KR102641464B1 - Wireless communication system and methdo based on deep learning - Google Patents

Abstract

This technology discloses a deep learning-based wireless communication system and method. A specific implementation example of this technology is based on deep learning, which simultaneously estimates the time offset and carrier frequency offset of the received signal and compensates for the estimated time offset and carrier frequency offset, thereby improving system performance depending on the channel environment.

Description

Deep learning-based wireless communication system and method {WIRELESS COMMUNICATION SYSTEM AND METHDO BASED ON DEEP LEARNING}

The present invention relates to a deep learning-based wireless communication system and method. More specifically, the present invention relates to a deep learning-based wireless communication system and method, and more specifically, to estimate the time and carrier frequency offset of a received signal based on deep learning and to improve system performance by correcting the difference between the estimated time and carrier frequency offset. It's about technology that allows you to improve.

In a typical OFDM wireless LAN system, when errors in time and carrier frequency offset occur, inter-channel interference (ICI) and inter-symbol interference (ISI) occur, making data recovery impossible at the receiver.

Accordingly, the time offset TO (Time Offset) is caused by propagation delay due to the clock error and distance between the transmitter and receiver, and the carrier frequency offset CFO (Carrier Frequency Offset) is caused by the frequency error and Doppler offset between carriers due to the allowable deviation of the oscillators of the transmitter and receiver. It is generated depending on frequency. Here, offset is the error.

Accordingly, the transmitter adds and transmits a legacy preampule before the data packet in the transmission signal, and the legacy preampule includes a short training field (STF) and a long training field (LTF).

That is, the receiver uses the STF (Short Training Field) to detect whether a valid packet has arrived, then estimates the approximate carrier frequency, and uses the LTF to estimate precise time error and precise carrier frequency error.

However, in the case of these existing time offset and carrier frequency offset estimation algorithms using STF and LTF, errors in each offset occur depending on the channel environment, and this has reached a limit where system performance deteriorates.

Accordingly, we would like to propose a method to improve system performance by simultaneously estimating the time offset TO and carrier frequency offset CFO of the received signal based on deep learning according to the applicant's channel environment and compensating for the estimated time offset and carrier frequency offset. do.

Therefore, the present invention is a deep learning-based wireless communication that improves system performance by simultaneously estimating the time offset and carrier frequency offset of the received signal based on deep learning according to the channel environment and compensating for the estimated time offset and carrier frequency offset. The purpose is to provide a system and method.

The object of the present invention is not limited to the object mentioned above, and other objects and advantages of the present invention that are not mentioned can be understood through the following description and will be more clearly understood through the examples of the present invention. In addition, it will be readily apparent that the objects and advantages of the present invention can be realized by means and combinations thereof as indicated in the claims.

A deep learning-based wireless communication system according to an embodiment of the present invention,

In a wireless communication system that performs wireless communication between a transmitter and a receiver,

The receiver is

A preprocessor that divides the received signal at predetermined frequency intervals, then performs cross-correlation on the STF (Short Training Field) of the legacy preamble of each divided received signal, converts it into a two-dimensional matrix, and normalizes it;

a learning unit that performs convolutional neural network-based learning on the received signal of a two-dimensional matrix to estimate time offset and carrier frequency offset; and

It may include a post-processing unit that compensates for the estimated time offset and the estimated carrier frequency offset for the received signal, then adjusts channel characteristics and performs data recovery.

Preferably, the predetermined frequency interval is

It can be set within the maximum possible carrier frequency offset range between the transmitter and receiver.

Preferably, the preprocessor,

A compensation module that divides the received signal into predetermined frequency intervals;

a cross-correlation module that performs cross-correlation on each of the divided received signals and converts them into a two-dimensional matrix; and

It may include a normalization module that normalizes each received signal of the two-dimensional matrix.

Preferably, the learning unit,

A time offset CNN module that derives an estimated time offset by performing training based on a convolutional neural network on a two-dimensional received signal; and

It may include a time offset learning evaluation module that evaluates the learning performance of the estimated time offset CNN module with respect to the derived estimated time offset.

Preferably, the time offset learning evaluation module

It is equipped to evaluate the learning results with a cost function for the time offset based on the cross-entropy for the actual time offset and the estimated time offset, and to control the parameters of the convolutional neural network of the time offset CNN module based on the performance results,

The cost function for the time offset can be expressed as Equation 11 below.

[Equation 11]

Here, t _k is the actual time offset, is the estimated time offset, and N is the batch size.

Preferably, the learning unit,

An estimated carrier frequency offset CNN module that performs convolutional neural network-based learning on two-dimensional received signals to derive an estimated carrier frequency offset; and

It includes a carrier frequency offset learning evaluation module that evaluates the learning performance of the convolutional neural network for the derived estimated carrier frequency offset,

The carrier frequency offset learning evaluation module,

The learning results are evaluated using a cost function for the carrier frequency offset based on the MSE (Means Square Error) between the actual carrier frequency offset and the estimated carrier frequency offset, and based on the performance results, the convolution neural network of the estimated carrier frequency CNN module is evaluated. It is provided to control parameters, and the cost function for the carrier frequency offset can be expressed as Equation 12 below.

[Equation 12]

here, is the estimated CFO, p is the actual CFO, and N is the batch size.

According to another embodiment of the present invention, a deep learning-based wireless communication method,

A preprocessing step of dividing the received signal at predetermined frequency intervals, then performing cross-correlation on the STF (Short Training Field) of the legacy preamble of the divided received signal, converting it into a two-dimensional matrix, and then normalizing it;

A learning step of performing convolutional neural network-based learning on the received signal of a two-dimensional matrix to estimate the time offset and carrier frequency offset; and

A post-processing step may include compensating for an estimated time offset and an estimated carrier frequency offset for the received signal, then adjusting channel characteristics, and then performing data recovery.

Preferably, the predetermined frequency interval is

It can be set within the range of the maximum possible frequency offset between the transmitter and receiver.

According to one embodiment, by simultaneously estimating the time offset and carrier frequency offset of the received signal based on deep learning and compensating for the estimated time offset and carrier frequency offset, the effect of improving system performance depending on the channel environment is obtained. .

The following drawings attached to this specification illustrate preferred embodiments of the present invention, and together with the detailed description of the invention described later, serve to further understand the technical idea of the present invention. Therefore, the present invention includes the matters described in such drawings. It should not be interpreted as limited to only .
1 is a configuration diagram of a deep learning-based wireless communication system according to an embodiment.
Figure 2 is an exemplary diagram showing a received signal in one embodiment.
Figure 3 is a detailed configuration diagram of a receiver of a system of one embodiment.
Figure 4 is a detailed configuration diagram of a preprocessing unit of a receiver according to an embodiment.
Figure 5 is an exemplary diagram showing output signals from each module of the preprocessor in one embodiment.
Figure 6 is a detailed configuration of a learning unit of a receiver according to an embodiment.
Figure 7 is an example diagram showing parameters of the TO CNN module in one embodiment.
Figure 8 is an example diagram showing parameters of the CFO CNN module in one embodiment.
Figure 9 is an example diagram showing parameters for each channel model in one embodiment.
Figure 10 is a graph showing the learning performance of estimated TO and estimated CFO for each channel model in one embodiment.
Figure 11 is a graph showing system performance for estimated TO for each channel model in one embodiment.
Figure 12 is a graph showing system performance for estimated CFO for each channel model in one embodiment.
Figure 13 is an overall flowchart of a deep learning-based wireless communication method according to another embodiment.

Specific structural or functional descriptions of the embodiments are disclosed for illustrative purposes only and may be modified and implemented in various forms. Accordingly, the embodiments are not limited to the specific disclosed form, and the scope of the present specification includes changes, equivalents, or substitutes included in the technical spirit.

Terms such as first or second may be used to describe various components, but these terms should be interpreted only for the purpose of distinguishing one component from another component. For example, a first component may be named a second component, and similarly, the second component may also be named a first component.

When a component is referred to as being “connected” to another component, it should be understood that it may be directly connected or connected to the other component, but that other components may exist in between.

Singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, terms such as “comprise” or “have” are intended to designate the presence of the described features, numbers, steps, operations, components, parts, or combinations thereof, and are intended to indicate the presence of one or more other features or numbers, It should be understood that this does not exclude in advance the possibility of the presence or addition of steps, operations, components, parts, or combinations thereof.

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the relevant technical field. Terms such as those defined in commonly used dictionaries should be interpreted as having meanings consistent with the meanings they have in the context of the related technology, and unless clearly defined in this specification, should not be interpreted in an idealized or overly formal sense. No.

Hereinafter, the present invention will be described in detail by explaining preferred embodiments of the present invention with reference to the attached drawings.

FIG. 1 is a configuration diagram of a deep learning-based wireless communication system according to an embodiment, FIG. 2 is an exemplary diagram showing a received signal according to an embodiment shown in FIG. 1, and FIG. 3 is a detailed configuration diagram of the receiver shown in FIG. 1. , FIG. 4 is a detailed configuration diagram of the pre-processing unit shown in FIG. 3, FIG. 5 is an example diagram showing the output signal of each module shown in FIG. 3, and FIG. 6 is a detailed configuration diagram of the learning unit shown in FIG. 3. Figure 7 is an example diagram showing the parameters of the TO CNN module shown in Figure 6, and Figure 8 is an example diagram showing the parameters of the CFO CNN module shown in Figure 6.

Referring to FIGS. 1 to 8, a deep learning-based wireless communication system according to an embodiment uses a deep learning-based timing offset (Timing Offset, hereinafter abbreviated TO) based on the preamble comprised in the front end of the data packet of the received transmission signal. ) and a carrier frequency offset (hereinafter abbreviated as CFO) are simultaneously estimated, and the system may include a transmitter 100 and a receiver 200.

The transmitter 100 transmits a transmission signal to which a legacy preamplifier for estimating TO and CFO is added to the front end of the data packet to the receiver 200. For example, the legacy pre-amplifier is configured with 10 Short Training Fields (STFs) and 2 Long Training Fields (LTFs), as shown in FIG. 2. Here, valid packets are detected using 7 STFs, CFO is estimated using the remaining 3 STFs, and precise TO and CFO are estimated using 2 LFTs.

Here, let us assume that a valid data packet is detected using 7 STFs, and that time and frequency synchronization is not achieved between the transmitter 100 and the receiver 200.

Accordingly, the receiver 200 performs preprocessing on the STF (Short Training Field) among the legacy preambles in the detected valid packet, performs learning based on a convolutional neural network on the preprocessed preamble, estimates TO and CFO, and produces estimated TO and estimated CFO. is configured to derive and compensate for the estimated TO and estimated CFO of the derived received signal, then adjust the channel characteristics and then perform data return. As shown in FIG. 3, the receiver 200 has a preprocessor ( 210), a learning unit 220, and a post-processing unit 230.

Here, the preprocessor 210 is provided to divide the received signal at a certain frequency interval and then perform cross-correlation on the STF among the legacy preambles of the divided received signal to convert it into a two-dimensional matrix and then normalize it. The preprocessor ( 210) may include at least one of a compensation module 211, a cross-correlation module 212, and a normalization module 213, as shown in FIG. 4.

The compensation module 211 divides the received signal into specific frequency intervals. Here, the specific frequency interval is set within the maximum carrier frequency offset range that can occur between the transmitter 100 and the receiver 200. For example, assuming that the maximum allowable offset of the oscillator is A ppm and the range of the maximum carrier frequency offset CFO is -f _A to f _A , the received signal is divided into a frequency interval f _o set within the range of the maximum carrier frequency offset CFO and , Also, the number K of divided received signals must always be an odd number. Accordingly, each received signal x _i (n) can be expressed as the following equation 11.

[Equation 1]

Here, x _i is the frequency-divided ith received signal.

For example, the received signal is divided into a frequency interval f _o set within the range of CFO and is transmitted to the cross-correlation module 212.

The cross-correlation module 212 performs a cross-correlation function for each divided received signal x _i (n) and the preamplifier s. Accordingly, the output signal y _i (n) of the cross-correlation function can be expressed as the following equations 2 to 4.

[Equation 2]

[Equation 3]

[Equation 4]

Here, M is the length of the preamble, y _i is the output of the ith cross-correlation module 212, y is a two-dimensional matrix for the output signal y _i of each cross-correlation module 212, and B is the cross-correlation module ( 212), and Y is the row vector of the two-dimensional matrix y .

And the received signal of the two-dimensional matrix is transmitted to the normalization module 213, whereby the received signal of the secondary matrix is normalized and converted to a value between 1 and 0, and the received signal of the converted secondary matrix is sent to the learning unit 220 ) is transmitted.

For example, in the case of a wireless local area network (WLAN) system, referring to FIG. 5, as shown in (a), the received signal with a frequency of -212 kHz is divided into 11 at a specific frequency interval f _O and (b) As shown, a cross-correlation function s is performed on each of the divided received signals based on a cross-correlation module with a window size of 864. At this time, the normalized received signal is converted into a two-dimensional matrix of 11 x 864 as shown in (c).

The learning unit 220 estimates TO and CFO based on a convolutional neural network (CNN) for the received signal of the preprocessed two-dimensional matrix and derives the estimated TO and the estimated CFO. Accordingly, the learning unit 220 As shown in FIG. 6, it may include a TO CNN module 221 and a CFO CNN module 222.

Referring to FIG. 7 as an example, the TO CNN module 221 is designed with five convolution layers, two skip connections, and one fully combined layer, as shown in (a), where the The parameters are as shown in (b). Accordingly, the TO CNN module 221 operates as a supervised learning-based classification network, and the received signal of the two-dimensional matrix is classified into a finite number of classes and TO is estimated based on the classified classes.

And the learning unit 220 further includes a TO learning evaluation module 223, and the TO learning evaluation module 223 evaluates the learning performance for the estimated TO of the TO CNN module 221 with a cost function for TO, The cost function for TO can be expressed in Equation 5 below.

[Equation 5]

Here, t _k is the actual time error, is the estimated time error, and N is the batch size. The higher the output value of this cost function, the better the learning performance. Accordingly, to improve learning performance, the parameters of the convolutional neural network of the TO CNN module 221 can be adjusted based on the cost function for TO.

Here, the parameters of the convolutional neural network of the TO CNN module 221 are as shown in Table 1 below.

[Table 1]

And, referring to FIG. 8, the CFO CNN module 222 has a total of 5 convolution layers, 2 skip connections, and one fully connected layer, as shown in (a). It is designed as, and the parameters of each layer are as shown in (b). Here, the CFO CNN module 222 derives features of the received signal of a two-dimensional matrix based on regression and estimates continuous CFO using the derived features of the received signal.

Meanwhile, the learning unit 220 further includes a CFO learning evaluation module 224, and the TO learning evaluation module 223 evaluates the learning performance for the estimated CF0 of the CFO CNN module 221 as a cost function for CFO. , the cost function for CFO can be expressed in Equation 6 below.

[Equation 6]

here, is the estimated CFO, p is the actual CFO, and N is the batch size. Accordingly, to improve learning performance, the parameters of the convolutional neural network of the CFO CNN module 222 can be adjusted based on the cost function for the CFO.

The parameters of the convolutional neural network of the CFO CNN module 222 are shown in Table 2 below.

[Table 2]

Meanwhile, each of the estimated TO and estimated CFO of the learning unit 220 is transmitted to the post-processing unit 230.

The post-processing unit 230 compensates for the received estimated TO and estimated CFO, then adjusts channel characteristics and performs data recovery. In one embodiment, a series of processes for performing data recovery after compensating for the received estimated TO and estimated CFO and then adjusting channel characteristics should be understood by those skilled in the art.

Accordingly, one embodiment simultaneously estimates the time offset and carrier frequency offset of the received signal based on deep learning and compensates for the estimated time offset and carrier frequency offset, thereby achieving the effect of improving system performance depending on the channel environment.

Figure 9 is an example diagram showing parameters for each channel model in an embodiment, Figure 10 is a graph showing the learning performance of the estimated TO and estimated CFO of each channel model A, C, and E shown in Figure 9, and Figure 11 is These are graphs showing the system performance for the estimated TO of each channel model A, C, and E shown in FIG. 9, and FIG. 12 shows the system performance for the estimated CFO of each channel model A, C, and E shown in FIG. 9. These are graphs.

Referring to FIGS. 9 to 12, it can be seen that system performance is improved when TO and CFO estimated based on deep learning in one embodiment are compensated.

For example, for a 50,000 test data set of received signals divided at 1dB intervals between -10Db and 15dB, the system performance is derived based on the signal-to-noise ratio and the false detection probability (FDR), and the respective channel environments in Figure 8 There is some ambiguity in the system performance for each TO because for channel model A, the estimated TO is accurate for the arrival of the first received signal, and for channel models C and E, the delay spread time for each channel is long.

As another example, if an estimated TO exists within the maximum multi-path fading delay time -LCP/2 to +LCP/2 for a divided received signal, the TO for each channel environment is judged to be accurate, and system performance is improved based on the accurate TO. .

Meanwhile, the system performance is derived based on RMSE (Root Means Square Error) for the received signals of 50,000 test data sets divided at 1 dB intervals between -10 dB and 15 dB. Referring to FIG. 12, medium and low signal-to-noise ratios are obtained. It can be seen that the RMSE is lower than that of the existing method.

FIG. 13 is a flowchart showing the operation process of the receiver shown in FIG. 3, and a deep learning-based wireless communication method according to another embodiment is explained with reference to FIG. 12.

First, in steps (S11) to (S13), the preprocessor 210 in one embodiment receives the transmission signal from the transmitter 100, divides the received signal into specific frequency intervals, and for each received signal in the divided specific frequency interval. Cross-correlation is performed with a predetermined window size, a received signal of a two-dimensional matrix is derived, the received signal of the derived two-dimensional matrix is normalized, and the received signal of the normalized two-dimensional matrix is provided to the learning unit 220.

In steps S14 and S15, the learning unit 220 of one embodiment performs learning based on a convolutional neural network for each received signal of the two-dimensional matrix of the preprocessor 210 to estimate each TO and CFO to obtain an estimated TO. and derive the estimated CFO.

Next, in step S16, the post-processing unit 230 of one embodiment compensates for the derived estimated TO and estimated CFO, then adjusts channel characteristics and performs data recovery.

One embodiment can improve system performance depending on the channel environment by simultaneously estimating the time offset and carrier frequency offset of the received signal based on deep learning and compensating for the estimated time offset and carrier frequency offset.

The embodiments described above may be implemented with hardware components, software components, and/or a combination of hardware components and software components. For example, the devices, methods, and components described in the embodiments may include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, and a field programmable gate (FPGA). It may be implemented using one or more general-purpose or special-purpose computers, such as an array, programmable logic unit (PLU), microprocessor, or any other device capable of executing and responding to instructions. A processing device may execute an operating system (OS) and one or more software applications that run on the operating system. Additionally, a processing device may access, store, manipulate, process, and generate data in response to the execution of software. For ease of understanding, a single processing device may be described as being used; however, those skilled in the art will understand that a processing device includes multiple processing elements and/or multiple types of processing elements. It can be seen that it may include. For example, a processing device may include a plurality of processors or one processor and one controller. Additionally, other processing configurations, such as parallel processors, are possible.

Software may include a computer program, code, instructions, or a combination of one or more of these, which may configure a processing unit to operate as desired, or may be processed independently or collectively. You can command the device. Software and/or data may be used on any type of machine, component, physical device, virtual equipment, computer storage medium or device to be interpreted by or to provide instructions or data to a processing device. , or may be permanently or temporarily embodied in a transmitted signal wave. Software may be distributed over networked computer systems and stored or executed in a distributed manner. Software and data may be stored on one or more computer-readable recording media.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. Computer-readable media may include program instructions, data files, data structures, etc., singly or in combination. Program instructions recorded on a computer-readable medium may be specially designed and configured for an embodiment or may be known and usable by those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. Includes magneto-optical media and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, etc. Examples of program instructions include machine language code, such as that produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. 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.

Although the embodiments have been described with limited drawings as described above, those skilled in the art can apply various technical modifications and variations based on the above. For example, the described techniques are performed in a different order than the described method, and/or components of the described system, structure, device, circuit, etc. are combined or combined in a different form than the described method, or other components are used. Alternatively, appropriate results may be achieved even if substituted or substituted by an equivalent.

100: transmitter
200: receiver
210: preprocessing unit
220: Learning Department
230: Post-processing unit

Claims (10)

In a wireless communication system that performs wireless communication between a transmitter and a receiver,
The receiver is
A preprocessor that divides the received signal at predetermined frequency intervals, then performs cross-correlation on the legacy preamble of each divided received signal, converts it into a two-dimensional matrix, and normalizes it;
a learning unit that performs convolutional neural network-based learning on the received signal of a two-dimensional matrix to estimate time offset and carrier frequency offset; and
It includes a post-processing unit that compensates for the estimated time offset and the estimated carrier frequency offset for the received signal, then adjusts the channel characteristics and performs data recovery,
The learning department,
A time offset CNN module that derives an estimated time offset by performing training based on a convolutional neural network on a two-dimensional received signal; and
A time offset learning evaluation module that evaluates the learning performance of the estimated time offset CNN module with respect to the derived estimated time offset,
It is equipped to evaluate the learning results with a cost function for the time offset based on the cross-entropy for the actual time offset and the estimated time offset, and to control the parameters of the convolutional neural network of the time offset CNN module based on the performance results,
A deep learning-based wireless communication system, characterized in that the cost function for the time offset is expressed in Equation 11 below.
[Equation 11]

Here, t _k is the actual time offset, is the estimated time offset, and N is the batch size. The method of claim 1, wherein the determined frequency interval is
A deep learning-based wireless communication system characterized in that it is set within the maximum possible carrier frequency offset range between the transmitter and receiver. The method of claim 1, wherein the preprocessing unit,
A compensation module that divides the received signal into predetermined frequency intervals;
a cross-correlation module that performs cross-correlation on each of the divided received signals and converts them into a two-dimensional matrix; and
A deep learning-based wireless communication system comprising a normalization module that normalizes each received signal of the two-dimensional matrix. delete delete The method of claim 1, wherein the learning unit,
An estimated carrier frequency offset CNN module that performs convolutional neural network-based learning on two-dimensional received signals to derive an estimated carrier frequency offset; and
A deep learning-based wireless communication system comprising a carrier frequency offset learning evaluation module that evaluates the learning performance of a convolutional neural network for the derived estimated carrier frequency offset. The method of claim 6, wherein the carrier frequency offset learning evaluation module,
The learning results are evaluated using a cost function for the carrier frequency offset based on the MSE (Means Square Error) between the actual carrier frequency offset and the estimated carrier frequency offset, and based on the performance results, the convolution neural network of the estimated carrier frequency CNN module is evaluated. A deep learning-based wireless communication system equipped to control parameters, and wherein the cost function for the carrier frequency offset is expressed in Equation 12 below.
[Equation 12]

here, is the estimated CFO, p is the actual CFO, and N is the batch size. In a deep learning-based wireless communication method performed based on a wireless communication system that performs wireless communication between the transmitter and receiver of claim 1,
At least one processor included in a wireless communication system that performs wireless communication between the transmitter and the receiver,
A preprocessing step of dividing the received signal at predetermined frequency intervals, then performing cross-correlation on the legacy preamble of the divided received signal, converting it into a two-dimensional matrix, and then normalizing it;
A learning step of performing convolutional neural network-based learning on the received signal of a two-dimensional matrix to estimate the time offset and carrier frequency offset; and
A post-processing step of compensating for an estimated time offset and an estimated carrier frequency offset for the received signal, then adjusting channel characteristics, and then performing data recovery,
The learning stage is,,
Deriving an estimated time offset by performing learning based on a convolutional neural network on a two-dimensional received signal; and
Comprising the step of evaluating the learning performance of the estimated time offset CNN module with respect to the derived estimated time offset,
It is equipped to evaluate the learning results with a cost function for the time offset based on the cross-entropy for the actual time offset and the estimated time offset, and to control the parameters of the convolutional neural network of the time offset CNN module based on the performance results,
A deep learning-based wireless communication method, characterized in that the cost function for the time offset is expressed in Equation 11 below.
[Equation 11]

Here, t _k is the actual time offset, is the estimated time offset, and N is the batch size. The method of claim 8, wherein the predetermined frequency interval is
A deep learning-based wireless communication method, characterized in that it is set within the range of the maximum possible frequency offset between the transmitter and the receiver. A computer-readable recording medium on which a program for executing the deep learning-based wireless communication method of claim 8 or 9 is recorded.