KR20190111587A - Apparatus and method for estimating location of user terminal based on deep learning - Google Patents

Abstract

The present invention proposes a deep-learning-based user terminal location estimation apparatus and method for estimating a location of a user terminal having no GPS function or a GPS function not normally operating based on training information learned using deep learning. According to the present invention, the apparatus comprises: a first information generation unit for generating location information and first reference signal reception power of at least one first base station located in an operation area based on information on a radio signal in the operation area; a parameter acquisition unit for obtaining a parameter including a weight and bias obtained through machine learning based on the location information of a first user terminal; and a terminal location estimation unit for estimating a location of a second user terminal located in the operation area based on the location information of the first base station, the first reference signal reception power, and the parameter.

Description

Apparatus and method for estimating location of user terminal based on deep learning}

The present invention relates to an apparatus and method for estimating the position of a user terminal. More particularly, the present invention relates to an apparatus and method for estimating the position of a user terminal based on trained information using machine learning.

The user equipment position estimation system estimates the position of the user equipment by using various radio signals and surrounding environment elements transmitted to the base stations to which the user equipment is connected in a wireless communication situation.

Conventional techniques for estimating user equipment location in wireless communication situations include Enhanced Cell ID (ECID), Assisted Global Navigation Satellite Systems (A-GNSS), Observed Time Difference Of Arrival (OTDOA), and the like.

However, ECID technology affects the round trip time calculation due to the uncertainty of the reception time, and there is a possibility of error due to multiple reflections.

In addition, A-GNSS works very well in outdoor environments, but the performance is not good in indoor environments or very dense cities, and there is a problem in that the user cannot turn off the GNSS function of the terminal.

In addition, in the case of OTDOA technology, in order to calculate the location of user equipment, the network must have the location of the base station transmit antenna and the transmission time of each cell, and it is very accurate to measure the reference signal of the neighboring cell. It should be possible.

Korean Patent Publication No. 2014-0076945 (Publication date: June 23, 2014)

The present invention has been made to solve the above-described problems, deep learning based on the training information learned using deep learning deep learning to estimate the location of the user terminal without the GPS function or the GPS function does not operate normally An object of the present invention is to propose an apparatus and method for estimating a location of a user terminal.

However, the object of the present invention is not limited to the above-mentioned matters, and other objects not mentioned will be clearly understood by those skilled in the art from the following description.

The present invention has been made to achieve the above object, the location information and the first reference signal reception power of at least one first base station located in the operating area based on the information on the radio signal in the operating area (Reference A first information generator for generating a signal received power; A parameter obtaining unit obtaining a parameter including a weight and a bias obtained through machine learning based on the location information of the first user terminal; And a terminal location estimator configured to estimate a location of a second user terminal located in the operating area based on the location information of the first base station, the first reference signal reception power, and the parameter. We propose a user terminal location estimation device.

In addition, the present invention is based on the information on the radio signal in the operating area location information of the at least one first base station located in the operating area and the first information for generating a first reference signal received power (Reference Signal Received Power) Generation step; A parameter obtaining step of obtaining a parameter including a weight and a bias obtained through machine learning based on the location information of the first user terminal; And a terminal location estimating step of estimating a location of a second user terminal located in the operating area based on the location information of the first base station, the first reference signal reception power, and the parameter. We propose a method for estimating a location based on a user terminal.

The present invention can achieve the following effects through the configuration for achieving the above object.

First, deep learning-based algorithms can more accurately model the relationship between input and output parameters.

Second, the position estimation accuracy can be improved by additionally using the information of the past time and the future time, and as the training data accumulates, the position estimation accuracy can be further improved.

1 is a conceptual diagram of an entire system for deep learning based user equipment location estimation according to an embodiment of the present invention.
2 is a detailed configuration diagram of the training parameter generator provided in the entire system according to an embodiment of the present invention.
3 is an overall schematic diagram of a network used in a deep neural network training unit provided in the entire system according to an embodiment of the present invention.
4 is a detailed configuration diagram of a test parameter generation unit provided in the entire system according to an embodiment of the present invention.
5 is a block diagram schematically illustrating an internal configuration of a deep learning based user terminal position estimation apparatus according to an exemplary embodiment of the present invention.
FIG. 6 is a block diagram schematically illustrating an internal configuration of a first information generator included in a deep learning based user terminal position estimation apparatus. Referring to FIG.
FIG. 7 is a block diagram schematically illustrating an internal configuration of a second information generator included in a deep learning based user terminal position estimation apparatus. Referring to FIG.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. First, in adding reference numerals to the components of each drawing, it should be noted that the same reference numerals are assigned to the same components as much as possible even if displayed on different drawings. In addition, in describing the present invention, when it is determined that the detailed description of the related well-known configuration or function may obscure the gist of the present invention, the detailed description thereof will be omitted. In addition, the following will describe a preferred embodiment of the present invention, but the technical idea of the present invention is not limited thereto and may be variously modified and modified by those skilled in the art.

Techniques for estimating user equipment location in a wireless communication situation include enhanced cell ID (ECID), assisted global navigation satellite systems (A-GNSS), and Observed Time Difference Of Arrival (OTDOA).

ECID technology extends from simple but very inaccurate Cell ID (CID) technology, which simply assumes that the location of the connected base station is the location of the user equipment, thereby providing a user with a round trip time (RTT) between the base station and the user equipment. It is used to estimate the distance to the equipment, and the angle of Arrival (AOA) from the user equipment is used as the direction information.

The round trip time is determined by analyzing the timing advance (TA) measurement of the radio signal, and the angle of arrival is determined from the known configuration of the base station antenna arrangement and the user equipment. The ECID technique performs better than the conventional CID technique, but the uncertainty of the reception time affects the round trip time calculation, and there is a possibility of error due to multiple reflections.

GNSS refers to multiple satellite systems, such as GPS or GLONASS. A-GNSS technology is augmented with data called asstance data provided from the network in the GNSS function. Earth data includes information that a mobile GNSS receiver can use to speed up the satellite signal acquisition process.

A-GNSS works very well in outdoor environments, but it does not perform well in indoor environments or very dense cities, and is not available when the user turns off the GNSS feature of the device.

OTDOA technology uses user equipment to measure the time difference between signals from two or more base stations. The signals used in the OTDOA generally use a reference signal (RS) and reference the measured time difference between a reference signal from a serving cell and a reference signal from a neighboring cell. This is called a reference signal time difference (RSTD).

However, in order to calculate the location of the user equipment, the network must have the location of the base station transmit antenna and the transmission time of each cell. The biggest problem with OTDOA technology is that it must be able to measure the reference signal of a neighboring cell very accurately.

Conventional user equipment position estimation techniques have difficulties as described for each technique, and since the wireless signals vary greatly depending on the surrounding environment, conventional statistical based techniques allow complex nonlinearity between the radio signal and the position of the user equipment. Modeling relationships is not easy

The present invention relates to an apparatus and method for estimating a location of user equipment using a deep learning based algorithm in an LTE wireless communication environment. More specifically, when there is a user equipment that does not have the GPS function turned on and the exact location is not known, the deep learning input parameters of the wireless measurements transmitted by the user equipment to the base stations connected to the user equipment and the latitude and longitude of the base stations are determined. And the latitude and longitude of the user equipment as deep learning output parameters to model complex and nonlinear relationships between inputs and outputs as deep learning based algorithms, as well as input past and future time together with current time input parameters. The present invention relates to a deep learning-based user equipment location technique that improves estimation performance by applying a context window that also bundles parameters.

DETAILED DESCRIPTION Hereinafter, a deep learning based user equipment position estimation technique using location information and a reference signal received power (RSRP) of a base station connected to a user equipment (UE) will be described with reference to the drawings.

1 is a conceptual diagram of an entire system for deep learning based user equipment location estimation according to an embodiment of the present invention.

The system 100 proposed by the present invention uses a deep learning-based user equipment location estimation algorithm rather than the conventional statistical method, and thus the complex and nonlinear relationship between the reference signal reception power, the latitude of the base station, the longitude and the latitude of the user equipment, and the longitude Model relationships effectively.

In addition, the system 100 proposed in the present invention can improve the estimation accuracy by making a more generalized model as training data is accumulated through deep learning based user equipment position estimation algorithm.

In addition, the system 100 proposed by the present invention applies a context window that uses not only an input parameter of a current time but also an input parameter of a past time and a future time closely related to an output parameter of a current time. By using more information to estimate the position of the equipment, the estimation accuracy can be improved.

Referring to FIG. 1, the deep learning based user equipment position estimation technique proposed by the present invention uses a model training step (S10) of training a deep neural network model for position estimation, and a database of user equipment to estimate a position. And estimating the position of the user equipment through the trained deep neural network model.

The model training step (S10) generates a training parameter for training the deep neural network model from the database so that the deep neural network model can estimate the position of the user equipment, and trains the deep neural network model using the training parameter. Step.

In order to train the deep neural network model to better estimate the location of the user equipment, it is important to select and extract parameters that are highly correlated with the location of the user equipment from a database containing a large amount of data. This process is performed in the training parameter generator 110.

2 is a detailed configuration diagram of the training parameter generator provided in the entire system according to an embodiment of the present invention.

The training parameter generator 110 generates training parameters for training the deep neural network. In detail, the training parameter generator 110 is a block for generating an input parameter to be input to the input layer of the deep neural network required for training the deep neural network in the training area wireless signal database 120 and an output parameter to be output from the output layer. In the present invention, the input parameter and the output parameter are defined as training parameters.

The number of base stations connected to a wireless signal varies according to the surrounding environment. However, in order to train the deep neural network, since the number of input parameters must be uniform, a first data extracting unit 210 which extracts only data when N base stations are connected is required. For example, in case of extracting only data when three base stations are connected, data when only one base station is connected and data when only two base stations are connected are not extracted.

The training parameter extracting unit 220 extracts a PCI (Physical Cell Id) used to know the reference signal reception power and the latitude and longitude of the base station among the extracted data, and the latitude and longitude of the user equipment. The reference signal reception power is input to the first input parameter generator 230, the PCI is input to the first base station location information extractor 240, and the latitude and longitude of the user equipment are directly the first data normalizer 250. Is entered.

The first base station location information extraction unit 240 is a block for extracting the latitude and longitude of the base station using the received PCI and the base station database 130. The base station database 130 contains data such as PCI, latitude and longitude of the base station. The first base station location information extracting unit 240 extracts the latitude and longitude of the base station mapped by comparing the received base station PCI with the base station database 130. Latitude and longitude of the base station is input to the first input parameter generator 230.

The first input parameter generator 230 is a block for generating an input parameter used to train the deep neural network. The first input parameter generator 230 receives the reference signal reception power from the training parameter extractor 220, and receives the latitude and longitude of the base station from the first base station location information extractor 240.

The first input parameter generator 230 combines the received input parameters of the current time into a vector with the input parameters of the past time and the future time by using a context window, and finally input layer of the deep neural network. Create an input parameter to be input to.

The first data normalizer 250 receives input parameters from the first input parameter generator 230 and receives latitude and longitude of the user equipment from the training parameter extractor 220.

Each parameter is very different from each other in the magnitude or spread of the values. In order for a deep neural network to be trained well, each parameter should have a similar distribution with each other near zero. This change in parameter distribution is called data normalization.

In the present invention, the first data normalization unit 250 subtracts one parameter into the average value of the parameter and divides it by the standard deviation value of the parameter to normalize the normalized parameter to have a mean value of 0 and a standard deviation value of 1. Work on each parameter.

Finally, the training parameter generator 110 outputs the reference signal reception power of the normalized past, present, and future time, the latitude and longitude of the base station as input parameters, and the latitude and longitude of the normalized user equipment as output parameters.

3 is an overall schematic diagram of a network used in a deep neural network training unit provided in the entire system according to an embodiment of the present invention.

The deep neural network training unit 140 trains the deep neural network using the training parameters generated by the training parameter generator 110. To this end, the deep neural network training unit 140 receives an input parameter and an output parameter from the training parameter generator 110. The input parameters are used in the input layer 310 of the deep neural network, and the output parameters are used in the output layer 320 of the deep neural network.

The deep neural network training unit 140 trains the deep neural network in the hidden layer 330 through a feedforward process and a backpropagation process when an input parameter and an output parameter are input. The deep neural network model 150 trained through this process can estimate the location of the user equipment using a wireless signal database of the user equipment whose location is not known.

Position estimation step (S20) to be described below is a process for estimating the position of the user equipment in the above situation. The following is a description of the position estimation step (S20).

In the location estimation step S20, the location of the user equipment is estimated using the test area wireless signal database 160 that does not have the location information of the user equipment because the user does not use the GPS function of the equipment. Generate the input parameters used as the input of the deep neural network model 150 from the database 160, and put the input into the deep neural network model 150 trained in the model training step (S10), the deep neural network model 150 is trained As shown, the latitude and longitude of the user equipment are estimated and output.

Since the output parameter output from the deep neural network model 150 is a normalized value, the data denormalization unit 170 denormalizes the output parameter in order to return it to an accurate value and finally estimates latitude and longitude of the user equipment.

4 is a detailed configuration diagram of a test parameter generation unit provided in the entire system according to an embodiment of the present invention.

The test parameter generator 180 generates a test parameter necessary in a database. The test parameter generator 180 plays a role similar to that of the training parameter generator 110 (see FIG. 2) in the model training step S10.

The training parameter generator 110 generates an input parameter consisting of reference signal reception power of the past, present, and future time, the latitude and longitude of the base station, and an output parameter consisting of the latitude and longitude of the user equipment.

On the other hand, since the test parameter generator 180 does not have information about latitude and longitude of the user equipment in the test area wireless signal DB 160 and is an object to be estimated, only the input parameter is generated without generating the output parameter. do. Therefore, like the first data extractor 210 of FIG. 2, the second data extractor 410 extracts only data when N base stations are connected.

Similar to the training parameter extracting unit 220 of FIG. 2, the test parameter extracting unit 420 extracts a PCI used to know the reference signal receiving power, the latitude and longitude of the base station, and the reference signal receiving power is reduced. 2 is input to the input parameter generator 430, PCI is input to the second base station location information extraction unit 440.

However, as mentioned earlier, the latitude and longitude of the user equipment used as output parameters are not extracted.

The second base station location information extractor 440 plays the same role as the first base station location information extractor 240 of FIG. 2.

The second input parameter generator 430 plays the same role as the first input parameter generator 230 of FIG. 2.

Like the first data normalization unit 250 of FIG. 2, the second data normalization unit 450 performs a task of normalizing each parameter of the input parameter.

However, in this case, the training area wireless signal database calculated by the first data normalization unit 250 of FIG. 2 is not normalized using average values and standard deviation values of the input parameters extracted from the test area wireless signal database 160. Normalization is performed using the mean values and standard deviation values of the input parameters extracted in step 120).

That is, in each of the input parameters generated by the second input parameter generator 430 of FIG. 4 using the average value and the standard deviation value of the respective input parameters calculated by the first data normalization unit 250 of FIG. 2. The average value is subtracted and normalized by dividing by the standard deviation value.

Finally, the test parameter generator 180 outputs normalized past, present, and future reference signal reception power, latitude, and longitude of the base station as input parameters.

When the input parameter generated by the test parameter generator 180 is input to the trained deep neural network model 150, the input parameter generated by the training neural network model 150 may be assigned to weight values and bias values determined through the training of the deep neural network model 150. The position and hardness of the user equipment are calculated (estimated) and output.

In this case, when the deep neural network model 150 is trained by using the normalized input parameter and the output parameter, the output parameter output above has a normalized scale. The data denormalization unit 170 is included in order to find the correct value by changing the scale to the original scale.

The data denormalizer 170 denormalizes the normalized output parameter and restores the original data distribution. The data denormalization unit 170 outputs from the deep neural network using average values and standard deviation values of the output parameters extracted from the training area wireless signal database 120 calculated by the first data normalization unit 250 of FIG. 2. Denormalize the parameter.

Inverse normalization proceeds through operations exactly the opposite of normalization. That is, each of the output parameters output from the deep neural network is multiplied by the above standard deviation values, and added to the above average values, respectively, and denormalized.

Finally, the data denormalization unit 170 outputs the estimated latitude and longitude of the user equipment.

Conventional user equipment position estimation techniques are all statistically based techniques, so it is difficult to properly model the complex nonlinear relationship between the radio signals and the position of the user equipment which are greatly changed by the surrounding environment. In addition, the location and signal characteristics of the base station are slightly different for each region, and one model is difficult to actively cope with various environments.

By applying the deep learning method to the user equipment location estimation technology, a deep learning model is trained using a variety of data to model the positional relationship between the wireless signal and the user equipment and actively cope with various environments. Aim to provide.

In addition, since the wireless signal is time series data, the information of the current time is deeply correlated with the information of the adjacent time. Therefore, the present invention may further improve performance by using not only the information of the current time but also the information of the past time and the future time. Suggest ways to train.

The main features of the present invention are summarized as follows.

First, the present invention includes a method for generating an input parameter and an output parameter for use in an input layer and an output layer of a deep neural network, and a normalization process of parameters used in the deep neural network. Deep learning based user equipment location estimation technology.

Secondly, the present invention relates to a method of using latitude, longitude information and reference signal reception power of a base station as an input parameter in an input parameter generation process.

Third, the present invention relates to a method of using latitude and longitude information of a user terminal as an output parameter in an output parameter generation process.

Fourth, the present invention relates to a training method that uses a context window in the input layer in the process of generating an input parameter and uses not only the current parameter but also the past and future time.

As mentioned above, one Embodiment of this invention was described with reference to FIGS. Hereinafter, the preferable form of this invention which can be inferred from such one Embodiment is demonstrated.

5 is a block diagram schematically illustrating an internal configuration of a deep learning based user terminal position estimation apparatus according to an exemplary embodiment of the present invention.

According to FIG. 5, the user terminal position estimation apparatus 500 estimates the position of the user terminal based on the training information learned using deep learning, and includes a first information generator 510 and a parameter acquirer 520. The terminal position estimator 530 includes a power supply unit 540 and a main controller 550.

The power supply unit 540 supplies power to each component of the user terminal position estimation apparatus 500.

The main controller 550 performs a function of controlling the overall operation of each component of the user terminal position estimation apparatus 500.

The first information generator 510 generates location information and first reference signal received power of at least one first base station located in the operating area based on the information on the radio signal in the operating area. It performs the function. The first information generator 510 is a concept corresponding to the test parameter generator 180 of FIG. 1.

As illustrated in FIG. 6, the first information generator 510 may include an operation region information generator 511, a second base station location detector 512, and a second location and received power updater 513. . FIG. 6 is a block diagram schematically illustrating an internal configuration of a first information generator included in a deep learning based user terminal position estimation apparatus. Referring to FIG.

The operating area information generation unit 511 generates a physical cell id (PCI) and a first reference signal reception power of the first base station based on the information on the radio signal in the operating area. The operation region information generator 511 is a concept corresponding to the test parameter extractor 420 of FIG. 4.

The second base station location detection unit 512 detects the location information of the first base station by comparing the PCI of the first base station generated by the operation region information generation unit 511 and the PCI stored in the database. The second base station position detector 512 is a concept corresponding to the second base station position information extractor 440 of FIG. 4.

The second location and reception power updating unit 513 updates the location information and the first reference signal reception power of the first base station detected and generated at the current time based on the information detected and generated at another time. . The second location and reception power updater 513 is a concept corresponding to the second input parameter generator 430 of FIG. 4.

The second position and the received power updater 513 may include a vector component associated with the position information of the first base station detected at the current time, a vector component associated with the first reference signal received power generated at the current time, and a second component detected at another time. Location information of the first base station detected at the current time by applying a vector component associated with the location information of the base station and a context window for mutually coupling the vector component associated with the first reference signal received power generated at another time. And update the first reference signal reception power generated at the current time.

The second position and the received power updater 513 may include the position information of the first base station detected at a previous time of the current time with the information detected and generated at another time, the first reference signal received power generated at the previous time, and the present time. The location information of the first base station detected at a later time of time and the first reference signal received power generated at a later time may be used.

The first information generator 510 may further include at least one of the second wireless signal information extractor 514 and the second normalizer 515.

The second radio signal information extractor 514 determines the number of first base stations and performs a function of extracting information on radio signals in an operating area based on the number of first base stations. The second wireless signal information extractor 514 is a concept corresponding to the second data extractor 410 of FIG. 4.

The second normalizer 515 performs data normalization of the position information of the first base station and the first reference signal received power updated by the second position and the received power updater 513. The second normalizer 515 is a concept corresponding to the second data normalizer 450 of FIG. 4.

Meanwhile, the first information generator 510 may generate the latitude of the first base station and the longitude of the first base station using the location information of the first base station.

This will be described with reference to FIG. 5 again.

The parameter obtaining unit 520 performs a function of obtaining a parameter including a weight and a bias obtained through machine learning based on the position information of the first user terminal.

The terminal location estimator 530 operates based on the location information of the first base station and the first reference signal reception power generated by the first information generator 510, and the parameters obtained by the parameter acquirer 520. A function of estimating a location of a second user terminal located in an area is performed. The terminal position estimator 530 is a concept corresponding to the deep neural network model 150 of FIG. 1.

The terminal location estimator 530 may estimate the location of the second user terminal when the second user terminal does not have a GPS function or the GPS function does not operate normally in the second user terminal.

The terminal location estimating unit 530 is based on the location information of at least one second base station located in the training area and the information obtained by calculating the second reference signal reception power in the training area. The location of the second user terminal may be estimated by denormalizing the information, the data normalized first reference signal reception power, and the data normalized parameter in order. The terminal location estimator 530 may be a concept including the data denormalization unit 170 of FIG. 1 in consideration of such a function.

The terminal location estimator 530 may estimate the latitude of the second user terminal and the longitude of the second user terminal based on the location of the second user terminal.

The terminal location estimator 530 may estimate the location of the second user terminal after a predetermined time has elapsed based on the current time.

The user terminal location estimation apparatus 500 may further include a second information generator 560 and a parameter generator 570.

The second information generator 560 may include location information of the first user terminal located in the training area, location information of at least one second base station located in the training area, and the first information based on the information about the radio signal in the training area. 2 generates a reference signal reception power. The second information generator 560 is a concept corresponding to the training parameter generator 110 of FIG. 1.

As illustrated in FIG. 7, the second information generator 560 may include a training area information generator 561, a first base station location detector 562, and a first location and received power updater 563. . FIG. 7 is a block diagram schematically illustrating an internal configuration of a second information generator included in a deep learning based user terminal position estimation apparatus. Referring to FIG.

The training area information generation unit 561 generates a location information of the first user terminal, a physical cell id (PCI) of the second base station, and a second reference signal reception power based on the information on the radio signal in the training area. Do this. The training area information generator 561 is a concept corresponding to the training parameter extractor 220 of FIG. 2.

The first base station location detector 562 compares the PCI of the second base station generated by the training area information generator 561 with the PCI stored in the database to detect the location information of the second base station. The first base station location detector 562 is a concept corresponding to the first base station location information extractor 240 of FIG. 2.

The first location and reception power updater 563 updates the location information and the second reference signal reception power of the second base station detected and generated at the current time based on the information detected and generated at another time. . The first location and reception power updater 563 is a concept corresponding to the first input parameter generator 230 of FIG. 2.

The first position and the received power updater 563 may include a vector component associated with the position information of the second base station detected at the current time, a vector component associated with the second reference signal received power generated at the current time, and a second component detected at another time. Location information of the second base station detected at the current time by applying a vector component associated with the location information of the base station 2 and a context window for mutually coupling the vector component associated with the second reference signal received power generated at another time. And update the second reference signal reception power generated at the current time.

The first position and the received power updater 563 are position information of the second base station detected at a previous time of the current time with the information detected and generated at another time, the second reference signal received power generated at the previous time, the current The location information of the second base station detected at a later time of time and the second reference signal received power generated at a later time may be used.

The second information generator 560 may further include at least one of the first wireless signal information extractor 564 and the first normalizer 565.

The first radio signal information extractor 564 determines the number of second base stations and performs a function of extracting information on radio signals in a training area based on the number of second base stations. The first wireless signal information extractor 564 is a concept corresponding to the first data extractor 210 of FIG. 2.

The first normalizer 565 performs a function of data normalizing the location information of the first user terminal, the location information of the updated second base station, and the updated second reference signal reception power. The first normalizer 565 is a concept corresponding to the first data normalizer 250 of FIG. 2.

This will be described with reference to FIG. 5 again.

The parameter generator 570 is based on deep learning based on the location information of the first user terminal, the location information of the second base station, and the second reference signal received power generated by the second information generator 560. It performs the function of generating parameters through machine learning. The parameter generator 570 is a concept corresponding to the deep neural network training unit 140 of FIG. 1.

The parameter generator 570 uses the position information of the second base station and the second reference signal reception power as input values, and the position information of the first user terminal as output values, and uses the forward and output values based on the input values. Parameters may be generated based on backpropagation.

Next, a method of operating the user terminal position estimation apparatus 500 will be described.

First, the first information generator 510 may obtain location information and at least one reference signal received power of the first base station located in the operating area based on the information on the radio signal in the operating area. (STEP A).

Thereafter, the parameter obtaining unit 520 obtains a parameter including a weight and a bias obtained through machine learning based on the position information of the first user terminal (STEP B).

Thereafter, the terminal location estimator 530 is configured based on the location information of the first base station generated by the first information generator 510, the first reference signal reception power, and the parameters obtained by the parameter acquirer 520. The location of the second user terminal located in the operating area is estimated (STEP C).

On the other hand, between STEP A and STEP B, the second information generator 560 is based on the information on the radio signal in the training area, the location information of the first user terminal located in the training area, at least one located in the training area The location information of the second base station and the second reference signal can be generated power (STEP D).

After STEP D Before STEP B, the parameter generator 570 performs deep learning based machine learning based on the location information of the first user terminal, the location information of the second base station, and the second reference signal received power. Parameters can be created via (STEP E).

STEP D can be embodied as follows.

First, the training area information generation unit 561 may generate location information of the first user terminal, physical cell id (PCI) of the second base station, and second reference signal reception power based on the information on the radio signal in the training area. (STEP Da).

Thereafter, the first base station location detector 562 may compare the PCI of the second base station with the PCI stored in the database and detect the location information of the second base station (STEP Db).

Thereafter, the first location and reception power updater 563 may update the location information and the second reference signal reception power of the second base station detected and generated at the current time based on the information detected and generated at another time ( STEP Dc).

Meanwhile, before STEP Da, the first radio signal information extractor 564 may determine the number of second base stations and extract information on radio signals in the training area based on the number of second base stations (STEP). Dd).

Meanwhile, after STEP Dc, the first normalization unit 565 may normalize the location information of the first user terminal, the updated location information of the second base station, and the updated second reference signal reception power (STEP De).

STEP A can be embodied as follows.

First, the operation region information generation unit 511 may generate the PCI (Physical Cell Id) and the first reference signal reception power of the first base station based on the information on the radio signal in the operation region (STEP Aa).

Thereafter, the second base station location detector 512 may compare the PCI of the first base station with the PCI stored in the database and detect the location information of the first base station (STEP Ab).

Thereafter, the second location and reception power updating unit 513 may update the location information and the first reference signal reception power of the first base station detected and generated at the current time based on the information detected and generated at another time ( STEP Ac).

Meanwhile, before STEP Aa, the second radio signal information extractor 514 may determine the number of first base stations, and extract information on radio signals in an operating area based on the number of the first base stations.

On the other hand, after STEP Ac, the second normalizer 515 may normalize the updated position information of the first base station and the updated first reference signal reception power.

Although all components constituting the embodiments of the present invention described above are described as being combined or operating in combination, the present invention is not necessarily limited to these embodiments. In other words, within the scope of the present invention, all of the components may be selectively operated in combination with one or more. In addition, although all of the components may be implemented in one independent hardware, each or some of the components of the components are selectively combined to perform some or all of the functions combined in one or a plurality of hardware It may be implemented as a computer program having a. In addition, such a computer program is stored in a computer readable medium such as a USB memory, a CD disk, a flash memory, and the like, and is read and executed by a computer, thereby implementing embodiments of the present invention. The recording medium of the computer program may include a magnetic recording medium, an optical recording medium and the like.

In addition, all terms including technical or scientific terms have the same meaning as commonly understood by a person of ordinary skill in the art unless otherwise defined in the detailed description. Terms used generally, such as terms defined in a dictionary, should be interpreted to coincide with the contextual meaning of the related art, and shall not be interpreted in an ideal or excessively formal sense unless explicitly defined in the present invention.

The above description is merely illustrative of the technical idea of the present invention, and various modifications, changes, and substitutions may be made by those skilled in the art without departing from the essential characteristics of the present invention. will be. Accordingly, the embodiments disclosed in the present invention and the accompanying drawings are not intended to limit the technical spirit of the present invention but to describe the present invention, and the scope of the technical idea of the present invention is not limited by the embodiments and the accompanying drawings. . The scope of protection of the present invention should be interpreted by the following claims, and all technical ideas within the scope equivalent thereto should be construed as being included in the scope of the present invention.

Claims (16)

A first information generator configured to generate location information of at least one first base station and a first reference signal received power based on information on a radio signal in an operating area;
A parameter obtaining unit obtaining a parameter including a weight and a bias obtained through machine learning based on the location information of the first user terminal; And
A terminal location estimator for estimating a location of a second user terminal located in the operating area based on the location information of the first base station, the first reference signal reception power, and the parameter;
Deep learning-based user terminal position estimation device comprising a. The method of claim 1,
The terminal location estimating unit estimates the location of the second user terminal when the second user terminal does not have a GPS function or the GPS function does not operate normally in the second user terminal. Estimation device. The method of claim 1,
Generate location information of a first user terminal located in the training area, location information of at least one second base station located in the training area, and a second reference signal reception power based on the information on the radio signal in the training area. A second information generator; And
A parameter generator configured to generate the parameter through deep learning based machine learning based on the location information of the first user terminal, the location information of the second base station, and the second reference signal reception power;
Deep learning-based user terminal position estimation apparatus further comprises. The method of claim 3, wherein
The second information generator,
A training area information generation unit configured to generate location information of the first user terminal, physical cell id (PCI) of the second base station, and the second reference signal reception power based on the information on the radio signal in the training area;
A first base station location detector for detecting location information of the second base station by comparing the PCI of the second base station with the PCI stored in the database; And
A first position and reception power updater for updating the position information of the second base station and the second reference signal received power detected and generated at a current time based on the information detected and generated at another time
Deep learning-based user terminal position estimation device comprising a. The method of claim 4, wherein
The first position and the received power updating unit detect a vector component related to the position information of the second base station detected at the current time, a vector component related to the second reference signal received power generated at the current time, and the other time. Detected at the current time by applying a vector window associated with the location information of the second base station and a vector window associated with the second reference signal received power generated at another time. Deep learning-based user terminal position estimation device, characterized in that for updating the position information of the second base station and the second reference signal received power generated at the current time. The method of claim 4, wherein
The first position and the received power updater are position information of the second base station detected at a previous time of the current time with information detected and generated at the other time, and the second reference signal received power generated at the previous time. And using the location information of the second base station detected at a later time of the current time and the second reference signal received power generated at the later time. The method of claim 4, wherein
The second information generator,
A first radio signal information extraction unit for determining the number of the second base stations and extracting information on radio signals in the training area based on the number of the second base stations; And
A first normalizer for data normalizing the location information of the first user terminal, the updated location information of the second base station, and the updated second reference signal reception power;
Deep learning-based user terminal position estimation apparatus further comprises. The method of claim 3, wherein
The parameter generation unit uses the position information of the second base station and the second reference signal reception power as input values, and the position information of the first user terminal as output values, and forwards and outputs the input values. Deep learning based user terminal position estimation apparatus, characterized in that for generating the parameter on the basis of backpropagation. The method of claim 1,
The first information generator,
An operating region information generator configured to generate a physical cell id (PCI) of the first base station and the first reference signal reception power based on the information on the radio signal in the operating region;
A second base station location detector for detecting location information of the first base station by comparing the PCI of the first base station with the PCI stored in the database; And
A second location and reception power updating unit for updating the location information of the first base station and the first reference signal reception power detected and generated at a current time based on the information detected and generated at another time
Deep learning-based user terminal position estimation device comprising a. The method of claim 9,
The second position and the received power updating unit detect a vector component associated with the position information of the first base station detected at the current time, a vector component associated with the first reference signal received power generated at the current time, and the other time. Detected at the current time by applying a context component for mutually combining the vector component associated with the location information of the first base station and the vector component associated with the first reference signal received power generated at another time. Deep learning-based user terminal position estimation apparatus, characterized in that for updating the location information of the first base station and the first reference signal received power generated at the current time. The method of claim 9,
The second position and the received power updater are position information of the first base station detected at a previous time of the current time with information detected and generated at the other time, and the first reference signal received power generated at the previous time. And using the position information of the first base station detected at a later time of the current time, and the first reference signal received power generated at the later time. The method of claim 9,
The first information generator,
A second radio signal information extraction unit for determining the number of the first base stations and extracting information on radio signals in the operating area based on the number of the first base stations; And
A second normalizer for data normalizing the updated position information of the first base station and the updated first reference signal reception power;
Deep learning-based user terminal position estimation apparatus further comprises. The method of claim 1,
The terminal location estimator is based on the position information of at least one second base station located in the training area and the information obtained by calculating the second reference signal reception power in the training area, the position information of the first base station data normalized And estimating the position of the second user terminal by sequentially denormalizing the data normalized first reference signal reception power and the data normalized parameter. The method of claim 1,
The first information generator generates the latitude of the first base station and the longitude of the first base station using the location information of the first base station,
And the terminal location estimating unit estimates the latitude of the second user terminal and the longitude of the second user terminal as the position of the second user terminal. The method of claim 1,
The terminal location estimator is a deep learning based user terminal position estimation apparatus, characterized in that for estimating the position of the second user terminal after a predetermined time on the basis of the current time. Generating a first reference signal received power and location information of at least one first base station located in the operating area based on the information on the radio signal in the operating area;
A parameter obtaining step of obtaining a parameter including a weight and a bias obtained through machine learning based on the location information of the first user terminal; And
Estimating a location of a second user terminal located in the operating area based on the location information of the first base station, the first reference signal reception power, and the parameter;
Deep learning-based user terminal position estimation method comprising a.

Images