Communication quality prediction device, communication quality prediction method, and program

Abstract

A communication quality prediction device 10 comprises: an acquisition unit 11 that acquires point cloud data about an area 100 in a time series; a preprocessing unit 12 that converts the point cloud data into two-dimensional images; and a prediction unit 13 that predicts, through calculation, the communication quality of a wireless terminal 50 in the area 100 from the two-dimensional images in a time series.

Description

Technical Field

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

Background Art

[0002] The realization of the Internet of Things (IoT) where various devices are connected to the Internet is progressing, and various devices such as automobiles, drones, and construction machinery vehicles are being wirelessly connected. With the increase in Internet, IoT, and Machine to Machine (M2M) traffic, the wireless communication band is becoming congested, and the utilization of higher frequencies is being considered. In next-generation mobile communications (5G, 6G), the realization of high-speed large-capacity communication using frequencies of 30 GHz or higher, called millimeter waves, is expected. On the other hand, communication using high frequencies above the sub-6 GHz band is strongly affected by the surrounding environment. In particular, in millimeter-wave communication and terahertz-wave communication, the communication quality drops sharply due to shielding by the human body or the like. Also, changes in the propagation environment due to the movement of reflectors and Doppler shifts caused by the movement of reflectors are known to affect communication. Such sharp changes in communication quality are factors that significantly degrade the perceived communication quality.

[0003] Since sharp changes in communication quality are factors that significantly degrade the perceived communication quality, it is necessary to detect large changes in communication quality in advance and take countermeasures. In Non-Patent Document 1 of the existing technology, a device that predicts the communication quality when the radio communication path of millimeter-wave communication is shielded due to the passage of an object using physical space information obtained from a depth camera and performs handover control and transmission power control has been proposed. Thus, it has been shown that physical space information is effective for communication quality prediction.

[0004] In recent years, with the development of autonomous driving technology for devices such as robots and cars, the installation of sensors such as Light Detection And Ranging (LiDAR) and cameras has increased in order to recognize the physical space information around the devices.

Prior Art Documents

[0005] [Non-Patent Document 1] T. Nishio, H. Okamoto, K. Nakashima, Y. Koda, K. Yamamoto, M. Morikura, Y. Asai, and R. Miyatake, "Proactive Received Power Prediction Using Machine Learning and Depth Images for mmWave Networks," IEEE Journal on Selected Areas in Communications, vol. 37, no. 11, pp. 2413-2427, Nov. 2019. [Overview of the project] [Problems that the invention aims to solve]

[0006] LiDAR allows for detailed acquisition of three-dimensional spatial information about the surrounding environment. However, the point cloud data obtained by LiDAR is enormous in volume, making it difficult to handle in terms of data storage and processing. Furthermore, machine learning models (deep learning models) for point cloud data are still less mature compared to technologies for handling two-dimensional image data.

[0007] This disclosure is made in view of the above and aims to reduce computational costs in predicting communication quality using physical spatial information. [Means for solving the problem]

[0008] A communication quality prediction device according to one aspect of the present disclosure is a communication quality prediction device for predicting communication quality, comprising: an acquisition unit that acquires point cloud data of a wireless communication area in a time series; a preprocessing unit that converts the point cloud data into a two-dimensional image; and a prediction unit that predicts and calculates the communication quality of wireless terminals in the wireless communication area from the time series of the two-dimensional image. The preprocessing unit sets a range in the height direction from the point cloud data to be used for creating a bird's-eye view, and converts the point cloud data within that height direction range into a bird's-eye view of the wireless communication area as seen from a high position. .

[0009] One aspect of the present disclosure is a communication quality prediction method using a communication quality prediction device, comprising: acquiring point cloud data of a wireless communication area in a time series; converting the point cloud data into a two-dimensional image; and predicting and calculating the communication quality of wireless terminals within the wireless communication area from the time-series two-dimensional image. Furthermore, in the conversion of the point cloud data to a two-dimensional image, a height range is set within the point cloud data to be used for creating a bird's-eye view, and the point cloud data within that height range is converted into a bird's-eye view of the wireless communication area as seen from a high position. . [Effects of the Invention]

[0010] According to this disclosure, computational costs can be reduced in predicting communication quality using physical spatial information. [Brief explanation of the drawing]

[0011] [Figure 1] Figure 1 shows an example of the configuration of the communication quality prediction device according to this embodiment. [Figure 2] Figure 2 shows an example of point cloud data. [Figure 3] Figure 3 shows an example of a bird's-eye view. [Figure 4] Figure 4 is a flowchart showing an example of the process for training a predictive model. [Figure 5] Figure 5 is a flowchart showing an example of the communication quality prediction process. [Figure 6] Figure 6 shows the indoor experimental environment. [Figure 7] Figure 7 shows an example of the hardware configuration of a communication quality prediction device. [Modes for carrying out the invention]

[0012] Embodiments of this disclosure will be described below with reference to the drawings.

[0013] Figure 1 shows an example of the configuration of the communication quality prediction device 10 of this embodiment. The communication quality prediction device 10 is a device that predicts the communication quality of the wireless terminal 50 from physical spatial information obtained by the LiDAR 30 installed in the wireless terminal 50. Communication quality refers to, for example, the throughput and received signal strength (RSSI) of wireless communication between the base station 70 and the wireless terminal 50. The LiDAR 30 may be built into the wireless terminal 50, or the LiDAR 30 may be a separate device from the wireless terminal 50. Area 100 is the range in which the communication quality prediction device 10 predicts the communication quality of the wireless terminal 50. Area 100 may be determined based on the wireless communication area formed by the base station 70. Multiple base stations 70 may be arranged, and a portion of the wireless communication area formed by each base station 70 may overlap. In Figure 1, the solid arrows indicate the data flow during inference, and the dashed arrows indicate the data flow during learning.

[0014] The communication quality prediction device 10 comprises an acquisition unit 11, a preprocessing unit 12, a prediction unit 13, a learning unit 14, and a data storage unit 15.

[0015] The acquisition unit 11 acquires physical spatial information around the wireless terminal 50 in a time series. Specifically, the acquisition unit 11 acquires point cloud data, which is a collection of points in the three-dimensional space around the wireless terminal 50, in a time series from the LiDAR 30 that moves together with the wireless terminal 50, as physical spatial information. The acquisition unit 11 may also place other sensors in addition to the LiDAR 30 and integrate the point cloud data obtained from the LiDAR 30 with sensor data from the other sensors.

[0016] The acquisition unit 11 may acquire location information of the wireless terminal 50 in addition to point cloud data. For example, the acquisition unit 11 communicates with the wireless terminal 50 and receives location information of the wireless terminal 50 from the wireless terminal 50.

[0017] The preprocessing unit 12 converts the physical space information (point cloud data) acquired by the acquisition unit 11 into a 2D image. As a method for converting point cloud data into a 2D image, for example, there are a method of parallel projection so as to overlook the point cloud data from a high viewpoint and convert it into a 2D image (bird's-eye view), and a method of perspective projection so as to view the point cloud data from the position of the LiDAR 30 (wireless terminal 50) and convert it into a 2D image. An example of point cloud data is shown in FIG. 2, and an example of a bird's-eye view converted from the point cloud data is shown in FIG. 3. By preprocessing the physical space information, for example, point cloud data with 30,000 elements and 700 kilobytes can be compressed into a 2D image with 60×45 elements and 3 kilobytes. Hereinafter, converting the physical space information into a 2D image is also referred to as generation or compression.

[0018] When generating a bird's-eye view from the point cloud data, the preprocessing unit 12 generates a bird's-eye view along the local coordinate system of the LiDAR 30 with the position of the LiDAR 30 as the center. For example, when the LiDAR 30 is built into the wireless terminal 50 and the LiDAR 30 moves and rotates together with the wireless terminal 50, the point cloud data is obtained in the local coordinate system of the LiDAR 30. The bird's-eye view generated by the preprocessing unit 12 is an image rotated around the position of the LiDAR 30 according to the movement and rotation of the LiDAR 30. When the LiDAR 30 is fixed, the bird's-eye view generated by the preprocessing unit 12 does not rotate, but the pattern of the bird's-eye view changes corresponding to the objects (including the wireless terminal 50) moving within the area 100. When the LiDAR 30 moves and rotates, the preprocessing unit 12 may convert the point cloud data into the global coordinate system and generate a bird's-eye view from the point cloud data converted into the global coordinate system. In this case, the pattern of the bird's-eye view changes corresponding to the objects moving within the area 100.

[0019] The preprocessing unit 12 may generate an aerial view from the point cloud data that satisfies the condition in the height direction (Z-axis direction) (hereinafter referred to as the Z-axis range condition). The Z-axis range condition is a condition indicating the range of the point cloud data used for generating the aerial view. For example, when using the LiDAR 30 indoors, since the laser light is reflected by the ceiling or the floor, the obtained point cloud data includes points corresponding to the ceiling or the floor. When generating an aerial view from the point cloud data including the ceiling and the floor, it becomes difficult to obtain the characteristics of the interior. Therefore, the preprocessing unit 12 sets the Z-axis range condition of the point cloud data used for generating the aerial view, and generates an aerial view from the point cloud data that satisfies the Z-axis range condition. The Z-axis range condition is set so as not to include the reflected point cloud from the ceiling or the floor according to the field of view of the LiDAR 30 and the distance between the position of the LiDAR 30 and the ceiling or the floor. For example, while changing the Z-axis range condition, generate an aerial view from the point cloud data actually acquired by the LiDAR 30, observe the change in the aerial view, and empirically set the Z-axis range condition. Alternatively, when the LiDAR 30 moves, the preprocessing unit 12 acquires the position information of the LiDAR 30, and sets the Z-axis range condition based on the distance from the position of the LiDAR 30 to the ceiling and the distance from the position of the LiDAR 30 to the floor. When the LiDAR 30 moves on a sloping floor or moves indoors where the height of the ceiling changes, the optimal Z-axis range condition can be set each time.

[0020] The preprocessing unit 12 may integrate the point cloud data obtained by a plurality of point cloud sensors to generate a two-dimensional image. For example, in addition to the LiDAR 30 incorporated in the wireless terminal 50, a fixed point cloud sensor is arranged in the area 100. The preprocessing unit 12 integrates the point cloud data acquired from the LiDAR 30 and the point cloud data acquired from the fixed point cloud sensor, and generates a two-dimensional image from the integrated point cloud data. The sensor data of sensors other than the point cloud data and the point cloud sensor may be integrated.

[0021] The preprocessing unit 12 may integrate the point cloud data with the static information of the area 100 to generate a two-dimensional image. The static information is, for example, a 3D map or a 2D map of the area 100.

[0022] The prediction unit 13 predicts and calculates the future communication quality of the wireless terminal 50 from a time-series two-dimensional image. Specifically, the prediction unit 13 inputs the time-series two-dimensional image converted by the preprocessing unit 12 into a prediction model and predicts and calculates the future communication quality of the wireless terminal 50. An existing machine learning model for image processing can be used as the prediction model. The prediction model is a machine learning model that infers future communication quality when a time-series two-dimensional image is input.

[0023] The prediction unit 13 may also input location information of the base station 70 in the 2D image into the prediction model, in addition to the 2D image. Since the location of the base station 70 within area 100 is known, if the location of the wireless terminal 50 is known, the relative position of the base station 70 as seen from the wireless terminal 50 can be determined, and the location of the base station 70 in the 2D image can be identified.

[0024] The learning unit 14 uses the time-series 2D images and communication quality held by the data storage unit 15 as training data, and learns a predictive model that predicts communication quality when a time-series 2D image is input. In addition to the above training data, the learning unit 14 may also use the location information of the base station 70 as training data.

[0025] The data storage unit 15 stores time-series 2D images and communication quality used for training the prediction model. The 2D images are images converted from physical space information by the preprocessing unit 12. The communication quality is the measured value of the wireless communication quality between the wireless terminal 50 and the base station 70, and is acquired from either the wireless terminal 50 or the base station 70. When collecting training data, the physical space information acquired by the acquisition unit 11 is converted into a 2D image by the preprocessing unit 12, and the 2D images taken at the same time or close in time are associated with the communication quality and stored in the data storage unit 15.

[0026] The data storage unit 15 may store location information of the base station 70 in the two-dimensional image, associated with the two-dimensional image.

[0027] The data storage unit 15 may store the prediction model (parameters) learned in the learning unit 14.

[0028] Next, we will explain an example of the process of training a predictive model, referring to the flowchart in Figure 4.

[0029] In step S11, the acquisition unit 11 acquires point cloud data from the LiDAR 30.

[0030] In step S12, the preprocessing unit 12 converts the point cloud data into a two-dimensional image and stores it in the data storage unit 15. When converting point cloud data into a two-dimensional image by applying a Z-axis range condition, the preprocessing unit 12 extracts point clouds that satisfy the Z-axis range condition from the point cloud data and converts them into a two-dimensional image. When setting the Z-axis range condition according to the position of the LiDAR 30, in step S11, the acquisition unit 11 acquires the position information of the LiDAR 30. The preprocessing unit 12 sets the Z-axis range condition from the position information of the LiDAR 30.

[0031] In step S13, the data storage unit 15 stores the communication quality received from at least one of the wireless terminal 50 and the base station 70, associating it with two-dimensional images taken at the same time or close to it. The data storage unit 15 may also receive the communication quality via the acquisition unit 11.

[0032] When the location information of the base station 70 is used to predict communication quality, the data storage unit 15 acquires the location information of the wireless terminal 50, calculates the location of the base station 70 in a two-dimensional image, and stores the location information of the base station 70 in association with the two-dimensional image.

[0033] In step S14, the learning unit 14 acquires time-series 2D images and communication quality from the data storage unit 15, and uses the time-series 2D images and communication quality as training data to train a prediction model. In addition to the above training data, the learning unit 14 may also use the location information of the base station 70 as training data.

[0034] Next, we will refer to the flowchart in Figure 5 and explain an example of a process for predicting communication quality.

[0035] In step S21, the acquisition unit 11 acquires point cloud data from the LiDAR 30 in a time series. If a Z-axis range condition is set, the acquisition unit 11 acquires the position information of the LiDAR 30. If the position information of the base station 70 is used to predict communication quality, the acquisition unit 11 acquires the position information of the wireless terminal 50 itself from the wireless terminal 50 and determines the relative position information of the base station 70.

[0036] In step S22, the preprocessing unit 12 converts the point cloud data into a two-dimensional image. When converting point cloud data into a two-dimensional image by applying a Z-axis range condition, the preprocessing unit 12 determines the Z-axis range condition from the position information of the LiDAR 30, extracts point clouds that satisfy the Z-axis range condition from the point cloud data, and converts them into a two-dimensional image.

[0037] The preprocessing unit 12 may integrate sensor data from other sensors into the point cloud data, and then convert the point cloud data into a two-dimensional image. Sensor data from other sensors may be, for example, point cloud data obtained from a point cloud sensor fixed in area 100.

[0038] In step S23, the prediction unit 13 inputs a time-series two-dimensional image into the prediction model to predict the future communication quality of the wireless terminal 50. In addition to the time-series two-dimensional image, the prediction unit 13 may also input location information of the base station 70 in the two-dimensional image into the prediction model.

[0039] Next, we will describe the experimental results of the communication quality prediction device 10 of this embodiment in the experimental environment shown in Figure 6.

[0040] The experimental environment shown in Figure 6 is an area 100 of approximately 20m x 6m installed indoors. A wireless terminal 50 equipped with LiDAR 30 randomly moves between locations marked with square marks 110 and triangular marks 120 within area 100. Specifically, the wireless terminal 50 moves according to the arrows, always passing through the locations marked with square marks 110, and skipping the locations marked with triangular marks 120 with a 50% probability.

[0041] The wireless terminal 50 communicates with the base station 70 using the IEEE 802.11ac wireless communication standard. The frequency used for wireless communication is 5.6 GHz, and the bandwidth is 20 MHz. The antenna of the wireless terminal 50 is located 50 cm from the floor. The antenna of the base station 70 is located 70 cm from the floor. The transmission power is 10 dBm. The measurement frequency for RSSI and throughput was set to 100 ms. The acquisition frequency for point cloud data was set to 100 ms.

[0042] Predictive models were trained using Gradient Boosting Decision Tree (GBRT) and Neural Network (NN), respectively, with approximately 80,000 training samples (equivalent to 2 hours) of data, approximately 10,000 validation samples (equivalent to 15 minutes) of data, and approximately 10,000 test samples (equivalent to 15 minutes) of data. The 2D images used for training were bird's-eye views converted from point cloud data obtained by the LiDAR 30 equipped on the wireless terminal 50.

[0043] Table 1 shows the root mean square error (RMSE) when predicting RSSI and throughput after 1 second using a prediction model with GBRT or NN.

[0044] [Table 1]

[0045] Table 1 shows that all prediction models can accurately predict communication quality.

[0046] As described above, the communication quality prediction device 10 of this embodiment, which predicts communication quality, comprises an acquisition unit 11 that acquires point cloud data of area 100 in a time series, a preprocessing unit 12 that converts the point cloud data into a two-dimensional image, and a prediction unit 13 that predicts and calculates the communication quality of wireless terminals 50 within area 100 from the time-series two-dimensional image. By compressing the point cloud data into a two-dimensional image, the computational cost of communication quality prediction can be reduced. Furthermore, since the compressed data is a two-dimensional image, existing image processing algorithms can be applied to the prediction model.

[0047] The communication quality prediction device 10 described above can use, for example, a general-purpose computer system as shown in Figure 7, which includes a central processing unit (CPU) 901, memory 902, storage 903, communication device 904, input device 905, and output device 906. In this computer system, the communication quality prediction device 10 is realized when the CPU 901 executes a predetermined program loaded onto the memory 902. This program can be recorded on a computer-readable non-temporary recording medium such as a magnetic disk, optical disk, or semiconductor memory, or it can be distributed via a network. [Explanation of Symbols]

[0048] 10. Communication quality prediction device 11 Acquisition Department 12 Pre-processing section 13 Prediction Section 14. Learning Department 15 Data Storage Unit 30 LiDAR 50 Wireless terminals 70 base station

Claims

1. A communication quality prediction device that predicts communication quality, An acquisition unit that acquires point cloud data of the wireless communication area in a time series, A preprocessing unit that converts the point cloud data into a two-dimensional image, The system includes a prediction unit that predicts and calculates the communication quality of wireless terminals within the wireless communication area from the time-series two-dimensional image, The preprocessing unit sets a range in the height direction from the point cloud data to be used for creating a bird's-eye view, and converts the point cloud data within that height direction range into a bird's-eye view of the wireless communication area as seen from a high position. Communication quality prediction device.

2. A communication quality prediction device according to claim 1, The prediction unit inputs the location of the base station in the bird's-eye view. Communication quality prediction device.

3. A communication quality prediction device according to claim 1, The acquisition unit acquires the point cloud data from a sensor that moves together with the wireless terminal. Communication quality prediction device.

4. A communication quality prediction device according to claim 1, The acquisition unit acquires sensor data from other sensors, The preprocessing unit integrates the point cloud data and the sensor data to convert them into a two-dimensional image. Communication quality prediction device.

5. A communication quality prediction device for predicting communication quality, An acquisition unit that acquires point cloud data of the wireless communication area in a time series, A preprocessing unit that converts the point cloud data into a bird's-eye view of the wireless communication area from a high position, The system includes a time-series bird's-eye view and a prediction unit that predicts and calculates the communication quality of wireless terminals within the wireless communication area based on the location of base stations in the bird's-eye view. Communication quality prediction device.

6. A method for predicting communication quality using a communication quality prediction device, Point cloud data of the wireless communication area is acquired in time series. The aforementioned point cloud data is converted into a two-dimensional image, The communication quality of wireless terminals within the wireless communication area is predicted and calculated from the time-series two-dimensional image. In the conversion of the point cloud data to a two-dimensional image, a height range is set from the point cloud data to be used for creating a bird's-eye view, and the point cloud data within that height range is converted into a bird's-eye view of the wireless communication area as seen from a high position. A method for predicting communication quality.

7. A program for operating a computer as each part of the communication quality prediction device according to any one of claims 1 to 5.

Images