Prediction system

The prediction system uses a radio wave intensity estimation and sensor alignment to achieve high-accuracy communication quality prediction with minimal data, addressing the need for extensive data in conventional methods.

WO2026234802A1PCT designated stage Publication Date: 2026-11-12NT T INC
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
PCT/JP2025/017048
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Filing Date
2025-05-09
Publication Date
2026-11-12

AI Technical Summary

Technical Problem

Conventional wireless communication quality prediction methods using deep learning require a large amount of learning data and suffer from decreased accuracy when data is scarce, necessitating separate model preparation for each environmental change.

Method used

A prediction system that includes a radio wave intensity estimation unit to determine the direction of arrival, a measurement device control unit to align sensors with this direction, and a learning device to predict communication quality using spatial information, allowing for high-accuracy predictions with minimal data.

Benefits of technology

Enables accurate communication quality prediction even with a small amount of training data, adapting to environmental changes without extensive retraining.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure JP2025017048_12112026_PF_FP_ABST
    Figure JP2025017048_12112026_PF_FP_ABST
Patent Text Reader

Abstract

This prediction system comprises: a radio signal strength estimation unit that estimates a direction of arrival of a radio wave arriving at a base station from a terminal; a measurement device control unit that controls the orientation of an acquisition unit, which acquires spatial information pertaining to the presence or absence of shielding between the base station and the terminal, so as to face the same direction as the direction of arrival of the radio wave estimated by the radio signal strength estimation unit; and a training device that, on the basis of the spatial information in a specific direction obtained by the acquisition unit, performs training to obtain a trained model for predicting future communication quality by using the spatial information and quality information pertaining to communication quality as inputs. 
Need to check novelty before this filing date? Find Prior Art

Description

Prediction system

[0001] The present invention relates to a prediction system.

[0002] In recent years, high frequency bands have been used in wireless communications. Since high frequency bands have strong radio wave straightness, communication quality is significantly degraded by shielding from surrounding objects. Therefore, proactive control based on wireless communication quality prediction is effective for stabilizing communication quality. Accordingly, conventionally, wireless quality prediction methods using deep learning that utilize spatial information such as images captured by cameras have been proposed (see, for example, Non-Patent Documents 1 and 2).

[0003] H. Nagata, R. Kudo, K. Takahashi, T. Ogawa and K. Takasugi, "Two-step wireless link quality prediction using multi-camera images", Proc. IEEE Annu. Int. Symp. Pers. Indoor Mobile Radio Commun., pp. 509-514, 2022. Y. Feng et al., "Vision-aided Reference Signal Receiving Power Prediction for Smart Factory", IEEE Wireless Communications and Networking Conference (WCNC), 2024.

[0004] The conventional technology uses deep learning which is supervised learning, and enables high-accuracy quality prediction, but it requires collecting a huge amount of learning data to construct a prediction model. The prediction model needs to be prepared for each different environment (for example, for each environmental change where the number of surrounding objects such as people changes). Therefore, although it is desirable that the number of learning data per environment is small, there has been a problem that the prediction accuracy decreases when the learning data is small.

[0005] In view of the above circumstances, the present invention aims to provide a technology capable of predicting communication quality with high accuracy using a small amount of learning data.

[0006] One aspect of the present invention is a prediction system comprising: a radio wave intensity estimation unit that estimates the direction of arrival of radio waves arriving from a terminal to a base station; a measurement device control unit that controls the orientation of an acquisition unit that acquires spatial information regarding the presence or absence of shielding between the base station and the terminal so as to face the same direction as the direction of arrival of the radio waves estimated by the radio wave intensity estimation unit; and a learning device that learns a learning model that predicts future communication quality based on the spatial information of a specific direction obtained by the acquisition unit, taking the spatial information and quality information regarding communication quality as input.

[0007] This invention makes it possible to predict communication quality with high accuracy using a small amount of training data.

[0008] This figure shows an example of the configuration of the prediction system in the embodiment. This is a sequence diagram showing the flow of the learning model generation process in the prediction system in the embodiment. This is a flowchart showing the flow of the learning model update process performed by the learning device in the embodiment. This figure shows an example of an experiment. This figure shows the results of comparing the mean absolute error with respect to the number of training data between the conventional method and the proposed method of the present invention.

[0009] One embodiment of the present invention will be described below with reference to the drawings.

[0010] Figure 1 shows an example configuration of the prediction system 100 in an embodiment. The prediction system 100 is a system that predicts communication quality. The prediction system 100 comprises a base station device 10, a terminal device 20, a measuring device 30, and a learning device 40. The measuring device 30 is installed in a location as close as possible to the base station device 10. For example, the measuring device 30 may be installed outside the base station device 10, such as directly below or directly above the base station device 10, or it may be installed inside the base station device 10 if it can be incorporated into the base station device 10. If the measuring device 30 cannot be installed outside the base station device 10, such as directly below or directly above the base station device 10, the measuring device 30 should be installed in a location as close as possible to the base station device 10.

[0011] The base station device 10 transmits and receives radio waves in order to perform wireless communication. The base station device 10 transmits and receives radio waves to and from, for example, the terminal device 20.

[0012] Terminal device 20 is a terminal whose quality is to be predicted. Terminal device 20 is, for example, a terminal device used by a user.

[0013] The measuring device 30 is a device that collects spatial information using a camera and sensors. Here, spatial information includes images captured by the camera and information such as distance and object shape measured by LiDAR (Light Detection and Ranging). Thus, spatial information represents information about the presence or absence of obstruction between the base station device 10 and the terminal device 20. For example, if an obstruction exists between the base station device 10 and the terminal device 20, the obstruction can be detected based on information such as images and object shape. On the other hand, if there is no obstruction between the base station device 10 and the terminal device 20, the obstruction will not be detected based on information such as images and object shape. The spatial information collected by the measuring device 30 is used as training data for the learning device 40. The measuring device 30 also has a function to estimate radio wave strength and collects spatial information by pointing the camera and sensors in the direction from which the radio waves are coming.

[0014] The measuring device 30 includes a camera and sensors (not shown), a radio wave intensity estimation unit 31, and a measuring device control unit 32. The camera and sensors acquire spatial information. The radio wave intensity estimation unit 31 estimates the shielding of the terminal device 20 from the sensor by aligning the sensor direction with the direction in which the radio wave intensity of the base station device 10 is strong. For example, the radio wave intensity estimation unit 31 acquires quality information for each beam of the base station device 10 in response to the signal transmitted by the terminal device 20.

[0015] Quality information includes, for example, RSSI (Received Signal Strength Indicator) and RSRP (Reference Signal Received Power). The radio wave strength estimation unit 31 identifies the beam with the highest quality based on the quality information of each beam of the base station device 10. The radio wave strength estimation unit 31 then identifies the direction of the identified beam as the direction of arrival of the radio waves from the terminal device 20 relative to the base station device 10. In other words, the radio wave strength estimation unit 31 identifies the direction in which the terminal device 20 is located relative to the base station device 10.

[0016] The measurement device control unit 32 adjusts the orientation of the measurement device 30 so that the camera and sensors are pointed in the direction identified by the radio wave intensity estimation unit 31. For example, the measurement device control unit 32 adjusts the orientation of the measurement device 30 so that the camera's field of view is pointed in the direction identified by the radio wave intensity estimation unit 31. Furthermore, if the direction of strong radio wave intensity changes, the measurement device control unit 32 performs control to adjust the orientation of the measurement device 30 accordingly (for example, control to move the camera's field of view). This ensures that shielding can always be determined.

[0017] The learning device 40 is a device that generates a learning model used for predicting communication quality. The learning device 40 includes a spatial information acquisition unit 41, a spatial information preprocessing unit 42, a quality information acquisition unit 43, a learning data generation unit 44, a learning model generation unit 45, a learning model storage unit 46, an environmental change determination unit 47, and a learning model update unit 48.

[0018] The spatial information acquisition unit 41 acquires spatial information such as the location information of the terminal device 20 and spatial information collected by the measuring device 30. The location information of the terminal device 20 may be acquired from the GPS (Global Positioning System) of the terminal device 20, or it may be acquired based on images collected by the measuring device 30.

[0019] The spatial information preprocessing unit 42 performs preprocessing on the spatial information acquired by the spatial information acquisition unit 41. For example, the spatial information preprocessing unit 42 performs preprocessing so that it acquires only the information necessary for generating training data from the spatial information. Here, the information necessary for generating training data is, for example, information of the region including the terminal device 20. For example, the spatial information preprocessing unit 42 extracts bounding box information of the terminal device 20 from the image obtained from the measuring device 30. Furthermore, the spatial information preprocessing unit 42 preprocesses and calculates the distance between the base station device 10 and the terminal device 20, the angle, the velocity in the direction of travel, the rotational velocity, etc., based on the position information of the terminal device 20.

[0020] The quality information acquisition unit 43 acquires quality information such as the throughput of the terminal device 20 and RSSI at the base station device 10.

[0021] The training data generation unit 44 generates training data for training a predictive model based on the preprocessed spatial information obtained by the spatial information preprocessing unit 42 and the quality information obtained by the quality information acquisition unit 43. The training data includes time-series image data as preprocessed spatial information.

[0022] The learning model generation unit 45 generates a learning model using machine learning with the learning data generated by the learning data generation unit 44. The learning model is a pre-trained model that has been trained to predict future communication quality using input data that combines pre-processed spatial information and quality information. For example, the learning model generation unit 45 may generate the learning model using deep learning. Here, learning refers to optimizing the coefficients used in the machine learning model. For example, learning refers to adjusting the coefficients used in the machine learning model so that the loss function is minimized. The coefficients used in the machine learning model are, for example, weight values ​​and bias values.

[0023] The learning model storage unit 46 stores the learning models generated by the learning model generation unit 45. The learning models stored in the learning model storage unit 46 are used when predicting communication quality.

[0024] The environmental change detection unit 47 detects environmental changes in the vicinity of the base station device 10. For example, the environmental change detection unit 47 detects environmental changes such as the number of people or other objects in the vicinity of the base station device 10 based on the learned data. Alternatively, the environmental change detection unit 47 may detect environmental changes such as the number of people or other objects based on time-series image data included in the spatial information obtained by the spatial information acquisition unit 41. If the environmental change detection unit 47 detects an environmental change, it instructs the learning model update unit 48 to update the learning model. If the environmental change detection unit 47 does not detect an environmental change, it does nothing in particular.

[0025] When the environmental change detection unit 47 detects an environmental change, the learning model update unit 48 retrains the learning model stored in the learning model storage unit 46. In other words, when the environmental change detection unit 47 detects an environmental change, the learning model update unit 48 updates the parameters of the learning model stored in the learning model storage unit 46.

[0026] [First Operation] Figure 2 is a sequence diagram showing the flow of the learning model generation process in the prediction system 100 in the embodiment. Note that the process in Figure 2 is executed at the start of processing. Here, we will explain the case where the measurement device 30 is equipped with a camera. The radio wave intensity estimation unit 31 of the measurement device 30 estimates the radio wave intensity for each beam in the base station device 10 (step S101). As a result of the estimation, the radio wave intensity estimation unit 31 identifies the direction of the beam with the highest radio wave intensity. The radio wave intensity estimation unit 31 outputs information indicating the direction of the identified beam to the measurement device control unit 32.

[0027] The measurement device control unit 32 controls the direction of the measurement device 30 based on information indicating the direction of the beam output from the radio wave intensity estimation unit 31 (step S102). For example, the measurement device control unit 32 controls the direction of the measurement device 30 so that the camera's field of view is oriented in the direction specified by the information indicating the direction of the beam output from the radio wave intensity estimation unit 31. Subsequently, the measurement device 30 collects spatial information (step S103). Specifically, the measurement device 30 collects image data obtained by capturing an image with a camera (not shown) as spatial information. The measurement device 30 transmits the collected spatial information to the learning device 40 (step S104).

[0028] The spatial information acquisition unit 41 acquires spatial information transmitted from the measuring device 30 (step S105). Furthermore, the spatial information acquisition unit 41 acquires the location information of the terminal device 20 as spatial information from the image data included in the spatial information transmitted from the measuring device 30. The spatial information acquisition unit 41 outputs the acquired spatial information to the spatial information preprocessing unit 42. The spatial information preprocessing unit 42 performs preprocessing on the spatial information output from the spatial information acquisition unit 41 (step S106). The spatial information preprocessing unit 42 outputs the preprocessed spatial information to the learning data generation unit 44. The preprocessed spatial information also includes the location information of the terminal device 20.

[0029] The quality information acquisition unit 43 acquires quality information (step S107). The quality information acquisition unit 43 outputs the acquired quality information to the learning data generation unit 44. The learning data generation unit 44 acquires the pre-processed spatial information output from the spatial information preprocessing unit 42 and the quality information output from the quality information acquisition unit 43. The learning data generation unit 44 generates learning data based on the acquired pre-processed spatial information, quality information, and correct answer data (step S108). Here, the correct answer data is throughput. The learning data generation unit 44 outputs the generated learning data to the learning model generation unit 45. The learning model generation unit 45 generates a learning model using the learning data output from the learning data generation unit 44 (step S109). After that, the learning model generation unit 45 stores the generated learning model in the learning model storage unit 46 (step S110).

[0030] After the learning model is stored in the learning model storage unit 46 through the above process, the learning device 40 does not need to generate a learning model until the learning model update unit 48 outputs an instruction to update the learning model. The measurement device 30 executes the processes from step S101 to step S103. For example, as the terminal device 20 moves, the radio wave intensity for each beam of the base station device 10 also changes. The measurement device 30 repeatedly executes the processes from step S101 to step S103 to respond to the change in radio wave intensity. In this way, spatial information is collected only in a specific direction (the direction in which the radio wave intensity is strong).

[0031] [Second Operation] Figure 3 is a flowchart showing the flow of the learning model update process performed by the learning device 40 in the embodiment. The process in Figure 3 is executed at a predetermined timing after the process in Figure 2 is completed. The predetermined timing may be a predetermined period, or it may be the timing at which spatial information is obtained from the measuring device 30.

[0032] The environmental change determination unit 47 determines, based on the learned data, whether or not there has been an environmental change, such as the number of people or other objects in the vicinity of the base station device 10 (step S201). If the environmental change determination unit 47 determines that there has been no environmental change (step S201-NO), the learning device 40 terminates the process shown in Figure 3.

[0033] On the other hand, if the environmental change determination unit 47 determines that an environmental change has occurred (step S201-YES), the environmental change determination unit 47 instructs the learning model update unit 48 to update the learning model (step S202). The learning model update unit 48 updates the learning model in accordance with the instruction from the environmental change determination unit 47 (step S203). Specifically, it updates the parameters of the learning model stored in the learning model storage unit 46 using the learning data generated by the learning data generation unit 44. The learning model update unit 48 stores the updated learning model in the learning model storage unit 46 (step S204).

[0034] Next, we will describe specific scenarios in which the prediction system 100 described above can be applied. The first application scenario is assumed to be on-site, such as logistics centers and factories. On-site, such as logistics centers and factories, it is expected that environmental changes will occur frequently due to changes in the arrangement of equipment, including base stations, and the introduction of new equipment. In such situations, the prediction system 100 can make highly accurate predictions even when environmental changes occur, using only a small amount of training data based on spatial information in a specific direction (the direction of strong radio wave intensity), or without additional training.

[0035] A second application scenario is an office. In an office, frequent coming and going of people is expected. In such a scenario, the prediction system 100 can make highly accurate predictions based on a small amount of training data, or without additional training, even when environmental changes occur that cause changes in the number of people, using spatial information only from a specific direction (the direction with strong radio wave intensity).

[0036] Next, we will describe the configuration of proactive control based on quality prediction using the prediction system 100 described above. Here, proactive control based on quality prediction means dynamically controlling the communication environment to maintain or improve communication quality based on the communication quality predicted by the learning model in the prediction system 100.

[0037] (Proactive control pattern 1 based on quality prediction) If the learning model predicts that the quality will be low, the base station equipment 10 controls RIS (Reconfigurable Intelligent Surface) and beamforming. This maintains wireless quality.

[0038] (Proactive control pattern 2 based on quality prediction) When using a real-time video streaming application, if the learning model predicts that the quality will be low, the application will proactively control the bitrate (for example, by changing the video quality from 4K to 2K). This prioritizes real-time performance while maintaining application quality.

[0039] (Proactive control pattern 3 based on quality prediction) In an environment where multiple automated mobile robots (AMRs) move around, a learning model is used to predict future temporal fluctuations in the communication quality of wireless terminals carried by workers and robots. Based on the prediction results, the AMRs are routed optimally, or the prediction results are notified to users and the system. This enables monitoring and management.

[0040] (Proactive control pattern 4 based on quality prediction) When aggregating data to the cloud or server, if the learning model predicts that the wireless quality is good (high), it will transmit videos and other data captured by surveillance cameras and cameras mounted on each robot via high-capacity communication. Conversely, if the wireless quality is predicted to be poor (low), it will perform proactive control such as not performing high-capacity communication. This reduces the number of retransmissions and shortens communication time. As a result, power saving can be achieved.

[0041] (Proactive control pattern 5 based on quality prediction) In places where a large number of people come and go, such as event venues and conference rooms, the learning model predicts routes that ensure stable communication and leads to pedestrian flow estimation.

[0042] (Experimental Results) Next, the effects of the present invention will be explained based on the experimental results. Figure 4 shows an example of the experiment. In the experiment, two robots, A and B, were prepared in place of pedestrians. Robot A was positioned at the location indicated by point P1, holding a terminal (equivalent to terminal device 20) in its left hand. Robot B was positioned at the location indicated by point P2 as an obstacle. As shown in Figure 4(A), each robot was made to travel back and forth as indicated by arrows A1 and A2. A camera (equivalent to measuring device 30) was installed at the viewpoint of the base station (equivalent to base station device 10) shown in Figure 4(A), and the orientation of the camera was the same as the direction of the radio waves from the base station (equivalent to base station device 10), as indicated by arrow A.

[0043] In this environment, the learning device 40 acquires image data and terminal location information from the camera as spatial information, and acquires the terminal's RSSI and throughput as wireless quality information. Based on the image data, the learning device 40 extracts the terminal's bounding box information and preprocesses and calculates the distance l, the terminal's angle θ, the terminal's velocity in the direction of travel Vr, and the terminal's rotational velocity Vω based on the terminal's location information. The learning device 40 generates input data by combining the preprocessed spatial information and the RSSI of the wireless quality information. By training the input data using deep learning, the learning device 40 trains and generates a learning model that outputs the throughput one second ahead.

[0044] FIG. 4(B) illustrates a scenario where a pre-trained learning model is applied. As shown in FIG. 4(B), robot C is newly placed at the position indicated by point P3. Note that robot C travels back and forth as indicated by arrow A3. In such an environment, the throughput of the terminal was estimated using the pre-trained learning model. The results are shown in FIG. 5. FIG. 5 is a diagram showing the results of a comparison of the Mean Absolute Error (MAE) with respect to the number of learning data points between the conventional method and the proposed method of the present invention. As shown in FIG. 6, it is demonstrated that the proposed method enables highly accurate prediction even when the number of learning data points is 0 or 100 samples, as compared with the conventional method in which a prediction model is constructed for each environmental change. In particular, with the proposed method, an accuracy improvement of 24.3% in terms of Mean Absolute Error was confirmed when the number of learning data points was 100 samples.

[0045] According to the prediction system 100 configured as described above, the prediction system 100 includes: a radio wave intensity estimation unit 31 that estimates the arrival direction of a radio wave arriving at the base station apparatus 10 from the terminal apparatus 20; a measurement apparatus control unit 32 that controls the orientation of a camera or a sensor so that the camera or sensor faces the same direction as the estimated arrival direction of the radio wave; and a learning apparatus 40 that learns a learning model for predicting future communication quality by taking spatial information and quality information related to communication quality as inputs, based on the spatial information in a specific direction obtained by the camera or the sensor.

[0046] In the millimeter wave band, the presence or absence of blocking of direct waves significantly affects the quality. Therefore, in the prediction system 100, the measurement apparatus 30 equipped with a camera and a sensor aligns the angle of view of the camera with the direction of strong radio wave intensity, whereby blocking can be estimated with high accuracy from an image based on the presence or absence of the terminal apparatus 20. Note that if the angle of view of the camera is not aligned, blocking cannot be accurately determined solely based on the presence or absence of the terminal apparatus 20. Then, the prediction system 100 performs learning focused on radio wave propagation paths that are susceptible to the influence of blocking. As a result, even when an environmental change such as a change in the number of surrounding objects such as people occurs, highly accurate prediction can be achieved with a small amount of learning data or no learning data, without preparing a huge amount of learning data for each environment to construct a learning model.

[0047] (Modification) In the above-described embodiment, the radio wave strength estimation unit 31 is configured to estimate the arrival direction of radio waves based on the quality information for each beam of the base station apparatus 10. In contrast, the radio wave strength estimation unit 31 may acquire beam information for specifying the beam selected by the base station apparatus 10, and estimate the arrival direction of radio waves based on the acquired beam information. In this case, the radio wave strength estimation unit 31 estimates the direction of the beam indicated by the acquired beam information as the arrival direction of the radio wave.

[0048] The measurement device 30 and the learning device 40 included in the prediction system 100 are configured using, for example, a processor such as a CPU (Central Processing Unit), a memory, and a communication interface. Each of the measurement device 30 and the learning device 40 functions as a communication device including a functional unit that performs specific processing when the processor executes a program.

[0049] The above program provides respective functions for causing an information processing apparatus to function as either the measurement device 30 or the learning device 40. Note that all or part of each function may be implemented using hardware including an electronic circuit or circuitry using, for example, an LSI (Large Scale Integrated circuit), an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), or an FPGA (Field Programmable Gate Array).

[0050] The above program may be recorded on a computer-readable non-transitory recording medium. The computer-readable non-transitory recording medium refers to, for example, portable media such as a flexible disk, a magneto-optical disk, a ROM (Read Only Memory), and a CD-ROM (Compact Disc Read Only Memory), and non-transitory recording media such as a storage device including a hard disk built into a computer system.

[0051] 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.

[0052] This invention can be applied to techniques for predicting communication quality.

[0053] 10...Base station equipment, 20...Terminal equipment, 30...Measurement device, 31...Radio wave intensity estimation unit, 32...Measurement device control unit, 41...Spatial information acquisition unit, 42...Spatial information preprocessing unit, 43...Quality information acquisition unit, 44...Learning data generation unit, 45...Learning model generation unit, 46...Learning model storage unit, 47...Environmental change determination unit, 48...Learning model update unit, 100...Prediction system

Claims

1. A prediction system comprising: a radio wave intensity estimation unit that estimates the direction of arrival of radio waves arriving from a terminal to a base station; a measurement device control unit that controls the orientation of an acquisition unit that acquires spatial information regarding the presence or absence of shielding between the base station and the terminal so that it faces the same direction as the direction of arrival of the radio waves estimated by the radio wave intensity estimation unit; and a learning device that learns a learning model that predicts future communication quality based on the spatial information of a specific direction obtained by the acquisition unit, taking the spatial information and quality information regarding communication quality as input.

2. The prediction system according to claim 1, wherein the radio wave intensity estimation unit estimates the direction of arrival of the radio waves based on the quality information of each beam of the base station.

3. The prediction system according to claim 1, wherein the radio wave intensity estimation unit acquires beam information relating to the beam selected by the base station and estimates the direction of arrival of the radio waves based on the acquired beam information.

4. The prediction system according to any one of claims 1 to 3, wherein the radio wave intensity estimation unit and the measurement device control unit are located closer to the base station.