KR20200119061A - A method for operating of a wireless terminal providing dynamic changing connections to wireless networks based on learning models of quality expectations - Google Patents

Abstract

The present invention is to provide a wireless terminal device which provides a dynamic wireless network variable access based on a quality prediction learning model and an operating method thereof. According to an embodiment of the present invention, a method of operating a wireless terminal device for transmitting and receiving data through a plurality of wireless networks comprises the steps of: obtaining a data packet to be transmitted; determining a model application variable of the wireless terminal device; obtaining quality prediction information for each accessible network by applying the model application variable to a quality prediction learning model for each wireless network; determining a first wireless network to transmit the data packet from among the plurality of wireless networks based on the quality prediction information for each network; and wirelessly transmitting the data packet to a target device using the determined first wireless network.

Description

A method of operating a wireless terminal device providing dynamic wireless network variable access based on a quality prediction learning model {A METHOD FOR OPERATING OF A WIRELESS TERMINAL PROVIDING DYNAMIC CHANGING CONNECTIONS TO WIRELESS NETWORKS BASED ON LEARNING MODELS OF QUALITY EXPECTATIONS}

The present invention relates to a wireless terminal device and a method of operation thereof. More specifically, the present invention relates to a wireless terminal device providing a dynamic wireless network variable access based on a quality prediction learning model and a method of operating the same.

With the rapid development of electronic and communication technologies, a service using a wireless network is developing into a communication service that transmits and receives data wirelessly, such as packet data.

Accordingly, the data usage of mobile terminals and the number of devices connected to the Internet are increasing exponentially, and the importance of the wireless Internet is becoming more and more prominent in the lives of modern people. If technologies such as VR, AR, IoT, and autonomous driving are actively commercialized with the advent of 5G mobile mobile communication and 6G Wi-Fi (802.11 ax, etc.) in the future, the dependence on the wireless network and expectations for the cost and quality of the network Will continue to increase.

With the increase in wireless Internet demand, wireless networks are provided in various forms, protocols, and communication methods. In particular, base station-based mobile communication networks and WIFI standard-based wireless networks based on AP (ACCESS POINT) Due to infrastructure investment in networks, wireless networks can now be used anywhere in the world.

However, due to the rapid increase in usage of wireless networks, the quality of the wireless network is deteriorating day by day. On the other hand, mobile communication companies and the like are investing excessive infrastructure construction costs to build a new communication network, which leads to a cost burden on consumers.

Accordingly, wireless network users are choosing and using a wireless network in consideration of cost burden. In particular, a WIFI standard network that has no cost or is cheaper than a high-cost wireless network based on a mobile communication network is intended to be used. However, in the case of a WIFI standard network, there is a problem in that the quality of the network cannot be stably maintained due to decentralized ownership and operation. For example, when connected to a WIFI AP with a weak signal or low performance, wireless Internet may not be used several times a day.

Due to this problem, the user directly observes the access speed of wireless networks, selects and uses the fastest and most cost-effective network, and sets to switch back to the mobile communication network when experiencing a decrease in speed.

However, this method has a problem in that the user experiences a process of deteriorating quality at least temporarily, and it is difficult to provide a seamless (SEAMLESS) wireless Internet use due to a reconnection process and connection delay that occur during network switching.

In addition, the quality of a wireless network should be determined not only by a simple speed, but also by various variables such as cost, responsiveness, and device performance, but currently, there is a limit in which the network can only be determined by only the signal strength, which is a primary variable experienced by the user.

The present invention was devised to solve the above-described problems, and generates a quality prediction learning model for each network corresponding to a plurality of accessible wireless networks in advance, and calculates quality prediction information for each network compared to the model application variable based on this. Thus, a wireless terminal device that provides a dynamic wireless network variable access based on a quality prediction learning model that enables fast network switching before the user experiences actual quality degradation and minimizes access delay caused by network switching. And it is an object to provide a method of operation.

A method according to an embodiment of the present invention for solving the above-described problems, in the operating method of a wireless terminal device capable of accessing a plurality of wireless networks, the method comprising: obtaining a data packet to be transmitted; Refining and determining model application parameters of the wireless terminal device; Obtaining quality prediction information for each accessible network by applying the model application variable to a quality prediction learning model for each wireless network; Determining a first wireless network to transmit the data packet from among the plurality of wireless networks based on the quality prediction information for each network; And wirelessly transmitting the data packet to a target device using the determined first wireless network.

An apparatus according to an embodiment of the present invention for solving the above-described problems, in a wireless terminal device capable of accessing a plurality of wireless networks, includes: a control unit for obtaining a data packet to be transmitted; A model application variable processor for refining and determining model application parameters of the wireless terminal device; A learning-based quality prediction module for obtaining quality prediction information for each accessible network by applying the model application variable to a quality prediction learning model for each wireless network; A network determining unit configured to determine a first wireless network to transmit the data packet from among the plurality of wireless networks based on the network-specific quality prediction information; And a wireless communication unit for wirelessly transmitting the data packet to a target device by using the determined first wireless network.

Meanwhile, the method according to an embodiment of the present invention for solving the above-described problems may be implemented with a program for executing the method on a computer and a recording medium in which the program is recorded.

According to an embodiment of the present invention, a quality prediction learning model for each network corresponding to a plurality of accessible wireless networks may be generated in advance, and quality prediction information for each network may be calculated based on the current model application variable. By predicting quality degradation in advance through a learning process based on various variables, it is possible to provide a dynamic variable access of a wireless network.

Accordingly, according to an embodiment of the present invention, it is possible to enable a fast network switch before the user experiences an actual quality deterioration, and minimize an access delay caused by the network switch.

1 is a conceptual diagram schematically showing an entire system according to an embodiment of the present invention.
2 is a block diagram showing in more detail a wireless terminal device according to an embodiment of the present invention.
3 and 4 are flowcharts illustrating a method of operating a wireless terminal device according to an embodiment of the present invention.
5 is a conceptual diagram schematically illustrating a system based on an apparatus for providing quality prediction information according to an embodiment of the present invention.
6 is a block diagram specifically showing an apparatus for providing quality prediction information according to an embodiment of the present invention.
7 is a flowchart illustrating an operation of an apparatus for providing quality prediction information according to an embodiment of the present invention.
8 is a block diagram illustrating a learning-based quality prediction module according to an embodiment of the present invention in more detail.
9 to 10 are diagrams for explaining a data merging and sharing process of a learning-based quality prediction model according to an embodiment of the present invention.

The following content merely illustrates the principles of the present invention. Therefore, those skilled in the art can implement the principles of the present invention and invent various devices included in the concept and scope of the present invention, although not clearly described or illustrated herein. In addition, it is understood that all conditional terms and examples listed in this specification are, in principle, expressly intended only for the purpose of making the concept of the present invention understood, and are not limited to the embodiments and states specifically listed as such. Should be.

In addition, it is to be understood that all detailed descriptions listing specific embodiments as well as principles, aspects and embodiments of the present invention are intended to include structural and functional equivalents of these matters. It should also be understood that these equivalents include not only currently known equivalents, but also equivalents to be developed in the future, that is, all devices invented to perform the same function regardless of structure.

Thus, for example, the block diagrams herein are to be understood as representing a conceptual perspective of exemplary circuits embodying the principles of the invention. Similarly, all flowcharts, state transition diagrams, pseudocodes, etc. are understood to represent various processes performed by a computer or processor, whether or not the computer or processor is clearly depicted and that can be represented substantially in a computer-readable medium. Should be.

In addition, the explicit use of terms presented as processor, control, or similar concepts should not be interpreted exclusively by referring to hardware capable of executing software, and without limitation, digital signal processor (DSP) hardware, ROM for storing software. It should be understood to implicitly include (ROM), RAM, and non-volatile memory. Other commonly used hardware may also be included.

The above-described objects, features, and advantages will become more apparent through the following detailed description in connection with the accompanying drawings, whereby those of ordinary skill in the technical field to which the present invention pertains can easily implement the technical idea of the present invention. There will be. In addition, in describing the present invention, when it is determined that a detailed description of a known technology related to the present invention may unnecessarily obscure the subject matter of the present invention, a detailed description thereof will be omitted.

Hereinafter, a preferred embodiment according to the present invention will be described in detail with reference to the accompanying drawings.

1 is a diagram schematically showing an entire system according to an embodiment of the present invention.

Referring to FIG. 1, a system according to an embodiment of the present invention includes a learning-based quality prediction module 200, a wireless terminal device 100, and a connection target device 300, and a wireless terminal device 100 and access The target device 300 may be configured to be connectable through one or more wireless networks among a plurality of wireless networks 200a, 200b, ... 200n.

First, each wireless network includes a local area network (LAN), a wide area network (WAN), a value added network (VAN), a personal area network (PAN), and a mobile communication network. Mobile radiocommunication network) or a satellite communication network may be implemented as various types of wireless networks, and each base station or access point (AP) device for providing the same may be provided in a remote location.

In particular, in order to provide a local area communication network, the wireless network may include a WIFI wireless network based on a WIFI standard, and each access point may broadcast device information for operating a Wi-Fi service coverage.

Meanwhile, the wireless terminal device 100 may be a wireless terminal device of a user who requests transmission/reception of a wireless data packet by determining a first wireless network 200a to be connected to the access target device 300 among wireless networks. Here, the wireless terminal device 100 may be implemented as various devices such as a smart phone, a tablet computer, a notebook computer, a personal digital assistant (PDA), a portable multimedia player (PMP), and smart glasses.

In addition, according to an embodiment of the present invention, the wireless terminal device 100 may collect quality model generation variables for each wireless network and transfer them to the learning-based quality prediction module 200, and the learning-based quality prediction module 200 According to the relational learning operation of the model generation variable, the quality prediction learning model for each wireless network may be generated.

Here, the learning-based quality prediction module 200 may learn and calculate relationship information between variables by collecting various variables for generating a quality prediction learning model, and each variable is physical variable information for each communication section for quality prediction, an invariant element. It can be diversified into information and variable element information.

In addition, the learning process may include a supervised or non-supervised artificial neural network learning process, regression analysis learning, deep learning analysis, and the like, which will be described in more detail later.

In addition, the wireless terminal device 100 determines a model application variable of the wireless terminal device 100 in response to a data packet to be transmitted wirelessly, and applies the model application variable to a quality prediction learning model for each wireless network, Obtains quality prediction information for each network, determines a first wireless network 200a to transmit the data packet from among the plurality of wireless networks based on the quality prediction information for each network, and uses the determined first wireless network The packet may be wirelessly transmitted to the connection target device 300.

Accordingly, the wireless terminal device 100 can more accurately predict in advance the quality deterioration due to the variation of the model application variable, based on the relationship learning process using the network speed, signal strength, or preset model generation variables. As a result, it is possible to change the network quickly and efficiently while minimizing the experience of quality deterioration of the user through dynamic variable control of the wireless network.

2 is a block diagram showing in more detail a wireless terminal device according to an embodiment of the present invention.

Referring to FIG. 2, a wireless terminal device 100 according to an embodiment of the present invention includes a control unit 110, a communication unit 120, a model generation variable processing unit 130, a model application variable processing unit 140, and a network determination unit. 150, a learning-based quality prediction module 200, a database 160, and a user setting unit 170.

The controller 110 overall controls the operation of each component of the wireless terminal device 100 and execution of functions. For example, the control unit 110 may be implemented as a processor for controlling all or part of the acquisition of packet data to be transmitted, processing of model generation variables, processing of model application variables, network determination, and user setting functions, or a program for executing the same. have.

In addition, all or part of the functions of the control unit 110 may be implemented as a software module installed in the wireless terminal device 100, and the software module operates in an application layer to perform network variable control through network socket binding control, or Alternatively, it is possible to perform network variable control that operates in the OS layer to drive the network interface.

Further, the communication unit 120 connects to the wireless networks 200a, 200b, ... 200n under the control of the control unit 110, or connects to a wired network to connect to/from the network where the connection target device 300 is located. It may include one or more communication modules that enable wireless data packet communication.

In particular, the communication unit 120 may include a wireless Internet module for accessing at least one of the wireless networks 200a, 200b, ... 200n and a wireless Internet protocol, and the wireless Internet module is a WLAN (Wireless LAN) ( Wi-Fi), Wireless broadband (Wibro), World Interoperability for Microwave Access (Wimax), and a network module based on a High Speed Downlink Packet Access (HSDPA) scheme may be exemplified.

In addition, the communication unit 120 may include a short-range communication module for connecting to at least one of the wireless networks 200a, 200b, ... 200n and short-range communication through a short-range communication protocol, and short-range communication technology Examples of examples include Bluetooth, Radio Frequency Identification (RFID), infrared data association (IrDA), Ultra Wideband (UWB), ZigBee, and the like.

On the other hand, the communication unit 120 may include a mobile communication module for mobile communication access to at least one of the wireless networks (200a, 200b, ... 200n) and a mobile communication protocol, for this purpose, a base station on a mobile communication network, an external It is possible to transmit and receive a wireless signal with at least one of the terminal, the connection target device or the server The wireless signal may include a voice call signal, a video call signal, or various types of data according to transmission/reception of text/multimedia messages.

Further, the communication unit 120 can selectively variably control the communication module for wireless transmission of data packets under the control of the control unit 110, which predicts the quality of each network analyzed by the learning-based quality prediction module 200 It can be determined according to the information.

In order to determine the quality prediction information for each network, the learning-based quality prediction module 200 collects the model generation variables obtained from the model generation variable processor 130, and a quality prediction learning model for each wireless network based on the collected model generation variables. Can be created and stored in the database 160.

Here, the model generation variable processing unit 130 may collect, refine, and transmit various variables for generating a quality prediction learning model for each wireless network in the learning-based quality prediction module 200, and this is a learning-based quality prediction module. It may be determined differently according to the learning model of 200 and the learning processes. As the learning process, artificial intelligence or artificial neural network learning can be exemplified, and variable setting and learning model configuration for this are important.

For example, the learning-based quality prediction module 200 acquires refined variable information from the quality model generation variable processed by the model generation variable processor 130, and according to a relational learning operation based on the refined variable information, You can create a learning model for predicting quality for each network.

In addition, the learning-based quality prediction module 200 can collect and process variables for model generation and application from internal information of the wireless terminal device 100, as well as a data set received from an external device at a remote location. ) Is collected and processed as a variable, or data input from the user is collected and processed as a variable to generate a quality prediction learning model for each wireless network from the variables collected and processed through various paths, and respond to the generated model and situation It can handle various application variables.

Here, the variable information may be divided into a model generation variable for generating a learning base of the prediction model and a model application variable for applying to the generated prediction model. Each variable information may include basic variable information that is primarily obtained for generating or applying a model for network quality prediction information, and may include a model generation variable refined from the basic variable information.

In addition, the model generation variable processing unit 130 refines the basic variable information or uses the basic variable information together to generate a model generation variable for generating a quality learning model and transmits it to the learning-based quality prediction module 200, Followed learning operations can be processed.

In addition, the model application variable processing unit 130 may collect and process model application variables that can be applied to one or more quality learning models to obtain a predicted value by refining the basic variable information or using the basic variable information together. have.

Accordingly, the model generation variable and the model application variable according to an embodiment of the present invention may be composed of variables that can be obtained by being refined from basic variable information, respectively, and the predictive model generation process and the learning model generation method and type , Application time, application target, required output value, etc. can be determined independently and differently.

Such variable information may include variable information for each communication section, for example. The variable information for each communication section is a physical variable corresponding to at least one of a wired section between a target device and an AP (Access Point) access point, an AP internal section, a wireless section between the AP and a wireless terminal device, and a wireless terminal device internal section. May contain information. Here, the AP access point may mean a wireless network access point such as a WIFI router or a base station.

More specifically, the physical variable information may include wired cable performance information corresponding to a wired section between the connection target device and the AP access point, and difference information between the IP of the connected device and the IP of the AP.

In addition, the physical variable information may include AP hardware capability information and number of connected devices corresponding to the AP internal section.

In addition, it may include at least one of protocol information, signal strength information, frequency band information, and channel/frequency congestion information corresponding to a radio section between the AP and the wireless terminal device.

Here, the protocol information may include a communication protocol, for example, in the case of a WIFI network, a protocol based on 802.11n and 802.11ac standards, and in the case of a mobile communication network, a standard-based protocol such as GSM, LTE-A, and LTE. This can be illustrated.

In addition, since the protocol information may include an encryption protocol, in the case of WIFI, an encryption protocol such as open type, WEP, and WPA2 may be exemplified.

Meanwhile, the physical variable information may include at least one of wireless module performance information, wireless module characteristic information, and temperature information corresponding to an internal section of the wireless terminal device 100, and the model generation variable processing unit 130 The control unit 110 can process the data request and refinement process for generating such model generation variables, and the control unit 110 obtains the requested variable information from the process information and sensor data being processed, It may be provided to the processing unit 130.

In addition, the model application variable processing unit 140 may also process a data request and refining process for generating model application variables to the control unit 110, and the control unit 110 receives the request from the process information and sensor data being processed. The defined variable information can be obtained and provided to the model application variable processing unit 140, and the definition and processing processes of the variable information to be described later are not limited to the model generation variable, and the model application variable processing by the model application variable processing unit 140 The same or similar process can be used.

In addition, as a performance index of a wireless network, RTT (round trip time) information of a TCP socket that can be computed from basic variable information, window size information, and transmission/reception data amount information can be exemplified as a refined model generation variable. The quality prediction module 200 performs a learning-based quality prediction by processing a learning process in which variables selected for model generation among basic variable information or refined model generation variable information are input values and the corresponding quality variables are output variables. It is possible to generate a quality prediction learning model for each wireless network.

Accordingly, the learning-based quality prediction module 200 may generate a quality prediction learning model for each wireless network from the quality model generation variable, and generate a relationship model between the variable element information and the quality prediction training model.

In addition, the variable information is classified into constant element information and variable element information, and may be used for selecting a model generation variable, generating a predictive model, or determining a model application variable.

Here, the variable factor is a factor in which the degree of influence on the quality of the wireless communication network is continuously variable, and for example, information on the number of AP access devices and information on signal strength (RSSI) may be exemplified, as a variable performance index. TCT socket destination IP information, RTT information, window size information, and transmission/reception data amount information may be exemplified.

In addition, the constant element is an element whose degree of influence on the quality of the wireless communication network is unchanged, and may include hardware performance information of an AP, information on the number of neighboring APs, and communication protocol information.

For example, the learning-based quality prediction module 200 processes each preset learning algorithm in response to model generation variables classified into constant and variable elements, and generates a quality prediction learning model for each wireless network accordingly. can do.

In addition, the learning-based quality prediction module 200 may obtain quality prediction information for each network by applying the model application variable to a variable element of the relationship model in obtaining quality prediction information for each network accessible at the current time point. have.

For example, in order to obtain quality prediction information for each network, first, the learning-based quality prediction module 200 uses hardware information of an AP providing a WIFI standard network and signal strength information with the wireless terminal as a model generation variable. You can build predictive learning models.

And when the quality prediction information for each network is required, the learning-based quality prediction module 200 acquires hardware information of the AP and signal strength information with the wireless terminal as the model application variable, and the quality prediction learning model for each wireless network From, the quality prediction information of the WIFI standard network usage point of the wireless terminal corresponding to the hardware information and signal strength information may be obtained and output.

For example, during wireless communication through Wi-Fi, the power strength (milliWatt) of radio waves emitted from a specific Wi-Fi AP may be generated according to the performance and settings of the AP, and the model generation variable processing unit 130 is a communication unit. The power intensity (milliWatt) of the divergent radio wave transmitted to 120 may be obtained as basic variable information.

The radiated power intensity may be different depending on the frequency of the AP (for example, 5GHz, 2.4GHz, etc.) and AP performance and settings, and the power intensity received by the wireless terminal device 100 is the wireless terminal device 100 and the AP. It may vary depending on the distance from the radio, the degree of isotropy (effective aperture) of the antenna of the communication unit 120, the inherent gain of the antenna of the wireless terminal device 100 and AP, and congestion of physical obstacles and surrounding frequencies.

Accordingly, the model generation variable processor 130 may obtain the refined model generation variable by calculating the received power as shown in Equation 1 below.

Here, P _Transmitted is AP-specific emission power information, and may be different according to the AP protocol, spatial trim, modulation, and coding rate. In addition, G _{Antenna gain1} represents the antenna gain information specific to the AP, A _Effective represents the antenna's inherent effective aperture or isotropic degree of the wireless terminal device 100, and α represents a physical obstacle existing in the radio path or a peripheral frequency congestion. It can indicate the degree to be reduced by.

Here, A _Effective can be obtained through polarization of radio waves, the law of thermodynamics, Rayleigh-Jeans Formula, and Johnson-Nyquist noise calculations. ) Received power can be calculated by arithmetic processing.

)를 취하여 데시벨 밀리와트(decibel milliwatt, dBm)로 단위를 변경하면 상기 수학식 2는 아래와 같은 식으로 정리된다.Here, G _{Antenna gain2} denotes antenna gain information specific to a terminal of the wireless terminal device 100. And, milliWatt unit is log (10*

) And changing the unit to decibel milliwatt (dBm), Equation 2 is summarized as follows.

와

는 각 접속점마다 갖는 고유의 값으로 연산될 수 있으며, 파장

는

로 연산되므로, 주파수의 값으로부터 획득될 있다.

는 무선 단말 장치(100)의 안테나 고유 특성 정보이므로 불변 요소로 분류될 수 있다.

는 단말기와 해당 접속점 사이 거리에 따라 변하는 가변 요소일 수 있으며, 이며

는 경로상 존재하는 물리적 장애물 혹은 주변 주파수 혼잡등으로 인해 감소되는 감소율에 대한 로그값으로서

가 1보다 작거나 같기에

은 0보다 작거나 같게 되며 환경에 따라 가변될 수 있다.here,

Wow

Can be calculated as a unique value for each connection point, and

Is

Since it is calculated as, it can be obtained from the value of the frequency.

Since is characteristic information of the antenna of the wireless terminal device 100, it may be classified as a constant element.

May be a variable element that changes depending on the distance between the terminal and the connection point, and

Is a logarithmic value of the reduction rate due to physical obstacles in the path or congestion of surrounding frequencies.

Is less than or equal to 1

Is less than or equal to 0 and may vary depending on the environment.

Accordingly, the learning-based quality prediction module 200 uses the Modulation Coding Scheme (MCS) Index, which is an AP-specific characteristic value, and RSSI data between the AP and the plurality of terminals as a model generation variable, to each AP and wireless terminal device 100 hardware. You can predict and learn performance.

값을 예측할 수 있으며, 이에 따라, 특정 AP와 복수 단말기 간의 RSSI 정보로부터 상기 수학식 3의 해를 연산하면, AP의 하드웨어 성능 지표인

과 무선 단말 장치(100)의 하드웨어 성능 지표인

을 유추 획득할 수 있다. 이에 따라, 학습 기반 품질 예측 모듈(200)은 AP의 MCS(Modulation Coding Scheme) 인덱스에 대응하여 결정되는 AP의 발산 파워 세기 정보로부터 산출되는 AP와 무선 단말 장치(100)의 하드웨어 성능 정보를 모델 생성 변수로 사용하여, 무선 네트워크별 품질 예측 모델 학습을 처리할 수 있다.Learning-based quality prediction module 200 from the MCS Index

The value can be predicted. Accordingly, if the solution of Equation 3 is calculated from RSSI information between a specific AP and a plurality of terminals, the hardware performance index of the AP

And the hardware performance index of the wireless terminal device 100

Can be obtained by analogy. Accordingly, the learning-based quality prediction module 200 generates a model of the hardware performance information of the AP and the wireless terminal device 100 calculated from the divergence power intensity information of the AP determined in response to the Modulation Coding Scheme (MCS) index of the AP. By using it as a variable, it can process the quality prediction model training for each wireless network.

In addition, the model generation variable processing unit 130 may collect frequency band, Modulation Coding Scheme (MCS) Index, Bandwidth, and Guard Interval values as variable information as unique characteristic information for each AP, and the learning-based quality prediction module 200 Can infer the minimum signal strength and maximum speed of a specific AP from such unique characteristic information, and can process the quality prediction model training for each wireless network by using the corresponding value as a model generation variable.

Meanwhile, the model generation variable processing unit 130 may perform refinement processing from the variable information processed through the communication unit 120 to collect RTT, window size, and transmission/reception data amount information of a TCP socket, and a learning-based quality prediction module The 200 may calculate a refined network-specific model generation variable from the processed variable information of the communication unit 120, and process a quality prediction model training for each wireless network corresponding thereto.

For example, in TCP communication, the output (Throughput) of the current socket may be defined in the form of a transmission packet (Packets in Flight), which is as follows.

The number of data sockets used by the communication unit 120 may be plural, and the speed at a specific time may be calculated as a sum or average of outputs of each socket. Accordingly, the learning-based quality prediction module 200 calculates the output information for each network as a model generation variable through an average calculation in which a weight is applied to the amount of output information and transmission/reception data or other factors, and the corresponding quality for each wireless network It can handle predictive model training.

According to the quality prediction model learning processing for each wireless network, maximum speed information corresponding to the minimum signal strength information for each wireless network may be calculated, and the predicted link speed may be determined in a variable limit.

Equation 5 below is an example of output when a data-weighted average variable is limited to a link speed for a signal strength (RSSI) at a specific point in time.

로 정의될 수 있다.here,

Can be defined as

Accordingly, the learning-based quality prediction module 200 may process the training of a prediction model for determining the quality information of the WIFI standard network, and the data collected by the model generation variable processor 130 as an input value as a learning process. The used logistic regression operation, etc. may be illustrated.

For example, the learning-based quality prediction module 200 is calculated from at least one of RTT, window size, and transmission/reception data amount information of a TCP socket based on a logistic relationship between signal strength (RSSI) and connection speed information. By using the output (Throughput) as an input value, a logistic regression function processing corresponding to the connection speed versus the signal strength (RSSI) can be performed. Accordingly, the learning-based quality prediction module 200 may pre-build a quality prediction model of a corresponding WIFI network for a specific signal strength (RSSI) and provide prediction information corresponding thereto.

On the other hand, the learning-based quality prediction module 200 performs a Decay function based on the rate samples of the latest n data collected in real time through the model generation variable processing unit 130 for quality learning and prediction of a mobile communication network. By processing, a moving average value may be obtained, and transmission performance information corresponding to the moving average value corresponding thereto may be subjected to regression learning to generate a quality prediction learning model corresponding to the mobile communication network.

Here, the decay function may be a process such as an exponential moving average, and the learning-based quality prediction module 200 assigns a weight to the latest data sample and It is possible to generate a quality prediction learning model for predicting the quality of a mobile communication network by using only the fields. Here, the decay function process may include not only an exponential moving average, but also a simple moving average, a cumulative moving average, a weighted moving average, or various processes including the same. .

For example, in the case of an exponential moving average, the learning-based quality prediction module 200 is an output calculated from the RTT, window size, and transmission/reception data amount of the TCP socket processed by the communication unit 120 through the model generation variable processing unit 130 By using the information (Throughput) as an input value, it is possible to perform regression learning processing of a function for a speed at a specific point in time to generate a quality prediction learning model corresponding to a mobile communication network.

where

의 근사치는 아래 수학식 8과 같이 정의할 수 있다.Here, if the Exponential Moving Average is calculated using the recent N random number, the coefficient

The approximate value of can be defined as in Equation 8 below.

As described above, the learning-based quality prediction module 200 may use an artificial intelligence algorithm as a learning process for generating a quality learning model for each network. Examples of artificial intelligence algorithms include an artificial neural network method in the form of supervised learning such as Logistic or Logarithmic Regression or Multilayer Perceptron (MLP), Support Vector Machine, Bayesian Networks, and Genetic Algorithm. I can.

In addition, an unsupervised learning method such as Dictionary Learning, Independent Component Analysis, Autoencoders, or various types of clustering may be used, and an analysis process such as Decision Trees and Anomaly Detection may be used.

In addition, the learning-based quality prediction module 200 may use a Reinforcement Learning algorithm using data collected in real time from each user as a model generation variable.

In the case of regression, methods such as Projection, Maximum Likelihood, Generalized Methods of Moments including Ordinary Least Squares method may be exemplified.In the above-described Supervised, Unsupervised, and Reinforcement processing, the least squares method and the gradient descent method Through regression model, backpropagation method, deep learning, entropy minimization process, etc. may be included.

More specifically, supervised learning and unsupervised learning processes of the learning-based quality prediction module 200 may be determined according to input values.

For example, as supervised learning to predict the performance of a specific AP with a specific index (e.g., speed), a classification method can be used, and the performance of a specific access point can be determined by specific criteria (e.g., Good, Normal, Bad) can be exemplified.

Here, Classification is a supervised learning algorithm that classifies output into specific classes for specific inputs, such as Logistic Regression, Naive Bayes, Stochastic Gradient Descent, K-Nearest Neighbors, Decision Tree, Random Forest, Support Vector Machine, etc. This can be illustrated.

Accordingly, the learning-based quality prediction module 200 substitutes the model generation variable into the supervised learning model, so that various APs (eg, Wi-Fi AP1, Wi-Fi AP2, LTE Cell Site, 5G Cell Site, etc.) Among them, a prediction model capable of outputting a result of determining which AP to use may be generated.

In addition, as non-supervised learning for predicting the performance of a specific AP with a specific index (e.g., speed), clustering can be used, and the performance of a specific access point is classified and classified by random clustered criteria. The way to do it can be illustrated.

Accordingly, the learning-based quality prediction module 200 substitutes the model generation variable to the non-supervised learning model, so that various APs (e.g., Wi-Fi AP1, Wi-Fi AP2, LTE Cell Site, 5G Cell Site, etc.) ), it is possible to create a learning model that can output the result of determining which AP to use.

Here, clustering is an unsupervised learning algorithm that divides the output into specific classes for specific inputs.K-means, Fuzzy K-means, Hierarchical clustering, Mixture of Gaussians, Mean-Shift Clustering, Density-Based Spatial Clustering of Applications with Noise (DBSCAN), etc. may be exemplified.

Meanwhile, the model application variable processor 140 determines a model application variable for quality prediction by the learning-based quality prediction module 200 and transmits it to the learning-based quality prediction module 200.

Here, the model application variable may be determined by collecting and refining the above-described variable information. The model application variable affects the current or future network quality in consideration of the wireless terminal device 100 and its communication environment and the persistence of the socket in response to a prediction function, a prediction method, and model generation information of a pre-trained network quality model. It can include various variables that affect

Accordingly, the learning-based quality prediction module 200 may input a model application variable to a pre-built quality prediction learning model, and output quality prediction information for each accessible network corresponding to the model application variable.

For example, the model application variable processor 140 may obtain model application variables from the wireless terminal device 100, and the model application variables include signal strength information applicable to a quality prediction learning model, time zone information, and terminal movement pattern. Information and the like may be illustrated.

For example, the learning-based quality prediction module 200 may apply a model application variable to an input layer of a quality prediction model, predict performance information corresponding to a network AP, and output the predicted performance information.

In addition, the learning-based quality prediction module 200 generates a relationship model f (E) between the model application variable (E) and the performance of the AP as a learning model, and the model application variable processor 140 applies a model applicable in response thereto. By determining a variable, a model result value according to the variation of the model application variable may be output as network performance information.

As described above, the model application variable information may include a variable element, and the constant element may also be used as a model application variable. These model application variables may be determined according to variables used to generate a predictive model.

For example, when a relationship model between the variable element information and the quality prediction learning model is constructed, the model application variable may be variable element information applicable to the relationship model.

In addition, for example, when a relationship model between the invariant element information and the quality prediction learning model is generated, the model application variable may include invariant element information applicable to the relationship model.

In addition, since the variable element and the constant element may be used in combination, it may be determined in various ways according to functions and variables used to generate a prediction model.

More specifically, for example, the model application variable processing unit 140 collects sensor information corresponding to a specific environment and environment information for each wireless communication network, calculates it as a model application variable, and the learning-based quality prediction module 200 Can be delivered to.

For example, the model application variable processor 140 refines the RSSI information, which is the signal strength of the wireless communication network, into a basic variable, and determines the distance information between the AP and the terminal as a model application variable, and the learning-based quality prediction module 200 Can be delivered to.

In addition, for example, the model application variable processing unit 140 may refine the speed (Velocity) or acceleration information of the wireless terminal device 100 as a basic variable to calculate movement pattern information. The movement pattern information may be determined as a model application variable and transmitted to the learning-based quality prediction module 200.

The learning-based quality prediction module 200 may calculate the relationship information between the distance information, the movement pattern information, and the learning model, and use it to predict the quality change for each network in the present or in the future.

Meanwhile, the network determination unit 150 compares the quality prediction information for each network according to preset user preference information or mode setting information, and transmits the optimal network according to the comparison result to a first wireless network to transmit a data packet. You can decide.

More specifically, the network determination unit 150 may determine a wireless network interface to be driven by the communication unit 120 or a network to which a socket generated for data packet transmission is bound according to a result of the comparison determination.

In addition, the network determiner 150 may compare and determine a network to be used based on at least one of quality information for each network, user preference information for speed and cost, and user setting mode information. The comparison determination process may be determined according to either a static determination or a learning determination.

The static determination method is determined, for example, according to variable comparison, and determination according to the user's speed and quality preference setting may be exemplified.

For example, when the user prefers speed, the network determination unit 150 may determine a network having a high quality variable, and when the user prefers cost, determine a WIFI network having a quality higher than a certain standard as the first network. However, if such a network does not exist, it can be processed to determine a mobile communication network. In addition, when quality and preference information is insufficient, the network determination unit 150 may set a plurality of sockets, bind to each connectable network, and determine a socket from which the first response comes earlier as the first network.

In addition, the network determination unit 150 The user preference can be determined based on the learning decision, and the network determination unit 150 performs actions defined as negative rewards such as a Wi-Fi Off action or a manual connection to another Wi-Fi from a user input. Is learned as an indicator that does not prefer a specific network, and thus learned user preference information may be used to determine the aforementioned first network.

In addition, the determination of the first wireless network by the network determination unit 150 may be periodically processed according to a certain time or may be processed at a specific time according to a certain condition in which quality deterioration of the current network is predicted. Through the processing, the user may feel little or no quality degradation.

In addition, the communication unit 120 may wirelessly transmit the data packet to the access target device 300 by using the determined first wireless network according to the determination of the network determination unit 150.

For such processing, the communication unit 120 may process network control, and process data transmission through the first wireless network in a manner that directly drives the network interface with hardware or controls the network layer by an application. can do.

For example, in the case of an application, data transmission through the first wireless network can be processed in the form of creating a socket and binding a specific network interface, which enables simultaneous use of a Wi-Fi wireless communication network and a cellular wireless communication network. .

Meanwhile, the user setting unit 170 can set a user mode for network switching, and thus mode information may be transmitted to the network determination unit 150 and used as a variable for determining the first network.

For example, the user mode may be any one of a speed mode, a balanced mode, or a cost mode.

The fast mode may be set when a terminal user who has no limit on the use of mobile communication data uses a short-range wireless communication network to amplify the speed. In this case, the user setting unit 170 may set a mobile communication network and a short-range wireless communication network to be mixed, thereby providing a faster speed and quality than a case of using a single wireless communication network.

The balance mode may be a setting in which a terminal user having a limitation in using mobile communication data uses a short-range wireless communication network to amplify speed, but preferentially uses a short-range wireless communication network based on a certain speed or quality. The user setting unit 170 is used only when the speed of the short-range wireless communication network is greater than or equal to the first speed, and when the speed is less than or equal to the first speed, the user setting unit 170 sets the mobile communication network to be mixed and provides the optimum speed and quality according to the situation. can do.

On the other hand, the price mode is a case in which a user with limited mobile communication data use uses a short-range wireless communication network as much as possible to save data, and the user setting unit 170 is a second speed that interferes with Internet use. Connection to the mobile communication network can be made only in the following cases.

3 and 4 are flowcharts illustrating a method of operating a wireless terminal device according to an embodiment of the present invention.

Referring to FIG. 3, first, the wireless terminal device 100 collects a wireless network quality model generation variable through the model generation variable processing unit 130 (S101).

Then, the wireless terminal device 100 performs refinement processing corresponding to the wireless network quality model generation variable through the model generation variable processing unit 130 (S102). And, the wireless terminal device 100 is a learning-based quality prediction module. Through 200, a quality prediction learning model for each wireless network is generated based on the model generation variable according to a preset relationship learning operation between variables (S103).

Thereafter, the wireless terminal device 100 processes the model application variable corresponding to the current state of the wireless terminal device 100 through the model application variable processing unit 140 (S105).

Then, the wireless terminal device 100 acquires quality prediction information for each accessible network according to the application of the quality prediction learning model of the model application variable through the learning-based quality prediction module 200 (S107).

Thereafter, the wireless terminal device 100 determines an optimal wireless network according to the user mode or preference setting of the wireless terminal through the network determination unit 150 (S109), and the control unit 110 determines data using the determined wireless network. Prepare for packet transmission (S111).

Then, the communication unit 120 transmits the data packet through the determined wireless network (S113).

Here, FIG. 4 is for describing a socket-based wireless network transmission process of the wireless terminal device 100 in more detail, and the controller 110 creates a socket for transmitting a wireless data packet (S151).

Then, the network determination unit 150 compares the network quality prediction information and user preference information that is statically determined or formed by learning, item by item, and determines one or more wireless networks (S153).

Here, the user preference information may be determined by the above-described user mode setting information, and a plurality of wireless networks may be determined.

Thereafter, the communication unit 120 binds and connects the generated socket to a network interface corresponding to the determined wireless network (S155).

Accordingly, the communication unit 120 may wirelessly transmit the data packet through the connected socket (S157).

5 is a conceptual diagram schematically showing a system based on an apparatus for providing quality prediction information according to another embodiment of the present invention, and FIG. 6 is a block diagram showing in detail an apparatus for providing quality prediction information according to an embodiment of the present invention.

5 and 6 illustrate a system in which the apparatus 400 for providing quality prediction information is configured as a separate device at a remote location. Referring to FIG. 5 first, a system according to another embodiment of the present invention includes learning-based quality prediction. A quality prediction information providing device 400 including a module 200, a wireless terminal device 100, and a connection target device 300, and the wireless terminal device 100 and the connection target device 300 are Networks (200a, 200b, ... 200n) may be configured to be connectable through one or more wireless networks, the quality prediction information providing apparatus 400 is a plurality of wireless networks (200a, 200b, ... 200n) And can be connected respectively. Also, the apparatus 400 for providing quality prediction information may provide a separate wired network connection.

Here, each of the wired/wireless networks that can be accessed is a local area network (LAN), a wide area network (WAN), a value added network (VAN), and a personal area network (PAN). ), a mobile radiocommunication network, a satellite communication network, and the like, and each base station or access point (AP) device for providing the same may be provided in a remote location.

In particular, in order to provide a local area communication network, the wireless network may include a WIFI wireless network based on a WIFI standard, and each access point may broadcast device information for operating a Wi-Fi service coverage.

Meanwhile, the wireless terminal device 100 may be a wireless terminal device of a user who requests transmission/reception of a wireless data packet by determining a first wireless network 200a to be connected to the access target device 300 among wireless networks. Here, the wireless terminal device 100 may be implemented as various devices such as a smart phone, a tablet computer, a notebook computer, a personal digital assistant (PDA), a portable multimedia player (PMP), and smart glasses.

In addition, according to an embodiment of the present invention, the apparatus 400 for providing quality prediction information may process a quality model generation variable for each wireless network and transmit it to the learning-based quality prediction module 200, and the learning-based quality prediction module 200 May generate a quality prediction learning model for each wireless network according to a relational learning operation of the model generation variable.

Here, the learning-based quality prediction module 200 may learn and calculate relationship information between variables by collecting various variables for generating a quality prediction learning model, and each variable is physical variable information for each communication section for quality prediction, an invariant element. It can be diversified into information and variable element information.

In addition, the wireless terminal device 100 determines a model application variable of the wireless terminal device 100 in response to a data packet to be transmitted wirelessly, and determines the model application variable for each wireless network provided from the quality prediction information providing device 400. Applying to a quality prediction learning model, obtaining quality prediction information for each accessible network, and determining a first wireless network 200a to transmit the data packet among the plurality of wireless networks based on the quality prediction information for each network, The data packet may be wirelessly transmitted to the access target device 300 using the determined first wireless network.

Here, the quality prediction information providing apparatus 400 may include the above-described learning-based quality prediction module 200, and the learning-based quality prediction module 200 is a model generation variable processor of the apparatus 400 for providing quality prediction information ( The above-described quality prediction model may be generated according to the variable processed at 430 ), and prediction information may be output according to the model application variable determined by the model application variable processor 440.

To this end, the information service provider 470 may provide a quality prediction learning model for each wireless network or may provide quality prediction information for each network corresponding to a model application variable according to a request from the wireless terminal device 100.

Accordingly, the wireless terminal device 100 receives an information service from the apparatus 400 for providing quality prediction information, and performs a relationship learning process between at least one of the network speed, signal strength, or preset model generation variables. Based on this, it is possible to more accurately predict the quality deterioration according to the model application variable in advance, and the dynamic variable control of the wireless network accordingly enables rapid and efficient network change while minimizing the user's experience of deteriorating quality.

More specifically, referring to FIG. 6, the apparatus 400 for providing quality prediction information according to an embodiment of the present invention includes a control unit 410, a communication unit 420, a model generation variable processing unit 430, and a model application variable processing unit ( 440), a learning-based quality prediction module 200, a database 460, and an information service provider 470.

The control unit 410 generally controls the operation of each component of the apparatus 400 for providing quality prediction information and execution of functions. For example, the controller 410 may be implemented as a processor for controlling all or part of a model generation variable processing, model application variable processing, and information service providing function, or a program for executing the same.

In addition, the communication unit 420 connects to the wireless terminal device 100 or wireless networks 200a, 200b, ... 200n, or connects to a wired network to enable wired/wireless data packet communication with each wireless network. It may include one or more communication modules.

In particular, the communication unit 420 may include a wireless Internet module for connecting to at least one of the wireless terminal device 100 or wireless networks 200a, 200b, ... 200n through a wireless Internet protocol, and a wireless Internet module Wireless LAN (WLAN) (Wi-Fi), Wireless broadband (Wibro), World Interoperability for Microwave Access (Wimax), and high speed downlink packet access (HSDPA) based network modules may be exemplified.

In addition, the communication unit 120 may include a short-range communication module for short-range communication access to at least one of the wireless terminal device 100 or wireless networks 200a, 200b, ... 200n, and a short-range communication protocol. As a short-range communication technology, Bluetooth, Radio Frequency Identification (RFID), infrared data association (IrDA), Ultra Wideband (UWB), ZigBee, and the like may be exemplified.

Meanwhile, the communication unit 120 may include a mobile communication module for mobile communication connection with at least one of the wireless terminal device 100 or wireless networks 200a, 200b, ... 200n, and For this reason, it is possible to transmit and receive wireless signals with at least one of a base station on a mobile communication network, an external terminal, a connection target device, and a server. The wireless signal may include a voice call signal, a video call signal, or various types of data according to transmission/reception of text/multimedia messages.

In addition, the communication unit 420 may provide the quality prediction information for each network to the wireless terminal device 100 under the control of the controller 410 or may provide prediction model information for determining the quality prediction information for each network. The information may be generated by the learning-based quality prediction module 200.

In order to determine the quality prediction information for each network, the learning-based quality prediction module 200 collects and refines the variable information for the model generation variable obtained from the model generation variable processor 430, and based on the processed model generation variable. A quality prediction learning model for each wireless network may be generated and stored in the database 460.

Here, the model generation variable processing unit 430 may collect and process various variables for generating a quality prediction learning model for each wireless network in the learning-based quality prediction module 200, and transmit them, and the model generation variable is a learning-based quality prediction. It may be determined differently according to the learning model and learning processes of the module 200. As the learning process, artificial intelligence or artificial neural network learning may be exemplified, and variable setting and learning model configuration for this are important, and for this, the learning-based quality prediction module 200 when provided in the wireless terminal device 100 described above is ) And its operation are similar, so a detailed description will be omitted.

In addition, the information service provider 470 may provide an information service for transmitting the quality prediction learning model information for each wireless network generated by the learning-based quality prediction module 200 to the wireless terminal device 100.

In addition, the information service providing unit 470 may transfer information of the wireless terminal device 100 according to the request of the wireless terminal device 100 to the model application variable processor 440, and the model application variable processor 440 is a wireless terminal. The device 100 determines a model application variable corresponding to the information, and transmits the determined model application variable to the learning-based quality prediction module 200, so that the learning-based quality prediction module 200 outputs the quality prediction information for each network. can do.

In addition, the information service provider 470 may transmit the obtained network-specific quality prediction information to the wireless terminal device 100 through the communication unit 420.

Accordingly, the wireless terminal device 100, which receives the quality prediction information for each network or obtained from the model, compares according to preset user preference information or mode setting information, and transmits a data packet to the optimal network according to the comparison result. It can be determined as the first wireless network to be performed.

In addition, since the wireless terminal device 100 can transmit a wireless data packet through the determined network, a detailed description thereof will be omitted since this is the same as described in FIG. 2.

7 is a flowchart illustrating an operation of an apparatus for providing quality prediction information according to an embodiment of the present invention.

Referring to FIG. 7, the apparatus 400 for providing quality prediction information according to an embodiment of the present invention collects a quality model generation variable for each wireless network through a model generation variable processor 430 (S201).

Then, the model generation variable processing unit 430 performs a variable refinement process on the model generation variable in order to generate a quality prediction learning model (S202).

In addition, the learning-based quality prediction module 200 of the apparatus 400 for providing quality prediction information generates a quality prediction learning model for each wireless network based on a model generation variable according to a preset relationship learning operation between variables (S203).

Thereafter, the apparatus 400 for providing quality prediction information processes the model application variable according to the request of the wireless terminal device 100 through the model application variable processing unit 440 (S205).

Then, the apparatus 400 for providing quality prediction information acquires quality prediction information for each network according to the application of the quality prediction learning model of the model application variable (S207).

Thereafter, the apparatus 400 for providing quality prediction information provides quality prediction information for each network to the wireless terminal (S209).

In this way, by providing the apparatus 400 for providing quality prediction information, it is possible to provide a dynamic wireless network variable access based on a quality prediction learning model even if the learning-based quality prediction module is not installed inside the wireless terminal device 100, It is possible to reduce infrastructure and cost by enabling the collection of learning data and building a model without consuming power and resources of the wireless terminal device 100.

8 is a block diagram illustrating a learning-based quality prediction module according to an embodiment of the present invention in more detail.

Referring to FIG. 8, the learning-based quality prediction module 200 according to an embodiment of the present invention includes one or more input layer processing units 210 and one or more first intermediate layer processing units 220a, 220b, ... 220n, and A feedback weight application unit 240 that includes a plurality of n-th intermediate layer processing units including one or more second intermediate layer processing units 230a, 230b, ... 230n and an output layer processing unit 240, and corresponding to an output layer It may optionally include.

The learning-based quality prediction module 200 converts the quality model generation variable among the variables collected by the model generation variable processing unit 430 into respective input layers, and performs a convolution operation on the processed input layers to produce an output associated with the network quality. It is possible to construct an artificial neural network that outputs to a layer. For this, each intermediate layer processor may process an individual multiplication operation according to a neural network network, and the feedback weight application unit 245 may selectively perform a weight adjustment process according to an evaluation of an output layer.

Through the iterative processing of such a learning process, a wireless network quality prediction model corresponding to a quality model generation variable for each wireless network may be built, and the built prediction model is stored in the wireless network quality prediction model storage unit 250, or It is transmitted to the database of the wireless terminal device 100 or the apparatus 400 for providing quality prediction information, and may be used to predict the quality of each network according to the determination of a model application variable later.

9 to 10 are diagrams for explaining a data merging and sharing process of a learning-based quality prediction model according to an embodiment of the present invention.

The data on the performance of each wireless terminal device 100 and the AP may be used to predict the performance of the AP even if the device is not directly connected. Therefore, merge sharing of prediction models enables more accurate prediction of AP performance.

Accordingly, referring to FIGS. 9 and 10, the learning-based quality prediction model according to an embodiment of the present invention may be formed by merging shared data, and data exchange for this may be performed by the wireless terminal device 100 or It may be processed by the prediction information providing device 400.

First, FIG. 9 shows a case where the learning-based quality prediction module 200 is provided in the wireless terminal device 100 so that the learning-based quality prediction model sharing data is shared among other wireless terminal devices 100a. The quality prediction model data or training data learned in (100) may be shared and merged to be used in another wireless terminal device (100a). For example, data on the performance of the wireless terminal device 100 and APs of each wireless network may be transmitted to other wireless terminal devices 100a.

For example, the wireless terminal device 100 is based on learning data about the performance between the wireless terminal device 100 and the AP generated by the learning-based quality prediction module 200 (for example, speed, specific variables and performance). Function data indicating a related relationship) to the other wireless terminal device 100a connected to the network.

Here, each wireless terminal device may be connected to a data sharing network such as a block chain network, and each wireless terminal device may store learning model data in a shared DB based on a block chain network.

Thereafter, the specific wireless terminal device 100 may request data related to the performance of a specific AP from the shared network to determine a wireless network, and the shared network may request quality related to the performance of the specific AP based on the previously stored merged shared data. Data can be transmitted to the wireless terminal device 100.

In addition, the wireless terminal device 100 receiving performance or quality data of a specific AP may determine and connect to a first wireless network based on this. In addition, when the wireless terminal device 100 collects the additional data of the AP, it may transmit it to the shared network and merge it into the training data for a learning-based quality prediction model.

In the data merging process, the additional data may be merged by weighting according to the similarity between variables of the previously stored data (eg, model, number of connected devices, time zone) and variables of the wireless terminal device 100, The average value, minimum value or maximum value during may be used. The data merging process may be distributed in a shared network, or data merged in the wireless terminal device 100 may be transmitted to the shared network.

In addition, since the use of shared data is not essential, the wireless terminal device 100 may perform learning and prediction processing using only individually collected data.

On the other hand, FIG. 10 shows a case where the learning-based quality prediction module 200 is provided in the quality prediction information providing apparatus 400 so that the learning-based quality prediction model sharing data is shared with other wireless terminal devices 100a. The quality prediction model data or training data learned by the terminal device 100 and other wireless terminal devices 100a may be shared and merged through the quality prediction information providing apparatus 400.

For example, the apparatus 400 for providing quality prediction information collects data on the performance of each AP of each wireless network from the wireless terminal device 100 and other wireless terminal devices 100a, and Learning data (eg, speed, function data representing a relationship between specific variables and performance) can be merged.

In addition, the quality prediction information providing apparatus 400 provides learning model data according to merge learning to the wireless terminal device 100 and other wireless terminal devices 100a connected to the network. The provision of model data can be processed according to a certain period.

For example, in order to predict performance information of a specific AP for determining a wireless network in a specific wireless terminal device 100, model data provided from the information service providing unit 470 of the quality prediction information providing device 400 may be used. have.

When the wireless terminal device 100 collects the additional data of the AP, it may transmit it to the quality prediction information providing device 400 and merge it into training data for a learning-based quality prediction model.

In the data merging process, the additional data may be merged by weighting according to the similarity between variables of the previously stored data (eg, model, number of connected devices, time zone) and variables of the wireless terminal device 100, The average value, minimum value or maximum value during may be used. The data merging process may be processed by the apparatus 400 for providing quality prediction information, or data merged by the wireless terminal device 100 may be transmitted to the apparatus 400 for providing quality prediction information.

Meanwhile, the above-described method according to various embodiments of the present invention may be implemented as a program and provided to each server or devices while being stored in various non-transitory computer readable media. Accordingly, the user terminal 100 can access the server or device and download the program.

The non-transitory readable medium refers to a medium that stores data semi-permanently and can be read by a device, not a medium that stores data for a short moment, such as a register, cache, or memory. Specifically, the above-described various applications or programs may be provided by being stored in a non-transitory readable medium such as a CD, DVD, hard disk, Blu-ray disk, USB, memory card, and ROM.

In addition, although the preferred embodiments of the present invention have been illustrated and described above, the present invention is not limited to the specific embodiments described above, and the technical field to which the present invention belongs without departing from the gist of the present invention claimed in the claims. In addition, various modifications are possible by those of ordinary skill in the art, and these modifications should not be individually understood from the technical spirit or prospect of the present invention.

Claims (19)

In the operating method of a wireless terminal device accessible to a plurality of wireless networks,
Obtaining a data packet to be transmitted;
Determining a model application variable of the wireless terminal device;
Obtaining quality prediction information for each accessible network by applying the model application variable to a quality prediction learning model for each wireless network;
Determining a first wireless network to transmit the data packet from among the plurality of wireless networks based on the quality prediction information for each network; And
Including the step of wirelessly transmitting the data packet to a target device using the determined first wireless network
A method of operating a wireless terminal device. The method of claim 1,
Collecting basic parameters for generating a quality model for each wireless network;
Refining the basic variable to obtain the model generation variable; And
Generating a quality prediction learning model for each wireless network according to a relationship learning operation of the refined model generation variable
A method of operating a wireless terminal device. The method of claim 2,
Generating the learning model,
Obtaining variable information for each communication section from the model generation variable; And
The step of generating a quality prediction learning model for each wireless network according to a relation learning operation for each communication section based on the variable information for each communication section.
A method of operating a wireless terminal device. The method of claim 3,
The variable information for each communication section is
Including physical variable information corresponding to at least one communication section among a wired section between an access target device and an AP (Access Point) access point, an AP internal section, a wireless section between the AP and a wireless terminal device, and a wireless terminal device internal section
A method of operating a wireless terminal device. The method of claim 4,
The physical variable information,
Wired cable performance information corresponding to the wired section between the connection target device and the AP access point, including information on the difference between the IP of the connection target device and the IP of the AP
A method of operating a wireless terminal device. The method of claim 4,
The physical variable information,
Including AP hardware capability information and number of connected devices corresponding to the AP inner section
A method of operating a wireless terminal device. The method of claim 4,
The physical variable information,
Including at least one of protocol information, signal strength information, frequency band information, and channel/frequency congestion information corresponding to a radio section between the AP and the wireless terminal device
A method of operating a wireless terminal device. The method of claim 4,
The physical variable information,
Including at least one of wireless module performance information, wireless module characteristic information, and temperature information corresponding to the internal section of the wireless terminal device
A method of operating a wireless terminal device. The method of claim 2,
Generating the learning model,
Obtaining constant element information and variable element information from the model generation variable;
Generating a quality prediction learning model for each wireless network according to a relational learning operation based on the constant element information; And
Including the step of generating a relationship model between the variable element information and the quality prediction learning model,
The step of obtaining the quality prediction information for each accessible network,
And obtaining quality prediction information for each network by applying the model application variable to the variable element of the relationship model.
A method of operating a wireless terminal device. The method of claim 9,
The constant element includes at least one of performance information of an access point (AP) providing a wireless network, information on the number of neighboring APs, and communication protocol information,
The variable element includes at least one of information on the number of devices connected to the AP, information on the strength of an access signal, information on a TCP socket destination, information on a round trip time (RTT), information on a window size, and information on a transmission/reception data
A method of operating a wireless terminal device. The method of claim 1,
The plurality of wireless networks include one or more WIFI standard networks,
Obtaining the quality prediction information,
Obtaining hardware information of an AP providing the WIFI standard network and signal strength information with the wireless terminal as the model application variable; And
Comprising the step of obtaining, from the quality prediction learning model for each wireless network, quality prediction information when using the WIFI standard network of the wireless terminal corresponding to the hardware information and signal strength information
A method of operating a wireless terminal device. The method of claim 11,
The quality prediction learning model for each wireless network,
Learning operation, logistic regression operation, artificial neural network operation, unsupervised learning operation, supervised learning with the hardware performance information calculated from the unique characteristic information of each AP and the unique characteristic information of each wireless terminal as an input variable and communication quality information as an output variable It is characterized in that generated according to at least one of the operations
A method of operating a wireless terminal device. The method of claim 12,
The unique characteristic information for each AP is calculated from the radiated power intensity information of the AP determined corresponding to the MCS (Modulation Coding Scheme) index of the AP.
A method of operating a wireless terminal device. The method of claim 12,
The unique characteristic information for each AP includes minimum signal strength information and maximum speed information calculated for each AP.
A method of operating a wireless terminal device. The method of claim 1,
The plurality of wireless networks includes one or more mobile communication networks,
Obtaining a moving average value by processing a Decay function based on velocity samples of the last n data obtained from the mobile communication network; And
And generating a quality prediction learning model corresponding to the mobile communication network by performing regression learning processing of transmission performance information corresponding to the moving average value.
A method of operating a wireless terminal device. The method of claim 1,
The quality prediction learning model for each wireless network is
Characterized in that it is formed to be shared according to exchange of learning data with other wireless terminal devices connected to the wireless terminal device
A method of operating a wireless terminal device. The method of claim 16,
Further comprising the step of sharing the variable data for exchanging the learning data to the other wireless terminal devices according to a predetermined period
A method of operating a wireless terminal device. The method of claim 1,
The step of determining the first wireless network,
Comparing the quality prediction information for each network according to preset user preference information or mode setting information; And
And determining an optimal network according to the comparison result as the first wireless network
A method of operating a wireless terminal device. A computer-readable nonvolatile recording medium in which a program for executing the method according to any one of claims 1 to 18 on a computer is recorded.

Images