Model building device and model building method, factor estimation device and factor estimation method, and recording medium

Abstract

A model construction device (4) comprises: an acquisition means (411) for acquiring, from a first wireless station (1), time series data (SRS) on the reception power (RP) of wireless communications between the first wireless station and a second wireless station (2); a generation means (412) for generating, on the basis of the time series data, input data (INP) including change data relating to the degree of change in the reception power per unit time; and a construction means (413) for constructing, by machine learning using the input data, an estimation model for estimating a degradation factor from the input data.

Description

[Technical Field]

[0001] This disclosure relates to, for example, the technical fields of a model construction device, a model construction method, and a recording medium capable of constructing an estimation model for estimating factors of degradation of communication quality between a first wireless station and a second wireless station, as well as a factor estimation device, a factor estimation method, and a recording medium capable of estimating factors of degradation of communication quality between a first wireless station and a second wireless station. [Background technology]

[0002] Patent Document 1 describes a quality estimation device that uses a neural network model to estimate communication quality in wireless communication (specifically, mobile communication). Other prior art documents related to this disclosure include Patent Documents 2 to 4. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 6696859 [Patent Document 2] International Publication No. 2020 / 217457 Brochure [Patent Document 3] Japanese Patent Application Laid-Open No. 2017-123124 [Patent Document 4] Japanese Patent Application Laid-Open No. 2008-278148 Summary of the Invention [Problem to be solved by the invention]

[0004] When communication quality deteriorates, it is desirable to take measures to restore the deteriorated communication quality. Here, the measures to restore the deteriorated communication quality often vary depending on the cause of the deterioration of the communication quality. However, the quality estimation device described in Patent Document 1 has a technical problem in that it cannot estimate the cause of the deterioration of the communication quality.

[0005] An object of the present disclosure is to provide a model construction device, a model construction method, a factor estimation device, a factor estimation method, and a recording medium that can solve the above-mentioned technical problems. As an example, an object of the present disclosure is to provide a model construction device, a model construction method, and a recording medium that can construct an estimation model for estimating factors of degradation of communication quality between a first wireless station and a second wireless station, and a factor estimation device, a factor estimation method, and a recording medium that can estimate factors of degradation of communication quality between the first wireless station and the second wireless station. [Means for solving the problem]

[0006] One aspect of the model construction device is a model construction device that constructs an estimation model for estimating factors of degradation of communication quality between a first wireless station and a second wireless station, and includes: an acquisition means that acquires time series data of received power of wireless communication between the first wireless station and the second wireless station from the first wireless station; a generation means that generates input data including change data regarding the degree of change in the received power per unit time based on the time series data; and a construction means that constructs an estimation model for estimating the factors of degradation from the input data by machine learning using the input data.

[0007] One aspect of the factor estimation device is a factor estimation device that estimates a factor of degradation of communication quality between a first wireless station and a second wireless station, and includes: an acquisition means that acquires time series data of received power of wireless communication between the first wireless station and the second wireless station from the first wireless station; a generation means that generates input data including change data regarding the degree of change in the received power per unit time based on the time series data; an estimation model that is constructed by machine learning and that estimates the degradation factor from the input data; and an estimation means that estimates the degradation factor using the input data.

[0008] One aspect of the model construction method is a model construction method for constructing an estimation model for estimating factors of degradation of communication quality between a first wireless station and a second wireless station, which method includes acquiring time series data of received power of wireless communication between the first wireless station and the second wireless station from the first wireless station, generating input data including change data regarding the degree of change in the received power per unit time based on the time series data, and constructing an estimation model for estimating the factors of degradation from the input data by machine learning using the input data.

[0009] One aspect of the factor estimation method is a factor estimation method for estimating a factor of degradation of communication quality between a first wireless station and a second wireless station, which includes acquiring time series data of received power of wireless communication between the first wireless station and the second wireless station from the first wireless station, generating input data including change data regarding the degree of change in the received power per unit time based on the time series data, and estimating the degradation factor using the input data and an estimation model constructed by machine learning for estimating the degradation factor from the input data.

[0010] A first aspect of the recording medium is a recording medium having recorded thereon a computer program that causes a computer to execute a model construction method for constructing an estimation model for estimating factors of degradation of communication quality between a first wireless station and a second wireless station, the model construction method acquiring time series data of received power of wireless communication between the first wireless station and the second wireless station from the first wireless station, generating input data including change data regarding the degree of change in the received power per unit time based on the time series data, and constructing an estimation model for estimating factors of degradation from the input data through machine learning using the input data.

[0011] A second aspect of the recording medium is a recording medium having recorded thereon a computer program that causes a computer to execute a factor estimation method for estimating factors of degradation of communication quality between a first wireless station and a second wireless station, the factor estimation method acquiring time series data of received power of wireless communication between the first wireless station and the second wireless station from the first wireless station, generating input data including change data regarding the degree of change in the received power per unit time based on the time series data, and executing the factor estimation method for estimating the degradation factors using an estimation model constructed by machine learning and for estimating the degradation factors from the input data and the input data. [Effects of the Invention]

[0012] According to the above-described aspects of the model construction device, model construction method, and recording medium, it is possible to construct an estimation model for estimating factors of degradation of communication quality between a first wireless station and a second wireless station. Also, according to the above-described aspects of the factor estimation device, factor estimation method, and recording medium, it is possible to estimate factors of degradation of communication quality between the first wireless station and the second wireless station. [Brief explanation of the drawings]

[0013] [Figure 1] FIG. 1 is a block diagram showing the overall configuration of a wireless communication system according to this embodiment. [Figure 2] FIG. 2 is a block diagram showing the configuration of the factor estimating device of this embodiment. [Figure 3] FIG. 3 is a block diagram showing the configuration of the model building device of this embodiment. [Figure 4] FIG. 4 is a flowchart showing the flow of the model construction operation performed by the model construction device. [Figure 5] FIG. 5 shows an example of the data structure of the training dataset. [Figure 6]Figure 6(a) is a graph showing the change in RSRP (Reference Signal Received Power) over time when communication quality begins to deteriorate due to distance attenuation, and Figure 6(b) is a graph showing the change in RSRP over time when communication quality begins to deteriorate due to obstruction. [Figure 7] Figure 7(a) is a graph showing the change in RSRP over time when communication quality that has deteriorated due to distance attenuation begins to recover, and Figure 7(b) is a graph showing the change in RSRP over time when communication quality that has deteriorated due to obstruction begins to recover. [Figure 8] Figure 8(a) is a graph showing an approximation curve that approximates RSRP when communication quality begins to deteriorate due to distance attenuation, and Figure 8(b) is a graph showing an approximation curve that approximates RSRP when communication quality begins to deteriorate due to obstruction. [Figure 9] Each of Figures 9(a) to 9(f) is a graph showing classes used to classify trends in RSRP changes over time. [Figure 10] FIG. 10 shows an example of the data structure of the input data. [Figure 11] Figure 11 shows the estimation results of the estimation model. [Figure 12] FIG. 12 is a flowchart showing the flow of the factor estimation operation performed by the factor estimation device. [Figure 13] FIG. 13 is a block diagram showing the configuration of a factor estimating device that can also function as a model building device. DETAILED DESCRIPTION OF THE INVENTION

[0014] Hereinafter, embodiments of a model construction device and a model construction method, a factor estimation device and a factor estimation method, and a recording medium will be described with reference to the drawings. Hereinafter, embodiments of a model construction device and a model construction method, a factor estimation device and a factor estimation method, and a recording medium will be described using a wireless communication system SYS to which the embodiments of the model construction device and the model construction method, the factor estimation device and the factor estimation method, and the recording medium are applied. However, the present invention is not limited to the embodiments described below. <1> Wireless communication system SYS configuration First, the configuration of the wireless communication system SYS of this embodiment will be described. <1-1> Overall configuration of the wireless communication system SYS

[0015] First, the overall configuration of the wireless communication system SYS of this embodiment will be described with reference to Fig. 1. Fig. 1 is a block diagram showing the overall configuration of the wireless communication system SYS of this embodiment.

[0016] As shown in FIG. 1, the wireless communication system SYS includes a wireless base station 1 which is a specific example of a "first wireless station," a user terminal 2 which is a specific example of a "second wireless station," a factor estimation device 3, and a model construction device 4. In the example shown in FIG. 1, the wireless communication system SYS includes a single wireless base station 1. However, the wireless communication system SYS may include multiple wireless base stations 1. Also, in the example shown in FIG. 1, the wireless communication system SYS includes multiple (specifically, N, where N is a constant indicating an integer equal to or greater than 2) user terminals 2. In the following description, as necessary, a user terminal 2 among the N user terminals 2 will be referred to as a "user terminal 2#n" (where n is a variable indicating an integer equal to or greater than 1 and equal to or less than the total number N of user terminals 2). However, the wireless communication system SYS may include a single user terminal 2.

[0017] The wireless base station 1 is capable of wireless communication with at least one user terminal 2 present within a cell of a predetermined size that expands from the wireless base station 1. The wireless base station 1 is capable of wireless communication with the user terminal 2 by transmitting radio waves to the user terminal 2 and receiving radio waves transmitted from the user terminal 2.

[0018] The user terminal 2 is capable of wireless communication with the radio base station 1 that covers the cell in which the user terminal 2 is located. The user terminal 2 is capable of wireless communication with the radio base station 1 by transmitting radio waves to the radio base station 1 and receiving radio waves transmitted from the radio base station 1. The user terminal 2 may also be referred to as a radio terminal.

[0019] The factor estimation device 3 is capable of performing a factor estimation operation to estimate a factor of deterioration in the communication quality between the wireless base station 1 and the user terminal 2#n (i.e., the quality of wireless communication between the wireless base station 1 and the user terminal 2#n).

[0020] The communication quality may include, for example, the throughput. In this case, the state in which the communication quality is degraded may include a state in which the throughput is lower than the allowable lower limit. Therefore, the factor estimation device 3 may perform a factor estimation operation to estimate the factor of the degradation of the throughput between the wireless base station 1 and the user terminal 2#n as the factor of the degradation of the communication quality.

[0021] The communication quality may include, for example, delay (i.e., delay time). In this case, a state in which the communication quality is degraded may include a state in which the delay is greater than the allowable upper limit (i.e., the delay time is longer than the allowable upper limit). Therefore, the factor estimation device 3 may perform a factor estimation operation to estimate the factor of the large delay between the wireless base station 1 and the user terminal 2#n as a factor of the degradation of the communication quality.

[0022] In this embodiment, the factor estimation device 3 estimates factors of degradation of communication quality using an estimation model M constructed by machine learning. An example of such an estimation model M is a model including a neural network. The estimation model M is a model that can output an estimation result of factors of degradation of communication quality between the wireless base station 1 and the user terminal 2#n when input data INP#n generated from time-series data SRS#n of received power RP#n of wireless communication performed between the wireless base station 1 and the user terminal 2#n is input. Therefore, the factor estimation device 3 acquires time-series data SRS#n of received power of wireless communication performed between the wireless base station 1 and the user terminal 2#n, generates input data INP#n based on the acquired time-series data SRS#n, and estimates factors of degradation of communication quality between the wireless base station 1 and the user terminal 2#n using the generated input data INP#n and the estimation model M.

[0023] Note that the "received power RP#n of the wireless communication performed between the wireless base station 1 and the user terminal 2#n" in this embodiment may refer to the strength of the radio waves transmitted from the wireless base station 1 at the location where the user terminal 2#n is located (i.e., the strength of the radio waves received from the wireless base station 1 by the antenna provided in the user terminal 2#n). In other words, the received power RP#n may refer to the received power in the downlink direction. Such received power RP#n is typically measured by the user terminal 2#n receiving the radio waves transmitted from the wireless base station 1. Furthermore, information on the received power measured by the user terminal 2#n may be transmitted from the user terminal 2#n to the wireless base station 1. In other words, information on the received powers RP#1 to RP#N measured by each of the N user terminals 2 may be transmitted from each of the N user terminals 2 to the wireless base station 1. In this case, the factor estimation device 3 may acquire at least one of the time series data SRS#1 to SRS#N of the received powers RP#1 to RP#N from the wireless base station 1 via the communication network 5. The communication network 5 may include a wireless network or a wired network. In this embodiment, the "received power RP#n of the wireless communication performed between the wireless base station 1 and the user terminal 2#n" may refer to the strength of the wireless radio wave transmitted from the user terminal 2#n at the location where the wireless base station 1 is located (i.e., the strength of the wireless radio wave received from the user terminal 2#n by an antenna provided in the wireless base station). In other words, the received power RP#n may refer to the received power in the uplink direction. Such received power RP#n is typically measured by the wireless base station 1 that receives the wireless radio wave transmitted from the user terminal 2#n. In other words, the wireless base station 1 measures the received powers RP#1 to RP#N by receiving the wireless radio waves transmitted from the user terminals 2#1 to 2#N. In this case, too, the factor estimation device 3 may acquire at least one of the time series data SRS#1 to SRS#N of the received power RP#1 from the wireless base station 1 via the communication network 5.

[0024] In this embodiment, an example will be described in which the wireless communication system SYS is a 3.9th generation mobile communication system (so-called 3.9G, a mobile communication system conforming to the LTE (Long Term Evolution) standard), a 4th generation mobile communication system (so-called 4G), or a 5th generation mobile communication system (so-called 5G). In this case, at least one of RSSI (Received Signal Strength Indicator), RSRP (Reference Signal Received Power), RSRQ (Reference Signal Received Quality), and SINR (Signal to Interference plus Noise Ratio) may be used as the received power RP#n. However, the wireless communication system SYS is not limited to a 3.9th generation mobile communication system, a 4th generation mobile communication system, or a 5th generation mobile communication system.

[0025] The model construction device 4 is capable of performing a model construction operation for constructing an estimation model M used by the factor estimation device 3 to estimate degradation of communication quality. As described above, the estimation model M is constructed by machine learning. Therefore, the model construction device 4 constructs the estimation model M by performing machine learning.

[0026] As described above, the estimation model M is a model that can output an estimation result of a factor in degradation of communication quality between the wireless base station 1 and the user terminal 2#n when input data INP#n generated from time-series data SRS#n of received power RP#n of wireless communication performed between the wireless base station 1 and the user terminal 2#n is input. For this reason, the model construction device 4 performs machine learning to construct the estimation model M using a learning dataset LDS that includes time-series data SRS#n of received power RP#n of wireless communication performed between the wireless base station 1 and the user terminal 2#n and learning data LD#n that includes ground truth labels y#n that indicate actual factors in degradation of communication quality between the wireless base station 1 and the user terminal 2#n under conditions in which the time-series data SRS#n was observed. Specifically, the model construction device 4 generates input data INP#n based on the time-series data SRS#n included in the learning data LD#n, inputs the generated input data INP#n to the estimation model M, and updates the parameters of the estimation model M based on a loss function related to the error between the correct label y#n included in the learning data LD#n and the output of the estimation model M. Note that when the estimation model M is a model including a neural network, an example of the parameters of the estimation model M is at least one of a bias and a weight.

[0027] The learning data set LDS may include not only learning data LD#n including time-series data SRS#n observed under conditions in which the communication quality of the wireless communication between the wireless base station 1 and the user terminal 2#n is degraded, but also learning data LD#n including time-series data SRS#n observed under conditions in which the communication quality of the wireless communication between the wireless base station 1 and the user terminal 2#n is not degraded. In this case, the correct label y#n included in the learning data LD#n may indicate that the communication quality is not degraded, instead of indicating the cause of the degradation of the communication quality.

[0028] The model construction device 4 may acquire time-series data SRS#n of received power RP#n from the wireless base station 1 via the communication network 5, and use the acquired time-series data SRS#n as part of the learning data LD#n. In this case, the model construction device 4 may acquire information about the correct label y#n in parallel with acquiring the time-series data SRS#n. Alternatively, the model construction device 4 may acquire a learning data set LDS prepared in advance. <1-2> Configuration of factor estimation device 3 Next, the configuration of the factor estimating device 3 will be described with reference to Fig. 2. Fig. 2 is a block diagram showing the configuration of the factor estimating device 3.

[0029] 2, the factor estimation device 3 includes a calculation device 31, a storage device 32, and a communication device 33. The factor estimation device 3 may further include an input device 34 and an output device 35. However, the factor estimation device 3 does not necessarily include at least one of the input device 34 and the output device 35. The calculation device 31, the storage device 32, the communication device 33, the input device 34, and the output device 35 may be connected via a data bus 36.

[0030] The arithmetic device 31 includes, for example, at least one of a central processing unit (CPU), a graphics processing unit (GPU), and a field programmable gate array (FPGA). The arithmetic device 31 loads a computer program. For example, the arithmetic device 31 may load a computer program stored in the storage device 32. For example, the arithmetic device 31 may load a computer program stored in a computer-readable, non-transitory storage medium using a storage medium reading device (not shown) included in the factor estimation device 3. The arithmetic device 31 may acquire (i.e., download or load) the computer program from a device (not shown) located outside the factor estimation device 3 via the communication device 33 (or another communication device). The arithmetic device 31 executes the loaded computer program. As a result, logical functional blocks for executing the operations to be performed by the factor estimation device 3 (for example, the above-mentioned factor estimation operation) are realized within the arithmetic device 31. That is, the arithmetic device 31 can function as a controller for realizing logical functional blocks for executing the operations (in other words, processing) that the factor estimating device 3 should perform.

[0031] 2 shows an example of logical functional blocks implemented in the calculation device 31 for performing the factor estimation operation. As shown in FIG. 2, the calculation device 31 implements a data acquisition unit 311, which is a specific example of an "acquisition means," a data aggregation unit 312, which is a specific example of a "generation means," and a factor estimation unit 313, which is a specific example of an "estimation means." While the operations of the data acquisition unit 311, the data aggregation unit 312, and the factor estimation unit 313 will be described in detail later, a brief overview will be provided here. The data acquisition unit 311 acquires time-series data SRS#n of received power RP#n from the wireless base station 1. The data aggregation unit 312 generates input data INP#n based on the time-series data SRS#n of received power RP#n acquired by the data acquisition unit 311. The factor estimation unit 313 estimates a factor of degradation in communication quality between the wireless base station 1 and the user terminal 2#n using the input data INP#n generated by the data aggregation unit 312 and an estimation model M.

[0032] The storage device 32 can store desired data. For example, the storage device 32 may temporarily store a computer program executed by the arithmetic device 31. The storage device 32 may temporarily store data that the arithmetic device 31 temporarily uses when the arithmetic device 31 is executing a computer program. The storage device 32 may store data that the factor estimation device 3 stores long-term. The storage device 32 may include at least one of a RAM (Random Access Memory), a ROM (Read Only Memory), a hard disk device, a magneto-optical disk device, an SSD (Solid State Drive), and a disk array device. In other words, the storage device 32 may include a non-temporary recording medium.

[0033] The communication device 33 is capable of communicating with the wireless base station 1 via the communication network 5. In this embodiment, the communication device 33 receives (i.e., acquires) time-series data SRS#n of received power RP#n from the wireless base station 1 via the communication network 5.

[0034] The input device 34 is a device that accepts information input to the factor estimation device 3 from outside the factor estimation device 3. For example, the input device 34 may include an operation device (for example, at least one of a keyboard, a mouse, and a touch panel) that can be operated by an operator of the factor estimation device 3. For example, the input device 34 may include a reading device that can read information recorded as data on a recording medium that can be externally attached to the factor estimation device 3.

[0035] The output device 35 is a device that outputs information to the outside of the factor estimation device 3. For example, the output device 35 may output information as an image. That is, the output device 35 may include a display device (a so-called display) that can display an image showing the information to be output. For example, the output device 35 may output information as sound. That is, the output device 35 may include an audio device (a so-called speaker) that can output sound. For example, the output device 35 may output information on paper. That is, the output device 35 may include a printing device (a so-called printer) that can print desired information on paper. <1-3> Configuration of model construction device 4 Next, the configuration of the model construction device 4 will be described with reference to Fig. 3. Fig. 3 is a block diagram showing the configuration of the model construction device 4.

[0036] 3, the model construction device 4 includes a calculation device 41, a storage device 42, and a communication device 43. The model construction device 4 may further include an input device 44 and an output device 45. However, the model construction device 4 does not necessarily have to include at least one of the input device 44 and the output device 45. The calculation device 41, the storage device 42, the communication device 43, the input device 44, and the output device 45 may be connected via a data bus 46.

[0037] The arithmetic device 41 includes, for example, at least one of a CPU, a GPU, and an FPGA. The arithmetic device 41 loads a computer program. For example, the arithmetic device 41 may load a computer program stored in the storage device 42. For example, the arithmetic device 41 may load a computer program stored in a computer-readable, non-transitory storage medium using a storage medium reading device (not shown) included in the model construction device 4. The arithmetic device 41 may acquire (i.e., download or load) the computer program from a device (not shown) located outside the model construction device 4 via the communication device 43 (or another communication device). The arithmetic device 41 executes the loaded computer program. As a result, logical functional blocks for executing operations to be performed by the model construction device 4 (e.g., the model construction operations described above) are realized within the arithmetic device 41. In other words, the arithmetic device 41 can function as a controller for realizing logical functional blocks for executing operations (in other words, processing) to be performed by the model construction device 4.

[0038] FIG. 3 shows an example of logical functional blocks implemented within the arithmetic device 41 for executing the model construction operation. As shown in FIG. 3, the arithmetic device 41 implements a data acquisition unit 411, which is a specific example of an "acquisition means," a data aggregation unit 412, which is a specific example of a "generation means," and a model construction unit 413, which is a specific example of a "construction means." While the operations of the data acquisition unit 411, the data aggregation unit 412, and the model construction unit 413 will be described in detail later, a brief overview will be provided here. The data acquisition unit 411 acquires a learning dataset LDS. The data aggregation unit 412 generates at least one of input data INP#1 to INP#N based on at least one of time-series data SRS#1 to SRS#N of received power RP#1 included in the learning dataset LDS acquired by the data acquisition unit 411. The model construction unit 413 performs machine learning to construct an estimation model M using at least one of the input data INP#1 to INP#N generated by the data aggregation unit 412 and at least one of the correct labels y#1 to y#N included in the learning dataset LDS.

[0039] The storage device 42 can store desired data. For example, the storage device 42 may temporarily store a computer program executed by the arithmetic device 41. The storage device 42 may temporarily store data that the arithmetic device 41 temporarily uses when the arithmetic device 41 is executing a computer program. The storage device 42 may store data that the model construction device 4 stores long-term. The storage device 42 may include at least one of a RAM, a ROM, a hard disk device, a magneto-optical disk device, an SSD, and a disk array device. In other words, the storage device 42 may include a non-temporary recording medium.

[0040] The communication device 43 is capable of communicating with the wireless base station 1 via the communication network 5. In the present embodiment, the communication device 43 may receive (i.e., acquire) at least one of the time series data SRS#1 of received power RP#1 to the time series data SRS#N of received power RP#N from the wireless base station 1 via the communication network 5.

[0041] The input device 44 is a device that accepts information input to the model construction device 4 from outside the model construction device 4. For example, the input device 44 may include an operation device (for example, at least one of a keyboard, a mouse, and a touch panel) that can be operated by an operator of the model construction device 4. For example, the input device 44 may include a reading device that can read information recorded as data on a recording medium that can be externally attached to the model construction device 4.

[0042] The output device 45 is a device that outputs information to the outside of the model construction device 4. For example, the output device 45 may output information as an image. That is, the output device 45 may include a display device (a so-called display) that can display an image showing the information to be output. For example, the output device 45 may output information as sound. That is, the output device 45 may include an audio device (a so-called speaker) that can output sound. For example, the output device 45 may output information on paper. That is, the output device 45 may include a printing device (a so-called printer) that can print desired information on paper. <2> Operations performed in the wireless communication system SYS

[0043] Next, the operations performed in the wireless communication system SYS will be described. As described above, in the wireless communication system SYS, the model construction device 4 performs a model construction operation to construct an estimation model M. Furthermore, in the wireless communication system SYS, the factor estimation device 3 performs a factor estimation operation to estimate a factor of degradation in communication quality, using the estimation model M constructed by the model construction operation. Therefore, the model construction operation and the factor estimation operation will be described in order below. <2-1> Model construction operation performed by the model construction device 4

[0044] First, the model construction operation performed by the model construction device 4 will be described with reference to Fig. 4. Fig. 4 is a flowchart showing the flow of the model construction operation performed by the model construction device 4. As shown in FIG. 4, the data acquisition unit 411 acquires a learning dataset LDS (step S11).

[0045] For example, the data acquiring unit 411 may acquire time-series data SRS#1 of received power RP#1 to time-series data SRS#N of received power RP#N from the wireless base station 1 via the communication network 5 as part of the learning data LD#1 to LD#N (i.e., part of the learning data set LDS). If the wireless base station 1 does not hold at least one of the time-series data SRS#1 to SRS#N, the data acquiring unit 411 may not acquire at least one of the time-series data SRS#1 to SRS#N that the wireless base station 1 does not hold. Furthermore, in parallel with acquiring the time-series data SRS#1 to SRS#N, the data acquiring unit 411 may acquire information on the correct labels y#1 to y#N as another part of the learning data LD#1 to #N (i.e., another part of the learning data set LDS). The correct labels y#1 to y#N may be input to the model construction device 4 by an operator of the model construction device 4.

[0046] The data acquisition unit 411 preferably acquires the time-series data SRS#1 to SRS#N from the radio base station 1 under circumstances where the radio base station 1 and the N user terminals 2 are in the same environment as the implementation environment of the radio base station 1 and the N user terminals 2 when the factor estimation device 3 actually estimates degradation factors. In other words, the data acquisition unit 411 preferably acquires the time-series data SRS#1 to SRS#N from the radio base station 1 under circumstances where the radio base station 1 and the N user terminals 2 are installed in a location where they are actually used. However, the data acquisition unit 411 may also acquire the time-series data SRS#1 to SRS#N from the radio base station 1 under circumstances where the radio base station 1 and the N user terminals 2 are installed in a location different from the location where they are actually used.

[0047] For example, the data acquiring unit 411 may acquire a training dataset LDS that has been prepared in advance. For example, if the training dataset LDS is stored in the storage device 42, the data acquiring unit 411 may acquire the training dataset LDS from the storage device 42. For example, if the training dataset LDS is recorded on a recording medium that can be attached externally to the model construction device 4, the data acquiring unit 411 may acquire the training dataset LDS from the recording medium using a recording medium reading device (e.g., the input device 44) provided in the model construction device 4. For example, if the training dataset LDS is recorded on a device (e.g., a server) external to the model construction device 4, the data acquiring unit 411 may acquire the training dataset LDS from the external device using the communication device 43.

[0048] The previously prepared learning data set LDS preferably includes time series data SRS#1 to SRS#N acquired under a situation where the radio base station 1 and N user terminals 2 are installed in a location where they are actually used. However, the previously prepared learning data set LDS may also include time series data SRS#1 to SRS#N acquired under a situation where the radio base station 1 and N user terminals 2 are installed in a location different from the location where they are actually used.

[0049] An example of the data structure of the training data set LDS is shown in Fig. 5. As shown in Fig. 5, the training data set LDS includes at least one training data LD#1, at least one training data LD#2, ..., at least one training data LD#N. However, the training data set LDS does not necessarily include at least one of the training data LD#1 to the training data LD#N.

[0050] Each learning data LD#n includes time-series data SRS#n of received power RP#n of wireless communication between the wireless base station 1 and the user terminal 2#n. Furthermore, each learning data LD#n includes a ground truth label y#n indicating the actual cause of degradation in communication quality between the wireless base station 1 and the user terminal 2#n under the conditions in which the time-series data SRS#n included in each learning data LD#n was observed.

[0051] 5, the training data set LDS includes training data LD#1 and training data LD#2. The training data LD#1 includes time-series data SRS#1 of received power RP#1 observed between time t#1-1 and time t#1-23. Furthermore, the training data LD#1 includes a correct answer label y#1 indicating an actual cause of degradation in communication quality between the radio base station 1 and the user terminal 2#1 between time t#1-1 and time t#1-23. Specifically, the training data LD#1 includes a correct answer label y#1 indicating that communication quality between the radio base station 1 and the user terminal 2#1 does not degrade between time t#1-1 and time t#1-11, and indicating that the actual cause of degradation in communication quality between the radio base station 1 and the user terminal 2#1 is "blocking" between time t#1-12 and time t#1-23. Furthermore, the learning data LD#2 includes time-series data SRS#2 of received power RP#2 observed between time t#2-1 and time t#2-36. Furthermore, the learning data LD#2 includes a correct answer label y#2 indicating an actual cause of degradation in communication quality between the wireless base station 1 and the user terminal 2#2 between time t#2-1 and time t#2-36. Specifically, the learning data LD#2 includes a correct answer label y#2 indicating that an actual cause of degradation in communication quality between the wireless base station 1 and the user terminal 2#2 between time t#2-1 and time t#2-21 is "fading," and indicating that an actual cause of degradation in communication quality between the wireless base station 1 and the user terminal 2#2 between time t#2-22 and time t#2-36 is "distance attenuation." In the example shown in FIG. 5, the correct label y#n in the learning data set LDS indicates a single degradation factor. However, the communication quality between the wireless base station 1 and the user terminal 2#n may be degraded due to multiple degradation factors. For example, the communication quality between the wireless base station 1 and the user terminal 2#n may be degraded due to fading and distance attenuation. Therefore, the correct label y#n may indicate multiple degradation factors as the actual degradation factors of the communication quality between the wireless base station 1 and the user terminal 2#n.

[0052] Note that the "state in which communication quality is degraded due to obstruction" may include a state in which communication quality is degraded due to the presence of a physical obstacle between the radio base station 1 and the user terminal 2 that blocks radio waves (i.e., prevents radio waves from propagating). Obstacles may include structures such as buildings and vehicles, or topographical features such as mountains and hills. The "state in which communication quality is degraded due to fading" may include a state in which communication quality is degraded due to the occurrence of at least one of coherent fading (so-called multipath), polarization fading, jump fading, absorption fading, selective fading, and K-shaped fading. The "state in which communication quality is degraded due to distance attenuation" may include a state in which communication quality is degraded due to the distance between the radio base station 1 and the user terminal 2 being longer than the appropriate distance for proper radio communication.

[0053] 4 again, thereafter, the data aggregation unit 412 generates input data INP#n based on each time-series data SRS#n included in the learning data set LDS acquired in step S11 (step S12). The input data INP#n is also data that the factor estimation device 3 generates based on the time-series data SRS#n as data to be input to the estimation model M when the factor estimation device 3 estimates a factor of degradation in communication quality. Therefore, the data aggregation unit 412 generates the input data INP#n based on the time-series data SRS#n, just like the factor estimation device 3. Therefore, the description regarding the generation of the input data INP#n in the model construction operation can also be used as the description regarding the generation of the input data INP#n in the factor estimation operation.

[0054] For example, the learning data set LDS includes time-series data SRS#n indicating instantaneous values ​​of received power RP#n at each time t. In this case, the data aggregation unit 412 may calculate, at each time t, an average value of received power RP#n during a certain period (e.g., 100 milliseconds) that includes each time t, and generate input data INP#n including information data related to the calculated average value. For example, the data aggregation unit 412 may generate input data INP#n including data related to an average RSSI, which is an example of received power RP#n. For example, the data aggregation unit 412 may generate input data INP#n including data related to an average RSRP, which is an example of received power RP#n. For example, the data aggregation unit 412 may generate input data INP#n including data related to an average RSRQ, which is an example of received power RP#n. For example, the data aggregation unit 412 may generate input data INP#n including data related to an average SINR, which is an example of received power RP#n.

[0055] Alternatively, the data aggregation unit 412 may generate input data INP#n that includes the time-series data SRS#n itself included in the learning data set LDS. That is, the data aggregation unit 412 may use the time-series data SRS#n itself as at least a part of the input data INP#n. For example, if the learning data set LDS includes time-series data SRS#n that indicates the average value of the received power RP#n at each time t, the data aggregation unit 412 may generate input data INP#n that includes the time-series data SRS#n itself of the average value of the received power RP#n included in the learning data set LDS. In this embodiment, a moving average value is typically used as the average value, but in addition to or instead of the moving average value, at least one of an arithmetic mean value (i.e., arithmetic mean value), a geometric mean value (i.e., geometric mean value), a weighted mean value, a harmonic mean value, and a logarithmic mean value may be used as the average value.

[0056] Particularly in this embodiment, the data aggregation unit 412 generates input data INP#n including change data relating to the degree of change per unit time of RSRP, which is an example of received power RP#n. When the input data INP#n including change data relating to the degree of change per unit time of RSRP is used, an estimation model M can be constructed that can more accurately estimate whether the cause of degradation in communication quality is distance attenuation or obstruction, compared to when the input data INP#n including change data relating to the degree of change per unit time of RSRP is not used. In other words, when the estimation model M is constructed using the input data INP#n including change data relating to the degree of change per unit time of RSRP, the model construction device 4 can construct an estimation model M that can more accurately estimate whether the cause of degradation in communication quality is distance attenuation or obstruction, compared to when the estimation model M is constructed without using the input data INP#n including change data relating to the degree of change per unit time of RSRP. As a result, when the factor of communication quality degradation is estimated using input data INP#n including variation data regarding the degree of change in RSRP per unit time, the factor estimation device 3 can more accurately estimate whether the factor of communication quality degradation is distance attenuation or shielding, compared to when the factor of communication quality degradation is estimated without using input data INP#n including variation data regarding the degree of change in RSRP per unit time. The reason for this will be explained below with reference to Figures 6(a) to 6(b) and Figures 7(a) to 7(b). In the following, an example will be described in which rate data relating to the rate of change of RSRP is used as the change data. However, data relating to any index value that is different from the rate of change of RSRP but that represents the degree of change of RSRP per unit time may also be used as the change data.

[0057] FIG. 6(a) is a graph showing the change in RSRP over time when communication quality begins to deteriorate due to distance attenuation. Meanwhile, FIG. 6(b) is a graph showing the change in RSRP over time when communication quality begins to deteriorate due to obstruction. When communication quality begins to deteriorate due to distance attenuation, it is highly likely that communication quality will deteriorate relatively slowly due to the gradual increase in the distance between the wireless base station 1 and the user terminal 2. On the other hand, when communication quality begins to deteriorate due to obstruction, it is highly likely that communication quality will deteriorate relatively rapidly due to a physical obstacle that appears between the wireless base station 1 and the user terminal 2. As a result, as shown in FIGS. 6(a) and 6(b), when communication quality begins to deteriorate due to distance attenuation, the rate of change in RSRP (in this case, the rate of decrease) is slower than when communication quality begins to deteriorate due to obstruction. In other words, the degree of change in RSRP per unit time is smaller. Specifically, because RSRP decreases relatively slowly in accordance with the relatively gradual deterioration in communication quality, the rate of decrease in RSRP is relatively slow. In other words, when communication quality begins to deteriorate due to obstruction, the rate of change in RSRP (in this case, the rate of decrease) is faster than when communication quality begins to deteriorate due to distance attenuation. That is, the degree of change in RSRP per unit time is greater. Specifically, because RSRP decreases relatively rapidly in response to a relatively rapid deterioration in communication quality, the rate of decrease in RSRP is relatively fast. Therefore, the rate of change in RSRP (in this case, the rate of decrease) can be used as an index value for estimating whether the cause of deterioration in communication quality is distance attenuation or obstruction.

[0058] 7(a) is a graph showing the change in RSRP over time when communication quality that has deteriorated due to distance attenuation starts to recover. Meanwhile, FIG. 7(b) is a graph showing the change in RSRP over time when communication quality that has deteriorated due to obstruction starts to recover. When communication quality that has deteriorated due to distance attenuation starts to recover, it is highly likely that the communication quality will recover relatively slowly due to the gradual shortening of the distance between the wireless base station 1 and the user terminal 2. On the other hand, when communication quality that has deteriorated due to obstruction starts to recover, it is highly likely that the communication quality will recover relatively rapidly due to the disappearance of a physical obstacle that existed between the wireless base station 1 and the user terminal 2. As a result, as shown in FIGS. 7(a) and 7(b), when communication quality that has deteriorated due to distance attenuation starts to recover, the rate of change in RSRP (in this case, the rate of increase) is slower than when communication quality that has deteriorated due to obstruction starts to recover. In other words, the degree of change in RSRP per unit time is smaller. Specifically, since RSRP increases relatively slowly as communication quality recovers relatively slowly, the rate at which RSRP increases is relatively slow. In other words, when communication quality that has deteriorated due to obstruction begins to recover, the rate at which RSRP changes (in this case, the rate of increase) is faster than when communication quality that has deteriorated due to distance attenuation begins to recover. In other words, the degree of change in RSRP per unit time is greater. Specifically, since RSRP increases relatively rapidly as communication quality recovers relatively rapidly, the rate at which RSRP increases is relatively fast. Therefore, the rate at which RSRP changes (in this case, the rate of increase) can be used as an index value for estimating whether the cause of communication quality deterioration is distance attenuation or obstruction.

[0059] The data aggregation unit 412 may calculate an arbitrary index value capable of representing the rate of change of RSRP included as time-series data SRS#n in the learning data LD#n, and generate input data INP#n including data related to the calculated index value as rate data. Specifically, the data aggregation unit 412 may calculate an arbitrary index value capable of distinguishing between a state in which the rate of change of RSRP included in the learning data LD#n is relatively fast and a state in which the rate of change of RSRP included in the learning data LD#n is relatively slow, and generate input data INP#n including data related to the calculated index value as rate data. An example of rate data will be described below.

[0060] The data aggregation unit 412 may calculate coefficients of an approximation equation that approximates the RSRP included in the learning data LD#n as time-series data SRS#n (specifically, that approximates the change over time in the RSRP), and generate input data INP#n that includes data related to the calculated coefficients as speed data. For example, FIG. 8(a) shows a quadratic polynomial "a1×t" that approximates the RSRP when communication quality begins to deteriorate due to distance attenuation. 2 On the other hand, Fig. 8(b) shows the approximate curve representing the quadratic polynomial "a2 × t" that approximates the RSRP when communication quality starts to deteriorate due to shadowing. 28(a) and 8(b) are graphs showing an approximation curve representing "RSRP + b2 × t + c2". As shown in FIGS. 8(a) and 8(b), when communication quality begins to deteriorate due to distance attenuation, the absolute value of the quadratic coefficient of the quadratic polynomial representing the approximation curve is smaller than when communication quality begins to deteriorate due to shielding. That is, a relationship of "|a1|<|a2|" is established between a quadratic coefficient a1 of the quadratic polynomial approximating RSRP with a relatively slow rate of decrease and a quadratic coefficient a2 of the quadratic polynomial approximating RSRP with a relatively fast rate of decrease. Although not shown, a similar relationship is also established when deteriorated communication quality recovers. Therefore, the quadratic coefficient of the quadratic polynomial approximating RSRP can be used as any index value capable of representing the rate of change of RSRP. Specifically, the quadratic coefficient of the quadratic polynomial approximating RSRP can be used as any index value capable of distinguishing between a state in which the rate of change of RSRP is relatively fast and a state in which the rate of change of RSRP is relatively slow.

[0061] The data aggregating unit 412 may generate input data INP#n including data related to any coefficient of any approximation equation approximating RSRP as rate data, as long as the coefficient can distinguish between a state in which the rate of change of RSRP is relatively fast and a state in which the rate of change of RSRP included in the learning data LD#n is relatively slow, not limited to quadratic coefficients of a quadratic polynomial approximating RSRP. For example, the data aggregating unit 412 may calculate any coefficient of a k-th degree polynomial (where k is an integer equal to or greater than 1) approximating RSRP included in the learning data LD#n, and generate input data INP#n including data related to the calculated coefficient as rate data. For example, the data aggregating unit 412 may calculate any coefficient of an approximation equation approximating RSRP included in the learning data LD#n using at least one of a linear function, a nonlinear function, an exponential function, and a logarithmic function, and generate input data INP#n including data related to the calculated coefficient as rate data.

[0062] The data aggregation unit 412 may calculate the variance of RSRP included as time-series data SRS#n in the learning data LD#n, and generate input data INP#n including data related to the calculated variance as speed data. The data aggregation unit 412 may calculate the standard deviation of RSRP included as time-series data SRS#n in the learning data LD#n, and generate input data INP#n including data related to the calculated standard deviation as speed data. This is because the more rapidly RSRP changes, the greater the relative variation in RSRP becomes, and therefore the variance and standard deviation of RSRP also become relatively large. Therefore, when communication quality begins to deteriorate due to distance attenuation (i.e., communication quality deteriorates relatively gradually), the variance and standard deviation of RSRP become relatively smaller compared to when communication quality begins to deteriorate due to obstruction (i.e., communication quality deteriorates relatively rapidly). Although not shown, a similar relationship also holds when deteriorated communication quality recovers. Therefore, the RSRP variance and standard change can be used as any index value that can represent the rate of change of RSRP. Specifically, the RSRP variance and standard change can be used as any index value that can distinguish between a state in which the rate of change of RSRP is relatively fast and a state in which the rate of change of RSRP is relatively slow.

[0063] The data aggregation unit 412 may classify the trend of time-varying RSRP according to the rate of change of RSRP included as time-series data SRS#n in the learning data LD#n, and generate input data INP#n including data related to the classification result as rate data. For example, the data aggregation unit 412 may classify the trend of time-varying RSRP using a classification model that outputs a classification result of the trend of time-varying RSRP when the RSRP time-series data SRS#n is input. In this case, the data aggregation unit 412 may generate input data INP#n including data related to the output of the classification model as rate data. Note that the classification model is preferably a model that can input time-series data. An example of a model that can input such sequence data is a neural network including LSTM (Long Short Term Memory).

[0064] The data aggregation unit 412 may classify the trend of RSRP changes over time into one of T classes (where T is a constant representing an integer equal to or greater than 2). For example, the data aggregation unit 412 may classify the trend of RSRP changes over time into one of six classes, as shown in FIGS. 9(a) to 9(f). FIG. 9(a) shows class #1 in which RSPR decreases relatively slowly. Class #1 typically corresponds to the trend of RSRP changes over time when communication quality begins to deteriorate due to distance attenuation. FIG. 9(b) shows class #2 in which RSPR decreases relatively rapidly and then does not increase immediately. Class #2 typically corresponds to the trend of RSRP changes over time when communication quality deteriorates due to obstruction (particularly obstruction due to terrain). FIG. 9(c) shows class #3 in which RSPR decreases relatively rapidly and then increases rapidly. Class #3 typically corresponds to the trend of RSRP changes over time when communication quality deteriorates due to obstruction (particularly obstruction due to structures such as buildings). FIG. 9(d) shows class #4, in which RSPR does not change much. Class #4 typically corresponds to the trend of RSRP changes over time when communication quality is not degraded. FIG. 9(e) shows class #5, in which RSPR increases relatively slowly. Class #5 typically corresponds to the trend of RSRP changes over time when communication quality that has deteriorated due to distance attenuation is beginning to recover. FIG. 9(f) shows class #6, in which RSPR increases relatively rapidly. Class #6 typically corresponds to the trend of RSRP changes over time when communication quality that has deteriorated due to obstruction (especially obstruction due to terrain) is recovering.

[0065] The data aggregator 412 may calculate an index value representing the rate of change of RSRP for each predetermined time interval within a certain period based on the RSRP time-series data SRS#n for the certain period. That is, the data aggregator 412 may divide the certain period into predetermined time intervals and calculate an index value representing the rate of change of RSRP for each time interval using a data portion representing the RSRP for each time interval among the RSRP time-series data SSR#n for the certain period.

[0066] The time interval for calculating the index value representing the rate of change may be fixed. Alternatively, the time interval for calculating the index value representing the rate of change may be variable. That is, the data aggregation unit 412 may set a time interval of a desired length as the time interval for calculating the index value representing the rate of change. As an example, when delay (specifically, end-to-end delay, or delay time) is used as the communication quality, the data aggregation unit 412 may set a time interval twice the delay time as the time interval for calculating the index value representing the rate of change. In this case, it is possible to relatively accurately extract a time interval in which the delay is likely to be large (i.e., the communication quality is degraded). As another example, when throughput is used as the communication quality, the data aggregation unit 412 may set a time interval of the same length as the throughput calculation period (i.e., the period from the previous throughput calculation to the next throughput calculation) as the time interval for calculating the index value representing the rate of change. As another example, when throughput is used as the communication quality, the data aggregation unit 412 may set a time interval of the same length as the throughput calculation time (i.e., the data volume aggregation period for calculating the throughput) as the time interval for calculating the index value representing the rate of change. In these cases, it is possible to extract relatively accurately a time interval in which the throughput is likely to be reduced (i.e., communication quality is degraded). As another example, if the user terminal 2 is installed on construction machinery at a construction site where an obstacle is present, the data aggregation unit 412 may set a time interval of a length calculated by the formula (size of obstacle + size of construction machinery) / moving speed of the construction machinery as the time interval for calculating an index value representing the rate of change. As another example, if the user terminal 2 is installed on construction machinery with a GPS function at a construction site where an obstacle is present, the data aggregation unit 412 may set a time interval of a length calculated by the formula (size of obstacle estimated from map information + size of construction machinery) / moving speed of the construction machinery estimated from GPS as the time interval for calculating an index value representing the rate of change. In these cases, it is possible to extract relatively accurately a time interval in which the wireless radio waves are likely to be blocked by an obstacle (i.e., communication quality is degraded due to the blocking).In this paragraph, the symbol " / " (slash mark) means division. In other words, the notation "A / B" means the mathematical formula "A÷B."

[0067] An example of input data INP#n including such speed data is shown in Fig. 10. In the example shown in Fig. 10, input data INP#1 is generated based on time series data SRS#1 included in learning data LD#1 shown in Fig. 5, and input data INP#2 is generated based on time series data SRS#2 included in learning data LD#2 shown in Fig. 5.

[0068] 4 again, thereafter, the model construction unit 413 performs machine learning to construct an estimation model M using the input data INP#1 to INP#N generated in step S12 and the correct labels y#1 to y#N included in the learning dataset LDS acquired in step S11 (step S13). Note that if the data aggregation unit 412 has not generated at least one of the input data INP#1 to INP#N, the model construction unit 413 may perform machine learning to construct an estimation model M without using at least one of the input data INP#1 to INP#N not generated by the data aggregation unit 412 (further, the correct label corresponding to at least one of the input data INP#1 to INP#N not generated by the data aggregation unit 412).

[0069] To perform machine learning, the model construction unit 413 inputs each of the input data INP#1 to INP#N to the estimation model M. As a result, the estimation model M outputs estimation results y'#1 to y'#N corresponding to the input data INP#1 to INP#N, respectively. Specifically, as shown in FIG. 11 which shows the estimation results y'#1 to y'#N of the estimation model M, the estimation model M outputs an estimation result y'#n of the degradation factor of communication quality between the radio base station 1 and the user terminal 2#n at each time t. FIG. 11 shows the estimation result y'#1 by the estimation model M to which the input data INP#1 has been input (i.e., the estimation result y'#1 of the degradation factor of communication quality between the radio base station 1 and the user terminal 2#1), and the estimation result y'#2 by the estimation model M to which the input data INP#2 has been input (i.e., the estimation result y'#2 of the degradation factor of communication quality between the radio base station 1 and the user terminal 2#2). 11 , the estimation model M outputs, as the estimation result y′#n, data in which, among a plurality of candidate factors assumed to be factors causing degradation of communication quality, at least one candidate factor that is estimated to be most likely to be the actual degradation factor is labeled “1” and at least one candidate factor that is unlikely to be estimated as the actual degradation factor is labeled “0.” Thereafter, the model construction unit 413 updates the parameters of the estimation model M based on a loss function related to the error between the correct label y#n and the estimation result y′#n output by the estimation model M. That is, the model construction unit 413 updates the parameters of the estimation model M based on a loss function related to the error between the correct label y#1 and the estimation result y′#1 output by the estimation model M, the error between the correct label y#2 and the estimation result y′#2 output by the estimation model M, and the error between the correct label y#N and the estimation result y′#N output by the estimation model M. At this time, for example, the model construction unit 413 may update the parameters of the estimation model M using an existing method for machine learning. An example of an existing method for machine learning is backpropagation. As a result, the estimation model M is constructed. <2-2> Factor estimation operation performed by the factor estimation device 3

[0070] Next, the factor estimation operation performed by the factor estimation device 3 will be described with reference to Fig. 12. Fig. 12 is a flowchart showing the flow of the factor estimation operation performed by the factor estimation device 3. Note that the factor estimation device 3 typically performs the factor estimation operation shown in Fig. 12 under a situation in which a wireless base station 1 installed in an actual location where it will be used and N user terminals 2 are actually performing wireless communication.

[0071] 12, when it is desired to estimate the cause of degradation in communication quality between the wireless base station 1 and the user terminal 2#n, the data acquisition unit 311 acquires time-series data SRS#n of the received power RP#n from the wireless base station 1 (step S21). Note that the time-series data SRS#n of the received power RP#n has already been explained when explaining the model construction operation, and therefore a detailed explanation will be omitted here.

[0072] Thereafter, the data aggregation unit 312 generates input data INP#n based on the time-series data SRS#n acquired in step S21 (step S22). Note that the operation of generating the input data INP#n in step S22 may be the same as the operation of generating the input data INP#n in step S12 of Fig. 4 described above (i.e., the operation of generating the input data INP#n in the model construction operation). Therefore, a detailed description of step S22 will be omitted.

[0073] Thereafter, the factor estimation unit 313 estimates a factor of degradation in communication quality between the wireless base station 1 and the user terminal 2#n using the input data INP#n generated in step S22 and the estimation model M constructed by the model construction device 4 (step S33). Specifically, the factor estimation unit 313 inputs the input data INP#n generated in step S22 to the estimation model M. As a result, the estimation model M outputs an estimation result y'#n of a factor of degradation in communication quality between the wireless base station 1 and the user terminal 2#n at each time t.

[0074] In this way, the factor estimation unit 313 estimates the factor of degradation of communication quality between the radio base station 1 and the user terminal 2#n. For example, when the estimation result y'#1 shown in FIG. 11 described in the model construction operation is output from the estimation model M, the factor estimation unit 313 estimates that the communication quality between the radio base station 1 and the user terminal 2#1 does not degrade between time t#1-1 and time t#1-11. Furthermore, the factor estimation unit 313 estimates that the actual factor of degradation of communication quality between the radio base station 1 and the user terminal 2#1 between time t#1-12 and time t#1-23 is "shadowing." Similarly, for example, when the estimation result y'#2 shown in FIG. 11 described in the model construction operation is output from the estimation model M, the factor estimation unit 313 estimates that the factor of degradation of communication quality between the radio base station 1 and the user terminal 2#2 between time t#2-1 and time t#2-22 is "fading." Furthermore, the factor estimation unit 313 estimates that the factor of the deterioration in communication quality between the wireless base station 1 and the user terminal 2#2 during the period from time t#2-22 to time t#2-36 is "distance attenuation." 11, the factor estimation unit 313 estimates a single degradation factor as the degradation factor of the communication quality between the radio base station 1 and the user terminal 2#n. However, as explained in the model construction operation, the communication quality between the radio base station 1 and the user terminal 2#n may be degraded due to multiple degradation factors. Therefore, the factor estimation unit 313 may estimate multiple degradation factors as the degradation factors of the communication quality between the radio base station 1 and the user terminal 2#n. When the factor estimation unit 313 estimates the degradation factor, the factor estimation unit 313 may store history data including information on the estimated degradation factor in the storage device 32 or the like. In this case, the factor estimation unit 313 may store history data in which information on the estimated degradation factor and information on the location where the communication quality deteriorated (for example, information on the location of the user terminal 2#n, which can be identified by a GPS device provided in the user terminal 2#n) are associated with each other in the storage device 32 or the like. The factor estimation unit 313 may store history data in which information on the estimated degradation factor and information on the time when the communication quality deteriorated are associated with each other in the storage device 32 or the like. The factor estimation unit 313 may store history data in which information on the estimated degradation factor and information on the location where the communication quality deteriorated and information on the time when the communication quality deteriorated are associated with each other in the storage device 32 or the like. <3> Technical effects of the wireless communication system SYS

[0075] As described above, in this embodiment, the model construction device 4 can construct an estimation model M capable of estimating factors of degradation of communication quality between the wireless base station 1 and the user terminal 2#n. As a result, the factor estimation device 3 can estimate factors of degradation of communication quality between the wireless base station 1 and the user terminal 2#n using the estimation model M constructed by the model construction device 4. As a result, an operator or device that restores degraded communication quality can select an appropriate measure to restore the degraded communication quality based on the factors of degradation of communication quality estimated by the factor estimation device 3, and implement the selected measure. For example, if the factor estimation device 3 estimates that the factor of degradation of communication quality between the wireless base station 1 and the user terminal 2#n is distance attenuation, the operator or device that restores the degraded communication quality can select a measure of adjusting the parameters of the wireless base station 1 as an appropriate measure to restore the degraded communication quality, and implement the selected measure. For example, an appropriate measure to restore quality can be selected and implemented. For example, if the factor estimation device 3 estimates that the cause of degradation of communication quality between the wireless base station 1 and the user terminal 2#n is shielding, an operator or device that restores the degraded communication quality can select at least one of a measure of changing the installation location of the wireless base station 1 and a measure of changing the movement route of the user terminal 2 as an appropriate measure for restoring the degraded communication quality, and implement the selected measure. For example, if the factor estimation device 3 estimates that the cause of degradation of communication quality between the wireless base station 1 and the user terminal 2#n is fading, an operator or device that restores the degraded communication quality can select a measure of redoing the placement design of the wireless base station 1 as an appropriate measure for restoring the degraded communication quality, and implement the selected measure.

[0076] Particularly in this embodiment, the model construction device 4 constructs the estimation model M using input data INP#n including change data related to the degree of change in RSRP per unit time (e.g., rate data related to the rate of change in RSRP). As a result, as described above, the model construction device 4 can construct the estimation model M that can more accurately estimate whether the cause of degradation in communication quality is distance attenuation or obstruction. This allows the factor estimation device 3 to more accurately estimate whether the cause of degradation in communication quality is distance attenuation or obstruction. Specifically, the factor estimation device 3 typically estimates that the cause of degradation in communication quality is distance attenuation when the rate data indicates that the rate of change in RSRP is slower than a predetermined rate (i.e., RSRP is changing relatively slowly). In other words, the factor estimation device 3 estimates that the cause of degradation in communication quality is distance attenuation when the change data indicates that the degree of change in RSRP per unit time is relatively small. On the other hand, the factor estimation device 3 typically estimates that the cause of the degradation of communication quality is obstruction when the rate data indicates that the rate of change of RSRP is faster than a predetermined rate (i.e., the RSRP is changing relatively rapidly). In other words, the factor estimation device 3 estimates that the cause of the degradation of communication quality is obstruction when the change data indicates that the degree of change of RSRP per unit time is relatively large. <4> Examples of application of the wireless communication system SYS The wireless communication system SYS may be applied to a system including a mobile object and a control server that controls the movement of the mobile object. In this case, a user terminal 2 is mounted on the mobile object, and the control server may transmit a control signal for controlling the movement of the mobile object to the user terminal 2 (i.e., to the mobile object on which the user terminal 2 is mounted) via the wireless base station 1. An example of a mobile object is a construction machine used for work at a construction site. Another example of a mobile object is a transport device (e.g., an Automatic Guided Vehicle (AGV)) that transports objects in at least one of a factory and a warehouse. The control server may receive information about the location of the mobile object via the wireless base station 1 and the user terminal 2. For example, if the mobile object is equipped with a positioning device that can identify the location of the mobile object, the user terminal 2 mounted on the mobile object may transmit information about the location of the mobile object identified by the positioning device to the control server via the wireless base station 1. The positioning device may include a device capable of determining the position of a moving object by receiving GPS signals from GPS satellites. The positioning device may include a device capable of determining the position of a moving object by receiving a signal from a transmitter that emits a signal within a predetermined range (i.e., a device capable of determining the position of a moving object by using a beacon). The positioning device may include a device (e.g., radar or lidar) capable of determining the position of a moving object by detecting return light (i.e., return light from an object irradiated with the measurement light) of measurement light (e.g., laser light) emitted from the positioning device. The positioning device may include a device (e.g., radar or lidar) capable of determining the position of a moving object by detecting return sound waves of ultrasound emitted from the positioning device (i.e., return sound waves from an object irradiated with ultrasound). The positioning device may include a device capable of determining the position of a moving object using RFID (Radio Frequency Identifier) ​​placed at a predetermined position. The positioning device may include a device capable of determining the position of a moving object using video captured by a camera. The positioning device may include a device capable of determining the position of a moving object using geomagnetism. The position determining device may include a device capable of determining the position of a mobile object by analyzing the detection results of various sensors equipped on the mobile object. The position determining device may include a device capable of determining the position of a mobile object by analyzing radio waves received by a user terminal 2 mounted on the mobile object. The position determining device may include a device capable of determining the position of a mobile object using IMES (Indoor Messaging System). <5> Variations

[0077] In the above description, the model construction device 4 uses the input data INP#n including rate data relating to the rate of change of the RSRP in order to construct an estimation model M capable of estimating with higher accuracy whether the cause of degradation in communication quality is distance attenuation or obstruction. However, the model construction device 4 may construct an estimation model M capable of estimating with higher accuracy whether the cause of degradation in communication quality is distance attenuation or obstruction by using input data INP#n including rate data relating to the rate of change of the received power RP#n different from the RSRP. Specifically, the RSRP corresponding to the received power RP#n of wireless communication between the wireless base station 1 and the user terminal 2#n may be obtained from a signal (for example, a secondary synchronization signal (SS)) or a CSI reference signal (Channel State Information Reference Indicator (CSI)) that a wireless base station 1 other than the wireless base station 1 that wirelessly communicates with the user terminal 2#n can use to measure the received power RP#n. In other words, RSRP can be said to represent an example of pure received power RP#n between the radio base station 1 and the user terminal 2#n, excluding the influence of other radio base stations 1 different from the radio base station 1 that communicates wirelessly with the user terminal 2#n. For this reason, the model construction device 4 may construct an estimation model M that can estimate with high accuracy whether the cause of degradation in communication quality is distance attenuation or shielding, by using input data INP#n including rate data related to the rate of change in any received power RP#n between the radio base station 1 and the user terminal 2#n, excluding the influence of other radio base stations 1 different from the radio base station 1 that communicates wirelessly with the user terminal 2#n. In this case, the factor estimation device 3 may generate input data INP#n including rate data related to the rate of change in any received power RP#n between the radio base station 1 and the user terminal 2#n, excluding the influence of other radio base stations 1 different from the radio base station 1 that communicates wirelessly with the user terminal 2#n, and estimate the cause of degradation in communication quality between the radio base station 1 and the user terminal 2#n using the generated input data INP#n.

[0078] In the above description, the wireless communication system SYS separately includes the factor estimation device 3 and the model construction device 4. However, in addition to or instead of separately including the factor estimation device 3 and the model construction device 4, the wireless communication system SYS may also include a factor estimation device 3 that can function as the model construction device 4. In this case, as shown in FIG. 13 showing the configuration of the factor estimation device 3 that can also function as the model construction device 4, the factor estimation device 3 that can also function as the model construction device 4 may include a data acquisition unit 311 that can also function as a data acquisition unit 411, a data aggregation unit 312 that can also function as a data aggregation unit 412, a factor estimation unit 313, and a model construction unit 413. Note that the factor estimation device 3 that can also function as the model construction device 4 is substantially equivalent to the model construction device 4 that can also function as the factor estimation device 3. In the above description, the wireless communication system SYS includes a wireless base station 1 as a specific example of a "first wireless station" and a user terminal 2 as a specific example of a "second wireless station." However, the wireless communication system SYS may include any wireless station of a type different from the wireless base station 1 as a specific example of a "first wireless station." The wireless communication system SYS may include any wireless station of a type different from the user terminal 2 as a specific example of a "second wireless station." For example, if the wireless communication system SYS is a wireless LAN (Local Area Network) system conforming to IEEE802.11, the wireless communication system SYS may include an access point (a so-called parent device) as a specific example of a "first wireless station," and a wireless terminal (a so-called child device) capable of wireless communication with the access point as a specific example of a "second wireless station." <6> Additional notes [Appendix 1] A model construction device that constructs an estimation model for estimating a factor of degradation of communication quality between a first wireless station and a second wireless station, an acquisition means for acquiring, from the first wireless station, time series data of received power of wireless communication between the first wireless station and the second wireless station; generation means for generating input data including change data relating to the degree of change in the received power per unit time based on the time series data; a construction means for constructing an estimation model for estimating the deterioration factor from the input data by machine learning using the input data; A model construction device comprising: [Appendix 2] The variation data includes data relating to at least one of a coefficient of an approximation formula that approximates the received power, a variance value of the received power, and a standard deviation of the received power. 2. The model construction apparatus of claim 1. [Appendix 3] The change data includes data on a classification result obtained by classifying the received power according to the degree of change in the received power per unit time. 3. The model construction device according to claim 1 or 2. [Appendix 4] The received power includes received power between the first wireless station and the second wireless station, excluding the influence of another first wireless station different from the first wireless station. 4. A model construction device according to any one of claims 1 to 3. [Appendix 5] The received power includes at least one of RSSI (Received Signal Strength Indicator), RSRP (Reference Signal Received Power), RSRQ (Reference Signal Received Quality), and SINR (Signal to Interference plus Noise Ratio). 5. A model construction device according to any one of claims 1 to 4. [Appendix 6] The generating means generates the input data including the change data relating to the degree of change in the received power per unit time during a variable predetermined period. 6. A model construction device according to any one of appendices 1 to 5. [Appendix 7] The estimation model estimates whether the degradation factor is a first factor that the communication quality is degraded due to attenuation over distance or a second factor that the communication quality is degraded due to obstruction. 7. A model construction device according to any one of appendices 1 to 6. [Appendix 8] A factor estimation device that estimates a factor of degradation of communication quality between a first wireless station and a second wireless station, an acquisition means for acquiring, from the first wireless station, time series data of received power of wireless communication between the first wireless station and the second wireless station; generation means for generating input data including change data relating to the degree of change in the received power per unit time based on the time series data; an estimation model constructed by machine learning and for estimating the deterioration factor from the input data; and an estimation means for estimating the deterioration factor using the input data. A factor estimation device comprising: [Appendix 9] The variation data includes data relating to at least one of a coefficient of an approximation formula that approximates the received power, a variance value of the received power, and a standard deviation of the received power. 9. The factor estimation device according to claim 8. [Appendix 10] The change data includes data on a classification result obtained by classifying the received power according to the degree of change in the received power per unit time. 10. The factor estimation device according to claim 8 or 9. [Appendix 11] The received power includes received power between the first wireless station and the second wireless station, excluding the influence of another first wireless station different from the first wireless station. 11. The factor estimation device according to any one of appendices 8 to 10. [Appendix 12] The received power includes at least one of RSSI (Received Signal Strength Indicator), RSRP (Reference Signal Received Power), RSRQ (Reference Signal Received Quality), and SINR (Signal to Interference plus Noise Ratio). 12. The factor estimation device according to any one of appendixes 8 to 11. [Appendix 13] The generating means generates the input data including the change data relating to the degree of change in the received power per unit time during a variable predetermined period. 13. The factor estimation device according to any one of appendices 8 to 12. [Appendix 14] The estimation means estimates whether the degradation factor is a first factor that the communication quality is degraded due to attenuation over distance or a second factor that the communication quality is degraded due to obstruction. 14. The factor estimation device according to any one of appendixes 8 to 13. [Appendix 15] The change data is speed data relating to a speed of change of the received power, and the estimation means (i) estimates that the degradation factor is a first factor of the degradation of the communication quality due to distance attenuation when the speed data indicates that the speed of change is slower than a predetermined speed, and (ii) estimates that the degradation factor is a second factor of the degradation of the communication quality due to obstruction when the speed data indicates that the speed of change is faster than a predetermined speed. 15. The factor estimation device according to any one of appendices 8 to 14. [Appendix 16] A model construction method for constructing an estimation model for estimating a factor of degradation of communication quality between a first wireless station and a second wireless station, comprising: acquiring, from the first wireless station, time series data of received power of wireless communication between the first wireless station and the second wireless station; generating input data including change data relating to the degree of change in the received power per unit time based on the time series data; By machine learning using the input data, an estimation model for estimating the deterioration factor from the input data is constructed. Model building methods. [Appendix 17] A factor estimation method for estimating a factor of degradation of communication quality between a first wireless station and a second wireless station, comprising: acquiring, from the first wireless station, time series data of received power of wireless communication between the first wireless station and the second wireless station; generating input data including change data relating to the degree of change in the received power per unit time based on the time series data; The deterioration factor is estimated using the input data and an estimation model that is constructed by machine learning and that is used to estimate the deterioration factor from the input data. Factor estimation methods. [Appendix 18] On the computer, A model construction method for constructing an estimation model for estimating a factor of degradation of communication quality between a first wireless station and a second wireless station, comprising: acquiring, from the first wireless station, time series data of received power of wireless communication between the first wireless station and the second wireless station; generating input data including change data relating to the degree of change in the received power per unit time based on the time series data; By machine learning using the input data, an estimation model for estimating the deterioration factor from the input data is constructed. A recording medium on which a computer program for executing a model building method is recorded. [Appendix 19] On the computer, A factor estimation method for estimating a factor of degradation of communication quality between a first wireless station and a second wireless station, comprising: acquiring, from the first wireless station, time series data of received power of wireless communication between the first wireless station and the second wireless station; generating input data including change data relating to the degree of change in the received power per unit time based on the time series data; The deterioration factor is estimated using the input data and an estimation model that is constructed by machine learning and that is used to estimate the deterioration factor from the input data. A recording medium on which a computer program for executing a factor estimation method is recorded.

[0079] At least some of the constituent elements of each of the above-described embodiments can be appropriately combined with at least some of the other constituent elements of each of the above-described embodiments. Some of the constituent elements of each of the above-described embodiments may not be used. Furthermore, to the extent permitted by law, the disclosures of all documents (e.g., published patent applications) cited in this disclosure are incorporated by reference as part of the description of this disclosure.

[0080] This disclosure may be modified as appropriate within the scope of the claims and the technical idea that can be read from the entire specification. Model construction devices, model construction methods, factor estimation devices, factor estimation methods, and recording media that incorporate such modifications are also included in the technical idea of ​​this disclosure. [Explanation of symbols]

[0081] SYS Wireless Communication System 1. Radio base station 2. User terminal 3 Factor estimation device 31 Arithmetic unit 311 Data Acquisition Department 312 Data Collection Department 313 Factor Estimation Section 4. Model building equipment 41 Arithmetic device 411 Data Acquisition Department 412 Data Collection Department 413 Model Construction Department RP received power SRS time series data INP Input Data LDS training dataset LD training data

Claims

1. A factor estimation device for estimating a factor of degradation of communication quality between a first wireless station and a second wireless station, comprising: an acquisition means for acquiring, from the first wireless station, time series data of received power of wireless communication between the first wireless station and the second wireless station; generation means for generating input data including change data relating to the degree of change in the received power per unit time based on the time series data; an estimation model constructed by machine learning using learning data including the time-series data for learning and a correct label corresponding to the time-series data for learning, the correct label indicating that the communication quality has not deteriorated, and for estimating the factor(s) of deterioration of communication quality and that the communication quality has not deteriorated from the input data; and an estimation means for estimating the one or more factor(s) of deterioration of communication quality and that the communication quality has not deteriorated, using the input data; A factor estimation device comprising:

2. The variation data includes data relating to at least one of a coefficient of an approximation formula that approximates the received power, a variance value of the received power, and a standard deviation of the received power. The factor estimation device according to claim 1 .

3. The change data includes data on a classification result obtained by classifying the received power according to the degree of change in the received power per unit time. The factor estimation device according to claim 1 or 2.

4. The received power includes received power between the first wireless station and the second wireless station, excluding influences of other first wireless stations different from the first wireless station. The factor estimation device according to claim 1 .

5. The received power includes at least one of a received signal strength indicator (RSSI), a reference signal received power (RSRP), a reference signal received quality (RSRQ), and a signal to interference plus noise ratio (SINR). The factor estimation device according to claim 1 .

6. The generating means generates the input data including the change data relating to the degree of change in the received power per unit time during a variable predetermined period. The factor estimation device according to claim 1 .

7. The estimation means estimates whether the deterioration factor is at least one of a first factor that the communication quality deteriorates due to attenuation over distance, a second factor that the communication quality deteriorates due to obstruction, and a third factor that the communication quality deteriorates due to fading. The factor estimation device according to claim 1 .

8. the change data is rate data relating to a rate of change of the received power, The estimation means (i) estimates that the degradation factor is a first factor that causes degradation of the communication quality due to distance attenuation when the speed data indicates that the rate of change is slower than a predetermined rate, and (ii) estimates that the degradation factor is a second factor that causes degradation of the communication quality due to obstruction when the speed data indicates that the rate of change is faster than a predetermined rate. The factor estimation device according to claim 1 .

9. A factor estimation method for estimating a factor of degradation of communication quality between a first wireless station and a second wireless station, comprising: acquiring, from the first wireless station, time series data of received power of wireless communication between the first wireless station and the second wireless station; generating input data including change data relating to the degree of change in the received power per unit time based on the time series data; an estimation model constructed by machine learning using learning data including the time-series data for learning and a correct label corresponding to the time-series data for learning, the correct label indicating that the communication quality has not deteriorated, and for estimating the one or more factors of deterioration in communication quality and that the communication quality has not deteriorated from the input data, and the input data is used to estimate the one or more factors of deterioration in communication quality and that the communication quality has not deteriorated. Factor estimation methods.

10. On the computer, A factor estimation method for estimating a factor of degradation of communication quality between a first wireless station and a second wireless station, comprising: acquiring, from the first wireless station, time series data of received power of wireless communication between the first wireless station and the second wireless station; generating input data including change data relating to the degree of change in the received power per unit time based on the time series data; an estimation model constructed by machine learning using learning data including the time-series data for learning and a correct label corresponding to the time-series data for learning, the correct label indicating that the communication quality has not deteriorated, and for estimating the one or more factors of deterioration in communication quality and that the communication quality has not deteriorated from the input data, and the input data is used to estimate the one or more factors of deterioration in communication quality and that the communication quality has not deteriorated. A computer program that executes the factor estimation method.

Images