Device for generating a learning model for estimating propagation characteristics of a wireless signal

Abstract

To provide a technique for estimating propagation characteristics that takes into account the shape of the surface of a reflecting object.SOLUTION: A generation device includes learning means for generating a learning model for estimating propagation characteristics of a wireless signal on the basis of a plurality of pieces of training data. Each of the plurality of pieces of training data includes data indicating one or more types of normal maps, and correct data indicating the characteristic value at the receiving position of the wireless signal transmitted at the transmission position. Each of the one or more types of normal maps indicates a normal direction of an object located at each of a plurality of pixels within a range including at least one of the transmission position and the reception position.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a technique for estimating the propagation characteristics of wireless signals.

Background Art

[0002] Patent Document 1 discloses a configuration in which a CNN (Convolutional Neural Network) is applied to map data including a reception point to extract urban structure parameters, and the propagation characteristics of radio waves are estimated based on the urban structure parameters. Non-Patent Document 1 also discloses that when the uneven size of the object surface is larger than the diameter of the second Fresnel zone, the wireless signal diffuses and reflects on the object surface.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Non-Patent Documents

[0004]

Non-Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] In order to estimate the propagation characteristics of wireless signals, it is necessary to consider the influence of reflectors of wireless signals such as buildings existing in space. Here, as described in Non-Patent Document 1, depending on the size of the unevenness of the surface of the reflector, the diffuse reflection component becomes larger than the specular reflection component on the surface. Patent Document 1 discloses the utilization of map data for estimating propagation characteristics, but does not consider the unevenness of the surface of buildings or the like.

[0006] The present invention provides an estimation technique for propagation characteristics that takes into account the shape of the surface of a reflector.

Means for Solving the Problem

[0007] According to one aspect of the present invention, the generation device includes a learning means for generating a learning model for estimating the propagation characteristics of a wireless signal based on a plurality of pieces of teacher data. Each of the plurality of pieces of teacher data includes data indicating one or more types of normal maps and correct answer data indicating characteristic values at the reception position of the wireless signal transmitted at the transmission position. Each of the one or more types of normal maps indicates the normal direction of an object at the position of each of a plurality of pixels within a range including at least one of the transmission position and the reception position.

Effect of the Invention

[0008] According to the present invention, it is possible to estimate the propagation characteristics in consideration of the shape of the surface of the reflector.

Brief Description of the Drawings

[0009]

Figure 1

Figure 2

Figure 3

Figure 4

Mode for Carrying Out the Invention

[0010] Hereinafter, embodiments will be described in detail with reference to the accompanying drawings. Note that the following embodiments do not limit the invention according to the claims, and not all combinations of the features described in the embodiments are essential for the invention. Two or more of the plurality of features described in the embodiments may be arbitrarily combined. Also, the same or similar configurations are given the same reference numerals, and redundant explanations are omitted.

[0011] FIG. 1(A) is a configuration diagram of a generation device that generates a learning model for estimating propagation characteristics, and FIG. 1(B) is a configuration diagram of an estimation device that estimates propagation characteristics using the learning model generated by the generation device of FIG. 1(A). The generation device includes a learning unit. The learning unit generates a learning model using known machine learning techniques with teacher data including normal map data as input. The estimation device includes an estimation unit. The estimation unit has the learning model generated by the generation device. The estimation unit outputs an estimated value of the propagation characteristics using the learning model based on input data including normal map data. Note that the propagation characteristics to be estimated may be delay spread, angular spread, received power, etc. at a predetermined reception position, but may also be values of other propagation characteristics.

[0012] FIG. 2 is an explanatory diagram of the normal map indicated by the normal map data. The normal map is composed of a plurality of pixels arranged in two dimensions (U direction and V direction), similar to a normal still image. In a normal still image, each pixel indicates a color represented by three values of R (red), G (green), and B (blue). On the other hand, in the normal map, a pixel indicates the direction of the normal of the reflector at the position of the pixel. In FIG. 2, it shows that the pixel value of one pixel is (Xa, Yb, Zb). The pixel value (Xa, Yb, Zb) indicates the normal vector (direction of the normal) in a three-dimensional world coordinate system with a predetermined position in real space as the origin. Note that the normal vector of a pixel without a reflector is, for example, (0, 0, 0).

[0013] The normal map indicates the unevenness of the object surface. Diffuse reflection is a phenomenon in which radio waves are reflected in various directions due to the unevenness of the object surface, but the reflection at each minute area of the object surface can be approximated by specular reflection. Therefore, based on the propagation direction of the radio signal incident on the reflector at the position of a certain pixel and the normal direction of the reflector, the reflection direction of the radio signal is determined. In this way, by using the normal map, it is possible to estimate the propagation characteristics considering diffuse reflection, that is, considering the shape of the surface of the reflector.

[0014] Hereinafter, each example in this embodiment will be described with reference to FIG. 3. FIG. 3 shows the teacher data in each example. The teacher data has example data and correct answer data. The learning unit of the generation device generates a learning model by repeatedly adjusting the learning model so that the output of the learning model when the example data is input to the learning model approaches the correct answer data corresponding to the example data. As shown in FIG. 3, all of the correct answer data are characteristic values of the propagation characteristics to be estimated. For example, when the estimation target is the delay spread, the characteristic value is the delay spread, and when the estimation target is the angle spread, the characteristic value is the angle spread.

[0015] <Example 1> In this example, the normal map data (hereinafter, the first normal map) when looking from the reception position to the transmission position is used as the example data, and the characteristic value at the reception position at that time is used as the correct answer data. At this time, as shown by the black circles in FIG. 2, the transmission position is a predetermined position of the normal map, for example, the center. The teacher data includes a plurality of first normal map data and the correct answer data for each of the plurality of first normal map data. In the plurality of first normal maps shown by each of the plurality of first normal map data, the transmission position is the same position.

[0016] To determine the characteristic value at the reception position, the direction of the scattered wave in the reflector visible from the reception position is important. Since the first normal map represents the normal direction of the reflector visible from the reception position, it is possible to perform an estimation considering the diffuse reflection component arriving at the reception position by the first normal map. Further, by setting the transmission position to the same position in each of the first normal maps shown by the plurality of first normal map data used as the teacher data, the estimation considering the transmission position is achieved and the estimation accuracy is improved.

[0017] <Example 2> In this embodiment, the normal map data (hereinafter referred to as the second normal map) when viewing the reception position from the transmission position is used as the example data, and the characteristic value at the reception position at that time is used as the correct answer data. At this time, as shown by the black circle in FIG. 2, the reception position is a predetermined position of the normal map, for example, the center. In addition, in each of the second normal maps indicated by a plurality of second normal map data used as teacher data, the reception positions are the same positions.

[0018] The number of reflection times up to the reception position has a great influence on the propagation characteristics. The second normal map represents the direction of the scattered wave in the reflector visible from the transmission position and the reception position. Therefore, it is possible to estimate in consideration of the number of reflection times of the diffuse reflection component arriving at the reception position based on the second normal map data.

[0019] <Example 3> In this embodiment, both the first normal map data and the second normal map data are used as the example data. Specifically, the first normal map data indicating the first normal map when viewing a certain transmission position from a certain reception position, the second normal map data indicating the second normal map when viewing the reception position from the transmission position, and the characteristic value at the reception position at that time are used as the correct answer data. In addition, in each of the first normal maps indicated by a plurality of first normal map data used as teacher data, the transmission positions are the same positions, and in each of the second normal maps indicated by a plurality of second normal map data used as teacher data, the reception positions are the same positions. In this embodiment, it is possible to perform an estimation having the advantages of both Embodiment 1 and Embodiment 2.

[0020] <Example 4> In this embodiment, in addition to the first normal map data, the normal map data (hereinafter referred to as the third normal map) including the transmission position and the reception position is used as the example data. The first normal map is the same as in Embodiment 1. The transmission position and the reception position in the third normal map are arbitrary, and it is not necessary to set the transmission position and the reception position to the same position in each of the third normal maps indicated by a plurality of third normal map data used as teacher data. In this embodiment, it is possible to use the information of the reflector in the blind spot from the reception position, and the estimation accuracy is improved.

[0021] <Example 5> In this example, in addition to the second normal map data, the third normal map data is used as the example data. The second normal map is the same as that in Example 2, and the third normal map is the same as that in Example 4. In this example, the information of the reflector in the blind spot from the transmission position can be used, and the estimation accuracy is improved.

[0022] <Example 6> In this example, the first normal map data, the second normal map data, and the third normal map data are used as the example data. The first normal map is the same as that in Example 1, the second normal map is the same as that in Example 2, and the third normal map is the same as that in Example 4. In this example, the information of the reflector in the blind spot from the transmission position and the reception position can be used, and the estimation accuracy is improved.

[0023] <Example 7> In this example, in addition to the normal map data described in the above Examples 1 to 6, the image data of an image in the same range as the normal map is used as the example data. Each pixel shown in the image data indicates, for example, the luminance value of each of RGB. By inputting an image in the same range in addition to the normal map, for example, the estimation accuracy of the degree of reflection in the reflector can be improved.

[0024] As described above, each example of the example data has been explained. In FIG. 2, the normal map indicates the normal direction in a Cartesian coordinate system, but it may indicate the normal direction in a spherical coordinate system. In this case, since the information on the size is not necessary, the normal direction can be indicated by two angles, and the amount of information can be reduced.

[0025] When estimating with the estimation device, the input data uses the same data as the example data when generating the learning model to be used. For example, when estimating using the learning model generated in Example 1, the input data is the first normal map data when looking at the transmission position from the reception position.

[0026] <Modified Form> FIG. 4(A) and FIG. 4(B) are configuration diagrams of modified forms of the generation device and the estimation device shown in FIGS. 1(A) and 1(B). The generation unit of the generation device converts the data of the normal map included in the teacher data based on the transmission position and the reception position in the normal map, and generates first map data indicating the first map. The learning unit performs learning based on the teacher data including the first map. Similarly, the generation unit of the estimation device converts the data of the normal map included in the input data based on the transmission position and the reception position in the normal map, and generates first map data indicating the first map. The estimation unit outputs an estimated value of the propagation characteristics based on the first map data.

[0027] The generation unit determines the specular reflection direction at each pixel based on the normal direction of each pixel indicated by the normal map and the transmission position, and determines the pixels of the normal map whose specular reflection direction is toward the reception position. Then, the pixel value of the pixel of the first map that is the same as the pixel of the normal map whose specular reflection direction is toward the reception position in the normal map is set as the first value, and the pixel value of the other pixels is set as the second value to generate the first map data. That is, each pixel of the first map indicates whether the reflected light from the reflector corresponding to the position of the pixel is directed toward the reception position. The power of the reflected wave at the reception point when the specular reflection direction is toward the reception position is larger than the power of the reflected wave at the reception point when the direction other than the specular reflection direction is toward the reception position, and the reflected wave with a large power has a large influence on the propagation characteristics. Therefore, by generating a learning model based on such a first map and estimating the propagation characteristics based on the first map, accurate estimation can be performed.

[0028] Note that each of the generation device and the estimation device according to the present invention can be realized by a program that causes a computer to operate as the generation device and the estimation device. These computer programs are stored in a computer-readable storage medium or can be distributed via a network.

[0029] The invention is not limited to the above-described embodiments, and various modifications and changes are possible within the scope of the gist of the invention.

Claims

1. A learning means for generating a learning model for estimating the propagation characteristics of a wireless signal based on a plurality of teacher data, wherein each of the plurality of teacher data includes data indicating one or more types of normal maps and correct answer data indicating characteristic values at a reception position of the wireless signal transmitted at a transmission position, and each of the one or more types of normal maps indicates a normal direction of an object at a position of each of a plurality of pixels within a range including at least one of the transmission position and the reception position, a generating device.

2. The generating device according to claim 1, wherein the one or more types of normal maps include a first type of normal map viewed from the reception position in the direction of the transmission position.

3. The generating device according to claim 2, wherein the one or more types of normal maps include a second type of normal map viewed from the transmission position in the direction of the reception position.

4. The generating device according to claim 3, wherein the one or more types of normal maps include a third type of normal map including the transmission position and the reception position.

5. The generating device according to claim 1, wherein each of the plurality of teacher data includes image data of the same range viewed from the same position and in the same direction as each of the one or more types of normal maps.

6. The generating device according to claim 1, wherein each of the one or more types of normal maps indicates the normal direction by two angles in spherical coordinates.

7. A generating means for generating first data indicating a first map based on a transmission position of a wireless signal, a reception position of the wireless signal, and data indicating a normal map, and a learning means for generating a learning model for estimating the propagation characteristics of the wireless signal based on a plurality of teacher data, wherein each of the plurality of teacher data includes the first data indicating the first map and correct answer data indicating characteristic values at the reception position of the wireless signal transmitted at the transmission position, the normal map indicates a normal direction of an object at a position of each of a plurality of first pixels within a predetermined range including at least one of the transmission position and the reception position, and the first map includes a plurality of second pixels corresponding to each of the plurality of first pixels within the predetermined range. For each of the plurality of first pixels, the generation means determines a third pixel that reflects the radio signal from the transmission position toward the reception position based on the normal direction of the first pixel, sets the value of a fourth pixel corresponding to the third pixel among the plurality of second pixels of the first map to a first value, and generates the first data indicating the first map such that the values of pixels different from the fourth pixel among the plurality of second pixels of the first map are different from the first value. A generating device. **Claim 8** The generating device according to claim 1, wherein the characteristic value of the propagation characteristic is a delay spread. **Claim 9** The generating device according to claim 1, wherein the characteristic value of the propagation characteristic is an angular spread. **Claim 10** A program that, when executed by one or more processors of a device having one or more processors, causes the device to function as the generating device according to any one of claims 1 to 9.

Images