JP7252885B2 - Communication quality estimation device and program - Google Patents

Description

TECHNICAL FIELD The present invention relates to technology for estimating communication quality of communication between a base station apparatus and a user apparatus.

In mobile communication networks, a high frequency band called millimeter wave is used to increase the communication speed between a base station apparatus (BS) and a user equipment (UE). Communication using millimeter waves is line-of-sight communication, so if there is an obstacle in the communication path, the communication quality deteriorates. Therefore, Patent Literature 1 and Non-Patent Literature 1 disclose techniques for acquiring depth images and predicting/estimating the communication quality between the BS and the UE based on the depth images. A depth image is an image having distance information, and can be acquired by, for example, an RGB-D (RGB and Depth) camera.

JP 2018-148297 A

Yuta Oguma, et. al. , "Proactive Base Station Selection Based on Human Blockage Prediction Using RGB-D Cameras for mmWave Communications"

Patent Literature 1 and Non-Patent Literature 1 perform machine learning using a depth image, the communication quality obtained from the UE or BS when the depth image is captured, and the positions of the UE and BS as teacher data, thereby obtaining the depth image, the UE It estimates the communication quality between the UE and the BS for a combination of the location of the BS and the location of the BS. However, if there is something hidden behind an object in the depth image, the estimation accuracy is degraded. In addition, the configurations of Patent Document 1 and Non-Patent Document 1 use the depth image itself as teacher data, which increases the processing amount of the learning process.

The present invention provides a technology capable of reducing the amount of learning processing performed for estimating communication quality.

According to one aspect of the present invention, the estimating device includes generation means for generating a three-dimensional spatial model of a predetermined region based on a plurality of depth images acquired by a plurality of cameras, and communication from the base station device to the base station device. an acquisition means for acquiring information on the position of the user device and the communication quality of the communication;
extracting means for extracting the part related to the communication from the three-dimensional space model as a channel space model based on the position of the base station device and the position of the user device; determining a predetermined object in the channel space model; determining means for determining a feature quantity of the communication channel space model based on the determined object;
learning means for generating learning data for estimating communication quality from the feature quantity based on the feature quantity of the channel space model and the communication quality of the communication, wherein the extraction means is configured to determine the location of the base station apparatus; and the location of the user equipment as the communication channel space model, the determination means divides the Fresnel zone into a plurality of zones, and determines the predetermined object present in each of the plurality of zones A value proportional to the volume of is used as the feature amount .

According to the present invention, it is possible to reduce the amount of learning processing performed for estimating communication quality.

1 is a configuration diagram of a system including an estimating device according to an embodiment; FIG. The block diagram of the estimation apparatus by one Embodiment. The figure which shows an example of a channel space model. The figure which shows an example of a channel space model. FIG. 4 is a diagram showing an example of dividing a Fresnel zone into multiple zones; FIG. 4 is a diagram showing an example of feature quantities of a channel space model; FIG. 4 is a diagram showing an example of feature quantities of a channel space model; The block diagram of the estimation apparatus by one Embodiment.

Hereinafter, embodiments will be described in detail with reference to the accompanying drawings. It should be noted that the following embodiments do not limit the invention according to the claims, and not all combinations of features described in the embodiments are essential to the invention. Two or more of the features described in the embodiments may be combined arbitrarily. Also, the same or similar configurations are denoted by the same reference numerals, and redundant explanations are omitted.

FIG. 1 is a configuration diagram of a system including an estimation device 3 according to this embodiment. The system has multiple BSs 1 and multiple cameras 2 . BS1 may communicate with one or more UEs (not shown in FIG. 1) within its coverage. By deploying a plurality of BS1, it is possible to provide UE with communication services over a predetermined area. Also, the camera 2 is, for example, a TOF camera, a stereo camera, an RGB-D camera, or the like. In this embodiment, the camera 2 is an RGB-D camera, and can capture a depth image including color information (RGB) of each pixel and distance information of each pixel. Note that the pixel distance information is information indicating the distance from the camera 2 to the subject (object) corresponding to the pixel. A plurality of cameras 2 are arranged at different positions so as to cover predetermined areas where communication services are provided by a plurality of BSs 1 . More specifically, as will be described later, the positions and orientations of the multiple cameras 2 are determined so that a three-dimensional (3D) space model of the predetermined area can be generated from the depth images acquired by each of the multiple cameras 2. be done.

Each BS 1 and each camera 2 are configured to be able to communicate with the estimation device 3 . Each BS 1 repeatedly transmits location information and communication quality information to the estimation device 3 . The location information transmitted by BS1 is information indicating the location of BS1 and the location of UE with which BS1 is communicating. The communication quality information transmitted by BS1 is information indicating the communication quality between the BS1 and the UE when the UE is at the location indicated by the location information. Communication quality is, for example, error rate, SN ratio (or SNI ratio), and received power. When estimating downlink communication quality from BS1 to UE, BS1 acquires downlink communication quality from UE and transmits it to estimation device 3 . Similarly, when estimating the uplink communication quality from the UE to BS1, BS1 transmits the uplink communication quality to the estimation device 3. Also, when estimating the communication quality in both directions, the BS 1 transmits the communication quality in both directions to the estimation device 3 . On the assumption that the communication quality in the uplink direction and the communication quality in the downlink direction are the same, it is also possible to adopt a configuration in which the downlink direction and the uplink direction are not distinguished. In this case, the BS 1 can transmit only the downlink communication quality or only the uplink communication quality to the estimation device 3 . Alternatively, the BS 1 can send the downlink and uplink communication quality to the estimation device 3 without distinguishing between the directions. Each camera 2 repeatedly captures depth images and repeatedly transmits image information indicating the depth images to the estimation device 3 .

FIG. 2 is a configuration diagram of the estimation device 3 according to this embodiment. Note that FIG. 2 shows only functional blocks related to generation of learning data. The model generation unit 31 creates a 3D space model of a predetermined area where communication services are provided by a plurality of BSs 1, based on the depth images captured repeatedly by the plurality of cameras 2 and having a difference in imaging time within a predetermined value. Generate. A 3D space model indicates an area where an object exists in the predetermined area. For example, it is possible to give a synchronized timing signal to a plurality of cameras 2 and to have each camera 2 capture depth images at approximately the same time based on this timing signal. In this case, the model generator 31 generates a 3D space model based on depth images captured by the cameras 2 at approximately the same time. The model generation unit 31 outputs 3D space model data representing a 3D space model to the communication channel extraction unit 32 .

As described above, BS 1 repeatedly transmits location information and communication quality information to estimation device 3 . Here, the communication quality information indicates the communication quality of communication performed between BS1 and UE, and the position information indicates the positions of BS1 and UE when the communication quality is acquired. The BS 1 repeatedly transmits the position information and the communication quality information to the estimation device 3. In the processing described below, the position information and the communication quality at the imaging time of the depth image used by the model generation unit 31 to generate the 3D space model. information, or location information and communication quality information at a time when the time difference from the imaging time is short (the time difference is within a predetermined value). Note that if there is a range in the imaging times of the depth images used to generate the 3D space model, the processing described below is performed using the position information and communication quality information within this time range, or the central time of this time range. This is done using location information and communication quality information whose time difference is within a predetermined value.

The communication channel extraction unit 32 extracts a partial area of the 3D space model indicated by the 3D space model data based on the two positions indicated by the position information, and uses this as a communication channel space model. The communication channel space model is a model of the communication channel (radio signal propagation channel) space between the BS 1 and the UE whose position information indicates the position.

Two different types of channel space models can be considered. The first type is the Fresnel zone model shown in FIG. In FIG. 3, BS1 and UE4 are the BS and UE whose positions are indicated in the position information, respectively. Fresnel zone 5 is determined by the distance between BS1 and UE4 and the communication frequency. The second type is the ray model shown in FIG. In FIG. 4, BS1 and UE4 are the BS and UE whose positions are indicated in the position information, respectively. Since the propagation characteristics of radio signals in the millimeter wave band are close to those of light, the light ray model uses the path connecting the BS1 and the UE4 as the communication path space. For example, according to FIG. 4, a direct path 61 connecting BS1 and UE4 and a reflected path 62 via reflector 7 are defined as the communication path space. These paths can be determined by ray tracing. Instead of defining the path as a pure line, it is also possible to define the path as extending to a certain extent in a direction perpendicular to the direction of the path.

The channel extraction unit 32 outputs data representing the channel space model to the feature quantity calculation unit 33 . The channel space model shows objects existing in the radio channel between BS1 and UE4. Here, an object that is small when viewed from the position of BS1 or UE4 has little effect on communication quality. Therefore, the feature amount calculation unit 33 removes objects that have a small influence on communication quality from the communication channel space model in order to reduce the processing load of the learning process.

In this embodiment, individual objects are specified from the communication channel space model, and the volume thereof is defined as V. FIG. Also, the distance between BS1 and the object is d1, and the distance between UE4 and the object is d2. Then, let S be the larger one of V/(d1 ³ ) and V/(d2 ³ ). If S is smaller than the threshold, the object is removed from the channel space model. That is, the channel space model is corrected so that the object does not exist in space. On the other hand, if S is greater than or equal to the threshold, the object is left as is. This process corresponds to removing objects whose size as seen from the position of BS1 and UE4 is smaller than a threshold.

After removing objects that have a small effect on the communication quality from the channel space model, the feature amount calculation unit 33 determines attributes of each object in the channel space model. Attributes are information related to the type of object, such as whether the object is a person, a car, a road sign, or a building. Note that the types of attributes to be determined are determined in advance. Specifically, it is possible to specify in advance objects that affect millimeter wave communication, and classify the types of these objects as attribute types. The attributes are determined based on, for example, the shape and color. After determining each attribute, the feature quantity calculator 33 generates a feature quantity of the channel space model.

FIG. 5 is an explanatory diagram of a method of generating a feature amount of the communication channel space model when using the Fresnel zone model shown in FIG. 3 as the communication channel space model, and FIG. Examples of quantities are given. As shown in FIG. 5A, the Fresnel zone 5 is divided into a plurality of regions of different sizes from the center. In FIG. 5A, it is divided into three regions of n=1-3. FIG. 5B shows a cross-section of the Fresnel zone perpendicular to the straight line connecting BS1 and UE4. A central circular region (corresponding to n=1) and two circular regions surrounding the circular region (corresponding to n=2 and 3) are each divided into a plurality of regions along the circumferential direction. In FIG. 5B, it is divided into eight regions. That is, in this example shown in FIGS. 5A and 5B, one Fresnel zone 5 is divided into a total of 24 regions. Below, each of the 24 regions is expressed as a zone. Each zone is assigned a zone identifier. For example, as shown in FIG. 5B, eight zones with n=1 are given zone identifiers Z ₁₁ to Z ₁₈ . Note that the number of divisions of the Fresnel zone 5 is arbitrary.

As shown in FIG. 6, the feature amount calculation unit 33 uses the total value of the values proportional to the volume of each attribute of the object included in the zone as the feature amount for each zone. In this embodiment, a value proportional to the volume is used as the number of pixels of the channel space model. For example, according to FIG. 6, it can be seen that the zone with zone identifier Z ₁₁ has 200 pixels corresponding to attribute #1 (e.g., people) and 0 pixels corresponding to attribute #2 (e.g., car). It is shown.

FIG. 7 shows an example of feature amounts when the light ray model shown in FIG. 4 is used as the communication path space model. As shown in FIG. 7, the feature amount includes a path feature amount and an object feature amount. The object feature amount is the same as in the case of using the Fresnel zone, and is the total value for each attribute of values proportional to the volume of the object existing on the path. The path feature quantity is a value that characterizes a path, and in FIG. 4 has four values: type, BS side angle, UE side angle, and path length. The type is a value indicating whether it is a direct path or a reflection path. Also, the BS side angle is the angle of the path on the BS1 side with respect to the straight line connecting the BS1 and the UE4. Similarly, the UE side angle is the angle of the path on the UE4 side with respect to the straight line connecting BS1 and UE4. Path length is the length of the path. Since the BS-side angle and the UE-side angle being 0 means that the path is a direct path, it is possible to adopt a configuration in which the type is not used. Alternatively, one or both of the BS side angle, the US side angle, and the path length may be used as the feature amount. Furthermore, it is also possible to adopt a configuration in which only the object feature amount is used and the path feature amount is not used.

A combination of the feature amount and the communication quality is input to the learning unit 34 as training data. For example, it is assumed that the estimation device 3 receives the communication quality information of the communication between the BS1 and the UE4 and the position information at that time from the BS1. In this case, from the 3D space model generated based on the depth image captured at the time when the communication quality was measured or at a time when the difference from the time when the communication quality was measured is short (below a predetermined value), the A channel space model is extracted based on the position information, and a combination of the feature amount generated from the channel space model and the communication quality indicated by the communication quality information becomes training data. The estimation device 3 generates teacher data as described above each time it acquires communication quality information and location information from each BS 1 . Then, the learning unit 34 performs learning so that the communication quality indicated by the teacher data is output when the feature amount of the teacher data is input, and outputs the learning data.

FIG. 8 is a configuration diagram of the estimation device 3 according to this embodiment. Note that FIG. 8 shows only functional blocks related to estimation of communication quality based on learning data. The position of the UE and the position of the BS whose communication quality is to be estimated are input to the channel extraction unit 32 . The channel extraction unit 32 extracts a channel space model from the 3D space model generated based on the latest depth image, and the feature quantity calculation unit 33 calculates the feature quantity of the channel space model. The processing in the channel extraction unit 32 and the feature amount calculation unit 33 is the same as that in generating learning data. If the learning data is generated using the Fresnel zone model, the Fresnel zone model is also used during estimation, and if the learning data is generated using the light ray model, the estimation Sometimes we use the ray model.

The estimation unit 35 has learning data, and when the feature amount is input, outputs communication quality as an estimation result.

For example, if there are multiple BSs to which the UE is to be handed over, the BS to which the UE is to be handed over can be determined based on the estimation result. Further, based on the movement status of a UE communicating with a certain BS, the position of the UE after a predetermined time is determined, the communication quality between the US at the determined position and the BS is estimated, and if the communication quality deteriorates below a predetermined value. If so, it can be controlled to perform handover.

In this embodiment, BS1 includes the position of BS1 in the position information, but unlike UE, the position of BS1 does not change. can be configured to transmit only the position of the UE to the estimation device 3 as position information. Also, the estimation device 3 of the present embodiment generates a 3D space model of a predetermined area serviced by a plurality of BSs 1, but may be configured to generate a 3D space model for each BS1. In this case, the estimation device 3 can be configured to be provided in the corresponding BS1.

Note that in the present embodiment, the feature amount is obtained for each attribute of the object, but the total volume of an object having a predetermined size (where S is equal to or greater than the threshold value) can be used as the feature amount without determining the attribute. Also, it is possible to adopt a configuration in which small objects are not deleted by comparing S with the threshold value.

As described above, by generating a 3D space model based on a plurality of depth images, it is possible to prevent degradation of estimation accuracy due to objects hidden behind the subject in each depth image. Also, a communication channel space model is extracted from the 3D space model, a feature amount of the communication channel space model is calculated, and learning processing is performed based on the feature amount. With this configuration, the processing amount of the learning process can be reduced compared to performing the learning process using the depth image or the channel space model itself. Furthermore, by deleting small objects that do not affect communication from the channel space model, it is possible to improve the estimation accuracy and reduce the amount of processing.

Note that the estimating device 3 according to the present invention can be realized by, for example, a program that, when executed by one or more processors of a device such as a computer, causes the one or more processors to function and operate as the estimating device 3. . These computer programs can be stored in a computer-readable storage medium or distributed via a network.

The invention is not limited to the above embodiments, and various modifications and changes are possible within the scope of the invention.

31: model generation unit, 32: communication channel extraction unit, 33: feature amount calculation unit, 34: learning unit

Claims (8)

generating means for generating a three-dimensional spatial model of a predetermined area based on a plurality of depth images acquired by a plurality of cameras;
Acquisition means for acquiring from a base station apparatus the position of a user apparatus communicating with the base station apparatus and information on the communication quality of the communication;
extracting means for extracting, as a channel space model, a portion related to the communication from the three-dimensional space model based on the position of the base station device and the position of the user device;
determining means for determining a predetermined object in the communication path space model and determining a feature amount of the communication path space model based on the determined object;
learning means for generating learning data for estimating communication quality from the feature quantity based on the feature quantity of the channel space model and the communication quality of the communication;
with
the extracting means extracts a Fresnel zone between the position of the base station apparatus and the position of the user apparatus as the channel space model;
The estimating device, wherein the determining means divides the Fresnel zone into a plurality of zones, and sets a value proportional to the volume of the predetermined object existing in each of the plurality of zones as the feature quantity. 2. The determining means determines a plurality of objects of different types, and sets a value proportional to the volume of each type of the predetermined object existing in each of the plurality of zones as the feature quantity . The estimating device described in . generating means for generating a three-dimensional spatial model of a predetermined area based on a plurality of depth images acquired by a plurality of cameras;
Acquisition means for acquiring from a base station apparatus the position of a user apparatus communicating with the base station apparatus and information on the communication quality of the communication;
extracting means for extracting, as a channel space model, a portion related to the communication from the three-dimensional space model based on the position of the base station device and the position of the user device;
determining means for determining a predetermined object in the communication path space model and determining a feature amount of the communication path space model based on the determined object;
learning means for generating learning data for estimating communication quality from the feature quantity based on the feature quantity of the channel space model and the communication quality of the communication;
with
The extracting means extracts one or more path routes connecting the position of the base station apparatus and the position of the user apparatus as the communication path space model,
The estimating device, wherein the determining means uses a value proportional to the volume of the predetermined object present in each of the one or more paths as the feature amount. The determination means determines a plurality of objects of different types, and sets a value proportional to the volume of each type of the predetermined object present in each of the one or more paths as the feature amount. Item 4. The estimation device according to item 3 . The feature amount includes, for each of the one or more paths, a path length, an angle of the path at the base station device with respect to a straight line connecting the base station device and the user device, the base station device and the 5. The estimating device according to claim 3 , comprising at least one of: and an angle of the path at the user device with respect to a straight line connecting the user device. The determination means deletes an object satisfying a predetermined condition from the communication path space model extracted by the extraction means, determines the predetermined object in the communication path space model after deletion, and determines the determined object based on the determined object. 6. The estimating device according to any one of claims 1 to 5 , wherein the feature quantity of the channel space model is determined. 7. The estimation apparatus according to claim 6 , wherein the object satisfying the predetermined condition is an object whose size when viewed from the position of the base station apparatus and the position of the user apparatus is smaller than a threshold. A program, characterized in that, when executed on said one or more processors of a device having one or more processors, it causes said device to function as an estimating device according to any one of claims 1 to 7 .

Images