Learning method, wireless quality estimation method, learning device, wireless quality estimation device, and program

Abstract

A learning method with which a learning device trains a model for estimating wireless quality by machine learning. The method comprises: a training step for training the model using the measured data of wireless quality obtained by a plurality of terminals and thereby generating a pretrained model; and a retraining step for retraining the pretrained model using the measured data that satisfies a prescribed condition.

Description

Technical Field

[0001] The present invention relates to a technique for estimating radio quality in a terminal that performs wireless communication.

Background Art

[0002] In wireless communication, the radio quality in a terminal varies due to various factors such as the received power at the radio receiver, interference from the surroundings, and the traffic of other terminals communicating on the same channel.

[0003] On the other hand, in applications such as autonomous driving of automobiles, autonomous delivery robots, and control of drones, when wireless communication is used for transmission of control signals, etc., a continuously stable communication quality is required. In such applications, prior estimation of radio quality becomes effective. For example, when the radio wave deteriorates and the communication quality decreases at the future driving destination of an autonomous vehicle, it is possible to perform control such as reducing the speed to prevent a major accident.

[0004] Regarding the estimation of radio quality in such applications, as one of the conventional methods, there is a method of detecting quality degradation based on measurement results of quality such as the radio wave intensity on the spot (this technique is referred to as "Conventional Technique 1").

[0005] However, making a judgment only based on the measurement results means that actions will be taken after the actual degradation, so there is a possibility that the judgment will be delayed. Therefore, it is effective to predict the degradation of radio quality and cause an action before the degradation occurs.

[0006] Also, as a method of pre-estimating and visualizing radio quality, there is a method of using a terminal for measurement in advance to measure the received power of radio waves and communication quality at the actual place of use, and constructing a database in combination with the position information. Such a database is called a heat map.

[0007] In the estimation of quality using such a heat map, due to time constraints, physical constraints, etc., the measurement points may not necessarily be continuous. In such a case, calculations are performed to interpolate between the measurement points. The quality estimation value on the actual terminal used is obtained by predicting the future movement position and obtaining the value corresponding to that position from the heat map (this technique is referred to as "Prior Art 2").

Prior Art Documents

Non-Patent Documents

[0008]

Non-Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0009] In Prior Art 1, there is a problem that the deterioration of wireless quality cannot be known in advance. Also, in Prior Art 2, although wireless quality deterioration can be predicted to some extent in advance, it is necessary to create a map of quality measurements in advance. At this time, the more measurement points there are, the more accurate the heat map becomes. Also, using the results measured on the actual terminal to be used eliminates the difference in characteristics between terminals, making it more accurate. However, it is often difficult to perform measurements in advance using the actual terminal to be used.

[0010] It is also conceivable to perform the estimation of wireless quality using a model obtained by machine learning. In machine learning, it is desirable to perform learning using measurement data obtained under conditions close to the conditions (desired conditions) when actually performing the estimation. However, it is often difficult to prepare a large number of such measurement data. Therefore, measurement data obtained under conditions different from the desired conditions will be used. However, learning using measurement data obtained under different conditions is inefficient, and the accuracy of the estimation result of wireless quality also decreases.

[0011] The present invention has been made in view of the above points, and an object thereof is to provide a technique that enables accurate estimation of radio quality even when the number of measurement data obtained under desired conditions is small.

Means for Solving the Problems

[0012] According to the disclosed technique, there is provided a learning method in which a learning device learns a model for estimating radio quality by machine learning, a learning step of generating a pre-trained model by learning the model using measurement data of radio quality obtained by a plurality of terminals, the measurement data having position information, terminal information, and radio quality information; from the measurement data of the aforesaid the plurality of terminals, extracting measurement data of a specific terminal to be the target of radio quality estimation or measurement data of a terminal close to the specific terminal, and the aforesaid performing re-learning using the extracted the aforesaid measurement data. the aforesaid using the measurement data , for the aforesaid pre-trained model to perform re-learning a re-training step and comprising A learning method is provided.

Effects of the Invention

[0013] According to the disclosed technique, there is provided a technique that enables accurate estimation of radio quality even when the number of measurement data obtained under desired conditions is small.

Brief Description of the Drawings

[0014]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Embodiments for Carrying Out the Invention

[0015] Hereinafter, embodiments (these embodiments) of the present invention will be described with reference to the drawings. The embodiments described below are merely examples, and the embodiments to which the present invention is applied are not limited to the following embodiments.

[0016] In the following description, wireless communication qualities such as radio wave reception intensity and throughput are collectively referred to as "wireless quality".

[0017] In the following description, a neural network model (which may also be called a learning device) is used for estimating wireless quality, but this is just an example. A model obtained by machine learning other than a neural network may also be used.

[0018] Also, in the following description, transfer learning is used as a method of re-learning for a pre-trained model, but this is also just an example. For example, fine-tuning may be used as a method of re-learning.

[0019] (Outline of the Embodiment) FIG. 1 shows an example of a case where the estimation result of wireless quality is used for assisting autonomous driving. FIG. 1 shows an image of estimating the wireless quality of the destination by a heatmap of wireless quality.

[0020] As shown in Fig. 1, on the planned route of the autonomous vehicle, the wireless quality is predicted. If such a heat map can be created with high precision, it becomes possible to estimate the wireless quality at the destination of the autonomous vehicle with high precision. However, in order to create a high-precision heat map, a lot of pre-measurement data under desired conditions is required. The desired conditions are, for example, to perform measurements using the terminal actually used for wireless communication in the autonomous vehicle.

[0021] However, as described above, it is often difficult to obtain a lot of pre-measurement data using the actually used terminal. For example, in applications such as automating crop harvesting by autonomous driving of a tractor, when remotely controlling and monitoring using wireless communication, creating a heat map of wireless quality in advance using the tractor requires, for example, performing it in advance during a period when there are no crops, and the operation is difficult.

[0022] Therefore, in the present embodiment, first, based on a lot of measurement data collected from various terminals other than the actually used terminal (the actually used terminal may be included), pre-training by machine learning is performed to create a model (referred to as a "pre-trained model").

[0023] Furthermore, using the measurement data collected from the actually used terminal (or a terminal close to the actually used terminal), transfer learning is performed on the pre-trained model, and the model obtained by transfer learning (referred to as a "transfer-trained model") is used to estimate the wireless quality for the actually used terminal.

[0024] As a result, even if the measurement data collected from the actually used terminal (or a terminal close to the actually used terminal) is small, it is possible to estimate the wireless quality with high accuracy. Hereinafter, the device configuration and processing content in the present embodiment will be described in more detail.

[0025] (Example of device configuration) Fig. 2 shows a configuration example of the wireless quality estimation device 100 in the present embodiment. The wireless quality estimation device 100 is a device that estimates the wireless quality in a terminal actually used for wireless communication (for example, a terminal mounted on an autonomous vehicle for wireless communication).

[0026] The wireless quality estimation device 100 may be provided in an autonomous vehicle or the like together with the terminal actually used for wireless communication, or the wireless quality estimation device 100 may be provided as a function within the terminal actually used for wireless communication. Further, the wireless quality management unit 110 within the wireless quality estimation device 100 may be the terminal actually used for wireless communication.

[0027] As shown in Fig. 2, the wireless quality estimation device 100 in the present embodiment has a wireless quality management unit 110 and a wireless quality estimation unit 120. Note that the "wireless quality estimation device" may be referred to as a "wireless quality estimation system".

[0028] The wireless quality estimation device 100 in the present embodiment is installed, for example, in an autonomous vehicle or the like as described above and is connected via an operation system and a network (communication line). However, such a form is merely an example. The wireless quality estimation device 100 may be provided in a general vehicle other than an autonomous vehicle, or the function of the wireless quality estimation device 100 may be provided in a terminal that performs wireless communication.

[0029] Hereinafter, an explanation will be given assuming that the wireless quality estimation device 100 and the terminal actually used as the target of wireless quality estimation (referred to as "terminal A") are provided in a close location (for example, inside an autonomous vehicle).

[0030] In the wireless quality estimation device 100, the wireless quality management unit 110 and the wireless quality estimation unit 120 may be implemented in separate devices and connected using a communication line. When the wireless quality management unit 110 and the wireless quality estimation unit 120 are implemented in separate devices, the "wireless quality management unit" may be referred to as the "wireless quality management device", and the "wireless quality estimation unit" may be referred to as the "wireless quality estimation device". Also, the "wireless quality estimation unit" or the "wireless quality estimation device" may be referred to as the "learning device".

[0031] Also, in this embodiment, the wireless quality estimation unit 120 performs both model learning and wireless quality estimation using the model. However, a functional unit for performing model learning and a functional unit for performing wireless quality estimation using the model may be provided separately. For example, a functional unit for performing model learning (which may also be referred to as a learning device) and a functional unit for performing wireless quality estimation using the model (which may also be referred to as a wireless quality estimation device) may be connected by a communication line, and a pre-trained model may be transmitted from the functional unit for performing model learning (learning device) to the functional unit for performing wireless quality estimation using the model (wireless quality estimation device).

[0032] <Regarding the wireless quality management unit 110> As shown in FIG. 2, the wireless quality management unit 110 includes a position information acquisition unit 111, an estimated position designation unit 112, a wireless communication unit 113, and a wireless terminal information acquisition unit 114.

[0033] The position information acquisition unit 111 acquires the geographical position (latitude, longitude, altitude), speed information, and moving direction information of terminal A using a sensor such as GPS. Also, the estimated position designation unit 112 determines the geographical position (latitude, longitude, altitude) for which wireless quality is to be acquired. The estimated position designation unit may be referred to as a position estimation unit.

[0034] As this determination method, there are determination methods such as estimating and designating the moving destination position after a predetermined time from the current position of terminal A acquired by the position information acquisition unit 111, the speed information, and the moving direction information, or having the moving route information of terminal A in advance and determining the moving destination after a predetermined time from the comparison with the current position.

[0035] The wireless communication unit 113 is a device unit that performs wireless communication according to various wireless communication standards. The wireless terminal information acquisition unit 114 is a functional unit that acquires and stores terminal information (information such as terminal model, wireless communication standard information, antenna characteristic information, etc.) of terminal A.

[0036] <Regarding the wireless quality estimation unit 120> FIG. 3 shows a functional configuration example of the wireless quality estimation unit 120. As shown in FIG. 3, the wireless quality estimation unit 120 includes a learning unit 121, a database unit 122, a transfer learning control unit 123, an input unit 124, an inquiry information input unit 125, and an estimated value output unit 126. The transfer learning control unit may be referred to as a learning control unit.

[0037] As described above, in this embodiment, machine learning technology is used for estimating the wireless quality. The learning unit 121 is a functional unit that implements a machine learning algorithm. Specifically, it performs learning of a model composed of a neural network. The model composed of a neural network is more specifically composed of weight parameters, functions, etc., and is implemented by software. By performing learning of the model, the weight parameters are adjusted to optimal values.

[0038] The database unit 122 includes a database unit 122 that accumulates learning data. The input unit 124 inputs learning data and stores it in the database unit 122.

[0039] The inquiry information input unit 125 inputs inquiry information (position information and terminal information to be estimated), and passes the inquiry information to the learning unit 121. The learning unit 121 inputs the inquiry information into the transfer learned model, and obtains an estimated value of the wireless quality as the output from the model.

[0040] The estimated value is passed from the learning unit 121 to the estimated value output unit 126. The estimated value output unit 126 outputs the estimated value of the wireless quality. The output estimated value is transmitted to, for example, an operating system and used for operating control in an autonomous vehicle equipped with terminal A.

[0041] Here, the estimated value of the radio quality is an estimated value of the radio quality such as the estimated received power value, throughput, delay, jitter, etc. at the position indicated by the input position information. These estimated values may be average or median estimated values, may be maximum or minimum estimated values, or may be other than these.

[0042] <Details of learning> Hereinafter, the learning of the model in the present embodiment will be described in more detail in relation to the transfer learning control unit 123.

[0043] The transfer learning control unit 123 performs control for causing the learning unit 121 to execute transfer learning on a pre-trained model.

[0044] FIG. 4 shows a functional configuration example of the transfer learning control unit 123. As shown in FIG. 4, the transfer learning control unit 123 includes a re-learning data selection unit 1231 for a specific terminal and a re-learning layer setting unit 1232.

[0045] The re-learning data selection unit 1231 for a specific terminal selects data of a terminal close to terminal A, which is the object of radio quality estimation, from the data stored in the database unit 122. Further, the re-learning layer setting unit 1232 sets, according to predetermined conditions, the range of neurons to which re-learning is applied in the pre-trained neural network model.

[0046] With reference to FIG. 5, an example of the learning operation in the learning unit 121 will be described. Note that the example shown in FIG. 5 shows an example in the case of using a neural network model including an input layer, an output layer, and one or more intermediate layers. Each layer has one or more neurons (which may also be called nodes).

[0047] In this embodiment, first, as shown in FIG. 5(a), the learning unit 121 performs model learning using the measurement data of all terminals stored in the database unit 122. This learning is called pre-learning. All terminals refer to all the terminals that have performed the measurement of the collected measurement data, not limited to terminal A.

[0048] Here, as features, by inputting a set of "location information, terminal information, and radio quality information" for each terminal into the learning unit 121, the weight parameters of each layer of the neural network model are optimized through learning. More specifically, the location information and terminal information are input into the model, and the weight parameters are adjusted so as to minimize the error between the output and the radio quality information which is the correct data.

[0049] The above-mentioned location information is the information of the location measured by the terminal, and is the information represented by latitude, longitude information, etc. obtained by GPS or the like. The terminal information is the model information of the terminal (which may also be called type information), the standard information of the wireless interface, the antenna type information, etc. Also, the radio quality information is the measured quality information such as the received power information, throughput, delay, jitter, etc. for each wireless base station to which the terminal is connected.

[0050] Through the above learning, a model corresponding to the location and terminal information of the terminal is created. In this learning, the more learning data (measurement data) of terminal A, which is the target of radio quality estimation, the more accurate the estimation result by this model will be.

[0051] Therefore, it is effective to obtain and learn a large amount of measurement data of the radio quality of terminal A in advance. However, in reality, it is difficult to accumulate a huge amount of measurement data in advance for terminal A whose estimation is desired.

[0052] Therefore, in this embodiment, as a terminal other than terminal A, a general terminal such as a smartphone is utilized to collect measurement data from the terminals of a large number of users and store it in the database unit 122. As a result, a large amount of measurement data can be collected, but the data has variations such as differences in the types of m wireless terminals, differences in antenna performance, and the way the terminal is held during measurement. Therefore, an error occurs between the wireless quality estimated by the model learned thereby and the actual wireless quality of terminal A which is the estimation target of the wireless quality.

[0053] Therefore, in this embodiment, transfer learning is performed on a model that has been pre-trained using measurement data obtained from the terminals of a large number of users to create a transfer learning completed model (re-trained model).

[0054] In transfer learning, the transfer learning control unit 123 selects the measurement data of a terminal close to terminal A which is the estimation target of the wireless quality from among the measurement data stored in the database unit 122, and sets the range of the layers in the neural network model that becomes the re-learning part in the transfer learning in the learning unit 121.

[0055] In the example of Fig. 5(b), the three intermediate layers before the output layer are set as the re-learning part of the transfer learning. In this way, the setting of the re-learning part may be performed in units of layers, or more finely, the re-learning part may be set in units of neurons in the layer. In any case, the range of neurons in the neural network is specified as the re-learning part.

[0056] Regarding which part in the pre-trained model is set as the re-learning part, it may be a predetermined part (e.g., the three intermediate layers before the output layer), or experiments may be conducted to determine the optimal re-learning part, or it may be determined by other methods.

[0057] In transfer learning (relearning), data (location information, terminal information) of a terminal close to terminal A, which is the target of wireless quality estimation, is input into the pre-trained model, and the weight parameters are adjusted so that the output becomes the correct data (wireless quality) of the terminal.

[0058] More specifically, the weight parameters of the pre-trained model are set as the initial values, and only the weight parameters of the part set as the relearning part are adjusted, while the weight parameters of the parts other than the part set as the relearning part remain the initial values. Alternatively, the learning rate of the parts other than the part set as the relearning part is set low to reduce the update amount of the weight parameters.

[0059] Note that the above example is an example of the learning method in relearning such as transfer learning, and is not limited to the above example.

[0060] Also, in transfer learning, a pre-trained model for each terminal information (terminal type) may be generated. For example, from the same pre-trained model, a pre-trained model A1 is generated using data A1 of a terminal close to terminal A1, and a pre-trained model A2 is generated using data A2 of a terminal close to terminal A2. And, for example, when terminal A2 is used as the terminal to be actually used, the wireless quality is estimated using the pre-trained model A2.

[0061] The terminal close to terminal A, which is the target of wireless quality estimation, is, for example, terminal A itself, a terminal of the same model as terminal A, a terminal of the same shape as terminal A (smartphone type, mobile router type, etc.), a terminal equipped with a wireless communication chip with the same model number as the wireless communication chip installed in terminal A, a terminal using the same wireless standard as the wireless standard used by terminal A, a terminal whose antenna performance (gain, half-power width, etc.) is within a difference of a certain range or less from that of terminal A, a terminal used under the same usage conditions (walking, in-vehicle, etc.) assumed by terminal A, and so on.

[0062] The measurement position of the terminal close to terminal A does not necessarily have to be geographically close to the position of terminal A (the position to be estimated). Measurements may be performed by a terminal close to terminal A at a location different from the location where terminal A is present.

[0063] By using the model after transfer learning obtained by re-learning with the measurement data of the terminal close to terminal A, the wireless quality of terminal A can be estimated more accurately.

[0064] It is assumed that the data for this re-learning is less data compared to other data. By performing re-learning using the data from the terminal close to terminal A on the pre-trained model using the measurement data of other terminals, it is possible to estimate the wireless quality for terminal A with high accuracy even if the data from the terminal close to terminal A is small.

[0065] (Example of operation flow) With reference to FIGS. 6 to 8, an example of the operation flow of the wireless quality estimation apparatus 100 in the present embodiment will be described. FIG. 6 shows the flow during pre-learning using all terminal data.

[0066] In S101, "position information, terminal information, wireless quality information" for each terminal is input as learning data to the wireless quality estimation unit 120 and recorded in the database unit 122. Here, the measurement data obtained from various terminals is recorded in the database unit 122.

[0067] In S102, the learning unit 121 reads out a data group from the database unit 122 and performs model learning by a machine learning algorithm. As a result, a pre-trained model is generated. The pre-trained model is held by the learning unit 121.

[0068] FIG. 7 shows the flow during learning using the data of a specific terminal.

[0069] In S201, the transfer learning control unit 123 extracts the data of the target specific terminal (that is, the data of the terminal close to terminal A) from the data recorded in the database unit 122.

[0070] In S202, the transfer learning control unit 123 designates, to the learning unit 121, the range of the layers of the neural network that is the relearning part for the pre-trained model.

[0071] In S203, the transfer learning control unit 123 inputs the data extracted in S201 into the learning unit 121 in which the relearning part is set. The learning unit 121 executes relearning for the pre-trained model using the data extracted in S201. Thereby, a transfer learning-completed model is generated, and the learning unit 121 holds the model.

[0072] FIG. 8 shows a flow at the time of estimating the radio quality for terminal A.

[0073] In S301, the position information acquisition unit 111 acquires the current position information of terminal A, and the estimated position designation unit 112 selects, from the current position information, the position for which it is desired to estimate the radio quality for terminal A.

[0074] In S302, the radio terminal information acquisition unit 114 acquires the terminal information of terminal A from the radio communication unit 113. Here, it is assumed that the radio communication unit 113 wirelessly acquires the terminal information from terminal A.

[0075] In S303, the radio quality management unit 110 inputs, as inquiry information, the information of the position for which it is desired to estimate the radio quality for terminal A and the terminal information into the radio quality estimation unit 120.

[0076] In S304, the radio quality estimation unit 120 refers to the input terminal information, and estimates the radio quality at the designated position of terminal A using the transfer learning-completed model that has been transfer-learned by a terminal close to terminal A having the terminal information, and answers the estimated value.

[0077] (Hardware configuration example) The devices (terminal, learning device, radio quality estimation device) described in this embodiment can all be realized, for example, by causing a computer to execute a program that describes the processing contents described in this embodiment.

[0078] The above program can be recorded on a computer-readable recording medium (such as a portable memory), saved, or distributed. It is also possible to provide the above program through a network such as the Internet or e-mail.

[0079] FIG. 9 is a diagram showing an example of the hardware configuration of the above computer. The computer in FIG. 9 includes a drive device 1000, an auxiliary storage device 1002, a memory device 1003, a CPU 1004, an interface device 1005, a display device 1006, an input device 1007, an output device 1008, etc., which are mutually connected by a bus B.

[0080] The program that realizes the processing on the computer is provided, for example, by a recording medium 1001 such as a CD-ROM or a memory card. When the recording medium 1001 storing the program is set in the drive device 1000, the program is installed from the recording medium 1001 to the auxiliary storage device 1002 via the drive device 1000. However, the installation of the program does not necessarily have to be performed from the recording medium 1001, and it may be downloaded from another computer via a network. The auxiliary storage device 1002 stores the installed program and also stores necessary files, data, etc.

[0081] When there is an instruction to start a program, the memory device 1003 reads and stores the program from the auxiliary storage device 1002. The CPU 1004 realizes the functions related to the device (learning device, wireless quality estimation device, etc.) according to the program stored in the memory device 1003. The interface device 1005 is used as an interface for connecting to the network. The display device 1006 displays a GUI (Graphical User Interface) etc. according to the program. The input device 1007 is composed of a keyboard, a mouse, buttons, or a touch panel, etc., and is used to input various operation instructions. The output device 1008 outputs the calculation result.

[0082] (Effect of the embodiment) According to the technology related to the present embodiment described above, even when the number of measurement data and performance data obtained under desired conditions is small, it is possible to accurately estimate the wireless quality.

[0083] (Summary of the embodiment) This specification describes at least the learning method, wireless quality estimation method, learning device, wireless quality estimation device, and program described in each of the following items. (Item 1) A learning method in which a learning device learns a model for estimating wireless quality by machine learning, A learning step of generating a pre-trained model by learning the model using measurement data of wireless quality obtained by a plurality of terminals, A re-learning step of re-learning the pre-trained model using measurement data that satisfies a predetermined condition A learning method comprising: (Item 2) In the re-learning step, measurement data of a specific terminal to be the target of wireless quality estimation or measurement data of a terminal close to the specific terminal is extracted from the measurement data of the plurality of terminals, and the re-learning is performed using the extracted measurement data The learning method according to Item 1. (Item 3) The model is a neural network model, and in the relearning step, a range of neurons for performing relearning in the pre-trained model is specified. The learning method according to claim 1 or 2. (Claim 4) A wireless quality estimation method executed by a wireless quality estimation device, comprising: estimating a position of a terminal to be subject to wireless quality estimation; estimating wireless quality at the estimated position of the terminal to be subject to wireless quality estimation, using a model re-learned by the learning method according to any one of claims 1 to 3. A wireless quality estimation method comprising the above steps. (Claim 5) A learning device for learning a model for estimating wireless quality by machine learning, comprising: a learning unit that generates a pre-trained model by learning a model using measurement data of wireless quality obtained by a plurality of terminals; a learning control unit that causes the learning unit to perform re-learning on the pre-trained model using measurement data that satisfies a predetermined condition. A learning device comprising the above components. (Claim 6) a position estimation unit that estimates the position of a terminal to be subject to wireless quality estimation; a wireless quality estimation unit that estimates wireless quality at the estimated position of the terminal to be subject to wireless quality estimation, using a model generated by performing re-learning on a pre-trained model learned using measurement data of wireless quality obtained by a plurality of terminals, using measurement data that satisfies a predetermined condition. A wireless quality estimation device comprising the above components. (Claim 7) A program for causing a computer to function as each unit in the learning device according to claim 5. (Claim 8) A program for causing a computer to function as each unit in the wireless quality estimation device according to claim 6.

[0084] As described above, the present embodiment has been explained. However, the present invention is not limited to such a specific embodiment, and various modifications and changes are possible within the scope of the gist of the present invention described in the claims.

Explanation of Signs

[0085] 100 Wireless quality estimation device 110 Wireless quality management unit 111 Location information acquisition unit 112 Estimated location designation unit 113 Wireless communication unit 114 Wireless terminal information acquisition unit 120 Wireless quality estimation unit 121 Learning unit 122 Database unit 123 Transfer learning control unit 1231 Relearning data selection unit for specific terminals 1232 Relearning layer setting unit 124 Input unit 125 Inquiry information input unit 126 Estimated value output unit 1000 Drive device 1001 Recording medium 1002 Auxiliary storage device 1003 Memory device 1004 CPU 1005 Interface device 1006 Display device 1007 Input device 1008 Output device

Claims

1. A learning method in which a learning device learns a model for estimating radio quality by machine learning, comprising: a learning step of generating a pre-trained model by learning a model using measurement data of radio quality obtained by a plurality of terminals, the measurement data having position information, terminal information, and radio quality information; a re-learning step of extracting the measurement data of a specific terminal targeted for radio quality estimation or the measurement data of a terminal close to the specific terminal from the measurement data of the plurality of terminals, and performing re-learning on the pre-trained model using the extracted measurement data; A learning method comprising the above steps.

2. The model is a neural network model, and in the re-learning step, a range of neurons for performing re-learning in the pre-trained model is specified. The learning method according to Claim 1.

3. A radio quality estimation method executed by a radio quality estimation device, comprising: a step of estimating the position of the specific terminal targeted for radio quality estimation; a step of estimating the radio quality at the estimated position of the specific terminal targeted for radio quality estimation using a model re-learned by the learning method according to Claim 1 or 2. A radio quality estimation method comprising the above steps.

4. A learning device that learns a model for estimating radio quality by machine learning, comprising: a learning unit that generates a pre-trained model by learning a model using measurement data of radio quality obtained by a plurality of terminals, the measurement data having position information, terminal information, and radio quality information; a learning control unit that extracts the measurement data of a specific terminal targeted for radio quality estimation or the measurement data of a terminal close to the specific terminal from the measurement data of the plurality of terminals, and causes the learning unit to perform re-learning on the pre-trained model using the extracted measurement data. A learning device comprising the above components.

5. A position estimation unit that estimates the position of a specific terminal targeted for radio quality estimation; A radio quality estimation device comprising a radio quality estimation unit that estimates the radio quality at the estimated position of the specific terminal targeted for radio quality estimation using a model generated by performing re-learning on a pre-trained model learned using measurement data of radio quality obtained by a plurality of terminals, the measurement data having position information, terminal information, and radio quality information, using the measurement data that satisfies a predetermined condition. The measurement data satisfying the predetermined conditions is the measurement data of the specific terminal targeted for radio quality estimation or the measurement data of a terminal close to the specific terminal, which is extracted from the measurement data of the plurality of terminals. Radio quality estimation device. **Claim 6** A program for causing a computer to function as each part in the learning device according to claim 4. **Claim 7** A program for causing a computer to function as each part in the radio quality estimation device according to claim 5.

Images