KR101428921B1 - Method and Apparatus for Selective Transport using Machine Learning in Multi-radio Environments - Google Patents

Abstract

Disclosed are a method and an apparatus for selecting a modulation and coding scheme (MCS) class and a different system in consideration of power consumption of a system using machine learning in a multi-radio environment. According to an embodiment of the present invention, training data to select parameters maximizing energy efficiency is generated and feature values of the training data are determined. After that, an MCS class maximizing energy efficiency is mapped to power by comparing and analyzing the training data and features through a simulation, a radio satisfying a PER condition in a low MCS class of mapped MCS classes is selected, and data is transmitted through the selected radio.

Description

TECHNICAL FIELD The present invention relates to an adaptive transmission method and apparatus using machine learning in a multi-radio environment,

The present invention relates to a MCS (Modulation and Coding Scheme) class and a heterogeneous system selection method and apparatus considering power consumption of the system through machine learning in a multi-radio environment.

In a multi-radio environment, limited resources can be efficiently allocated. Therefore, power or frequency can be efficiently allocated. However, considering only power or frequency efficiency, the energy efficiency of the entire system may not be the maximum. Korean Patent Laid-open No. 10-20110090831 relates to a user equipment and a wireless radio frequency auxiliary device in a communication system supporting such a multi-input multiple output, wherein the user equipment includes a radio frequency signal transmitted to a user equipment and a system of a radio frequency auxiliary device The wireless radio frequency auxiliary device controls and activates the wireless radio frequency auxiliary device to convert the first frequency of the first frequency to the second frequency and the wireless radio frequency auxiliary device transmits the second frequency to the user equipment.

For example, if there are two radios for the first radio and the second radio, this parameter when selecting the parameter that maximizes energy efficiency in each radio may not be the parameter that maximizes the energy efficiency of the overall system. Therefore, it is necessary to select parameters that maximize energy efficiency for the entire system. The energy efficiency of the entire system can be increased by considering the energy efficiency of the entire system, rather than considering the energy efficiency of each radio.

According to the prior art, transmission parameters considering only frequency efficiency are selected without considering the energy efficiency of the entire system. In this case, the frequency efficiency may increase, but the energy efficiency of the overall system may not be maximum. In addition, due to the environmental problems of the earth and the battery capacity of the receiver, the need for energy-efficient transmission parameter selection has increased.

Energy efficiency in a multi-radio environment has greater energy consumption than a single radio environment. Since the battery capacity is limited on the terminal side, it is necessary to select the parameters that can receive or transmit the largest amount of data at the same time, while using the lowest power. In addition, on the base station side, more energy is consumed because multiple users must be supported at the same time as a conventional single dario system. Therefore, in a multi-radio environment, transmission considering the energy efficiency of the entire system is required. In addition, the system and Modulation and Coding Scheme (MCS) classes should be selected through adaptive transmission according to channel conditions.

The selection of algorithms is also important for this energy efficient transmission. The theoretical approach of constructing and using optimization problems has limitations. There are difficulties in mathematically modeling the link state, and it is impractical because a large number of assumptions are required when constructing a formula. Also, there is a limit to the empirical approach of using tables based on experimental results. Although it is more practical than the theoretical approach, there is a disadvantage that reconstruction of transmission parameters according to environment is required. Therefore, in order to overcome these problems, an algorithm that can be practically used by using machine learning is needed.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a method and apparatus for receiving or transmitting the most data with the smallest power in a multi-radio environment. Also, a method and apparatus for adaptive transmission according to a channel state are provided by considering energy for the entire system by proposing a realistic algorithm through machine learning.

In one aspect, an adaptive transmission method in a multi-radio environment proposed in the present invention includes generating training data for selecting a parameter that maximizes energy efficiency, Calculating a feature of training data; mapping (determining) a feature of the training data to an ideal parameter; computing a feature of the test data; and using a machine learning algorithm (K-NN) Analyzing the features of the data and the features of the test data, and determining the parameters of the test data.

The step of calculating a feature for the test data calculates a feature value in the same manner as the training data when the test data is input.

A feature for selecting a parameter that maximizes energy efficiency determines a feature space that is small in dimension and can reflect the link state.

For example, the step of calculating a feature at this time uses the value of the average capacity to determine the value of the feature space

Wherein the step of comparing parameters of the training data and the test data with the parameters of the test data through the machine learning algorithm (K-NN) comprises determining a parameter of the test data by using a machine learning algorithm (K-NN) To determine the power of the test data and the MCS class at the same time.

When applying the KNN algorithm, a lower MCS class is selected when there are the same number of MCS classes, and a higher power level is selected in case of power.

Further comprising the step of selecting a radio that satisfies the PER condition when transmitting the lowest power level and the lowest power level in the MCS class when selecting the power of the training data and the MCS class, , Only the radio with the highest energy efficiency among all the radios is selected.

A training data generator for generating training data for selecting a parameter that maximizes energy efficiency, and a controller for determining the type of the feature, the feature set value and the feature space for the training data, A mapping unit for mapping (determining) an ideal parameter of the training data with a feature in consideration of radio selection, a machine learning algorithm, and a feature value determination unit, A machine learning algorithm unit for comparing (analyzing) the training data with the features of the test data and mapping (determining) power to an MCS class for maximizing energy efficiency, and a transmission unit for transmitting data to the radio selected by the mapping unit do.

The machine learning algorithm part uses the machine learning algorithm to determine the parameter values of the test data when the features of the test data are input. The feature value determination unit is a feature for selecting a parameter that maximizes energy efficiency, and determines a feature space that is small in dimension and can reflect the link state.

The machine learning algorithm part determines the MCS class by applying the machine learning algorithm (K-NN) according to each power using the feature set value.

When the MCS class is selected as the power of the training data and the MCS class is used as the reference power level, the mapping unit does not use the radio that does not satisfy the reference condition when the PER condition is not satisfied.

According to embodiments of the present invention, transmission capable of increasing energy efficiency in a multi-radio system is possible. Therefore, the battery life of the terminal increases, and the energy consumption at the base station can be reduced. In addition, it can be applied in various environments through machine learning, and parameters can be selected with high accuracy using a machine learning algorithm (K-NN). In addition, since data is transmitted by selecting radio according to the channel state, it is possible to improve the accuracy of the algorithm.

Figure 1 shows a flow diagram of an adaptive transmission method in a multiple radio environment.
2 is a diagram for explaining a method of determining an MCS class using a feature value of training data according to an embodiment of the present invention
3 is a diagram for explaining a process of selecting a radio satisfying a condition in a multi-radio environment according to an embodiment of the present invention.
4 is a diagram illustrating the configuration of an adaptive transmission apparatus in a multi-radio environment.
5 is a diagram showing simulation results of an adaptive transmission apparatus according to an embodiment of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

Figure 1 shows a flow diagram of an adaptive transmission method in a multiple radio environment.

1, an adaptive transmission method in a multi-radio environment includes generating (110) training data for selecting parameters that maximize energy efficiency, generating training data (130) calculating a feature of the training data and mapping it to an ideal parameter, calculating (140) a feature of test data for comparison with the training data, and determining a machine learning algorithm (K -NN) to compare the features of the training data with the features of the test data, and to determine parameters of the test data.

And generates training data for selecting a parameter that maximizes energy efficiency according to an embodiment (110). For example, if W training data is generated, W channels are generated. When the training data is generated, an ideal parameter of the training data is determined through a simulation considering radio selection (120). A feature of the training data is then calculated and mapped 130 with the ideal parameters. Then, the feature of the test data for comparison with the training data is calculated (140). At this time, it is possible to decide what type of feature to use and to determine a feature space. When determining the feature space, the feature must be a value that reflects the link state. The feature space should be one that has a small dimension and can reflect the state of the physical layer. The larger the dimension of the feature space, the more computation is required, thus increasing the complexity. If the feature space does not reflect the link state well, the feature space may not be used. According to the prior art, it is known that the use of post processing SNR (SNR) is good for classification. In the present invention, the feature space is determined using an average capacity.

The average capacity can be expressed by Equation (1).

[Equation 1]

Since each system has one feature, k is the number of features when k is k. Proposed technique Next, we can find the MCS class and power of LTE and WLAN accurately when calculating the post processing SNR (post processing SNR) in the average capacity equation.

In addition, a feature of the training data is compared with a feature of the test data using a machine learning algorithm (K-NN), and parameters of the test data are determined (130). At this time, the MCS class and power are determined by applying the machine learning algorithm (K-NN) using the feature set value. If there are the same number of MCS classes when using the machine learning algorithm (K-NN), it is possible to select a lower MCS class and increase the stability by selecting a higher power level in the case of power.

Also, when the power of the training data and the MCS class are selected, the radio that satisfies the PER condition is selected when the MCS class is transmitted with the lowest power level as the reference power level. If all radios do not meet the PER requirement, select only the radio with the highest energy efficiency among all radios. In the worst-case scenario, using all the radios to transmit data can result in lower energy efficiency. Therefore, it is possible to increase the energy efficiency by removing the bad channel radio and selecting and transmitting the radio satisfying the PER condition in the low MCS class of the mapped MCS class. At this time, various algorithms can be applied to select the radio. In a low MCS class, the radio may not be used in an SNR region that does not satisfy the PER condition. For example, when transmission is performed with 0dB to MCS class 1, if PER is greater than 0.1, it may not be used. In addition, if all radios do not satisfy the PER shipbuilding, only the most energy efficient radio can be selected and used.

2 is a diagram for explaining a method of determining an MCS class using a feature value of training data according to an embodiment of the present invention.

The machine learning algorithm (K-NN) is an algorithm that selects the new data value by checking the nearest K data when new data is received.

For _{example, (x 1, y 1)} , ..., I have the training data of the (x _n, y _n), and, x _i When ∈ R ² , y _i ∈ {0,1}, we can estimate y _{n +1} when test data (test data) x _{n +1} , which is new data, is input. By comparing the nearest K data, we can choose y _i to be one of the selected values (0 or 1).

Referring to FIG. 2, there are a first training data 210 and a second training data 220. When the test data 230 is input, 1 training data 210 and one second training data. At this time, the test data 230 can select more by comparing the data values. Thus, the test data 230 may select the value of the first training data 210. [

3 is a diagram for explaining a process of selecting a radio satisfying a condition in a multi-radio environment according to an embodiment of the present invention.

In the presence of multiple radios, the user can use only the radio selected according to the channel condition, rather than transmitting all the data using the radio.

For example, when there are a first radio 310 and a second radio 320, a user terminal 330, and a control center, the user terminal 330 may communicate with the first radio 310 and the second radio 320, Information can be transmitted. The first radio 310 and the second radio 320 receiving the channel information transmit the transmission parameters P _{T , 1} , i ₁ , P _{T , 2} , i ₂ through the control center 340 and the control signal 350 ) Can be transmitted and received. The user terminal 330 is to be used only by selected radio in accordance with the channel state of the transmission parameters first radio 310 and the second radio 320 received a _{(P T, 1, i 1} , P T, 2, i 2) Data 360 may be transmitted.

Symbol period to t, coding rate (cording rate) of R _{k, m,} the modulation level (modulation level) when using the m-th MCS classes in the k-th heterogeneous systems of using the m-th MCS classes in the k-th heterogeneous systems the M _{k , m} , the throughput of the kth heterogeneous system may be equal to Equation (2).

&Quot; (2) "

At this time, the throughput in the entire system can be expressed by Equation (3).

&Quot; (3) "

When each throughput is determined, a reference PER (packet error rate) must be satisfied. This condition is the power used when transmitting in the PER is the k-th heterogeneous systems (packet error rate), P _k of using the m-th MCS Class, PER _{k, m} is a k-th heterogeneous systems can be the same as the equation (4) And the value of P _k can be selected.

&Quot; (4) "

All the power used in the transmission system can be expressed by Equation (5).

&Quot; (5) "

Assuming that the MCS class used in the kth system is i _k , Bit Per Joule in terms of energy efficiency can be expressed by Equation (6).

&Quot; (6) "

We can find the discrete power level and MCS class to satisfy the above equation. At this time, considering the radio selection, a system for maximizing Bit Per Joule in terms of energy efficiency and a coding and modulation class can be selected. This process allows you to compare and analyze training data and features, and to map MCS classes and power. In the mapped MCS class, a radio that satisfies the PER condition can be selected in the low MCS class to transmit data to the selected radio.

4 is a diagram illustrating the configuration of an adaptive transmission apparatus in a multi-radio environment.

In the multi-radio environment, the adaptive transmission apparatus includes a training data generation unit 410, a feature value determination unit 420, a test data input unit 430, a mapping unit 440, a machine learning algorithm unit 450, ).

The training data generation unit 410 generates training data for selecting a parameter that maximizes energy efficiency. The MCS class can be determined using feature values for the generated training data.

The feature value determination unit 420 calculates a feature type, a feature space determination, and a feature value for the training data, and calculates a feature of the test data.

When the feature value determining unit 420 determines the feature space, the feature should be a value reflecting the link status. The feature space should be one that has a small dimension and can reflect the state of the physical layer. The larger the dimension of the feature space, the more computation is required, thus increasing the complexity. If the feature space does not reflect the link state well, the feature space may not be used. In the present invention, the feature space is determined using an average capacity. In addition, the feature value determination unit 420 determines the feature value in the same manner as the test data when the test data is input.

The test data input unit 430 receives test data for comparison with the training data. The input test data is used to calculate a feature through the feature value determination unit 420.

The mapping unit 440 performs simulations to simultaneously select the ideal parameters of the training data and the radio to be used, and maps the features and parameters of the training data.

Also, in radio selection, when the power of the training data and the MCS class are selected, the radio that satisfies the PER condition is selected when transmitted in the lowest MCS class as a reference power level. If all radios do not meet the PER requirement, select only the radio with the highest energy efficiency among all radios.

In the worst-case scenario, using all the radios to transmit data can result in lower energy efficiency. Therefore, it is possible to increase the energy efficiency by removing the bad channel radio and selecting and transmitting the radio satisfying the PER condition in the low MCS class of the mapped MCS class. At this time, various algorithms can be applied to select the radio. In a low MCS class, the radio may not be used in an SNR region that does not satisfy the PER condition. For example, when transmission is performed with 0dB to MCS class 1, if PER is greater than 0.1, it may not be used. In addition, if all radios do not satisfy the PER shipbuilding, only the most energy efficient radio can be selected and used.

The machine learning algorithm unit 450 compares and analyzes the training data with the features of the test data through a machine learning algorithm (K-NN) to simultaneously determine the MCS class and power to maximize energy efficiency. If there are the same number of MCS classes when using the machine learning algorithm (K-NN), it is possible to select a lower MCS class and increase the stability by selecting a higher power level in case of power.

The transmitting unit 460 transmits data to the radio selected by the mapping unit 440.

5 is a diagram showing simulation results of an adaptive transmission apparatus according to an embodiment of the present invention.

Simulation was performed considering multiple radio environments where LTE and WLAN can be used at the same time. All four parameters were selected, and the SNR levels selectable by LTE and WLAN were set at 2dB, 8dB intervals (0, 2, 4, 6, 8, 16, 24, 32, 40, 48dB) respectively. In addition, 8 types of WCS MCS classes and 15 types of MCS classes of LTE are set.

The simulation environment may be as shown in Tables 1, 2 and 3.

<Table 1> MCS Class of WLAN

<Table 2> MCS Class of LTE

<Table 3> Noise power and bandwidth

The feture value is obtained using the average capacity of Equation (1). At this time, the feature may be affected by the channel. For example, if the channel condition is poor, the feature may be small.

Table 4 compares the simulation results with the ideal data. If energy efficiency is selected for each radio with parameters that have the highest energy efficiency (each), if two radio parameters are used without condition (Together), the radio is selected according to the channel condition The results of the selection are shown in Table 4.

<Table 4>

The energy efficiency for the entire system may not be the maximum even if a parameter with the highest energy efficiency is selected for each radio. If you select radio according to the channel condition, you can confirm that it is more efficient than using two radio without condition.

When using two radios to determine the parameters, the WLAN MCS class and the LTE MCS class can not select the MCS class well when the channel condition is bad. When the channel is not good, the WLAN MCS class and the LTE MCS class are not selected as one type but the selection is divided into various types of MCS classes in a selected power interval. This causes a lot of boundary between parameters, resulting in lower accuracy.

In other words, the machine learning algorithm (K-NN) did not work correctly. The parameters found using the machine learning algorithm do not satisfy the PER condition. If two radios are used to determine the parameters, there may be a lot of errors because the power is discrete, and this may cause repeated MCS classes. If the MCS class is repeated, the learning can not be done well and many errors can occur.

If you select radio according to the channel condition, one radio should be used unconditionally. At this time, you can select the radio with the highest energy efficiency. For example, if you transmit 0dB to MCS class 1, you can not use the radio if you can not satisfy PER. When the radio is used unconditionally at the same time, power must be increased because transmission is required even when the channel state is not good, and there is a problem in selecting an efficient MCS class. Therefore, a method of determining the features of the test data using the training data through the machine learning can be used. Energy efficiency can be improved by comparing and analyzing the determined features to map the power to the MCS class that maximizes energy efficiency and selecting the radio that satisfies the PER condition in the lower MCS class among the mapped MCS classes.

The simulation result when radio is selected according to the channel state will be described with reference to FIG.

5A is a graph showing an MCS class according to a feature of a WLAN.

You can see that the WLAN channel is not used when you have a feature that is smaller than 1. In addition, if the WLAN is more efficient than LTE, the WLAN can be used even in poor channel conditions.

5B is a graph showing an MCS class according to a feature of LTE.

If the channel status of LTE is not good, it can be confirmed that it is not used.

5C is a graph showing the power according to a feature of the WLAN.

If there is a part that does not use WLAN and channel condition is bad, it can be confirmed that data is transmitted by increasing power.

5D is a graph showing the power according to the feature of LTE.

There are places where data can not be accurately identified, but it is more efficient than using two radios simultaneously.

Depending on the channel state, the radio selection is reflected and the parameters found using the machine learning algorithm (K-NN) can satisfy most of the PER conditions. .

The apparatus described above may be implemented as a hardware component, a software component, and / or a combination of hardware components and software components. For example, the apparatus and components described in the embodiments may be implemented within a computer system, such as, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable array (FPA) A programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and one or more software applications running on the operating system. The processing device may also access, store, manipulate, process, and generate data in response to execution of the software. For ease of understanding, the processing apparatus may be described as being used singly, but those skilled in the art will recognize that the processing apparatus may have a plurality of processing elements and / As shown in FIG. For example, the processing unit may comprise a plurality of processors or one processor and one controller. Other processing configurations are also possible, such as a parallel processor.

The software may include a computer program, code, instructions, or a combination of one or more of the foregoing, and may be configured to configure the processing device to operate as desired or to process it collectively or collectively Device can be commanded. The software and / or data may be in the form of any type of machine, component, physical device, virtual equipment, computer storage media, or device , Or may be permanently or temporarily embodied in a transmitted signal wave. The software may be distributed over a networked computer system and stored or executed in a distributed manner. The software and data may be stored on one or more computer readable recording media.

The method according to an embodiment may be implemented in the form of a program command that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions to be recorded on the medium may be those specially designed and configured for the embodiments or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape; optical media such as CD-ROMs and DVDs; magnetic media such as floppy disks; Magneto-optical media, and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments. For example, it is to be understood that the techniques described may be performed in a different order than the described methods, and / or that components of the described systems, structures, devices, circuits, Lt; / RTI &gt; or equivalents, even if it is replaced or replaced.

Therefore, other implementations, other embodiments, and equivalents to the claims are also within the scope of the following claims.

Claims (11)

Generating training data for selecting a parameter that maximizes energy efficiency;
Determining an ideal parameter of the training data through simulation considering radio selection;
Calculating a feature of the training data and mapping the parameter with the ideal parameter;
Calculating features of test data for comparison with the training data; And
Analyzing a feature of the training data and a feature of the test data using a machine learning algorithm (K-NN), and determining parameters of the test data
/ RTI &gt; The method of claim 1, The method according to claim 1,
The calculation of the features of the training data
When the test data is input, the featur value is calculated in the same manner as the training data
Multi - radio environment adaptive transmission method. 3. The method of claim 2,
The calculation of the features of the training data
The value of the average capacity is used to determine the value of the feature space
Multi - radio environment adaptive transmission method. The method according to claim 1,
Comparing the features of the training data with the features of the test data through the machine learning algorithm (K-NN) to determine parameters of the test data
The MCS class is determined by applying a machine learning algorithm (K-NN) according to each power using a feature set value
Multi - radio environment adaptive transmission method. 5. The method of claim 4,
When applying the K-NN algorithm, a lower MCS class is selected when there are the same number of MCS classes, and a higher power level is selected when power is applied
Multi - radio environment adaptive transmission method. The method according to claim 1,
Further comprising the step of selecting a radio that satisfies the PER condition when transmitting the lowest power MCS class with a power level as a reference when selecting the power of the training data and the MCS class,
If the PER condition is not satisfied, only the radio having the highest energy efficiency among all the radios is selected
Multi - radio environment adaptive transmission method. A training data generation unit for generating training data for selecting a parameter that maximizes energy efficiency;
A feature value determination unit for determining a type of a feature, a feature set value, and a feature space with respect to the training data, and determining a feature of the test data;
A test data input unit for receiving test data;
A mapping unit for mapping an ideal parameter to a feature through a simulation considering radio selection;
A machine learning algorithm unit that compares and analyzes features of the training data and the test data through a machine learning algorithm to map an MCS class and power to maximize energy efficiency; And mapping
And a transmitter for transmitting data to a radio selected by the mapping unit,
Wherein the radio environment adaptive transmission apparatus comprises: 8. The method of claim 7,
The feature value determination unit determines the feature set value by confirming data of a near value when the test data is input
Multiple radio environment adaptive transmission device. 8. The method of claim 7,
The feature value determining unit is a feature for selecting a parameter that maximizes energy efficiency, and determines a feature space that is small in type and dimension and can reflect the link state
Multiple radio environment adaptive transmission device. 8. The method of claim 7,
The machine learning algorithm unit applies a machine learning algorithm (K-NN) according to each power using a feature set value to determine an MCS class
Multiple radio environment adaptive transmission device. 8. The method of claim 7,
When the MCS class and the power of the training data are selected, the mapping unit does not use the radio that does not satisfy the above-mentioned condition when the PER is not satisfied when the MCS class is transmitted with the lowest MCS class as the reference power level
Multiple radio environment adaptive transmission device.

Images