JP7653684B2 - Communication quality prediction device, communication quality prediction method, and communication quality prediction program - Google Patents

Description

The present invention relates to a communication quality prediction device, a communication quality prediction method, and a communication quality prediction program.

Next-generation mobile communications, such as 5G and 6G, are expected to realize high-speed, large-capacity wireless communications using frequencies above 30 GHz known as millimeter waves. However, wireless communications using high frequencies above the sub-6 GHz band are strongly affected by the surrounding environment. In particular, in millimeter wave and terahertz wave communications, communication quality drops sharply due to obstruction of the wireless communication path by the human body, etc. (Non-Patent Document 1).

Therefore, a control device is known that predicts the future communication quality of a wireless communication channel using images captured by a camera, and performs transmission power control and handover for a wireless communication terminal based on the prediction results (Patent Document 1, Non-Patent Document 2).

JP 2018-148297 A

S. Collonge and two others, “Influence of the human activity on wide-band characteristics of the 60 GHz in-door radio channel”, IEEE Trans. Wireless Commun., vol.3, no.6, November 2004, p.2396-p.2406 T. Nishio and 7 others, “Proactive Received Power Prediction Using Machine Learning and Depth Images for mmWave Networks”, IEEE Journal on Selected Areas in Communications, vol. 37, no. 11, November 2019, p.2413-p.2427

However, in Patent Document 1 and Non-Patent Document 2, because communication quality is predicted using camera images, it may be difficult to install cameras and use images, especially on the streets, in offices, and in public facilities, from the perspective of protecting privacy and confidential information.

The present invention has been made in consideration of the above circumstances, and the object of the present invention is to provide a technology that can predict future communication quality in various wireless communication areas without being restricted by the need to protect privacy, confidential information, etc.

A communication quality prediction device according to one aspect of the present invention is a communication quality prediction device that predicts future communication quality, and includes a collection unit that acquires environmental information about the vicinity of a wireless communication path in time series as point cloud data, a preprocessing unit that generates processed point cloud data by performing processing to reduce the amount of data in the point cloud data, a conversion unit that converts the processed point cloud data of the environmental information into environmental information in a coordinate system that can be recognized by the communication quality prediction device, and a prediction unit that predicts and calculates the future communication quality of the wireless communication path based on changes over time in the environmental information in the coordinate system.

A communication quality prediction method according to one aspect of the present invention is a communication quality prediction method for predicting future communication quality, in which a communication quality prediction device performs the steps of acquiring environmental information around a wireless communication path in the form of point cloud data in a time series, generating processed point cloud data by performing processing to reduce the amount of data of the point cloud data, converting the processed point cloud data of the environmental information into environmental information in a coordinate system that can be recognized by the communication quality prediction device, and predicting and calculating the future communication quality of the wireless communication path based on changes over time in the environmental information in the coordinate system.

A communication quality prediction program according to one aspect of the present invention causes a computer to function as the above-mentioned communication quality prediction device.

The present invention provides technology that can predict future communication quality in various wireless communication areas without being restricted by the need to protect privacy or confidential information.

FIG. 1 is a diagram showing a functional block configuration of a communication quality prediction device. FIG. 2 is a diagram illustrating a process flow of the communication quality predicting device. FIG. 3 is a diagram showing the results of prediction of communication quality. FIG. 4 is a diagram showing the predicted results of communication quality. FIG. 5 is a diagram illustrating a hardware configuration of the communication quality prediction device.

Below, an embodiment of the present invention will be described with reference to the drawings. In the description of the drawings, the same parts are given the same reference numerals and the description will be omitted.

Summary of the Invention
The present invention discloses a communication quality prediction device that predicts the future communication quality of a wireless communication path using point cloud data of a wireless communication area acquired by a sensor, not a camera or an image. Since point cloud data does not include privacy information or confidential information compared to images, there are fewer restrictions on the use of the data, and sensors can be installed in various environments.

However, point cloud data has a larger amount of data than images, making it difficult to use it as is. Therefore, the present invention performs noise removal, data volume reduction, and feature selection on the time-series point cloud data acquired by the sensor, and converts it into an appropriate format. After that, a communication quality prediction model is trained using machine learning to predict the future communication quality of the wireless communication channel.

As a result, it is possible to predict future communication quality in high-frequency wireless communication in various wireless communication areas. The present invention is characterized by its use of point cloud data. Therefore, it can also be applied to point cloud data obtained by converting depth information or video information, without using equipment that directly generates point cloud data, such as a camera or sensor.

[Configuration of communication quality prediction device]
1 is a diagram showing a functional block configuration of a communication quality prediction device 1 according to the present embodiment. The communication quality prediction device 1 includes a plurality of collection units 1-1-1 to 1-1-R, a preprocessing unit 1-2, a conversion unit 1-3, a generation unit 1-4, and a prediction unit 1-5.

The multiple collection units 1-1-1 to 1-1-R each have a function of acquiring point cloud data of environmental information in a time series from a sensor that acquires environmental information around the wireless communication path as point cloud data over time. The sensor is LiDAR (Light Detection And Ranging) or the like. Environmental information includes dynamic objects such as moving people and cars, growing plants, buildings under construction, rivers with fluctuating water levels, etc., and static objects such as benches, traffic lights, etc.

The pre-processing unit 1-2 has a function of generating processed point cloud data by performing processing to reduce the amount of data of the point cloud data. For example, the pre-processing unit 1-2 performs both outlier removal processing, which reduces the number of point cloud data by removing noise points that are outliers from the point cloud data, and downsampling processing, which reduces the number of point cloud data by lowering the spatial resolution of the point cloud data.

The conversion unit 1-3 has a function of converting the processed point cloud data of the environmental information into environmental information in a coordinate system that can be recognized by the communication quality prediction device 1. The point cloud data obtained by the LiDAR is reflection point information of light seen from the LiDAR, and needs to be converted into information in a coordinate system that the prediction unit 1-5 understands. Therefore, the conversion unit 1-3 converts the point cloud data measured by the LiDAR into information that can be recognized by the communication quality prediction device 1. Specifically, based on the position and orientation of the LiDAR, the point cloud data of the environmental information after data volume reduction is converted into environmental information having a predetermined coordinate system. For example, conversion processing to a three-dimensional coordinate system, conversion processing to a two-dimensional coordinate system, and conversion processing to a voxel are performed. Here, a voxel is a data format that represents a solid, and by giving a value (often 0, 1) to a cube that divides a three-dimensional space into a grid, it represents whether or not an object exists in a small area of the cube.

The generation unit 1-4 has a function of generating data and information other than point cloud data as auxiliary information to be input to the communication quality prediction model. For example, the generation unit 1-4 generates various communication information related to communications around the wireless communication path, temperature, air temperature, setting information of wireless communication terminals of users who are intruding or have intruded into the wireless communication path, setting information of wireless communication systems such as base stations, information related to the state of the wireless communication system, etc. as auxiliary information.

The prediction unit 1-5 has a function of predicting and calculating the future communication quality of the wireless communication channel based on the change over time in the environmental information of the above coordinate system. For example, the prediction unit 1-5 inputs the environmental information of the transformed coordinate system into a communication quality prediction model, and also inputs auxiliary information as necessary, and performs machine learning to predict and calculate communication quality data.

The communication quality prediction model is a machine learning model that predicts and calculates the communication quality of wireless communication in a wireless communication area based on environmental information and auxiliary information in the coordinate system of the wireless communication area. The communication quality prediction model may be built in the communication quality prediction device 1 or in another device.

The communication quality data includes, for example, received signal power, signal-to-noise power ratio (SNR), signal-to-interference-plus-noise power ratio (SINR), received signal strength indication (RSSI), received signal reference quality (RSRQ), packet error rate, number of arriving bits, number of arriving bits per unit time, modulation and coding scheme index (MCS), number of retransmissions, delay time, error correction method, frequency of the communication system, frequency conditions such as bandwidth of the resources used, etc. It may also be information on the differentiation of these values, an index calculated by substituting these values into a predetermined formula, a system setting item that affects these indexes, etc.

[Operation of communication quality prediction device]
2 is a diagram showing a processing flow of the communication quality predicting device 1. The communication quality predicting device 1 predicts future communication quality in a wireless communication area such as an office district.

Step S1:
First, the multiple collection units 1-1-1 to 1-1-R acquire point cloud data of environmental information (e.g., people, vehicles) around the wireless communication path propagating in the office district in a time series manner from a LiDAR installed in the office district.

Step S2:
Next, the preprocessing unit 1-2 performs outlier removal processing, downsampling processing, or both processing on the point cloud data acquired from the LiDAR, thereby generating processed point cloud data in which the data amount has been reduced.

Outlier removal processing is a process that removes noise points that are included in point cloud data when measuring with LiDAR. Removing noise points enables the prediction unit 1-5 to predict future communication quality more accurately. Methods for removing outliers include calculating the distance between a certain point and a certain number of neighboring points and removing neighboring points whose distance is greater than a threshold, and removing points in a sphere of a specified size centered on the point that contains fewer points.

Other methods of removing outliers include deleting point clouds other than the point cloud in the three-dimensional space formed between the wireless communication terminal and the base station, removing point clouds related to static structures known in advance in the three-dimensional space, and removing point clouds that do not exist continuously over time. For example, a method of removing point clouds other than the point cloud in the three-dimensional space formed between the wireless communication terminal and the base station includes a method of removing ranges other than the range within an angle φ from the base station when looking in the direction of the base station from the wireless communication terminal, or the range within an angle φ from the wireless communication terminal when looking from the base station, or both ranges, as conditions between the base station and the wireless communication terminal.

Downsampling is a process that reduces the number of input dimensions to be input to the communication quality prediction model by lowering the spatial resolution and reducing the total number of points in the point cloud data. This makes it possible to suppress the amount of calculations in the machine learning performed by the prediction unit 1-5. Downsampling methods include dividing the space containing the point cloud data into voxels of a certain size and narrowing down the multiple points present in the voxel to only one representative point, and thinning out the point data uniformly at regular intervals in the coordinate system.

Step S3:
Next, the conversion unit 1-3 converts the processed point cloud data into two-dimensional coordinates, three-dimensional coordinates, or voxels, which are formats that can be recognized by the communication quality prediction device 1 and can be input to the communication quality prediction model.

One method of conversion to a two-dimensional plane is to determine a specific two-dimensional plane from the six two-dimensional planes that form the three-dimensional space, and map the point cloud data in the three-dimensional space onto that two-dimensional plane. This makes it possible to use position information in the two-dimensional space, and reduce the amount of point cloud data. Alternatively, only information on intersections with any two-dimensional plane in the three-dimensional space may be extracted. When extracting intersection information with a certain two-dimensional plane, multiple two-dimensional spaces may be prepared. For example, two-dimensional planes with heights of 30 cm and 1 m may be used, and point cloud data information that exists in the three-dimensional coordinate system at those heights may be extracted.

In addition, to further reduce the amount of point cloud data, the conversion unit 1-3 may divide the three-dimensional space into voxel units of a predetermined size and convert the point cloud data into voxels (voxel data conversion). Methods for converting to voxels include a method of setting a voxel to 1 if a point of point cloud data exists in the voxel and 0 if no point exists in the voxel, and a method of setting the number of points in the voxel as the density and the voxel value. This makes it possible to reduce the amount of point cloud data. Using some kind of auxiliary information, values that have a strong effect on wireless communication can be set to 1, and values that have a small effect can be set to a value smaller than 1, thereby further improving the prediction accuracy of wireless communication.

Step S4:
Next, the generating unit 1-4 generates various communication information related to communication around the wireless communication path, temperature, air temperature, setting information of the wireless communication terminal of the user who is intruding into the wireless communication path, setting information of the wireless communication system such as a base station, information on the state of the wireless communication system, etc. as auxiliary information to be input to the communication quality prediction model. Note that the auxiliary information is secondary information for improving the prediction accuracy of the communication quality. Therefore, step S4 may be skipped so as not to use the auxiliary information.

Step S5:
Finally, the prediction unit 1-5 inputs the position information in the above-mentioned three-dimensional coordinate system (or the position information in the two-dimensional coordinate system, or the voxel data in the three-dimensional coordinate system) into a communication quality prediction model, and also inputs auxiliary information into the communication quality prediction model as necessary, and performs machine learning to predict and calculate communication quality data such as RSSI.

The machine learning process is almost the same as the learning method in Non-Patent Document 2, except that voxel data is used instead of a depth image. Below, we will explain the case where point cloud data is converted to voxel data in step S3.

First, the communication quality data measured simultaneously with the point cloud data is time-correlated with the voxel data and labeled. At this time, in order to predict future communication quality, "temporal difference labeling" (see Non-Patent Document 2), which shifts the time for labeling, is applied.

Any algorithm can be used as the machine learning algorithm. Neural networks, random forests, gradient boosting decision trees, etc. can be used.

In a neural network, by incorporating a convolutional layer, it is possible to use voxel data, which is 3D data, as input directly. Convolution extracts spatiotemporal features, making it possible to perform predictive calculations of communication quality.

On the other hand, random forests, gradient boosting decision trees, etc. do not support multidimensional inputs, so voxels cannot be used as inputs as is. Therefore, voxels are flattened into a one-dimensional array and used as features. A machine learning algorithm is run using voxels labeled with communication quality values (or a one-dimensional array flattened version of them) as a training dataset to obtain a communication quality prediction model. By inputting voxels generated from point cloud data into the obtained communication quality prediction model, future communication quality values can be predicted and calculated.

When the point cloud data is converted into position information in a two-dimensional coordinate system, rather than voxel data, the (x, y) coordinates of the point cloud data on the X-axis and Y-axis perpendicular to each other on a horizontal plane are input to the above machine learning algorithm. In this case, all data in the Z-axis direction may be included by simply deleting the Z-axis information that was originally held, or a range in the Z-axis direction may be specified, and the range in the Z-axis direction may be set to Zmin(x, y) to Zmax(x, y) based on the condition (x, y), for example, and point cloud information with other conditions may be deleted. Furthermore, (x, y) may be generated for two or more coordinate points or two or more ranges in the Z-axis information and used as input information. In addition, two-dimensional sizes obtained by extracting data at a certain height of the voxel data or calculating the sum or maximum value of the height direction may be generated and input to the above machine learning algorithm.

After step S5, the prediction result of the future communication quality of the wireless communication path predicted by the communication quality prediction device 1 is output to a control device provided in the wireless communication system. Based on the prediction result, the control device controls the settings of the wireless communication system, the settings and operations of the wireless communication terminal, or physically controllable objects in the wireless communication area so as to reduce the impact of communication quality in the wireless communication area or to prevent a decrease in communication quality. For example, the control device performs transmission power control for the wireless communication terminal to increase the transmission power, handover to switch communication to a base station with higher sensitivity, communication with multiple wireless communication base stations, and changes the settings of the modulation method and error correction code.

[Modification]
Up to this point, the case where LiDAR is used has been described as an example. On the other hand, instead of acquiring point cloud data using a sensor such as LiDAR, it is also possible to acquire the point cloud data by converting a depth image acquired by an RGB-D camera or the like. For example, in "Yan Wang and 5 others, "Pseudo-LiDAR From Visual Depth Estimation: Bridging the Gap in 3D Object Detection for Autonomous Driving", Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2019", a method is described in which a depth image is estimated from a landscape image and LiDAR point cloud data is pseudo-generated based on the depth image. Therefore, the communication quality prediction device 1 of the present invention can be applied not only in an environment where a sensor such as LiDAR is introduced, but also in an environment where an RGB-D camera is introduced. In this case, from the viewpoint of privacy, etc., restrictions on use can be alleviated by transmitting the data over a network after being converted into point cloud data.

[Prediction performance evaluation results]
3 and 4 are diagrams showing prediction results of communication quality from point cloud data performed using the communication quality prediction device 1 according to this embodiment. FIG. 3 shows the prediction result when a neural network is used. FIG. 4 shows the prediction result when a random forest is used. These prediction results are predictions of future communication quality (RSSI value of Frame signal) 500 ms ahead from a specific time based on point cloud data obtained at the specific time. The predicted values shown by the solid lines roughly match the actual measured values shown by the dashed lines.

The data used in this performance evaluation was the same as that used in Non-Patent Document 2, and the accuracy was comparable to that of Non-Patent Document 2. This shows that communication quality can be predicted with the same level of accuracy even when using sensors such as LiDAR, which are relatively easy to install, without using cameras, which have restrictions from the standpoint of privacy and confidential information protection.

[Industrial applications]
The present invention is a technology that stabilizes communication quality in wireless LANs (Local Area Networks), particularly in devices equipped with millimeter wave communication functions, and is a particularly useful invention for applications that require high-capacity, stable wireless networks, such as sensing data collection in factories, VR (Virtual Reality) applications, and AR (Augmented Reality) applications.

[effect]
According to this embodiment, the communication quality prediction device 1 predicts future communication quality, and includes a plurality of collection units 1-1-1 to 1-1-R that acquire environmental information around a wireless communication path as point cloud data in a time series, a pre-processing unit 1-2 that generates processed point cloud data by performing processing to reduce the amount of data of the point cloud data, a conversion unit 1-3 that converts the processed point cloud data of the environmental information into environmental information of a coordinate system that can be recognized by the communication quality prediction device, and a prediction unit 1-5 that predicts and calculates the future communication quality of the wireless communication path based on the change over time in the environmental information of the coordinate system. Therefore, the wireless communication quality can be predicted without using a camera or an image, and can be applied even in places where it is not desirable to install a camera from the viewpoint of protecting privacy, confidential information, etc. As a result, a technology that can predict future communication quality in various wireless communication areas can be provided without being restricted by the viewpoint of protecting privacy, confidential information, etc.

[others]
The present invention is not limited to the above embodiment. The present invention can be modified in many ways within the scope of the gist of the present invention. The communication quality prediction device 1 according to the present embodiment can be realized by a computer and a program, and the program can be recorded on a recording medium or provided through a network.

For example, the communication quality prediction device 1 according to this embodiment can be realized using a general-purpose computer system including a 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. 5. The memory 902 and the storage 903 are storage devices. In this computer system, the CPU 901 executes a predetermined program loaded onto the memory 902, thereby realizing each function of the communication quality prediction device 1.

The communication quality prediction device 1 may be implemented in one computer. The communication quality prediction device 1 may be implemented in multiple computers. The communication quality prediction device 1 may be a virtual machine implemented in a computer. The program for the communication quality prediction device 1 may be stored in a computer-readable recording medium such as an HDD, SSD, USB memory, CD, or DVD. The program for the communication quality prediction device 1 may also be distributed via a communication network.

1. Communication quality prediction device
1-1-1～1-1-R…Collection section
1-2…Pre-processing section
1-3…Conversion section
1-4…Generation section
1-5…Prediction section
901...CPU
902…Memory
903…Storage
904...Communication equipment
905…Input device
906...Output device

Claims (6)

A communication quality prediction device for predicting future communication quality, comprising:
a collection unit that acquires environmental information around the wireless communication path in time series as point cloud data;
a pre-processing unit that performs processing to reduce the amount of data of the point cloud data to generate processed point cloud data;
a conversion unit that converts the processed point cloud data of the environmental information into environmental information in a coordinate system recognizable by the communication quality prediction device;
a prediction unit that predicts and calculates a future communication quality of the wireless communication path based on a time-dependent change in environmental information of the coordinate system,
The conversion unit is
A communication quality prediction device that converts the processed point cloud data of the environmental information into voxel data. The pre-treatment unit includes:
The communication quality prediction device according to claim 1 , wherein the number of points in the point cloud data is reduced by removing noise points from the point cloud data. The pre-treatment unit includes:
The communication quality prediction device according to claim 1 , wherein the number of the point cloud data is reduced by lowering a spatial resolution of the point cloud data. The point cloud data is
The communication quality prediction device according to claim 1 , wherein the environmental information is point cloud data converted from a depth image captured by a camera. A communication quality prediction method for predicting future communication quality, comprising the steps of:
A communication quality prediction device
acquiring environmental information about a wireless communication path in the form of point cloud data in time series;
generating processed point cloud data by performing a process for reducing the amount of data of the point cloud data;
converting the processed point cloud data of the environmental information into environmental information in a coordinate system recognizable by the communication quality prediction device;
predicting and calculating a future communication quality of the wireless communication path based on a time-dependent change in environmental information of the coordinate system;
In the converting step,
The communication quality prediction method further comprises converting the processed point cloud data of the environmental information into voxel data. A communication quality prediction program that causes a computer to function as the communication quality prediction device according to any one of claims 1 to 4 .

Images