System, device, method and program that predict communication quality

1. A system comprising:

a processor; and

a storage medium having computer program instructions stored thereon, when executed by the processor, perform to:

generate environment information including at least one piece of information from a plurality of pieces of information regarding a position, a posture, a motion, a control command, a camera, and a sensor of a terminal that performs wireless communication;

store a plurality of communication prediction models obtained by learning a relationship between the environment information and communication quality of wireless communication under a frequency channel condition corresponding to a frequency or a frequency bandwidth of wireless communication or both of the frequency and the frequency bandwidth;

select one or more communication prediction models to be used to predict communication quality of the terminal from among the plurality of communication prediction models stored in the communication prediction model storage unit and use the selected one or more communication prediction models to generate a communication prediction model corresponding to a frequency channel condition of wireless communication used by the terminal; and

input the environment information generated by the environment information generation unit to the communication prediction model generated by the communication prediction model generation unit and predict current or future communication quality of the terminal.

2. The system according to claim 1 , wherein the computer program instructions further perform to

stores the communication prediction model in association with information regarding model accuracy, and

selects a communication prediction model that enhances model accuracy, from among the plurality of communication prediction models.

3. The system according to claim 1 or 2 , wherein the computer program instructions further perform to

stores information regarding similar frequency channel conditions for the plurality of communication prediction models, and

selects the communication prediction models having similar frequency channel conditions from among the plurality of communication prediction models.

4. The system according to claim 1 , wherein the computer program instructions further perform to

stores the communication prediction model by splitting into two parts that are a former part and a latter part, and

acquires two communication prediction models that are the former part and the latter part stored in the communication prediction model storage unit, based on a frequency channel condition for predicting communication quality and combines the two communication prediction models to generate the communication prediction model.

5. The system according to claim 4 , wherein the computer program instructions further perform to:

generate the former part communication prediction model and the latter part communication prediction model,

determines, in advance, a former part communication prediction model and a latter part communication prediction model to be used for any frequency channel condition,

at least any of a setting for preventing an updating coefficient from decreasing, a setting for preventing a regularization coefficient from increasing, and a setting for preventing a condition such as a weight, a bias, or a decision tree from being changed is set in the former part communication prediction model, and

in the latter part communication prediction model, an updating coefficient that is same as or greater than an updating coefficient for the former part communication prediction model is used to set to reduce or not to user a regularization coefficient, and a learning load is set lower than a learning load for the former part communication prediction model.

6. The system according to claim 1 wherein the computer program instructions further perform to:

generate the former part communication prediction model and the latter part communication prediction model,

trains the former part communication prediction model using measurement information including communication quality and environment information under a plurality of frequency channel conditions, and

trains the latter part communication prediction model using measurement information including communication quality and environment information obtained under more detailed frequency channel conditions than the plurality of frequency channel conditions that have used for training the former part communication prediction model.

7. An apparatus comprising:

an environment information generation unit configured to generate environment information including at least one piece of information from a plurality of pieces of information regarding a position, a posture, a motion, a control command, a camera, and a sensor of a terminal that performs wireless communication;

a communication prediction model storage unit configured to store a plurality of communication prediction models obtained by learning a relationship between the environment information and communication quality of wireless communication under frequency channel conditions corresponding to a frequency or a frequency bandwidth of wireless communication or both of the frequency and the frequency bandwidth;

a communication prediction model generation unit configured to select one or more communication prediction models to be used to predict communication quality of the terminal from among the plurality of communication prediction models stored in the communication prediction model storage unit and use the selected one or more communication prediction models to generate a communication prediction model corresponding to a frequency channel condition of the wireless communication used by the terminal; and

a communication prediction unit configured to input the environment information generated by the environment information generation unit to the communication prediction model generated by the communication prediction model generation unit and predict current or future communication quality of the terminal.

8. A method comprising:

by an environment information generation unit, generating environment information including at least one piece of information from a plurality of pieces of information regarding a position, a posture, a motion, a control command, a camera, and a sensor of a terminal that performs wireless communication;

by a communication prediction model generation unit, referring to a plurality of communication prediction models obtained by learning a relationship between the environment information and communication quality of wireless communication under a frequency channel condition corresponding to a frequency or a frequency bandwidth of wireless communication or both of the frequency and the frequency bandwidth, selecting one or more communication prediction models to be used to predict communication quality of the terminal from among the plurality of communication prediction models, and using the selected one or more communication prediction models to generate a communication prediction model corresponding to a frequency channel condition of the wireless communication used by the terminal; and

by a communication prediction unit, inputting the environment information generated by the environment information generation unit to the communication prediction model generated by the communication prediction model generation unit, and predicting current or future communication quality of the terminal.