 11 outputs the acquired spatial information to the spatial information preprocessing unit 12.

[0029] Furthermore, the spatial information acquisition unit 11 acquires spatial information from the measuring device 40 that indicates the position, speed, and direction of movement of the terminal device 3 and reflective objects (not shown), which are identified based on images captured by a camera mounted on the measuring device 40 and point cloud data measured by LiDAR. The spatial information acquisition unit 11 outputs the acquired spatial information to the spatial information preprocessing unit 12.

[0030] The spatial information preprocessing unit 12 acquires spatial information output from the spatial information acquisition unit 11. The spatial information preprocessing unit 12 performs preprocessing on the acquired spatial information. Preprocessing here refers to the process of calculating the distance and angle from the base station device 2 to the terminal device 3, and the distance and angle from the base station device 2 to the reflector, based on the values ​​of the position, movement speed, and movement direction of the base station device 2, terminal device 3, and reflector (not shown) included in the spatial information. The spatial information preprocessing unit 12 outputs the calculation results from the preprocessing as preprocessed spatial information to the training data generation unit 14.

[0031] The quality information acquisition unit 13 acquires information indicating the wireless communication quality between the base station device 2 and the terminal device 3 from at least one of the base station device 2 and the terminal device 3. The quality information acquisition unit 13 outputs the acquired wireless communication quality information to the training data generation unit 14.

[0032] The wireless communication quality referred to here includes, for example, the received signal strength (RSSI) and reference signal reception quality (RSSQ) at the base station device 2 or terminal device 3, the throughput in wireless communication between the base station device 2 and terminal device 3, the signal-to-noise power ratio, the signal-to-interference noise power ratio, the packet error rate, the number of bits received, the bit error rate, the number of bits received per unit time, the MCS (Modulation and Coding Scheme) index, the number of retransmissions, the packet arrival delay time, the error correction technology settings, and the user's contract information for terminal device 3.

[0033] Furthermore, the wireless communication quality may include values ​​obtained by differentiating any of the above wireless communication quality values ​​listed as examples, and index values ​​calculated from the above wireless communication quality values ​​using a predetermined calculation formula. In addition, the wireless communication quality may include the value of the radio wave frequency used in wireless communication between the base station device 2 and the terminal device 3, the value of the resource bandwidth, and information on various setting items of the wireless communication system 1 that affect these values.

[0034] The training data generation unit 14 acquires the pre-processed spatial information output from the spatial information preprocessing unit 12. The training data generation unit 14 also acquires information indicating wireless communication quality output from the quality information acquisition unit 13. The training data generation unit 14 sequentially generates training data by associating the pre-processed spatial information with input data (example problems) for the learning model and the information indicating wireless communication quality with output data (correct labels) from the learning model.

[0035] Furthermore, the input data (example) to the learning model may include not only the preprocessed spatial information, but also some information indicating other wireless communication quality (i.e., wireless communication quality other than the correct label), excluding the wireless communication quality used as output data.

[0036] The input and output data may also be time-series information. For example, the training data generation unit 14 may use the time-series values ​​of the distance and angle from the base station device 2 to the terminal device 3, the distance and angle from the base station device 2 to a reflector (not shown), the received signal strength (RSSI) at the base station device 2 or terminal device 3, and the MCS index for the past 5 seconds as input data, and generate training data that associates the throughput value between the base station device 2 and the terminal device 3 one second later with the throughput value. In this case, the learning device 10 can generate a learning model that can predict the throughput value between the base station device 2 and the terminal device 3 one second later by inputting the above spatial information and information indicating the wireless communication quality for the past 5 seconds.

[0037] The training data generation unit 14 comprehensively generates training data by associating input data (example questions) and output data (correct answer labels) in various combinations. The training data generation unit 14 outputs the comprehensively generated training data to the learning model generation unit 15.

[0038] The learning model generation unit 15 acquires the training data output from the training data generation unit 14. The learning model generation unit 15 also acquires a learning model that has not yet undergone machine learning and is pre-stored in the learning model storage unit 16. The learning model generation unit 15 performs machine learning using the acquired training data and the learning model, and generates a trained learning model with updated parameters. The learning model generation unit 15 overwrites and updates the learning model pre-stored in the learning model storage unit 16 with the trained learning model.

[0039] The learning model storage unit 16 stores the learning model. The learning model storage unit 16 is configured to include, for example, semiconductor memories such as RAM (Random Access Memory) and EEPROM (Electrically Erasable Programmable Read-Only Memory), flash memories such as SSD (Solid State Drive), magnetic disks such as HDD (Hard Disk Drive), storage media such as optical disks, or any combination of these storage media.

[0040] The environmental change determination unit 17 and the learning model update unit 18 are not functional units used at the time of generating the learning model (i.e., before the operation start of the wireless communication system 1), but are functional units used when additional machine learning is performed as needed after the operation start of the wireless communication system 1 to update the learning model.

[0041] The environmental change determination unit 17 determines whether it is necessary to update the learning model. The environmental change determination unit 17 acquires the training data generated by the training data generation unit 14. Further, the environmental change determination unit 17 acquires the learned learning model generated by the learning model generation unit 15. The environmental change determination unit 17 inputs the input data included in the training data to the acquired learned learning model to obtain a prediction result (predicted value) of the future wireless communication quality.

[0042] Then, the environmental change determination unit 17 compares the obtained predicted value with the output data included in the training data (i.e., the correct label which is the measured value of the wireless communication quality). The environmental change determination unit 17 determines that it is necessary to update the learning model when the difference between the predicted value and the measured value of the wireless communication quality exceeds a predetermined threshold. For example, the environmental change determination unit 17 determines that it is necessary to update the learning model when the difference between the predicted value and the measured value of the throughput is 30 [Mbps] or more, which is a predetermined threshold. When the environmental change determination unit 17 determines that it is necessary to update the learning model, it notifies the learning model update unit 18.

[0043] Note that the environmental change determination unit 17 may determine whether to update the learning model based on other conditions. For example, when the environmental change determination unit 17 detects that the terminal device 3 or a reflector (not shown) is not present at a position assumed based on the spatial information, it may determine that it is necessary to update the learning model. Alternatively, for example, when a predetermined period has elapsed since the generation of the learning model or the previous update, the environmental change determination unit 17 may determine that it is necessary to update the learning model. That is, the environmental change determination unit 17 may determine that it is necessary to update the learning model at a predetermined interval. Alternatively, for example, when the communication control system 4 is to predict the future wireless communication quality in a communication environment (location) different from the previous one, the environmental change determination unit 17 may determine that it is necessary to update the learning model.

[0044] When the learning model update unit 18 receives a notification from the environmental change determination unit 17 indicating that it is necessary to update the learning model, it acquires the training data generated by the training data generation unit 14. Also, when the environmental change determination unit 17 receives a notification from the environmental change determination unit 17 indicating that it is necessary to update the learning model, it acquires the learned learning model generated by the learning model generation unit 15.

[0045] The learning model update unit 18 performs machine learning using the acquired training data and the learned learning model, and further generates a relearned learning model with updated parameters. The learning model generation unit 15 overwrites and updates the learned learning model previously stored in the learning model storage unit 16 with the relearned learning model.

[0046] [Functional Configuration of Estimation Device] FIG. 3 is a block diagram showing the functional configuration of the estimation device 20 in an embodiment of the present invention. As shown in FIG. 3, the estimation device 20 includes a spatial information acquisition unit 21, a spatial information preprocessing unit 22, a learning model acquisition unit 23, a quality estimation unit 24, and an estimation result output unit 25.

[0047] The spatial information acquisition unit 21 acquires spatial information of the base station device 2 and the terminal device 3, which are the targets for wireless communication quality prediction, and spatial information of reflective objects (not shown) that exist between the base station device 2 and the terminal device 3, which are the targets for wireless communication quality prediction. As mentioned above, the spatial information includes information indicating the position, speed of movement, and direction of movement of the base station device 2, the terminal device 3, and the reflective objects (not shown). Furthermore, as mentioned above, reflective objects are objects that exist between the base station device 2 and the terminal device 3 and can reflect radio waves.

[0048] The spatial information acquisition unit 21 acquires spatial information indicating the position measured by a receiver of a positioning system such as GPS mounted on the terminal device 3 from the terminal device 3. The spatial information acquisition unit 21 also acquires spatial information indicating the speed and direction of movement measured by various sensors such as an acceleration sensor mounted on the terminal device 3 from the terminal device 3. The spatial information acquisition unit 21 outputs the acquired spatial information to the spatial information preprocessing unit 22.

[0049] Furthermore, the spatial information acquisition unit 21 acquires spatial information from the measuring device 40 that indicates the position, speed, and direction of movement of the base station device 2, terminal device 3, and reflective object (not shown), which are identified based on images captured by a camera mounted on the measuring device 40 and point cloud data measured by LiDAR. The spatial information acquisition unit 21 outputs the acquired spatial information to the spatial information preprocessing unit 22.

[0050] The spatial information preprocessing unit 22 acquires spatial information output from the spatial information acquisition unit 21. The spatial information preprocessing unit 22 performs preprocessing on the acquired spatial information. As mentioned above, the preprocessing involves calculating the distance and angle from the base station device 2 to the terminal device 3, and the distance and angle from the base station device 2 to the reflector, based on the position, movement speed, and movement direction values ​​of the base station device 2, terminal device 3, and reflector (not shown) included in the spatial information. The spatial information preprocessing unit 22 outputs the calculation results from the preprocessing as preprocessed spatial information to the quality estimation unit 24.

[0051] The learning model acquisition unit 23 acquires the learned model stored in the learning model storage unit 16 of the learning device 10. Note that the learned model here also includes the learned model that has been retrained. The learning model acquisition unit 23 outputs the acquired learned model to the quality estimation unit 24.

[0052] The quality estimation unit 24 acquires the pre-processed spatial information output from the spatial information pre-processing unit 22. The quality estimation unit 24 also acquires the trained model output from the trained model acquisition unit 23. The environmental change determination unit 17 inputs the pre-processed spatial information as input data into the trained model to obtain a prediction result of future wireless communication quality. The quality estimation unit 24 outputs information indicating the obtained prediction result of future wireless communication quality to the estimation result output unit 25.

[0053] The estimation result output unit 25 acquires information indicating the predicted future wireless communication quality output from the quality estimation unit 24. The estimation result output unit 25 outputs the acquired information indicating the predicted future wireless communication quality to the communication control device 30. The communication control device 30 then performs communication control, such as transmission bitrate control and buffer control, according to the predicted wireless communication quality based on the information acquired from the estimation result output unit 25, thereby enabling more accurate prediction of wireless communication quality.

[0054] [Operation of the Learning Device During Learning Model Generation] An example of the operation of the learning device 10 during learning model generation will be described below. Figure 4 is a flowchart showing the operation of the learning device 10 during learning model generation in one embodiment of the present invention. The operation of the learning device 10 shown in the flowchart of Figure 4 is started in advance at an arbitrary timing before the start of operation of the wireless communication system 1.

[0055] First, the spatial information acquisition unit 11 acquires spatial information of the base station device 2 and terminal device 3, which are the targets for prediction of wireless communication quality, and spatial information of reflective objects (not shown) existing between the base station device 2 and terminal device 3, which are the targets for prediction of wireless communication quality, from the base station device 2, terminal device 3, and measuring device 40 (step S001). As described above, the spatial information includes information indicating the position, speed of movement, and direction of movement of the base station device 2, terminal device 3, and reflective objects (not shown). Furthermore, as described above, reflective objects are objects that exist between the base station device 2 and terminal device 3 and can reflect radio waves.

[0056] Next, the spatial information preprocessing unit 12 acquires the spatial information output from the spatial information acquisition unit 11. The spatial information preprocessing unit 12 performs preprocessing on the acquired spatial information (step S002). As described above, preprocessing is the process of calculating the distance and angle from the base station device 2 to the terminal device 3, and the distance and angle from the base station device 2 to the reflector, based on the values ​​of the position, movement speed, and movement direction of the base station device 2, terminal device 3, and reflector (not shown) included in the spatial information. The spatial information preprocessing unit 12 outputs the calculation results from the preprocessing as preprocessed spatial information to the training data generation unit 14.

[0057] Next, the quality information acquisition unit 13 acquires information indicating the wireless communication quality between the base station device 2 and the terminal device 3 from at least one of the base station device 2 and the terminal device 3 (step S003). The quality information acquisition unit 13 outputs the acquired wireless communication quality information to the training data generation unit 14.

[0058] As mentioned above, wireless communication quality includes, for example, the received signal strength and reference signal reception quality at the base station device 2 or terminal device 3, throughput, signal-to-noise power ratio, signal-to-interference noise power ratio, packet error rate, number of bits received, bit error rate, number of bits received per unit time, MCS index, number of retransmissions, packet arrival delay time, error correction technology settings, and user contract information.

[0059] Next, the training data generation unit 14 acquires the pre-processed spatial information output from the spatial information preprocessing unit 12. The training data generation unit 14 also acquires information indicating wireless communication quality output from the quality information acquisition unit 13. The training data generation unit 14 generates training data by associating the pre-processed spatial information with the wireless communication quality information with the output data from the learning model (step S004). The training data generation unit 14 comprehensively and sequentially generates training data by associating the input data with the output data in various combinations. The training data generation unit 14 outputs the comprehensively generated training data to the learning model generation unit 15.

[0060] Next, the learning model generation unit 15 acquires the training data output from the training data generation unit 14. The learning model generation unit 15 also acquires the learning model stored in the learning model storage unit 16 before machine learning is performed. The learning model generation unit 15 performs machine learning using the acquired training data and the learning model, and generates a trained learning model with updated parameters (step S005). The learning model generation unit 15 overwrites and updates the learning model stored in the learning model storage unit 16 with the trained learning model.

[0061] This completes the operation of the learning device 10 during the generation of the learning model, as shown in the flowchart of Figure 4.

[0062] [Operation of the Learning Device During Learning Model Update] An example of the operation of the learning device 10 during learning model update will be described below. Figure 5 is a flowchart showing the operation of the learning device 10 during learning model update in one embodiment of the present invention. The operation of the learning device 10 shown in the flowchart of Figure 5 is started, for example, at predetermined intervals (for example, every hour, every day, etc.) after the start of operation of the wireless communication system 1.

[0063] First, the environmental change determination unit 17 acquires the training data generated by the training data generation unit 14 (step S101). Next, the environmental change determination unit 17 acquires the trained model generated by the trained model generation unit 15 (step S102). Next, the environmental change determination unit 17 inputs the input data contained in the training data into the acquired trained model to obtain a predicted value of the wireless communication quality (step S103).

[0064] Next, the environmental change determination unit 17 compares the obtained predicted value with the output data included in the training data (i.e., the measured value of the wireless communication quality) (step S104). If the difference between the predicted value and the measured value of the wireless communication quality is less than or equal to a predetermined threshold (step S105, NO), the environmental change determination unit 17 determines that updating the learning model is unnecessary (step S106). In this case, the operation of the learning device 10 during learning model update, as shown in the flowchart of Figure 5, is completed.

[0065] On the other hand, if the difference between the predicted value and the measured value of the wireless communication quality exceeds a predetermined threshold (step S105, YES), the environmental change determination unit 17 determines that the learning model needs to be updated (step S107). If the environmental change determination unit 17 determines that the learning model needs to be updated, it notifies the learning model update unit 18.

[0066] Next, when the learning model update unit 18 receives notification from the environment change determination unit 17 that an update to the learning model is necessary, it acquires the training data generated by the training data generation unit 14 (step S108). Next, when the environment change determination unit 17 receives notification from the environment change determination unit 17 that an update to the learning model is necessary, it acquires the trained learning model generated by the learning model generation unit 15 (step S109).

[0067] Next, the learning model update unit 18 performs machine learning using the acquired training data and the previously trained learning model to generate a retrained learning model with further updated parameters. The learning model generation unit 15 overwrites and updates the previously trained learning model stored in the learning model storage unit 16 with the retrained learning model (step S110).

[0068] This completes the operation of the learning device 10 during the update of the learning model, as shown in the flowchart of Figure 5.

[0069] [Operation of the Estimation Device] An example of the operation of the estimation device 20 will be described below. Figure 6 is a flowchart showing the operation of the estimation device 20 in one embodiment of the present invention. The operation of the estimation device 20 shown in the flowchart of Figure 6 is started, for example, at predetermined intervals (for example, every 5 minutes, every 1 hour, etc.) after the start of operation of the wireless communication system 1.

[0070] First, the spatial information acquisition unit 21 acquires spatial information of the base station device 2 and terminal device 3, which are the targets for prediction of wireless communication quality, and spatial information of reflective objects (not shown) existing between the base station device 2 and terminal device 3, which are the targets for prediction of wireless communication quality, from the base station device 2, terminal device 3, and measuring device 40 (step S201). As described above, the spatial information includes information indicating the position, speed of movement, and direction of movement of the base station device 2, terminal device 3, and reflective objects (not shown). Furthermore, as described above, reflective objects are objects that exist between the base station device 2 and terminal device 3 and can reflect radio waves.

[0071] Next, the spatial information preprocessing unit 22 acquires the spatial information output from the spatial information acquisition unit 21. The spatial information preprocessing unit 22 performs preprocessing on the acquired spatial information (step S202). As described above, preprocessing is the process of calculating the distance and angle from the base station device 2 to the terminal device 3, and the distance and angle from the base station device 2 to the reflector, based on the values ​​of the position, movement speed, and movement direction of the base station device 2, terminal device 3, and reflector (not shown) included in the spatial information. The spatial information preprocessing unit 22 outputs the calculation results from the preprocessing as preprocessed spatial information to the quality estimation unit 24.

[0072] Next, the learning model acquisition unit 23 acquires the learned model stored in the learning model storage unit 16 of the learning device 10 (step S203). Note that the learned model referred to here also includes the learned model that has been retrained. The learning model acquisition unit 23 outputs the acquired learned model to the quality estimation unit 24.

[0073] Next, the quality estimation unit 24 acquires the pre-processed spatial information output from the spatial information pre-processing unit 22. The quality estimation unit 24 also acquires the trained model output from the trained model acquisition unit 23. The environmental change determination unit 17 inputs the pre-processed spatial information as input data into the trained model to obtain a prediction result of future wireless communication quality (step S204). The quality estimation unit 24 outputs information indicating the obtained prediction result of future wireless communication quality to the estimation result output unit 25.

[0074] Next, the estimation result output unit 25 acquires information indicating the predicted future wireless communication quality output from the quality estimation unit 24. The estimation result output unit 25 outputs the acquired information indicating the predicted future wireless communication quality to the communication control device 30 (step S205).

[0075] This completes the operation of the estimation device 20 as shown in the flowchart of Figure 6.

[0076] As described above, the communication control system 4 in one embodiment of the present invention predicts future wireless communication quality based on changes in the communication environment between the base station device 2 and the terminal device 3. Then, the communication control system 4 performs communication control, such as transmission bitrate control and buffer control, according to the prediction result of wireless communication quality. As a result, the communication control system 4 can prevent situations from occurring where the wireless communication quality falls below the level required for the services and systems used by the terminal device 3.

[0077] Furthermore, in one embodiment of the present invention, the learning device 10 acquires information regarding the communication environment between the base station device 2 and the terminal device 3, and information regarding the wireless communication quality between the base station device 2 and the terminal device 3. The learning device 10 generates training data using the acquired data and performs machine learning in advance. As a result, the learning device 10 generates a learning model that takes information regarding the communication environment (spatial information, etc.) between the base station device 2 and the terminal device 3 as input and outputs information regarding future wireless communication quality.

[0078] At this time, the learning device 10 performs preprocessing on the acquired spatial information, calculating the distance and angle from the base station device 2 to the terminal device 3, and the distance and angle from the base station device 2 to the reflector, based on the values ​​of the position, movement speed, and movement direction of the base station device 2, terminal device 3, and reflector (not shown) included in the spatial information. Then, the learning device 10 performs machine learning using the preprocessed spatial information.

[0079] By having such a configuration, the communication control system 4 in one embodiment of the present invention can predict wireless communication quality with high accuracy, even in situations where the communication environment changes or in an environment different from the environment in which machine learning was performed, simply by performing machine learning again using only a small amount of training data, or without performing machine learning again at all.

[0080] Furthermore, radio waves, especially those in high-frequency bands such as the millimeter wave band, have strong directivity, and the quality of wireless communication varies greatly depending on the distance and angle from the base station 2 to the terminal device 3. In such cases, it is desirable to generate training data in advance for various combinations of distance and angle from the base station 2 to the terminal device 3 and perform machine learning.

[0081] For example, if the communication environment is indoors, the terminal device 3 can be moved randomly or systematically throughout the room at various locations, at various speeds, and in various directions to reproduce various combinations of distance and angle from the base station device 2 to the terminal device 3, thereby comprehensively collecting spatial information and information indicating wireless communication quality. This makes it possible to predict future wireless communication quality with higher accuracy in various communication environments.

[0082] Generally, position coordinates are used as spatial information for the base station device 2, terminal device 3, and reflective objects (not shown). However, since the setting of the axes and origin of the position coordinates is arbitrary, the axes and origin that serve as the basis for each piece of spatial information will differ from place to place. Therefore, generally, a machine learning model generated using spatial information for a specific location is difficult to use for predictions about other locations.

[0083] If we were to forcibly use general location coordinates, the prediction accuracy would be significantly reduced. Furthermore, if we were to perform machine learning using spatial information for other locations, a huge amount of training data would be required, and even more training data would be needed when performing machine learning again. To address these challenges, the learning device 10 in the embodiment described above includes a configuration that converts spatial information into distance and angle from the base station device 2 to the terminal device 3 as a preprocessing step, so that spatial information can be commonly used in any location.

[0084] Furthermore, even if environmental changes such as changes in equipment layout or new equipment introduction occur at logistics centers, factories, or other sites, the communication control system 4 in the embodiment of the present invention can predict wireless communication quality with high accuracy by performing machine learning again using only a small amount of training data, or without performing machine learning again at all. For example, if the terminal device 3 held by a person is the device to be predicted, since people generally make a variety of movements, such a variety of movements can be predicted with high accuracy by performing machine learning again using only a small amount of training data, or without performing machine learning again at all.

[0085] Furthermore, if a decline in future wireless communication quality is predicted, the communication control system 4 can maintain wireless communication quality by taking preventative measures, for example, by controlling RIS (Reconfigurable Intelligent Surface) or beamforming on the base station equipment 2 side. RIS is a technology that expands the communication area by reflecting radio waves behind obstacles such as buildings using reflectors.

[0086] Furthermore, if a decrease in wireless communication quality is predicted when using a real-time video distribution application (not shown), the communication control system 4 can take preventative measures, such as lowering the bitrate on the video distribution application side (for example, switching the video quality from 4K to 2K), thereby prioritizing real-time performance while maintaining the wireless communication quality of the video distribution application.

[0087] Furthermore, even in a communication environment where multiple autonomous mobile robots (AMRs) move around, the communication control system 4 can control the optimal path of the AMMRs by predicting future temporal fluctuations in the wireless communication quality of terminal devices held by workers and AMMRs. In addition, the communication control system 4 can provide an environment for monitoring and managing workers and AMMRs by notifying users and other systems of the prediction results.

[0088] For example, in situations where large amounts of data, such as videos captured by surveillance cameras or cameras mounted on each robot, are uploaded to the cloud or a server, the communication control system 4 can control the upload to allow data to be done in batches while future wireless communication quality is predicted to be good. Conversely, if future wireless communication quality is predicted to deteriorate, the communication control system 4 can perform proactive control, such as preventing the upload of such large amounts of data. Proactive control refers to taking preventative measures based on current situation analysis. This results in effects such as reducing the number of data retransmissions and shortening communication time, leading to power savings.

[0089] Furthermore, in places where a large number of people come and go, such as event venues and conference rooms, the communication control system 4 can predict future wireless communication quality and control the system to guide people along routes that ensure more stable communication, or to estimate pedestrian flow.

[0090] (Examples) The following describes examples of the present invention. Figure 7 is a diagram illustrating an example of the prediction of wireless communication quality by the communication control system 4 in one embodiment of the present invention.

[0091] This embodiment was carried out in a room. Figure 7 shows the location (room) where the wireless communication was actually performed. In Figure 7, the triangular mark indicates the location of the base station device 2. As shown in Figure 7, the typical position coordinates (x, y) of the base station device 2 are (-11.2, -4.2).

[0092] Furthermore, in the upper right of base station device 2 in the figure, there are 64 black dots arranged in an 8x8 grid. As shown in Figure 7, the position coordinates (x, y) of the black dot in the upper left corner are (-4.5, 3.5), the position coordinates (x, y) of the black dot in the upper right corner are (2.5, 3.5), the position coordinates (x, y) of the black dot in the lower left corner are (-4.5, -3.5), and the position coordinates (x, y) of the black dot in the lower right corner are (2.5, -3.5).

[0093] In this embodiment, a robot (humanoid robot) carrying the terminal device 3 is made to move sequentially across black dots randomly selected from among these 64 black dots. This allows the terminal device 3 to collect information indicating the wireless communication quality for various movements (movements). The shaded pattern shown around these 64 black dots represents the trajectory of the terminal device 3's movement.

[0094] As described above, the communication control system 4 in this embodiment performs preprocessing of spatial information to calculate the distance and angle from the base station device 2 to the terminal device 3 based on the values ​​of the positions of the base station device 2 and the terminal device 3, the movement speed and direction of the terminal device 3, respectively, included in the spatial information. As shown in Figure 7, for example, the communication control system 4 performs preprocessing of the spatial information so that the position coordinates of the base station device 2, (x, y) = (-11.2, -4.2), become (x', y') = (0, 0) (i.e., set as the origin).

[0095] Figure 8 is a diagram illustrating the spatial information of the terminal device 3 before preprocessing is performed. As shown in Figure 8, the position coordinates of the terminal device 3 (UE) mounted on the robot are expressed as (x[t], y[t]), for example, with a predetermined position (x, y) = (0, 0) as the origin. The velocity of the terminal device 3 (UE) in the direction of travel is expressed as v[t], for example. The rotational velocity of the terminal device 3 (UE) is expressed as ω[t], for example. The direction of movement of the terminal device 3 (UE) is expressed as θ[t], which is the angle with respect to the x-axis, for example.

[0096] Based on the above definition, the feature quantity φp of the spatial information of terminal device 3 before preprocessing is performed. １ [t] can be expressed as shown in equation (1) below.

[0097] φp １ [t] = (p[t], d[t], v[t]) ... (1)

[0098] Here, p[t] is p[t] = (x[t], y[t]), d[t] = (cosθ[t] / 2, sinθ[t] / 2), and v[t] = (v[t], ω[t]).

[0099] Figure 9 is a diagram illustrating the spatial information of the terminal device 3 after preprocessing. As shown in Figure 9, in the preprocessed spatial information, the position coordinates of the terminal device 3 (UE) mounted on the robot are expressed as (x'[t], y'[t]) with the position coordinates (x', y') = (0, 0) of the base station device 2 as the origin. The communication control system 4 in the embodiment converts the feature quantities of the spatial information of the terminal device 3 after preprocessing into a format that is expressed using the distance l[t] from the base station device 2 to the terminal device 3 and the angle θ'[t], based on the feature quantities of the spatial information of the terminal device 3 before preprocessing, as shown by equation (1) above.

[0100] φp, the spatial information feature quantity of terminal device 3 after preprocessing. ２ [t] is expressed as shown in equation (2) below.

[0101] φp ２ [t] = (l[t], θ'[t], v[t]) ... (2)

[0102] Here, l[t] = (x'[t] ２ , y'[t] ２ )^(1 / 2).

[0103] Thus, in the spatial information before preprocessing, the position reference differs at each location when the target device moves, making it unusable for predicting future wireless communication quality. Therefore, the communication control system 4 in this embodiment performs preprocessing to standardize the data so that the position of the base station device 2 is used as the reference. Specifically, the communication control system 4 converts the position coordinates (x, y) into position coordinates (x', y') with the position of the base station device 2 as the origin, and further converts the direction of movement of the terminal device 3 into distance and angle from the base station device 2.

[0104] The communication control system 4 then generates training data that associates, for example, the distance l and angle θ' from the base station device 2 to the terminal device 3, the velocity v and rotational velocity ω, the RSSI value which is wireless quality information, and the MCS index value, all of which are obtained by preprocessing spatial information. The communication control system 4 inputs this training data into a learning model and performs machine learning to generate a learning model that, for example, outputs the throughput value one second later for the input of preprocessed spatial information.

[0105] Figures 10 and 11 illustrate an example of the effectiveness of predicting wireless communication quality by the communication control system 4 in one embodiment of the present invention. As shown in Figure 10, here we take the example of sequentially moving three black dots from the 64 black dots shown in Figure 7: the upper right corner, the lower left corner, and the lower right corner. As mentioned above, the position coordinates (x, y) of the black dot in the upper right corner are (2.5, 3.5), the position coordinates (x, y) of the black dot in the lower left corner are (-4.5, -3.5), and the position coordinates (x, y) of the black dot in the lower right corner are (2.5, -3.5).

[0106] Figure 11 is a graph showing the effect of the prediction of wireless communication quality by the communication control system 4. In the line graph shown in Figure 11, the vertical axis represents the mean squared error (RMSE) (unit: Mbps) between the measured throughput and the predicted throughput, and the horizontal axis represents the number of training data samples used in the additional machine learning. The additional machine learning referred to here is the additional machine learning (retraining) performed by the learning model update unit 18 shown in Figure 2.

[0107] Of the four line graphs shown in Figure 11, if we label them (A), (B), (C), and (D) from top to bottom, then line graph (A) shows the calculation results of the mean squared error between the measured throughput and the predicted throughput for each number of training data samples, when no preprocessing or pre-training is performed. Here, preprocessing refers to the preprocessing performed by the spatial information preprocessing unit 12 shown in Figure 2, and pre-training refers to the machine learning performed by the learning model generation unit 15 shown in Figure 2.

[0108] Similarly, the line graph in (B) shows the calculation results of the mean squared error between the measured and predicted throughput for each sample size of the training data when preprocessing is performed but pretraining is not performed, the line graph in (C) shows the calculation results of the mean squared error between the measured and predicted throughput for each sample size of the training data when preprocessing is performed but pretraining is not performed, and the line graph in (D) shows the calculation results of the mean squared error between the measured and predicted throughput for each sample size of the training data when preprocessing is performed and pretraining is also performed.

[0109] The results of predicting future throughput using the communication control system 4 shown in Figure 11 demonstrate that, compared to the case where neither preprocessing nor pretraining is performed (as shown in line graph (A)), performing both preprocessing and pretraining (as shown in line graph (D)) improves the throughput prediction accuracy by 70.0% even with only additional machine learning using training data of about 100 samples. Furthermore, it was shown that even when the number of samples is further increased, performing both preprocessing and pretraining (as shown in line graph (D)) allows for more accurate throughput prediction.

[0110] Although not shown in Figure 11, it was demonstrated that even when both preprocessing and pretraining were performed, and no additional machine learning was performed (i.e., when the number of samples was 0), the mean squared error between the measured throughput and the predicted throughput could be kept to approximately 25.7 [Mbps].

[0111] Figures 12 and 13 illustrate another example of the effectiveness of predicting wireless communication quality by the communication control system 4 in one embodiment of the present invention. The embodiment of wireless communication quality prediction shown in Figures 12 and 13 was carried out in a different location (room) than the embodiment shown in Figures 10 and 11 described above.

[0112] Figure 12 shows the location (room) where wireless communication took place in this embodiment. As shown in Figure 12, in this embodiment, the robot equipped with the terminal device 3 was instructed to move sequentially over the three black dots shown in the figure. As shown in Figure 12, the position coordinates (x, y) of the upper left black dot are (-8.0, -5.0), the position coordinates (x, y) of the upper right black dot are (-1.5, 5.0), and the position coordinates (x, y) of the lower right black dot are (3.4, -5.6).

[0113] Figure 13 is a graph showing the effect of the prediction of wireless communication quality by the communication control system 4. Similar to the line graph shown in Figure 11, the vertical axis of the line graph shown in Figure 13 represents the mean squared error (RMSE) (unit: Mbps) between the measured throughput and the predicted throughput, and the horizontal axis represents the number of training data samples used in the additional machine learning. As mentioned above, the additional machine learning is the machine learning (retraining) performed by the learning model update unit 18 shown in Figure 2.

[0114] Of the four line graphs shown in Figure 13, if we label them (a), (b), (c), and (d) from top to bottom, then line graph (a) shows the calculation results of the mean squared error between the measured throughput and the predicted throughput for each number of training data samples, when no preprocessing or pre-training is performed. As mentioned above, the preprocessing referred to here is the preprocessing performed by the spatial information preprocessing unit 12 shown in Figure 2, and the pre-training is the machine learning performed by the learning model generation unit 15 shown in Figure 2.

[0115] Similarly, the line graph in (b) shows the calculation results of the mean squared error between the measured and predicted throughput for each sample size of the training data when preprocessing is performed but pretraining is not performed, the line graph in (c) shows the calculation results of the mean squared error between the measured and predicted throughput for each sample size of the training data when preprocessing is performed but pretraining is not performed, and the line graph in (d) shows the calculation results of the mean squared error between the measured and predicted throughput for each sample size of the training data when preprocessing is performed and pretraining is also performed.

[0116] The results of predicting future throughput using the communication control system 4 shown in Figure 13 demonstrate that, compared to the case where neither preprocessing nor pretraining is performed (as shown in line graph (a)), performing both preprocessing and pretraining (as shown in line graph (d)) improves the throughput prediction accuracy by 62.6% even with only additional machine learning using training data of about 100 samples. Furthermore, it was shown that even when the number of samples is further increased, performing both preprocessing and pretraining (as shown in line graph (d)) allows for more accurate throughput prediction.

[0117] Although not shown in Figure 13, it was demonstrated that even when both preprocessing and pretraining were performed, and no additional machine learning was performed (i.e., when the number of samples was 0), the mean squared error between the measured throughput and the predicted throughput could be kept to approximately 32.7 [Mbps].

[0118] Based on the prediction results of future throughput in each of the above embodiments, it has been shown that the communication control system 4 in one embodiment of the present invention can predict wireless communication quality with high accuracy, even in situations where the communication environment changes or in an environment different from the environment in which machine learning was performed, by performing machine learning again using only a small amount of additional training data, or by not performing machine learning again at all.

[0119] According to the embodiment described above, the learning device comprises a spatial information preprocessing unit, a data generation unit, and a learning model generation unit. For example, the learning device is the learning device 10 in the embodiment, the spatial information preprocessing unit is the spatial information preprocessing unit 12 in the embodiment, the data generation unit is the training data generation unit 14 in the embodiment, and the learning model generation unit is the learning model generation unit 15 in the embodiment. The spatial information preprocessing unit acquires spatial information between a first communication device and a second communication device that perform wireless communication, and performs preprocessing on the spatial information. For example, the first communication device is the base station device 2 in the embodiment, and the second communication device is the terminal device 3 in the embodiment. The data generation unit acquires wireless communication quality information, which is information regarding the quality of wireless communication, and generates training data to which the preprocessed spatial information and wireless communication quality information are associated. The learning model generation unit 15 performs machine learning in advance using the training data and generates a learning model that predicts the quality of wireless communication between the first communication device and the second communication device.

[0120] Furthermore, in the learning device described above, the spatial information preprocessing unit may perform preprocessing to convert the acquired spatial information into spatial information based on the position of the first communication device.

[0121] In addition, in the learning device described above, the spatial information preprocessing unit may acquire spatial information obtained by moving a second communication device that performs wireless communication at various locations, at various speeds, and in various directions.

[0122] The above-described learning device may further include an environmental change detection unit and a learning model update unit. For example, the environmental change detection unit is the environmental change detection unit 17 in the embodiment, and the learning model update unit is the learning model update unit 18 in the embodiment. In this case, the environmental change detection unit determines whether or not there has been an environmental change between the first communication device and the second communication device. If the learning model update unit determines that there has been an environmental change, it updates the learning model using new training data.

[0123] Furthermore, the learning device of the present invention can also be realized using a computer and a program, and the program can be recorded on a recording medium or provided via a network.

[0124] Some or all of the configurations of the devices in the wireless communication system 1 in the above-described embodiment may be implemented using a computer. In that case, the program for implementing this function may be recorded on a computer-readable recording medium, and the program recorded on this recording medium may be loaded into the computer system and executed. Here, "computer system" includes hardware such as the OS and peripheral devices. Furthermore, "computer-readable recording medium" refers to portable media such as flexible disks, magneto-optical disks, ROMs, CD-ROMs, and storage devices such as hard disks built into the computer system. Moreover, "computer-readable recording medium" may also include those that dynamically hold programs for a short period of time, such as communication lines used when transmitting programs via networks such as the Internet or communication lines such as telephone lines, and those that hold programs for a certain period of time, such as volatile memory inside the computer system that acts as a server or client in that case. Furthermore, the above-mentioned program may be for implementing some of the functions described above, or it may be a program that can implement the above-mentioned functions in combination with a program already recorded in the computer system, or it may be implemented using a programmable logic device such as an FPGA (Field Programmable Gate Array).

[0125] While embodiments of this invention have been described in detail above with reference to the drawings, the specific configuration is not limited to these embodiments and includes designs and the like that do not depart from the spirit of this invention.

[0126] 1 Wireless communication system 2 Base station equipment 3 Terminal equipment 4 Communication control system 10 Learning device 11 Spatial information acquisition unit 12 Spatial information preprocessing unit 13 Quality information acquisition unit 14 Training data generation unit 15 Learning model generation unit 16 Learning model storage unit 17 Environmental change determination unit 18 Learning model update unit 20 Estimation device 21 Spatial information acquisition unit 22 Spatial information preprocessing unit 23 Learning model acquisition unit 24 Quality estimation unit 25 Estimation result output unit 30 Communication control device 40 Measurement device

Claims

1. A learning device comprising: a spatial information preprocessing unit that acquires spatial information between a first communication device and a second communication device that perform wireless communication and preprocesses the spatial information; a data generation unit that acquires wireless communication quality information, which is information relating to the quality of the wireless communication, and generates training data in which the preprocessed spatial information and the wireless communication quality information are associated; and a learning model generation unit that performs machine learning in advance using the training data and generates a learning model that predicts the quality of the wireless communication between the first communication device and the second communication device.

2. The learning device according to claim 1, wherein the spatial information preprocessing unit performs the preprocessing to convert the acquired spatial information into spatial information based on the position of the first communication device.

3. The learning device according to claim 2, wherein the spatial information preprocessing unit acquires the spatial information obtained by moving the second communication device that performs wireless communication at various locations, at various speeds, and in various directions.

4. The learning device according to any one of claims 1 to 3, further comprising: an environment change determination unit that determines whether or not there has been an environmental change between the first communication device and the second communication device; and a learning model update unit that updates the learning model using the new training data when it is determined that there has been an environmental change.

Images