Wireless communication device and communication quality prediction method

Abstract

A wireless communication device 1 comprises: a wireless communication unit 11 that performs wireless communication with a communication terminal 2 and acquires wireless communication information including wireless communication quality; a spatial information acquisition unit 12 that acquires spatial information of the communication terminal 2 and a surrounding object; a data generation unit 13 that generates training data from the wireless communication information and the spatial information; a quality condition input unit 14 that inputs a quality condition for which prediction is important; a model generation unit 15 that performs machine learning by using the training data and the quality condition at the time of learning, and constructs a prediction model for predicting the wireless communication quality by receiving, as an input, the spatial information or the spatial information and the wireless communication information; and a utilization unit 17 that predicts the wireless communication quality of the communication terminal 2 by inputting the spatial information or the spatial information and the wireless communication information to the prediction model at the time of inference.

Description

Wireless communication device and communication quality prediction method

[0001] The present disclosure relates to a wireless communication device and a communication quality prediction method.

[0002] When using a device (communication terminal) equipped with wireless communication functions, the communication quality may change due to changes in the surrounding environment (such as the movement of surrounding objects), which may result in the communication quality not being able to meet the requirements of the services and systems of the communication terminal.

[0003] In particular, in the case of IEEE 802.11ad and 5G cellular communications, which use millimeter wave bands, quality is significantly reduced due to obstructions between the transmitter and receiver of wireless communications, so the impact of the surrounding environment on wireless communication quality is significant. In addition, Doppler shift caused by moving reflective objects is also known to affect communications.

[0004] By predicting wireless communication quality in advance, it may be possible to take measures before services and systems are affected. For example, in high-capacity wireless communication services such as Extended Reality (XR), transmission bit rate control and buffer control can be used to provide stable services. Wireless communication quality refers to the communication capacity and throughput required by applications.

[0005] However, when considering data collection, preprocessing, prediction of wireless communication quality using a prediction model, and even control of applications and wireless systems, predictions for a longer time horizon are required. For wireless communication quality predictions for 0.5 seconds or more in the future, the effectiveness of wireless communication quality prediction methods using spatial information such as the position, speed, and direction of communication terminals has been confirmed.

[0006] Known methods for predicting wireless communication quality include the techniques described in Non-Patent Documents 1 to 3. Non-Patent Document 1 discloses a technique for predicting wireless communication quality using spatial information obtained from a camera or LiDAR in a situation where the surrounding environment, such as obstructions, is fixed and a robot carrying a communication terminal is moving. Non-Patent Documents 2 and 3 disclose techniques for predicting wireless communication quality using spatial information obtained from a depth camera or LiDAR in a situation where the communication terminal and Access Point (AP) are fixed and two users are walking so as to shield the communication terminal and AP.

[0007] T. Zhang, J. Liu, and F. Gao, “Vision Aided Beam Tracking and Frequency Handoff for mmWave Communications,” Proc. IEEE Conf. Comput. Commun. Workshops, pp. 1-2, May 2022. T. Nishio et al., “Proactive Received Power Prediction Using Machine Learning and Depth Images for mmWave Networks,” in IEEE Journal on Selected Areas in Communications, vol. 37, no. 11, pp. 2413-2427, Nov. 2019S. Ohta, T. Nishio, R. Kudo, and K. Takahashi, “Millimeter-wave Received Power Prediction Using Point Cloud Data and Supervised Learning,” Proc. IEEE 95th Veh. Technol. Conf., pp. 1-5, Jun. 2022.

[0008] Conventional technologies do not evaluate quality conditions where prediction is important, such as sudden drops or rises in wireless communication quality. Therefore, even if prediction accuracy appears high using indicators such as the mean absolute error, the accuracy may be low for quality conditions where prediction is important. In particular, in high-frequency bands such as millimeter waves, quality fluctuations are large, and evaluation of specific quality conditions, such as points where quality drops, may be important.

[0009] The present disclosure has been made in consideration of the above, and aims to realize evaluation and improvement of accuracy in wireless communication quality prediction corresponding to use cases.

[0010] A wireless communication device of one aspect of the present disclosure includes a wireless communication unit that performs wireless communication with a communication terminal and acquires wireless communication information including wireless communication quality; a spatial information acquisition unit that acquires spatial information of the communication terminal and surrounding objects present around the communication terminal; a data generation unit that generates training data from the wireless communication information and the spatial information; a quality condition input unit that inputs quality conditions for which prediction is important; a model generation unit that, during learning, performs machine learning using the training data and the quality conditions and constructs a prediction model that predicts wireless communication quality by inputting the spatial information or the spatial information and the wireless communication information; and a utilization unit that, during inference, inputs the spatial information or the spatial information and the wireless communication information into the prediction model to predict the wireless communication quality of the communication terminal.

[0011] According to the present disclosure, it is possible to realize evaluation and accuracy improvement in wireless communication quality prediction corresponding to use cases.

[0012] FIG. 1 is a diagram showing an example of the configuration of a wireless communication device of this embodiment. FIG. 2 is a flowchart showing an example of the processing flow during learning of the wireless communication device. FIG. 3 is a flowchart showing an example of the processing flow during inference of the wireless communication device. FIG. 4 is a diagram showing an environment of an example. FIG. 5 is a graph showing the true positive rate of an example. FIG. 6 is a graph showing the true negative rate of an example. FIG. 7 is a diagram showing an example of the hardware configuration of a wireless communication device.

[0013] [Configuration of Wireless Communication Device] An example of the configuration of a wireless communication device 1 according to this embodiment will be described with reference to Fig. 1. The wireless communication device 1 shown in the figure includes a wireless communication unit 11, a spatial information acquisition unit 12, a data generation unit 13, a quality condition input unit 14, a model generation unit 15, a model selection unit 16, and a utilization unit 17. The dashed arrows in the figure indicate the flow of data during learning.

[0014] The wireless communication unit 11 performs wireless communication with the communication terminal 2 and acquires wireless communication information. For example, the wireless communication information includes the throughput of the communication terminal 2, received signal power, signal-to-noise power ratio, signal-to-interference plus noise power ratio, Received Signal Strength Indication (RSSI), Reference Signal Received Quality (RSRQ), packet error rate, number of arriving bits, bit error rate, number of arriving bits per unit time, Modular Code Index (MCS), number of retransmissions, packet arrival delay time, error correction technology settings, communication terminal user contract information, differential information of these values, and communication system setting items that affect the index calculated from these values ​​using a calculation formula. The wireless communication information includes at least the wireless communication quality predicted by the prediction model.

[0015] The communication terminal 2 is, for example, a mobile device such as a smartphone that can be carried by a user. Although only one communication terminal 2 is illustrated in Fig. 1, the wireless communication device 1 can communicate with a plurality of communication terminals 2.

[0016] The spatial information acquisition unit 12 acquires spatial information within a predetermined area in which the communication terminal 2 moves (for example, the periphery of the communication terminal 2). The spatial information is information such as the position, speed, and direction of the communication terminal 2 and moving or stationary objects (hereinafter referred to as peripheral objects) within the predetermined area. For example, the spatial information acquisition unit 12 acquires spatial information by receiving, from the communication terminal 2, information acquired by a GPS or sensor included in the communication terminal 2, acquiring image information acquired by an externally placed camera, or acquiring point cloud information from a LiDAR.

[0017] During learning, the data generation unit 13 generates training data from spatial information and wireless communication information. The training data is data used when constructing a prediction model by machine learning to predict the wireless communication quality of the communication terminal 2 in the near future (e.g., 0.5 seconds from now) or at the destination of the communication terminal 2. For example, the training data is data indicating the relationship between the wireless communication quality and spatial information such as the position, speed, and direction of the communication terminal 2 and surrounding objects. The training data may include wireless communication information and may be data indicating the relationship between the spatial information and wireless communication information and the wireless communication quality. In other words, the training data is data used in the machine learning of a prediction model to predict wireless communication quality from the positions and movements of the communication terminal 2 and surrounding objects, and the state and fluctuations of the wireless communication information.

[0018] During inference, the data generation unit 13 may generate input data to be input to the prediction model from the spatial information and wireless communication information.

[0019] The quality condition input unit 14 accepts input of quality conditions for which prediction is important from the user. Quality conditions for which prediction is important are conditions indicating events or fluctuations that require prediction in each use case, such as quality falling below a reference value or a sudden drop in quality. For example, for a use case requiring a bandwidth of 100 Mbps or more in a 4K high-resolution virtual reality (VR) application, quality conditions such as "predict with a true positive rate of 90% or more for less than 100 Mbps (positive)," "predict with a true positive rate of 90% or more for less than 100 Mbps (positive), including points where throughput drops suddenly by 40% or more," or "minimize the absolute mean error for points where throughput drops suddenly by 40% or more" are input. The quality conditions may include performance requirements to be met for the quality conditions and goals for improving accuracy.

[0020] The model generation unit 15 inputs training data and quality conditions, and constructs a prediction model by machine learning the relationship between spatial information and wireless communication quality or the relationship between spatial information, wireless communication information, and wireless communication quality while optimally tuning for the quality conditions. For example, the model generation unit 15 uses the training data to construct a prediction model that performs two-class classification based on the input spatial information and determines whether the wireless communication quality of the communication terminal 2 will be below a predetermined threshold in the near future. When constructing the prediction model, the model generation unit 15 performs optimal tuning for the quality conditions, such as adjusting a cutoff value in the two-class classification based on the quality conditions, adjusting the amount of training data that matches the quality conditions (e.g., by focusing learning on training data that matches the quality conditions), and adjusting the weighting of the loss function in learning for items that match the quality conditions (e.g., by increasing the weighting to focus learning on items that match the quality conditions).

[0021] The model generation unit 15 constructs a plurality of prediction models for each quality condition, and transmits the quality conditions and the constructed prediction models to the model selection unit 16.

[0022] The model selection unit 16 holds at least one prediction model generated by the model generation unit 15 and the quality conditions corresponding to that prediction model, and selects the prediction model that best matches the quality conditions input from the quality condition input unit 14. For example, the model selection unit 16 uses validation data to evaluate each prediction model against the quality conditions input from the quality condition input unit 14, and selects a prediction model that satisfies the performance requirements.

[0023] The utilization unit 17 inputs the spatial information or the spatial information and the wireless communication information into the prediction model selected by the model selection unit 16, and predicts the future wireless communication quality of the communication terminal 2. Based on the predicted wireless communication quality, the utilization unit 17 may take preventive measures before the wireless communication quality deteriorates. For example, the utilization unit 17 performs bit rate control and buffer control in a video transmission application, or performs handover and resource control. If peripheral objects can be controlled, the utilization unit 17 may control the peripheral objects so as to suppress deterioration of the wireless communication quality based on the predicted wireless communication quality.

[0024] [Operation of Wireless Communication Device] An example of the operation of the wireless communication device 1 will be described with reference to the flowcharts of FIGS.

[0025] First, an example of the operation of the wireless communication device 1 during learning will be described with reference to the flowchart of FIG.

[0026] In step S11, the wireless communication unit 11 performs wireless communication with the communication terminal 2 and acquires wireless communication information including wireless communication quality. As described above, the wireless communication quality is, for example, throughput, RSSI, RSRQ, and the like.

[0027] In step S12, the spatial information acquisition unit 12 acquires spatial information within a predetermined area around the communication terminal 2 from a sensor such as a camera or LiDAR. The spatial information acquisition unit 12 may acquire spatial information such as the position, speed, and direction of the communication terminal 2 acquired by a sensor of the communication terminal 2 from the communication terminal 2.

[0028] In step S13, the quality condition input unit 14 inputs quality conditions from the user.

[0029] The order of steps S11 to S13 does not matter. Step S13 may be performed in advance, or step S13 may be performed before generating a prediction model in step S15 (described later). Steps S11 and S12 may be performed in parallel, or steps S12 and S11 may be performed in that order.

[0030] In step S14, the data generating unit 13 generates training data from the spatial information and the wireless communication information.

[0031] In step S15, the model generation unit 15 inputs training data and quality conditions, and while optimally tuning the quality conditions, performs machine learning on the relationship between spatial information and wireless communication quality or the relationship between spatial information and wireless communication information and wireless communication quality to construct a prediction model. When multiple quality conditions are input, the model generation unit 15 constructs a prediction model for each of the multiple quality conditions.

[0032] Next, an example of the operation of the wireless communication device 1 during inference will be described with reference to the flowchart of FIG.

[0033] 2, in steps S21 to S23, the wireless communication unit 11 acquires wireless communication information, the spatial information acquisition unit 12 acquires spatial information, and the quality condition input unit 14 inputs quality conditions and performance requirements. During inference, the wireless communication device 1 may acquire only spatial information.

[0034] In step S24, the model selection unit 16 selects a prediction model according to the quality conditions.

[0035] In step S25, the utilization unit 17 inputs the spatial information or the spatial information and the wireless communication information into the selected prediction model to predict the wireless communication quality. The utilization unit 17 may control the communication terminal 2 or a peripheral object based on the prediction result. Note that the wireless communication device 1 may learn using the spatial information and the wireless communication information acquired during inference as training data.

[0036] [Example] Next, with reference to Figs. 4 to 6, an example of learning a prediction model for predicting the wireless communication quality (throughput in this case) of the communication terminal 2 will be described.

[0037] 4 shows an environment of the embodiment. Within the area of ​​the environment of the embodiment shown in the figure, a person carrying a communication terminal 2 moves along the arrows between points A, B, and C in the figure. The communication terminal 2 communicates with a base station 3.

[0038] Two cameras 4 are placed within the area to capture images within the area. The wireless communication device 1 receives images from the cameras 4 and acquires the position, speed, and direction of a person (communication terminal 2) as spatial information. The wireless communication device 1 also acquires wireless communication information including throughput from the communication terminal 2. The wireless communication device 1 may also acquire wireless communication information from a base station 3.

[0039] The quality condition "predict with a true positive rate of 90% or more where throughput will suddenly drop by 40% or more and become less than 100 Mbps (positive) from the current value" is input to the wireless communication device 1.

[0040] The wireless communication device 1 generates training data from spatial information and wireless communication information. The wireless communication device 1 uses the training data and quality conditions to construct a prediction model while optimally tuning the quality conditions. Specifically, when the wireless communication device 1 inputs spatial information, it constructs a prediction model that performs two-class classification, outputting a positive result when the wireless communication quality of the communication terminal 2 at the destination (future) is less than 100 Mbps and a negative result when the wireless communication quality is 100 Mbps or higher. During tuning, the cutoff value is adjusted so that a true positive rate of 90% is achieved for locations where throughput drops sharply by 40% or more.

[0041] Figures 5 and 6 show the true positive and true negative rates for the points where the rate of change in throughput is 0 to less than 10, 10 to less than 20, 20 to less than 30, 30 to less than 40, and 40 or greater when the cutoff value is changed from 100 Mbps to 200 Mbps in a prediction model that predicts points where throughput will fall below 100 Mbps. The dashed triangle lines in the figures (the [-inf, inf] lines) represent the average values ​​for all data. By adjusting the cutoff value of the prediction model to 146 Mbps, we were able to predict points where throughput drops sharply by 40% or more with a true positive rate of 90% or greater. Furthermore, by increasing the weighting in the loss function for data where throughput drops sharply by 40% or greater during training, we were able to improve the true negative rate to 92% without changing the true positive rate.

[0042] Previously, evaluations were limited to the average value of all data, as shown by the dashed triangular lines in Figures 5 and 6. Even when the cutoff value was set to 100 Mbps, the dashed triangular line appears to be highly accurate, with a true positive rate of over 88%, but where throughput suddenly drops by 40% or more, the true positive rate is only 42%, indicating a drop in accuracy. Such sudden drops in throughput occur in high-frequency bands such as millimeter waves, making it important to make highly accurate predictions for these.

[0043] As described above, the wireless communication device 1 of this embodiment includes a wireless communication unit 11 that performs wireless communication with the communication terminal 2 and acquires wireless communication information including wireless communication quality; a spatial information acquisition unit 12 that acquires spatial information about the communication terminal 2 and surrounding objects; a data generation unit 13 that generates training data from the wireless communication information and the spatial information; a quality condition input unit 14 that inputs quality conditions for which prediction is important; a model generation unit 15 that, during learning, performs machine learning using the training data and the quality conditions and constructs a prediction model that predicts wireless communication quality by inputting the spatial information or the spatial information and the wireless communication information; and a utilization unit 17 that, during inference, inputs the spatial information or the spatial information and the wireless communication information into the prediction model and predicts the wireless communication quality of the communication terminal 2. This enables highly accurate predictions for quality conditions according to the applicable use case. Since appropriate predictions can be made for each use case, service quality can be significantly improved by applications and wireless system control based on the predictions.

[0044] For example, in use cases where VR goggles are used to remotely support tasks such as assembly and inspection in indoor factories, communication quality can fluctuate depending on the presence and movement of people and objects in the vicinity. Using the technology disclosed herein, it is possible to predict the wireless communication quality of the VR goggles' wireless communication terminal and notify the user and the system. By predicting communication quality and proactively increasing the video resolution and frame rate when the quality is high and decreasing them when the quality is low, the service can be used stably.

[0045] For example, in use cases where an unspecified number of people enter and exit venues such as event venues or conference rooms, where the quality of communication may fluctuate depending on the presence and movement of people in the vicinity, the technology disclosed herein can be used to predict future fluctuations in the communication quality of a user's wireless terminal and notify the user or system. By predicting communication quality, for example, when the quality is low, it becomes possible to ensure stable communication quality by controlling the switching of access points or base stations and by using beamforming.

[0046] For example, in use cases where AMRs (Autonomous Mobile Robots) move back and forth between logistics centers and warehouses, the technology disclosed herein can be used to predict future fluctuations in the communication quality of wireless devices carried by workers and robots, and notify users and systems. By predicting communication quality and efficiently transferring heavy data such as video and log data acquired by the robot to a server when the quality is high, power consumption can be reduced, and by proactive control such as not transferring such data when the quality is low, while ensuring essential communications such as sending and receiving emergency stop signals to the robot, robotized logistics centers and warehouses can be used stably.

[0047] By providing a means for humans or systems to proactively respond (control) to these notifications, it may be possible to avoid problems caused by deterioration in wireless communication quality.

[0048] Furthermore, the prediction targets may include wireless communication quality such as data transfer volume, delay, jitter, packet loss rate, and received signal strength in addition to throughput.

[0049] The wireless communication device 1 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. 7. In this computer system, the wireless communication device 1 is realized by the CPU 901 executing a predetermined program loaded onto the memory 902. This program can be recorded on a non-transitory computer-readable recording medium such as a magnetic disk, an optical disk, or a semiconductor memory, or can be distributed via a network.

[0050] REFERENCE SIGNS LIST 1 wireless communication device 11 wireless communication unit 12 spatial information acquisition unit 13 data generation unit 14 quality condition input unit 15 model generation unit 16 model selection unit 17 utilization unit 2 communication terminal 3 base station 4 camera

Claims

1. A wireless communication device comprising: a wireless communication unit that performs wireless communication with a communication terminal and acquires wireless communication information including wireless communication quality; a spatial information acquisition unit that acquires spatial information of the communication terminal and surrounding objects present around the communication terminal; a data generation unit that generates training data from the wireless communication information and the spatial information; a quality condition input unit that inputs quality conditions for which prediction is important; a model generation unit that, during learning, performs machine learning using the training data and the quality conditions and constructs a prediction model that predicts wireless communication quality by inputting the spatial information or the spatial information and the wireless communication information; and a utilization unit that, during inference, inputs the spatial information or the spatial information and the wireless communication information into the prediction model and predicts the wireless communication quality of the communication terminal.

2. A wireless communication device according to claim 1, comprising a model selection unit that holds a prediction model for each quality condition and selects a prediction model that matches the input quality condition during inference.

3. A wireless communication device according to claim 1, wherein the quality condition is a condition under which wireless communication quality falls below a reference value or a condition under which wireless communication quality suddenly drops, and the model generation unit constructs the prediction model so as to satisfy predetermined performance requirements for the quality condition.

4. A communication quality prediction method for a wireless communication device, comprising: performing wireless communication with a communication terminal, acquiring wireless communication information including wireless communication quality; acquiring spatial information about the communication terminal and surrounding objects around the communication terminal; generating training data from the wireless communication information and the spatial information; inputting quality conditions for which prediction is important; during learning, performing machine learning using the training data and the quality conditions, and constructing a prediction model that predicts wireless communication quality by inputting the spatial information or the spatial information and the wireless communication information; and during inference, inputting the spatial information or the spatial information and the wireless communication information into the prediction model to predict the communication quality of the communication terminal.

Images