Calculation device and program
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- NT T INC
- Filing Date
- 2025-01-24
- Publication Date
- 2026-07-30
Smart Images
Figure JP2025002238_30072026_PF_FP_ABST
Abstract
Description
Computing device and program
[0006] ,
[0001] The present invention relates to a computing device and a program.
[0002] In order to cope with the increasing mobile traffic year by year, communication using a high frequency band such as millimeter wave that can utilize a wide bandwidth has attracted attention. On the other hand, since radio waves in a high frequency band are greatly attenuated by moving obstacles such as the human body and vehicles, the communication quality is likely to deteriorate. Therefore, it is effective to detect deterioration of communication quality in advance and perform communication control including transmission power control and handover.
[0003] Non-Patent Document 1 discloses a technique of constructing a machine learning model that takes the positions and speeds of a communication device and obstacles as inputs and outputs communication quality, and estimating communication quality using the machine learning model.
[0004] S. Ohta, T. Nishio, R. Kudo, K. Takahashi and H. Nagata, "Real2Sim2Real for Point Cloud-Based mmWave Link Quality Prediction: An Empirical Study," in IEEE Transactions on Vehicular Technology.
[0005] In the method disclosed in Non-Patent Document 1, learning is performed using a learning dataset measured in advance, but there is a problem that it is difficult to collect the learning dataset. In contrast, it is conceivable to generate a learning dataset by simulation. However, since the simulation of radio wave propagation has a large amount of calculation, there is a problem that it takes time to generate a learning dataset. An object of the present invention is to provide a computing device and a program capable of suppressing the amount of calculation for calculating communication quality based on simulation of the environment where a communication device exists.
[0006] One aspect of the present invention is a computing device comprising: a fixed model showing the shape of a fixed shield in an environment in which a first and second communication device exist; a first acquisition unit that acquires the results of a first simulation, which is a simulation of the path of radio waves between the first and second communication device in the environment, using a fixed model showing the shape of a fixed shield in the environment in which the first and second communication devices exist; a second acquisition unit that acquires the results of a second simulation, which is a simulation of the path of radio waves passing near a moving shield when a moving model showing the shape of a moving shield present in the environment is placed on the fixed model; and a calculation unit that calculates the quality of communication between the first and second communication device when the moving shield is present in the environment, based on the results of the first and second simulations.
[0007] According to the above embodiment, the computing device can reduce the computational complexity of calculating communication quality based on a simulation of the environment in which the communication device exists.
[0008] This is a schematic block diagram showing the configuration of a communication system according to the first embodiment. This is a schematic block diagram showing the configuration of a communication control device according to the first embodiment. This is a flowchart showing a communication control method by the communication control device according to the first embodiment. This is a schematic block diagram showing the configuration of a learning device according to the first embodiment. This is a flowchart showing the learning process by the learning device according to the first embodiment. This is a flowchart showing the learning process by the learning device according to the second embodiment. This is a schematic block diagram showing the configuration of a learning device according to the third embodiment. This is a flowchart showing the learning process by the learning device according to the third embodiment. This is a schematic block diagram showing the configuration of a computer according to at least one embodiment.
[0009] <First Embodiment> <Configuration of Communication System 1> The embodiments will be described in detail below with reference to the drawings. Figure 1 is a schematic block diagram showing the configuration of the communication system 1 according to the first embodiment. The communication system 1 according to the first embodiment is a system that performs communication using high-frequency radio waves such as millimeter waves. The communication system 1 comprises a plurality of access point devices 10, 1 or a plurality of communication terminals 20, an imaging device 30, a communication control device 40, and a learning device 50.
[0010] The access point device 10 provides a wireless communication access point. The communication terminal 20 can communicate via the access point device 10 by establishing a connection with the access point device 10. The imaging device 30 images the space in which the communication system 1 is provided. The imaging range of the imaging device 30 includes the communication range of the access point device 10. The imaging device 30 may be a device capable of imaging the three-dimensional shape of space, such as a LiDAR or a stereo camera. An imaging device 30 may be provided for each access point device 10. In this case, each imaging device 30 may image the communication range of the corresponding access point device 10.
[0011] The communication control device 40 controls communication between the access point device 10 and the communication terminal 20. For example, the communication control device 40 estimates the communication quality between the access point device 10 and the communication terminal 20, and outputs a control signal to hand over the communication terminal 20 to another access point device 10 if the communication quality deteriorates. The communication control device 40 estimates the communication status between the access point device 10 and the communication terminal 20 using a machine learning model. The learning device 50 trains the machine learning model that the communication control device 40 uses to estimate the communication status.
[0012] 《Configuration of the Communication Control Device 40》 Figure 2 is a schematic block diagram showing the configuration of the communication control device 40 according to the first embodiment. The communication control device 40 includes an image acquisition unit 41, an obstacle identification unit 42, a communication information collection unit 43, a communication quality prediction unit 44, and a control unit 45. The image acquisition unit 41 acquires image data captured by the imaging device 30. The obstacle identification unit 42 identifies the three-dimensional position and speed of moving obstacles M1 (people, vehicles, etc.) that appear in the acquired image data. The obstacle identification unit 42 can extract moving obstacles M1 by, for example, taking the difference between the image of the imaging range of the imaging device 30 when there are no moving obstacles M1 and the image data. The obstacle identification unit 42 may also extract moving obstacles M1 from the image data using, for example, pattern matching or a machine learning object detection model.
[0013] The communication information collection unit 43 collects information from the access point device 10 regarding the communication terminal 20 connected to the access point device 10. The information collected by the communication information collection unit 43 includes the ID and location of the communication terminal 20. The communication information collection unit 43 may determine the speed of movement from the change in the location of the communication terminal 20, or it may collect the speed of movement as communication information.
[0014] The communication quality prediction unit 44 inputs the time series of the position and velocity of the moving obstacle M1 and the position and velocity of the communication terminal 20 to a machine learning model trained by the learning device 50, and predicts the communication quality between the communication terminal 20 and the access point device 10. The machine learning model is trained to predict the communication quality considering the influence of the moving obstacle M1, based on the positional relationship between the access point device 10, the communication terminal 20, and the moving obstacle M1. In other embodiments, the machine learning model may take camera images or point cloud data as input and output the communication quality between the communication terminal 20 and the access point device 10. Examples of machine learning models include neural networks and random forest models. Models suitable for processing time series data, such as RNN (Recurrent Neural Network) and LSTM (Long Short Term Memory), are preferred. The method for training the machine learning model will be described later.
[0015] The control unit 45 controls communication between the access point device 10 and the communication terminal 20 based on the communication quality predicted by the communication quality prediction unit 44. For example, if the communication quality between the communication terminal 20 and the access point device 10 falls below a predetermined threshold, the control unit 45 performs handover control to switch the access point device 10 to which the communication terminal 20 is connected.
[0016] 《Control by Communication Control Device 40》 Figure 3 is a flowchart showing the communication control method by the communication control device 40 according to the first embodiment. The communication control device 40 repeatedly executes the process shown in Figure 3 at predetermined control cycles while the communication system 1 is operating. The video acquisition unit 41 acquires video data captured from the imaging device 30 during the current control cycle (step S1). The obstruction identification unit 42 identifies the position and movement speed of the moving obstruction M1 that is captured in the acquired video data (step S2). For each access point device 10, the obstruction identification unit 42 identifies the position of the moving obstruction M1 in a planar coordinate system with the position of the access point device 10 as the origin (step S3). The obstruction identification unit 42 may identify the position of the moving obstruction M1 that is within the communication range of the access point device 10 and ignore the moving obstruction M1 that is outside the communication range.
[0017] Next, the communication information collection unit 43 collects information (ID and location) of the communication terminal 20 connected to each access point device 10 from each access point device 10 (step S4). If the speed of the communication terminal 20 is not included, the communication information collection unit 43 calculates the speed from the collected change in location.
[0018] The communication quality prediction unit 44 performs the following processing for each combination of the access point device 10 and the communication terminal 20 detected in step S4. Specifically, the communication quality prediction unit 44 inputs the position and movement speed of the moving obstruction M1 identified in step S3 and the position and movement speed of the access point device 10 identified in step S4 into a machine learning model to estimate the communication quality when the access point device 10 and the communication terminal 20 communicate (step S5). At this time, the communication quality prediction unit 44 predicts the communication quality not only for combinations of access point device 10 and communication terminal 20 for which a connection has been established, but also for combinations of access point device 10 and communication terminal 20 for which a connection has not been established. The communication quality prediction unit 44 may omit predicting the communication quality for combinations of access point device 10 and communication terminal 20 for which a connection has not been established, in which case the communication terminal 20 is outside the communication range of the access point device 10.
[0019] The control unit 45 selects each of the communication terminals 20 detected in step S4 (step S6), and then executes the following processes from steps S7 to S11 for the selected communication terminal 20. First, the control unit 45 determines whether the communication quality between the communication terminal 20 selected in step S6 and the access point device 10 with which a connection has been established exceeds a predetermined threshold (step S7). If the communication quality exceeds the predetermined threshold (step S7: YES), the control unit 45 decides to maintain the connection between the communication terminal 20 and the access point device 10 (step S8). On the other hand, if the communication quality does not exceed the predetermined threshold (step S7: NO), the control unit 45 identifies the access point device 10 with the highest communication quality among the multiple access point devices 10 (step S9). The control unit 45 determines whether the access point device 10 with the highest communication quality is an access point device 10 with which a connection has already been established (step S10). If the communication terminal 20 has established a connection with the access point device 10 with the highest communication quality (step S10: YES), the control unit 45 decides to maintain the connection between the communication terminal 20 and the access point device 10 (step S8). If the communication terminal 20 has not established a connection with the access point device 10 with the highest communication quality (step S10: NO), the control unit 45 determines the access point device 10 as the handover destination and sends a handover instruction to the communication terminal 20 (step S11).
[0020] 《Configuration of the Learning Device 50》 The learning device 50 learns the parameters of a machine learning model used to predict communication quality by the communication control device 40 through communication simulation. Incidentally, the simulation for calculating communication quality can use the ray tracing method, which calculates the path of radio waves between the access point device 10 and the communication terminal 20. In the ray tracing method, it is necessary to calculate the reflection and diffraction of radio waves using a three-dimensional model that represents the shape of the shielding objects present in the communication environment. Therefore, it is known that the computational load increases when the three-dimensional model contains a lot of information about complex surfaces. Also, when obtaining a sufficient amount of time-series data of communication quality as a training dataset for machine learning for a communication environment that changes over time due to the movement of the communication terminal 20 and the moving shielding object M1, as in the first embodiment, it may take a great deal of time to generate the training dataset. Therefore, the learning device 50 according to the first embodiment calculates communication quality while suppressing the computational load. The learning device 50 is an example of a computing device.
[0021] Figure 4 is a schematic block diagram showing the configuration of the learning device 50 according to the first embodiment. The learning device 50 includes a model storage unit 51, a movement simulator unit 52, a communication simulator unit 53, a first acquisition unit 54, a second acquisition unit 55, a calculation unit 56, a learning unit 57, and a calculation result storage unit 58.
[0022] The model storage unit 51 stores a fixed model, which is a three-dimensional model representing the shape of fixed obstacles M2 (floors, ceilings, walls, columns, etc.) in the environment where the access point devices 10 and communication terminals 20 exist, and a mobile model, which is a three-dimensional model representing the shape of moving obstacles M1 (people, vehicles, etc.). The three-dimensional model may be, for example, point cloud data. The fixed model stored by the model storage unit 51 may represent the shape of fixed obstacles M2 in the entire environment encompassing multiple access point devices 10, or it may represent the shape of fixed obstacles M2 within the communication range of each access point device 10.
[0023] The movement simulator unit 52 simulates the movement of the moving obstacle M1 in the environment represented by the fixed model. The movement simulator unit 52 simulates the movement of the moving obstacle M1 according to behavioral rules that indicate the relationship with other movement models and fixed models for each movement model, for example. For example, the movement simulator unit 52 simulates the movement of the moving obstacle M1 according to behavioral rules that the moving obstacle M1 moves so as not to come into contact with the fixed obstacle M2 or other moving obstacles M1.
[0024] The communication simulator unit 53 simulates the radio wave path between the access point device 10 and the communication terminal 20 using a fixed model and a mobile model. The communication simulator unit 53 simulates the radio wave path using the ray tracing method.
[0025] The first acquisition unit 54 acquires multiple radio wave paths between the access point device 10 and the communication terminal 20 when no moving obstacles are present, as simulated by the communication simulator unit 53. The second acquisition unit 55 acquires radio wave paths that pass near moving obstacles, as simulated by the communication simulator unit 53.
[0026] The calculation unit 56 calculates the quality of communication between the access point device 10 and the communication terminal 20 based on the multiple radio wave paths acquired by the first acquisition unit 54 and the second acquisition unit 55. In other words, the calculation unit 56 replaces the paths that pass through the mobile model among the multiple paths acquired by the first acquisition unit 54 with the paths acquired by the second acquisition unit 55, and by combining the multiple paths, calculates the quality of communication, such as propagation loss, between the access point device 10 and the communication terminal 20.
[0027] The learning unit 57 uses a training dataset that includes the position and movement speed of the communication terminal 20 and the moving obstacle M1, as well as the quality of communication, to train the parameters of a machine learning model so that it takes the position and movement speed of the communication terminal 20 and the moving obstacle M1 as input and outputs the quality of communication between the access point device 10 and the communication terminal 20. The learning unit 57 trains a machine learning model for each access point device 10. The machine learning model trained by the learning unit 57 is output to the communication control device 40 and used to estimate the quality of communication.
[0028] The calculation result storage unit 58 stores the calculation result of the communication quality between the access point device 10 and the communication terminal 20, which has been simulated by the communication simulator unit 53.
[0029] <Learning process by the learning device 50> Before the communication system 1 is put into operation, the learning device 50 trains a machine learning model for estimating the quality of communication, corresponding to each of the multiple access point devices 10. In addition, when a new access point device 10 is added to the communication system 1, the learning device 50 trains a machine learning model corresponding to that access point device 10.
[0030] Figure 5 is a flowchart showing the learning process by the learning device 50 according to the first embodiment. The communication simulator unit 53 simulates the radio wave path between the access point device 10 and the communication terminal 20 when the communication terminal 20 is present at each of the multiple locations that divide the communication range of the access point device 10 into multiple meshes in a three-dimensional coordinate system with the position of the access point device 10 as the origin (step S21). This simulation makes it possible to identify the radio wave path in an environment without moving obstructions M1. The communication simulator unit 53 records the simulation results for each location of the communication terminal 20 in the calculation result storage unit 58 (step S22).
[0031] The learning device 50 repeatedly performs the following steps S23 to S33 until the amount of training data reaches a predetermined target amount.
[0032] The movement simulator unit 52 of the learning device 50 arranges multiple movement models in a fixed model stored in the model storage unit 51 and simulates the movement of the moving obstacle M1 represented by each movement model over a predetermined target time (step S23). The initial arrangement of the movement models may be determined randomly. The target time for simulating the movement of the moving obstacle M1 may be an integer multiple of two or more of the control cycle of the communication control device 40.
[0033] The first acquisition unit 54 determines one of the moving models or a random fixed position to be the location of the communication terminal 20 (step S24). By determining one of the moving models to be the location of the communication terminal 20, the learning device 50 can calculate the quality of communication at the communication terminal 20 being transported by the user. By determining a fixed position to be the location of the communication terminal 20, the learning device 50 can calculate the quality of communication at the communication terminal 20 being placed on a table or the like.
[0034] The learning device 50 selects one by one multiple time points obtained by dividing the target time for simulation into predetermined calculation cycles (step S25), and then executes the following processes from steps S26 to S32 for the selected time points.
[0035] The first acquisition unit 54 acquires the radio wave path in the case where there is no moving obstruction M1, which is recorded in the calculation result storage unit 58 in association with the position of the communication terminal 20 at the time selected in step S25 (step S26). The second acquisition unit 55 identifies the position of the moving obstruction M1 at the time selected in step S25 from the simulation results of the moving simulator unit 52 (step S27). The second acquisition unit 55 identifies the combination of the radio wave path acquired in step S26 that passes through the moving obstruction M1 and the moving obstruction M1 that the radio wave path passes through (step S28). The second acquisition unit 55 extracts a partial model from the overall model in which the fixed model and the moving model are arranged, which contains obstructions (fixed obstruction M2 and moving obstruction M1) that are located within a predetermined range centered on the position of the moving obstruction M1 identified in step S28 (step S29). The range to be extracted as a partial model may be represented by a sphere with a predetermined radius centered on the moving obstruction M1. By using a sphere as the range for extraction of the submodel, internal determination can be performed using only distance calculations, eliminating the need to consider rotation. This simplifies the calculations compared to other ranges such as cuboids or ellipses, and the range can be expanded simply by scaling the radius, thus reducing the number of parameters.
[0036] The communication simulator unit 53 uses the partial model extracted in step S29 to recalculate the path that passed through the moving obstruction M1 identified in step S28 (step S30). At this time, the path to be calculated is not the path of all radio waves emitted from the access point device 10 or communication terminal 20, but only a portion of the path identified in step S28, so the computational load of the simulation can be reduced. Also, since the amount of data included in the partial model is smaller than that of the overall model, the computational load of the simulation can be reduced.
[0037] The calculation unit 56 obtains the radio wave path between the access point device 10 and the communication terminal 20 at the time selected in step S25 by superimposing the radio wave path obtained in step S26 that does not pass through the moving obstruction M1 with the path recalculated in step S30 (step S31). The calculation unit 56 calculates the quality of communication between the access point device 10 and the communication terminal 20 by combining the paths obtained in step S31 (step S32). The calculation unit 56 calculates the quality in consideration of the radio wave reach rate and the effect of fading due to the relationship between path length difference and frequency.
[0038] By repeating the calculations from steps S26 to S32 described above, the learning device 50 can obtain a time series of communication quality during the target time simulated in step S23. The calculation unit 56 adds the combination of the position and speed of the moving obstacle M1 simulated in step S23 and the communication quality calculated in step S32 to the learning dataset (step S33).
[0039] When the amount of training dataset reaches the target amount, the learning unit 57 uses the training dataset to input the position and movement speed of the communication terminal 20 and the moving obstacle M1, and trains the parameters of the machine learning model so that it outputs the quality of communication between the access point device 10 and the communication terminal 20 (step S34).
[0040] As described above, the learning device 50 according to the first embodiment generates a learning dataset in the following procedure. First, the learning device 50 obtains the radio wave path between the access point device 10 and the communication terminal 20 when there is no moving obstruction M1, based on a fixed model. The fixed model does not include the moving obstruction M1. The learning device 50 recalculates the radio wave path that passes through the moving obstruction M1 or its vicinity from the obtained radio wave path. As a result, when the learning device 50 calculates the communication quality at multiple times in a time-changing communication environment, it can omit calculations for paths that are not affected by the moving obstruction M1. Therefore, the learning device 50 according to the first embodiment can reduce the amount of computation required for calculating the communication quality based on the simulation.
[0041] 《Second Embodiment》 In the learning device 50 according to the first embodiment, whenever there is a path that passes through a partial model, that is, a path that passes through the moving obstacle M1 or its vicinity, the learning device 50 performs a simulation using the partial model for that path each time. In contrast, the learning device 50 according to the second embodiment simulates the changes in the path due to the moving obstacle M1 for the path that passes through the partial model in advance, and uses the results to calculate the communication quality, thereby reducing the amount of computation.
[0042] 《Learning Process by Learning Device 50》 Figure 6 is a flowchart illustrating the learning process by the learning device 50 according to the second embodiment. The communication simulator unit 53 simulates the radio wave path between the access point device 10 and the communication terminal 20 when the communication terminal 20 is present at each of the multiple locations that divide the communication range of the access point device 10 into multiple meshes in a three-dimensional coordinate system with the position of the access point device 10 as the origin (step S41). The communication simulator unit 53 records the simulation results for each location of the communication terminal 20 in the calculation result storage unit 58 (step S42).
[0043] The communication simulator unit 53 calculates the relationship between the direction of the radio wave input to the partial model, the output direction of the radio wave, and the loss for each type of moving shield M1 (step S43). The direction of the radio wave is represented by a combination of the elevation angle and the azimuth angle. The communication simulator unit 53 simulates the radio wave path emitted from a plurality of directions toward the movement model by the ray tracing method. The communication simulator unit 53 records the relationship between the input direction, output direction, and loss of the radio wave calculated for each partial model in the calculation result storage unit 58 (step S44).
[0044] The learning device 50 repeatedly executes the processes from step S45 to step S54 until the amount of the learning data set reaches a predetermined target amount.
[0045] The movement simulator unit 52 of the learning device 50 arranges a plurality of movement models in the fixed model stored in the model storage unit 51 and simulates the movement of the moving shield M1 represented by each movement model at a predetermined target time (step S45). The first acquisition unit 54 determines one or a random fixed position of the movement model as the position of the communication terminal 20 (step S45).
[0046] The learning device 50 selects one by one a plurality of times obtained by dividing the target time of the simulation at a predetermined calculation cycle (step S47), and executes the processes from step S47 to step S52 for the selected time.
[0047] The first acquisition unit 54 acquires the radio wave path in the case where there is no moving shield M1, which is recorded in the calculation result storage unit 58 in association with the position of the communication terminal 20 at the time selected in step S47 (step S48). The second acquisition unit 55 identifies the position of the moving shield M1 at the time selected in step S47 from the simulation result of the movement simulator unit 52 (step S48). The second acquisition unit 55 identifies the combination of the path passing through the moving shield M1 among the radio wave paths acquired in step S47 and the moving shield M1 through which the radio wave path passes (step S49). The second acquisition unit 55 recalculates the path passing through the moving shield M1 based on the relationship between the input direction, output direction, and loss of the radio wave with respect to the moving shield M1 recorded in the calculation result storage unit 58 (step S50).
[0048] The calculation unit 56 can obtain the radio wave path between the access point device 10 and the communication terminal 20 at the time selected in step S46 by overlapping the path that does not pass through the moving shield M1 among the radio wave paths acquired in step S47 and the path recalculated in step S50 (step S51). The calculation unit 56 calculates the communication quality between the access point device 10 and the communication terminal 20 by synthesizing the path obtained in step S51 (step S52).
[0049] By repeating the calculations from step S47 to step S52 above, the learning device 50 can obtain the time series of the communication quality at the target time simulated in step S44. The calculation unit 56 adds the combination of the position and moving speed of the moving shield M1 simulated in step S44 and the communication quality calculated in step S52 to the learning dataset (step S53).
[0050] When the amount of the learning dataset reaches the target amount, the learning unit 57 uses the learning dataset to input the positions and moving speeds of the communication terminal 20 and the moving shield M1, and learns the parameters of the machine learning model so as to output the communication quality between the access point device 10 and the communication terminal 20 (step S54).
[0051] According to the learning device 50 of the second embodiment, by pre-simulating the input / output relationship of radio waves due to the moving obstruction M1, the calculation duplication can be prevented by reading out the results of the simulation. As a result, the learning device 50 of the second embodiment can reduce the amount of computation required for calculating communication quality based on the simulation. Furthermore, in situations where multiple moving obstructions M1 exist, the straight-line propagation characteristic of high-frequency bands such as millimeter waves allows for efficient quality calculation by superimposing the results of the read-out simulations.
[0052] In the learning device 50 according to the second embodiment, the results of the simulation of the input / output relationship of radio waves with respect to each moving shield M1 are recorded in the calculation result storage unit 58, but this is not limited to this. For example, the learning device 50 according to another embodiment may store the relationship between the positional relationship of the access point device 10, the communication terminal 20, and the moving shield M1, and the communication quality. In other words, the learning device 50 may store the loss (dBm) of radio waves passing through the moving shield M1 in association with the combination of the position of the moving shield M1 and the position of the communication terminal 20. In the process of creating a learning dataset, if the positional relationship between the communication terminal 20 and the moving shield M1 simulated by the moving simulator unit 52 is similar to past calculation results, the learning device 50 may read the recorded information and use it as the calculation result of the communication quality. In situations where there are multiple moving shields M1, the quality can be efficiently calculated by superimposing the read simulation results due to the straight-line propagation characteristic of high-frequency bands such as millimeter waves.
[0053] For example, if the calculation result storage unit 58 records the propagation loss when the communication terminal 20 is at the first position and the moving obstruction M1 is at the second position, the propagation loss when the communication terminal 20 is at the first position and the moving obstruction M1 is at the third position, and the propagation loss when the communication terminal 20 is at the first position and the moving obstruction M1 is at the fourth position, the calculation unit 56 can calculate the propagation loss when the communication terminal 20 is at the first position and the moving obstruction M1 is at each of the second, third, and fourth positions as follows. That is, the calculation unit 56 can calculate the propagation loss when the communication terminal 20 is at the first position and the moving obstruction M1 is at each of the second, third, and fourth positions by adding the propagation loss when the moving obstruction M1 is absent to the sum of the differences between the propagation loss when the moving obstruction M1 is absent and each of the propagation losses recorded in the calculation result storage unit 58 (i.e., the increase in propagation loss due to the presence of the moving obstruction M1).
[0054] 《Third Embodiment》 The computational complexity of the communication quality simulation varies depending on the simulation conditions. For example, in addition to the ray tracing method, the line-of-sight method is known as a method for simulating the path of radio waves. The line-of-sight method is a method for determining the quality of communication based on the presence or absence of an obstruction between two communication devices. In other words, the line-of-sight method ignores the effects of radio wave reflection and diffraction, as in the ray tracing method, which reduces accuracy but allows for the calculation of communication quality with less computation. Another example is the electromagnetic field analysis method, which is known as a method for simulating the path of radio waves. The electromagnetic field analysis method simulates the behavior of the electromagnetic field, not just the reflection direction based on the shape of the obstruction, as in the ray tracing method, which increases computational complexity but allows for the calculation of communication quality with greater accuracy. Furthermore, even with the same ray tracing method, the accuracy and computational complexity change depending on whether or not some calculations are omitted, as in the first and second embodiments, and also depending on the setting of the radius of the sphere used to obtain the partial model. The learning device 50 according to the third embodiment learns a machine learning model while changing the method for simulating communication accuracy.
[0055] 《Configuration of the Learning Device 50》 Figure 7 is a schematic block diagram showing the configuration of the learning device 50 according to the third embodiment. In addition to the configuration of the first embodiment, the learning device 50 according to the third embodiment further includes a condition setting unit 59.
[0056] The condition setting unit 59 receives input from the administrator of the communication system 1 as conditions for machine learning, including the time limit until the communication system 1 starts operation, the target estimation error, and the unit generation time. The condition setting unit 59 manages the learning process so that the creation of the training dataset is completed by the time limit until the system starts operation. The condition setting unit 59 manages the learning process so that the error between the prediction result of the communication quality by the machine learning model and the actual communication quality is smaller than the target estimation error. The condition setting unit manages the learning process so that the time required to generate one sample of the training dataset is shorter than the unit generation time.
[0057] 《Learning Process by Learning Device 50》 Figure 8 is a flowchart showing the learning process by the learning device 50 according to the third embodiment. The condition setting unit 59 receives input from the administrator of the communication system 1 as conditions for machine learning, namely the time limit until the start of operation of the communication system 1, the target estimation error, and the unit generation time (step S61). Next, the learning device 50 generates a training dataset of one sample in the same manner as in the first embodiment and trains the machine learning model (step S62).
[0058] The condition setting unit 59 determines whether the time limit set in step S61 has elapsed since the start of the learning process (step S63). If the time limit has not elapsed (step S63: NO), the condition setting unit 59 causes the communication control device 40 to perform a process to estimate the quality of communication between the access point device 10 and the communication terminal 20 using the learned machine learning model (step S64). The communication control device 40 estimates the quality of communication between the access point device 10 and the communication terminal 20 by executing the processes from steps S1 to S5 of the first embodiment. The condition setting unit 59 obtains the actual communication quality from the access point device 10 and calculates the error with the estimation result (step S65). The condition setting unit 59 determines whether the error between the estimated communication quality result and the actual communication quality is smaller than the target estimation error set in step S61 (step S66). If the error is larger than the target estimation error (step S66: NO), the condition setting unit 59 determines whether the accuracy of the estimation has improved (step S67). Specifically, the condition setting unit 59 determines whether the error calculated in step S65 is smaller than past errors.
[0059] If the accuracy of the estimation has not improved (step S67: NO), the condition setting unit 59 changes the simulation settings to improve accuracy and increase computational load (step S68). For example, to make the simulation settings closer to the real world, the condition setting unit 59 may increase the radius of the sphere used to extract the partial model, perform a simulation using the whole model without using the partial model, increase the maximum number of reflections in ray tracing, or add setting conditions such as transmittance according to the type of shielding. Alternatively, the type of simulation may be switched to another method that is more accurate than the ray tracing method, such as an electromagnetic field analysis method.
[0060] If the accuracy of the estimation improves (step S67: YES), the condition setting unit 59 saves the conditions of step S62 (step S69) and determines whether the time required to generate one sample is longer than the unit generation time set in step S61 (step S70). If the time required to generate one sample is longer than the unit generation time set in step S61 (step S70: YES), the condition setting unit 59 changes the simulation settings to reduce accuracy and computational load (step S71). For example, to further simplify the simulation settings, the condition setting unit 59 may, for example, reduce the radius of the sphere used to extract the submodel, or reduce the maximum number of reflections in ray tracing. The computational load can also be reduced by switching the simulation method from the method according to the first embodiment to the method according to the second embodiment. The type of simulation may also be switched to another method that is simpler than the ray tracing method, such as a line-of-sight determination method. In the line-of-sight determination method as well, making a determination that does not take into account the Fresnel zone requires less computation than making a determination that takes into account the Fresnel zone.
[0061] If the simulation conditions are changed or maintained by the processing from step S67 to step S71, the learning device 50 returns to step S62 and generates a single sample training dataset by simulating under the new conditions.
[0062] The learning device 50 terminates the learning process in step S63 if the time limit has elapsed since the start of the learning process (step S63: YES), or in step S66 if the error between the estimated communication quality and the actual communication quality becomes smaller than the target estimated error (step S66: YES).
[0063] <Other Embodiments> Although one embodiment has been described in detail above with reference to the drawings, the specific configuration is not limited to that described above, and various design changes can be made. That is, in other embodiments, the order of the above-described processing may be changed as appropriate. Also, some processing may be executed in parallel. The learning device 50 according to the above embodiment may be configured by a single computer, or the configuration of the learning device 50 may be divided and arranged on multiple computers, and the multiple computers may cooperate with each other to function as the learning device 50. For example, the part of the learning device 50 that generates the learning dataset and the part that executes the learning processing may be implemented on different computers. In other embodiments, the communication control device 40 and the learning device 50 may be configured as a single unit.
[0064] <Computer Configuration> Figure 9 is a schematic block diagram showing the configuration of a computer according to at least one embodiment. The computer 90 includes a processor 91, main memory 92, storage 93, and interface 94. The learning device 50 described above is implemented in the computer 90. The operation of each processing unit described above is stored in the storage 93 in the form of a program. The processor 91 reads the program from the storage 93, loads it into the main memory 92, and executes the above processing according to the program. The processor 91 also allocates storage areas in the main memory 92 corresponding to each of the above-mentioned storage units according to the program. Examples of the processor 91 include a CPU (Central Processing Unit), a GPU (Graphic Processing Unit), and a microprocessor.
[0065] The program may be for implementing a part of the functions to be performed by the computer 90. For example, the program may perform functions in combination with other programs already stored in storage, or in combination with other programs implemented on other devices. In other embodiments, the computer 90 may include a custom LSI (Large Scale Integrated Circuit) such as a PLD (Programmable Logic Device) in addition to, or instead of, the above configuration. Examples of PLDs include PAL (Programmable Array Logic), GAL (Generic Array Logic), CPLD (Complex Programmable Logic Device), and FPGA (Field Programmable Gate Array). In this case, some or all of the functions implemented by the processor 91 may be implemented by the integrated circuit. Such an integrated circuit is also included as an example of a processor. In other embodiments, the computer 90 may be virtualized on one or more computers.
[0066] Examples of storage 93 include magnetic disks, magneto-optical disks, optical disks, and semiconductor memory. Storage 93 may be an internal medium directly connected to the bus of the computer 90, or it may be an external medium connected to the computer 90 via an interface 94 or a communication line. Furthermore, if this program is distributed to the computer 90 via a communication line, the computer 90 that receives the distribution may expand the program into the main memory 92 and execute the above processing. In at least one embodiment, storage 93 is a tangible storage medium that is not temporary.
[0067] Furthermore, the program may be intended to implement some of the functions described above. In addition, the program may be a so-called differential file (differential program) that implements the functions described above in combination with other programs already stored in the storage 93.
[0068] 1...Communication system 10...Access point device 20...Communication terminal 30...Imaging device 40...Communication control device 50...Learning device 41...Image acquisition unit 42...Obstruction identification unit 43...Communication information collection unit 44...Communication quality prediction unit 45...Control unit 51...Model storage unit 52...Movement simulator unit 53...Communication simulator unit 54...First acquisition unit 55...Second acquisition unit 56...Calculation unit 57...Learning unit 58...Calculation result storage unit 59...Condition setting unit 90...Computer 91...Processor 92...Main memory 93...Storage 94...Interface M1...Moving obstruction M2...Fixed obstruction
Claims
1. A computing device comprising: a first acquisition unit that acquires the results of a first simulation, which is a simulation of the radio wave path between the first and second communication devices in the environment, using a fixed model showing the shape of a fixed shield in the environment in which the first and second communication devices exist, and information representing the positional relationship of the first and second communication devices in the environment; a second acquisition unit that acquires the results of a second simulation, which is a simulation of the radio wave path passing near a moving shield when a moving model showing the shape of a moving shield present in the environment is placed on the fixed model; and a calculation unit that calculates the quality of communication between the first and second communication devices when the moving shield is present in the environment, based on the results of the first and second simulations.
2. The computing device according to claim 1, wherein the first simulation is a simulation that tracks the path of radio waves between the first communication device and the second communication device using the fixed model, and the second simulation is a simulation that tracks the path of radio waves between the first communication device and the second communication device using a partial model consisting of the shape of a moving shield represented by the moving model and the shape of a fixed shield existing in a predetermined range near the moving shield represented by the fixed model.
3. The computing device according to claim 1, comprising a learning unit that takes information representing the positional relationship between the first communication device, the second communication device, and the moving obstruction as input, and outputs the quality of communication between the first communication device and the second communication device, and trains a machine learning model using a training dataset consisting of a combination of the positions of the first communication device, the second communication device, and the moving obstruction used to calculate the quality of communication between the first communication device and the second communication device, and the results of the calculation.
4. A program for causing a computer to perform the following: obtain the result of a first simulation, which is a simulation of the radio wave path between the first and second communication devices in the environment, using a fixed model showing the shape of a fixed shield in the environment in which the first and second communication devices exist, and information representing the positional relationship of the first and second communication devices in the environment; obtain the result of a second simulation, which is a simulation of the radio wave path passing near a moving shield when a moving model showing the shape of a moving shield present in the environment is placed on the fixed model; and calculate the quality of communication between the first and second communication devices when the moving shield is present in the environment, based on the results of the first and second simulations.