Information processing method and information processing

Abstract

An information processing device 10 comprises: an information acquisition unit 11 that acquires status information regarding the status of a target terminal 51, the communication quality of the target terminal 51, and surrounding space information regarding a surrounding space 50 of the target terminal 51; a prediction learning unit 12 that inputs at least one among the status information, the communication quality, and the surrounding space information in to a prediction model, and predicts whether the communication quality of the target terminal 51 will be equal to or smaller than a threshold value; and an index value calculation unit 15 that, on an individual communication quality level basis, uses a prediction result of the prediction model and the actual communication quality to calculate the area of a region under the ROC curve which has respective axes representing a true positive rate and a false positive rate, and defines the calculated area as the index value in the communication quality level.

Description

Information processing method and information processing device

[0001] The present disclosure relates to an information processing method and an information processing device.

[0002] As wireless communication becomes more widespread, ensuring stable communication quality is becoming an important issue. The state of the wireless signal propagation channel changes over time and space, which can affect wireless quality.

[0003] To address these issues, communication quality prediction algorithms have been proposed to guarantee service quality in wireless communication systems (Non-Patent Documents 1 and 2). Predicting communication quality and taking appropriate measures based on the prediction results, such as selecting high-quality links, can improve the reliability of data transmission (Non-Patent Document 3). Non-Patent Document 4 proposes a method for optimizing handover timing by constructing a deep learning model that predicts received power from camera images and predicting the received power of a terminal that fluctuates due to moving objects that block the communication path.

[0004] Gregor Cerar, Haili Yetgin, Mihael Mohorcic, and Carolina Fortuna, “Machine Learning for Link Quality Estimation: A Survey,” IEEE COMMUNICATIONS SURVEYS & TUTOLIALS, 2021, VOL. 23, NO. 2, pp. 696-728; Takashi Nagata, Riichi Kudo, Kaoruko Takahashi, Ramesh Eshan, Tomoaki Ogawa, Yuya Aoki, Tomoki Horise, and Yoshifumi Morihiro, “5G Throughput Prediction Technology Using Physical Spatial Information,” IEICE Technical Report, IEICE, 2023, Vol. 123, No. 31, pp. 36-41; Mingxiao Niu, Linlan Liu, and Jian Shu, “Link Quality Prediction for Wireless Networks: Current Status and Future Directions,” ICIIT '23, 2023, pp. 52-56; Yusuke Koda, Kota Nakashima, Koji Yamamoto, Takayuki Nishio, and Masahiro Morikura, “Handover Management for mmWave Networks With Proactive Performance Prediction Using Camera Images and Deep Reinforcement Learning,” IEEE TRANSACTIONS ON COGNITIVE COMMUNICATIONS AND NETWORKING, 2020, VOL. 6, NO. 2, pp. 802-816

[0005] In conventional technologies, given information is acquired to predict wireless communication quality, and terminals, etc. are controlled based on the prediction. However, it was unclear whether control based on the predicted wireless communication quality could guarantee the desired service quality in a real environment.

[0006] The present disclosure has been made in consideration of the above, and aims to more appropriately apply a learned predictive model in a real environment.

[0007] An information processing device of one embodiment of the present disclosure includes an acquisition unit that acquires status information regarding the status of a wireless communication terminal, communication quality of the wireless communication terminal, and surrounding space information regarding the space surrounding the wireless communication terminal; a prediction unit that inputs at least one of the status information, the communication quality, and the surrounding space information into a prediction model and solves a classification problem with a predetermined level of communication quality as a boundary; and a calculation unit that calculates, for each level of communication quality, an evaluation index for the classification problem using the prediction result of the prediction model at that level of communication quality and the actual measured value of communication quality, and sets it as an index value at that level of communication quality.

[0008] According to the present disclosure, a learned predictive model can be more appropriately applied in a real environment.

[0009] FIG. 1 is a diagram showing an example of the configuration of an information processing device of this embodiment. FIG. 2 is a flowchart showing an example of the flow of a process for calculating an index value. FIG. 3 is a diagram showing an example of a predicted value and an actual measured value of communication quality. FIG. 4 is a diagram showing an example of an ROC curve. FIG. 5 is a diagram showing an example of an index value for each index threshold. FIG. 6 is a diagram showing an example of a service usage environment. FIG. 7 is a diagram showing an example of an index value for each index threshold. FIG. 8 is a diagram showing an example of a service usage environment. FIG. 9 is a diagram showing an example of an index value for each index threshold. FIG. 10 is a diagram showing an example of an index value for each index threshold. FIG. 11 is a diagram showing an example of the hardware configuration of an information processing device.

[0010] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings.

[0011] An example of the configuration of an information processing device 10 according to this embodiment will be described with reference to Fig. 1. The information processing device 10 shown in the figure includes an information acquisition unit 11, a predictive learning unit 12, a control unit 13, a requirement information acquisition unit 14, and an index value calculation unit 15.

[0012] The information processing device 10 predicts the communication quality of the target terminal 51 and controls the target terminal 51. If the other terminals 52 and the object 53 are controllable, the information processing device 10 may control the other terminals 52 and the object 53. The target terminal 51 is a terminal to which a service is provided. The other terminals 52 are terminals other than the target terminal 51 that perform wireless communication. The object 53 is an object present in the surrounding space 50 of the target terminal 51.

[0013] Furthermore, the information processing device 10 determines the level of communication quality at which the prediction model has high prediction accuracy based on index values ​​that represent the performance of the prediction model for each level of communication quality in the real space, selects or learns a prediction model so that the prediction accuracy is high at the level of communication quality that meets the required requirements, or adjusts the surrounding space so that the prediction accuracy is high at the level of communication quality that meets the required requirements.

[0014] The information acquisition unit 11 acquires status information of the target terminal 51, the other terminal 52, and the object 53, as well as communication information of the target terminal 51, the other terminal 52, and the base station with which the target terminal 51 is communicating. The status information is information that indicates the status of the terminal or object, such as the position, speed, orientation, and type (model) of the terminal. The status information may be acquired using a sensor such as a camera or Light Detection and Ranging (LiDAR). The communication information is an index related to the quality of wireless communication between at least one communication unit in the communication device and an external communication device. Indicators related to QoE (Quality of experience) can be used, such as received power, RSSI (Received Signal Strength Indicator), RSRQ (Reference Signal Received Quality), SNR (Signal to noise ratio), SINR (Signal to interference noise ratio), packet loss rate, data rate, application quality, and indicators related to increases and decreases thereof, as well as indicators combining two or more of these by linear calculation, etc. Information about the target terminal 51 may be acquired as terminal information, or information about the states of other terminals 52 and objects 53 present in the surrounding space 50 may be acquired as surrounding space information.

[0015] The predictive learning unit 12 inputs the information obtained by the information acquisition unit 11 into the trained prediction model and predicts future communication information of the target terminal 51. More specifically, the prediction model inputs the information obtained by the information acquisition unit 11 and solves classification problems such as inferring communication quality in the near future (e.g., several seconds from now), determining whether the communication quality will be below a predetermined threshold (hereinafter referred to as the index threshold), or determining whether the communication quality will be above the threshold. When solving a classification problem, the prediction model may output a score close to 1 if the communication quality is below the index threshold, and may output a score close to 0 if the communication quality is not below the index threshold. The index threshold is set according to the requirements of the service 30.

[0016] The predictive learning unit 12 inputs at least one of the state information, surrounding space information, and communication information into the prediction model to predict the communication information. For example, the predictive learning unit 12 inputs the position, orientation, and speed of the target terminal 51 into the prediction model as state information, inputs the positions and speeds of the other terminals 52 and objects 53 into the prediction model as surrounding space information, and inputs past communication information into the prediction model as communication information.

[0017] The predictive learning unit 12 may hold a plurality of predictive models and use different predictive models depending on the situation of the surrounding space.

[0018] The prediction learning unit 12 may use the information obtained by the information acquisition unit 11 as learning data to machine-learn a prediction model.

[0019] The control unit 13 controls the target terminal 51, the other terminal 52, or the object 53 based on the prediction result of the prediction learning unit 12. For example, when the communication quality becomes worse than a predetermined threshold, the control unit 13 causes the target terminal 51 to hand over and connect to another base station.

[0020] The requirement information acquisition unit 14 acquires information about requirements required by the service 30. For example, the requirement information acquisition unit 14 acquires the throughput required by the service 30.

[0021] The index value calculation unit 15 calculates an index value representing the performance of the prediction model for each level of communication quality in the real space, using the prediction results of the prediction learning unit 12 and the actual measured values ​​of communication quality.

[0022] Furthermore, the index value calculation unit 15 determines the communication quality level with the highest prediction accuracy of the prediction model based on the calculated index value and notifies the service 30 of the level, selects or learns a prediction model so as to increase the prediction accuracy at the requested communication quality level, or adjusts the surrounding space so as to increase the prediction accuracy at the requested communication quality level. For example, the index value calculation unit 15 notifies the service 30 of the communication quality level with the highest calculated index value as a communication quality requirement for providing the service 30. The index value calculation unit 15 controls the prediction learning unit 12 to use the prediction model with the highest index value at the requested communication quality level. The index value calculation unit 15 causes the prediction learning unit 12 to relearn the prediction model so as to increase the index value at the requested communication quality level. The index value calculation unit 15 moves the position of the object 53 in the surrounding space 50, controls the movement path, direction, and speed of the target terminal 51, or moves the position of the base station so as to increase the index value at the requested communication quality level.

[0023] An example of the flow of a process for calculating an index value will be described with reference to the flowchart of FIG.

[0024] In step S11, the index value calculation unit 15 sets an index threshold T. When the requirement information acquisition unit 14 acquires requirements, the index value calculation unit 15 sets the index threshold T in accordance with the requirements. When an index value is used to determine the requirements, the index value calculation unit 15 sets the index threshold T while changing it in stages.

[0025] In step S12, the index value calculation unit 15 obtains, from the predictive learning unit 12, a prediction result and an actual measurement value indicating whether or not the communication quality will be equal to or less than the index threshold T. FIG. 3 shows an example of the predicted value and the actual measurement value of communication quality (throughput). In FIG. 3, the predicted value is indicated by a solid line, and the actual measurement value is indicated by a dashed line. Furthermore, in FIG. 3, the index threshold T=150 Mbps is illustrated by a dashed line. The index value calculation unit 15 obtains, at each time, a prediction result (a score that is equal to or less than the index threshold T) and an actual measurement value from the predictive learning unit 12.

[0026] In step S13, the index value calculation unit 15 calculates the Area Under the Curve (AUC) of the Receiver Operating Characteristic (ROC) curve in order to evaluate the accuracy of the prediction value when the actual measurement value is determined to be "positive" when it is equal to or less than the index threshold T. FIG. 4 shows an example of an ROC curve. The horizontal axis represents the false positive rate, and the vertical axis represents the true positive rate. AUC is the area under the ROC curve. The function of the ROC curve of the index threshold T is expressed as f ROC (T), AUC(T) is expressed by the following formula:

[0027]

[0028] AUC(T) is the index value at the index threshold T.

[0029] In step S14, the index value calculation unit 15 determines whether or not to set another index threshold value T.

[0030] When setting another index threshold T, the index value calculation unit 15 returns the process to step S11, sets the next index threshold T, and repeats the process.

[0031] After calculating the index value in the desired range, in step S15, the index value calculation unit 15 calculates the maximum index threshold T max The index threshold T is calculated as the variable T x AUC(T x ) is the index value function f ROC_AUC (T x ) is shown in Figure 5. ROC_AUC (T x 5 shows an example of the index value for each index threshold obtained by the above method. In the example of FIG. 5, the index value is maximum when the index threshold T is 50 Mbps. Therefore, the index value calculation unit 15 calculates the maximum index threshold T max is set to 50Mbps.

[0032] In this embodiment, an ROC curve is used to calculate the index value, but a Precision-Recall (PR) curve may also be used to calculate the index value.

[0033] First Embodiment In a first embodiment, an information processing device 10 is used to propose service requirements suited to a service usage environment.

[0034] We provide services with guaranteed throughput conditions based on a service level agreement (SLA). If the throughput conditions are not met, we detect this in advance and change the base station to which the service user terminal is connected to another base station to continue providing service. However, since the cost of using the new base station is high, we want to reduce the number of base station changes as much as possible while still meeting the throughput conditions.

[0035] The information processing device 10 of this embodiment is used to determine this throughput condition. Specifically, the information processing device 10 determines the maximum index threshold in the service usage environment and proposes the maximum index threshold as the throughput condition. A specific example will be described below.

[0036] An example of a service usage environment is shown in Figure 6. In the service usage environment of Figure 6, a robot holding a target terminal 51 moves along the arrow in the figure. The target terminal 51 communicates with a base station 55 using UDP over the high bandwidth (28 GHz band) of the fifth-generation mobile communication system (5G).

[0037] The information processing device 10 measures the throughput and the position and orientation of the target terminal 51, calculates the index value while gradually changing the throughput condition from 25 Mbps to 275 Mbps, and obtains the maximum index threshold.

[0038] Figure 7 shows an example of the results of calculating the index value while gradually changing the throughput condition from 25 Mbps to 275 Mbps. As shown in the figure, the index value is maximum when the throughput condition is 100 Mbps. Therefore, the information processing device 10 proposes to set the throughput condition to 100 Mbps.

[0039] When the throughput condition was 100Mbps, the frequency of unexpected drops in throughput (false negative rate) was 0.84%, the frequency of unnecessary base station switching (false positive rate) was 11.5%, and the overall error rate was 12.34%.

[0040] On the other hand, when the throughput condition was set to 150Mbps, which is lower than 100Mbps, the frequency of unexpected drops in throughput (false negative rate) was 3.03%. The frequency of unnecessary base station switching (false positive rate) was 9.73%. The overall error rate was 12.76%.

[0041] Fig. 8 shows an example of another service usage environment. In the service usage environment of Fig. 8, a robot holding a target terminal 51 travels along the arrows in the figure, and another robot holding another terminal 52 travels along the arrows in the figure. The target terminal 51 communicates with a base station 55 via UDP using the high bandwidth (28 GHz band) of the fifth-generation mobile communication system (5G).

[0042] The information processing device 10 measures the throughput and the positions and orientations of the target terminal 51 and other terminals 52, calculates the index value while gradually changing the throughput condition from 25 Mbps to 300 Mbps, and obtains the maximum index threshold.

[0043] Figure 9 shows an example of the results of calculating the index value while gradually changing the throughput condition from 25 Mbps to 300 Mbps. As shown in the figure, the index value is maximum when the throughput condition is 50 Mbps. Therefore, the maximum index threshold is 50 Mbps, and it is proposed that the information processing device 10 set the throughput condition to 50 Mbps.

[0044] When the throughput condition was set to 50Mbps, the frequency of unexpected drops in throughput (false negative rate) was 0.07%, the frequency of unnecessary base station switching (false positive rate) was 17.5%, and the overall error rate was 17.57%.

[0045] On the other hand, when the throughput condition was set to 25Mbps, which is lower than the index value of 50Mbps, the frequency of unexpected drops in throughput (false negative rate) was 0.001%. The frequency of unnecessary base station switching (false positive rate) was 33.75%. The overall error rate was 33.751%.

[0046] In this way, the index values ​​can be used to present conditions suitable for the service usage environment when using a prediction model.

[0047] Second Embodiment In a second embodiment, the information processing device 10 is used to select a prediction model that satisfies the requirements of a service provider.

[0048] We provide a video streaming service that requires a throughput of 100Mbps. We predict throughput, detect one second in advance when the requirement will not be met, and reduce the data rate before throughput drops below 100Mbps. We also want to keep unexpected throughput degradation to within 1.5%.

[0049] The information processing device 10 of this embodiment is used to predict throughput. The information processing device 10 holds a plurality of trained prediction models. Specifically, the information processing device 10 calculates an index value for each prediction model using an index threshold of 100 Mbps, and uses the prediction model with the highest index value to predict throughput. A specific example will be described below.

[0050] In the service usage environment, there exists a robot A having a target terminal 51 and a robot B moving around the environment.

[0051] The information processing device 10 holds a prediction model (AB_fromABT) that predicts throughput one second later using the position information and past throughput of robots A and B, and a prediction model (AB_fromAT) that predicts throughput one second later using the position information and past throughput of robot A.

[0052] The information processing device 10 measures the throughput and the positions of robots A and B, and calculates index values ​​for each of the two prediction models (AB_fromABT, AB_fromAB) with an index threshold of 100 Mbps.

[0053] 10 shows an example of the results of calculating index values ​​using each of the two prediction models (AB_fromABT, AB_fromAB). The figure also shows index values ​​for index thresholds other than 100 Mbps.

[0054] The prediction model (AB_fromABT), which uses the location information and past throughput of robots A and B, achieved an index value of 0.903 when the index threshold was set to 100 Mbps. At this time, the frequency of unexpected drops in throughput (false negative rate) was 1.60%.

[0055] The prediction model (AB_fromAT) that uses the location information and past throughput of Robot A achieved an index value of 0.920 when the index threshold was set to 100 Mbps. At this time, the frequency of unexpected drops in throughput (false negative rate) was 1.48%.

[0056] The information processing device 10 selects and uses the prediction model (AB_fromAT) that has a large index value when the index threshold is set to 100 Mbps. The prediction model (AB_fromAT) also satisfies the requirement for the frequency of unexpected throughput drops.

[0057] Furthermore, the information processing device 10 may re-learn the prediction model so that the index value increases.

[0058] In this way, the index values ​​can be used to select a prediction model that is suitable for the service usage environment. While the information processing device 10 is operating to control the target terminal 51, the other terminal 52, or the object 53 in the surrounding space 50 based on the prediction results, the index value calculation unit 15 calculates the index values ​​and selects a prediction model, thereby enabling predictions that follow the dynamically changing environment.

[0059] Third Embodiment In a third embodiment, the surrounding space 50 is controlled so that the index value calculated by the information processing device 10 becomes larger.

[0060] The information processing device 10 calculates an index value using the required conditions of the service 30 as an index threshold.

[0061] To increase the index value, the position of an object 53 existing in the surrounding space 50 is changed, the moving path, speed or direction of the target terminal 51 is changed, or the position of the base station is moved.

[0062] The information processing device 10 may calculate the maximum index threshold and control the surrounding space 50 so that the maximum index threshold becomes larger.

[0063] In this way, the index value can be used to configure the service usage environment so that the accuracy of the prediction model is improved.

[0064] As described above, the information processing device 10 of this embodiment includes an information acquisition unit 11 that acquires status information regarding the status of the target terminal 51, the communication quality of the target terminal 51, and surrounding space information regarding the surrounding space 50 of the target terminal 51; a prediction learning unit 12 that inputs at least one of the status information, the communication quality, and the surrounding space information into a prediction model and predicts whether the communication quality of the target terminal 51 will be equal to or lower than a threshold; and an index value calculation unit 15 that calculates, for each level of communication quality, the area under an ROC curve with the true positive rate and the false positive rate as axes using the prediction result of the prediction model and the actual communication quality, and sets the calculated area as an index value for the level of communication quality. This makes it possible to quantify the evaluation of the prediction model at a desired level of communication quality in the real space, thereby enabling selection of a required level of communication quality, selection of a prediction model to be used, and improvement of the configuration of the real space.

[0065] The information processing device 10 described above can be, for example, a general-purpose computer system including a central processing unit (CPU) 901, a memory 902, a storage 903, a communication device 904, an input device 905, and an output device 906, as shown in Fig. 11. In this computer system, the information processing device 10 is realized by the CPU 901 executing a predetermined program loaded onto the memory 902. This program can be recorded on a computer-readable non-transitory recording medium such as a magnetic disk, an optical disk, or a semiconductor memory, or can be distributed via a network.

[0066] REFERENCE SIGNS LIST 10 Information processing device 11 Information acquisition unit 12 Prediction learning unit 13 Control unit 14 Requirement information acquisition unit 15 Index value calculation unit

Claims

1. An information processing method in which a computer acquires status information relating to the status of a wireless communication terminal, communication quality of the wireless communication terminal, and surrounding space information relating to the space surrounding the wireless communication terminal, inputs at least one of the status information, communication quality, and surrounding space information into a prediction model, solves a classification problem with a predetermined level of communication quality as a boundary for the wireless communication terminal, and calculates, for each level of communication quality, an evaluation index for the classification problem using the prediction result of the prediction model at that level of communication quality and the actual measured value of communication quality, and sets the index value at that level of communication quality.

2. An information processing method according to claim 1, comprising: selecting one of a plurality of prediction models to predict the communication quality of the wireless communication terminal; calculating the index value for each of the plurality of prediction models; and selecting the prediction model with the best index value for the desired communication quality level.

3. An information processing method according to claim 1, further comprising controlling the wireless communication terminal and the space surrounding the wireless communication terminal so as to improve the index value of the desired communication quality level.

4. An information processing device comprising: an acquisition unit that acquires status information regarding the status of a wireless communication terminal, communication quality of the wireless communication terminal, and surrounding space information regarding the space surrounding the wireless communication terminal; a prediction unit that inputs at least one of the status information, the communication quality, and the surrounding space information into a prediction model and solves a classification problem with a predetermined level of communication quality as a boundary; and a calculation unit that, for each level of communication quality, calculates an evaluation index for the classification problem using the prediction result of the prediction model at that level of communication quality and an actual measured value of communication quality, and sets it as an index value at that level of communication quality.

Images