KR20230090293A - Method and apparatus for communication based on machine learning - Google Patents

Abstract

In one embodiment of a communication system, a method of operating a first device includes outputting a first learning model obtained by repeatedly performing a machine learning operation based on a result of transmitting and receiving a virtual wireless signal, and performing the machine learning operation. The step of doing is: inputting the (normalized) virtual channel information in the first 2-dimensional domain to the first machine learning structure, obtaining corresponding information of the first window, based on the obtained information of the first window generating a virtual transmission signal corresponding to one or more transmission symbols, and obtaining a virtual received signal and one or more reconstruction symbols corresponding to the virtual received signal based on the virtual channel information and the virtual transmission signal. , and updating the first machine learning structure based on a loss function defined based on the one or more transmission symbols and the one or more recovery symbols.

Description

Machine learning-based communication method and apparatus {METHOD AND APPARATUS FOR COMMUNICATION BASED ON MACHINE LEARNING}

The present disclosure relates to a machine learning-based communication technology, and more specifically, to a technology for improving wireless signal transmission and reception performance by generating a window used for wireless signal transmission and reception in a communication system based on machine learning.

Along with the development of information and communication technology, various wireless communication technologies are being developed. Representative wireless communication technologies include long term evolution (LTE) and new radio (NR), which are defined in the 3rd generation partnership project (3GPP) standard. LTE may be one wireless communication technology among 4th generation (4G) wireless communication technologies, and NR may be one wireless communication technology among 5th generation (5G) wireless communication technologies. A wireless communication technology after 5G (eg, 6th generation (6G), etc.) may be referred to as a beyond 5G (B5G) wireless communication technology.

In one embodiment of the communication system, an orthogonal frequency division multiplexing (OFDM) method or an orthogonal frequency division multiple access (OFDMA) method may be used for modulation and demodulation. The OFDM scheme may have advantages of relatively excellent frequency efficiency and simple implementation. However, the OFDM scheme may have a problem in that communication performance is greatly deteriorated due to inter-carrier interference (ICI) that may occur in a high-speed moving environment due to a high Doppler shift or the like. When the sub-carrier spacing (SPS) is increased to alleviate the problem of the OFDM scheme, the ratio of the guard interval is increased, which may cause transmission efficiency to deteriorate.

In order to improve the problems of the OFDM scheme, an orthogonal time frequency space (OTFS) scheme (or an OTFS transmission scheme) is being studied. In the OTFS scheme, unlike the OFDM scheme, resources in the delay-Doppler domain can be used in a process of transmitting and receiving data symbols. The OTFS scheme can have a long channel coherence time in the delay-Doppler domain, so channel feedback can not be required with high frequency, and high transmission efficiency can be achieved. However, when radio resources are limited or non-ideal pulses are used in a communication environment, performance of the OTFS scheme may deteriorate. A technique for improving the performance of the OTFS method by efficiently mitigating this problem may be required.

Matters described in this background art section are prepared to enhance understanding of the background of the invention, and may include matters that are not prior art already known to those skilled in the art to which this technique belongs.

An object of the present disclosure for achieving the above needs is to provide a machine learning-based communication method and apparatus for improving wireless signal transmission and reception performance by generating a window used for wireless signal transmission and reception in a communication system based on machine learning. there is.

In one embodiment of a communication system for achieving the above object, a method of operating a first device includes performing a machine learning operation based on a result of transmitting and receiving a virtual wireless signal, and repeatedly performing the machine learning operation. and outputting a first learning model obtained by performing the machine learning operation, wherein the performing of the machine learning operation includes generating virtual channel information in a first 2-dimensional domain, and a virtual channel in the first 2-dimensional domain. Obtaining an output value for a first window corresponding to the virtual channel information by inputting information to a first machine learning structure, virtual transmission corresponding to one or more transmission symbols based on the output value for the first window Generating a signal, obtaining a virtual received signal and one or more reconstruction symbols corresponding to the virtual received signal based on the virtual channel information and the virtual transmission signal, and obtaining the one or more transmission symbols and the one or more reconstruction symbols. It may include updating the first machine learning structure based on a loss function defined based on recovery symbols.

According to an embodiment of the machine learning-based communication method and apparatus, the first machine learning apparatus may repeatedly perform a machine learning operation based on a result of transmitting and receiving a virtual wireless signal. The first machine learning device generates a first learning model for adaptively generating a window used for wireless signal transmission in a first two-dimensional domain (eg, a two-dimensional domain used in a transmission scheme such as the OTFS scheme) can create One or more communication nodes of the communication system may generate a window suitable for a real-time communication environment based on the first learning model generated by the first machine learning device, and perform real-time communication based on the created window. As a result, performance of transmitting and receiving radio signals can be efficiently improved.

1 is a conceptual diagram illustrating an embodiment of a communication system.
2 is a block diagram illustrating an embodiment of a communication node constituting a communication system.
3 is a block diagram for explaining an embodiment of a communication system including a transmitting communication node and a receiving communication node.
4 is a block diagram for explaining an embodiment of a modulator of a transmission node in a communication system.
5 is a conceptual diagram for explaining embodiments of a data symbol and reference signal (RS) mapping method for resources on a first two-dimensional (2D) domain in a communication system.
6A and 6B are conceptual diagrams for explaining an embodiment of a relationship between data symbols, resources on a first 2D domain, and resources on a second 2D domain in a communication system.
7 is a block diagram for explaining an embodiment of a demodulation unit of a receiving node in a communication system.
8 is a conceptual diagram for explaining an embodiment of a machine learning structure.
9 is a flowchart for explaining an embodiment of a machine learning-based communication method.
10 is a flowchart for explaining an embodiment of a machine learning method.

Since the present disclosure can make various changes and have various embodiments, specific embodiments are illustrated in the drawings and described in detail. However, this is not intended to limit the present disclosure to specific embodiments, and should be understood to include all modifications, equivalents, and substitutes included in the spirit and scope of the present disclosure.

Terms such as first and second may be used to describe various components, but the components should not be limited by the terms. These terms are only used for the purpose of distinguishing one component from another. For example, a first element may be termed a second element, and similarly, a second element may be termed a first element, without departing from the scope of the present disclosure. The terms and/or include any combination of a plurality of related recited items or any of a plurality of related recited items.

It is understood that when an element is referred to as being "connected" or "connected" to another element, it may be directly connected or connected to the other element, but other elements may exist in the middle. It should be. On the other hand, when an element is referred to as “directly connected” or “directly connected” to another element, it should be understood that no other element exists in the middle.

Terms used in the present disclosure are only used to describe specific embodiments, and are not intended to limit the present disclosure. Singular expressions include plural expressions unless the context clearly dictates otherwise. In the present disclosure, terms such as "comprise" or "having" are intended to indicate that there is a feature, number, step, operation, component, part, or combination thereof described in the specification, but one or more other features It should be understood that the presence or addition of numbers, steps, operations, components, parts, or combinations thereof is not precluded.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related art, and unless explicitly defined in the present disclosure, they should not be interpreted in an ideal or excessively formal meaning. don't

A communication system to which embodiments according to the present disclosure are applied will be described. A communication system to which embodiments according to the present disclosure are applied is not limited to the content described below, and embodiments according to the present disclosure may be applied to various communication systems. Here, the communication system may be used in the same sense as a communication network.

Throughout the specification, a network refers to, for example, wireless Internet such as WiFi (wireless fidelity), portable Internet such as WiBro (wireless broadband internet) or WiMax (world interoperability for microwave access), and GSM (global system for mobile communication). ) or CDMA (code division multiple access) 2G mobile communication networks, WCDMA (wideband code division multiple access) or CDMA2000 3G mobile communication networks, HSDPA (high speed downlink packet access) or HSUPA (high speed uplink packet access) It may include a 3.5G mobile communication network, a 4G mobile communication network such as a long term evolution (LTE) network or an LTE-Advanced network, a 5G mobile communication network, a B5G mobile communication network (6G mobile communication network, etc.).

Throughout the specification, a terminal includes a mobile station, a mobile terminal, a subscriber station, a portable subscriber station, a user equipment, and an access terminal. It may refer to a terminal, a mobile station, a mobile terminal, a subscriber station, a mobile subscriber station, a user device, an access terminal, or the like, and may include all or some functions of a terminal, a mobile station, a mobile terminal, a subscriber station, a mobile subscriber station, a user equipment, an access terminal, and the like.

Here, a desktop computer capable of communicating with a terminal, a laptop computer, a tablet PC, a wireless phone, a mobile phone, a smart phone, and a smart watch (smart watch), smart glass, e-book reader, PMP (portable multimedia player), portable game console, navigation device, digital camera, DMB (digital multimedia broadcasting) player, digital voice digital audio recorder, digital audio player, digital picture recorder, digital picture player, digital video recorder, digital video player ) can be used.

Throughout the specification, a base station includes an access point, a radio access station, a node B, an evolved nodeB, a base transceiver station, and an MMR ( It may refer to a mobile multihop relay)-BS, and may include all or some functions of a base station, access point, wireless access station, NodeB, eNodeB, transmission/reception base station, MMR-BS, and the like.

Hereinafter, with reference to the accompanying drawings, preferred embodiments of the present disclosure will be described in more detail. In order to facilitate overall understanding in describing the present disclosure, the same reference numerals are used for the same components in the drawings, and redundant descriptions of the same components are omitted.

1 is a conceptual diagram illustrating an embodiment of a communication system.

Referring to FIG. 1, a communication system 100 includes a plurality of communication nodes 110-1, 110-2, 110-3, 120-1, 120-2, 130-1, 130-2, 130-3, 130-4, 130-5, 130-6). A plurality of communication nodes are 4G communication (eg, long term evolution (LTE), advanced (LTE-A)), 5G communication (eg, new radio (NR)) specified in the 3rd generation partnership project (3GPP) standard ), etc. can be supported. 4G communication may be performed in a frequency band of 6 GHz or less, and 5G communication may be performed in a frequency band of 6 GHz or more as well as a frequency band of 6 GHz or less.

For example, for 4G communication and 5G communication, a plurality of communication nodes may use a code division multiple access (CDMA)-based communication protocol, a wideband CDMA (WCDMA)-based communication protocol, a time division multiple access (TDMA)-based communication protocol, FDMA (frequency division multiple access)-based communication protocol, OFDM (orthogonal frequency division multiplexing)-based communication protocol, Filtered OFDM-based communication protocol, CP (cyclic prefix)-OFDM-based communication protocol, DFT-s-OFDM (discrete Fourier transform-spread-OFDM) based communication protocol, OFDMA (orthogonal frequency division multiple access) based communication protocol, SC (single carrier)-FDMA based communication protocol, NOMA (Non-orthogonal multiple access), GFDM (generalized frequency) division multiplexing)-based communication protocol, FBMC (filter bank multi-carrier)-based communication protocol, UFMC (universal filtered multi-carrier)-based communication protocol, SDMA (Space Division Multiple Access)-based communication protocol, etc. can be supported. .

In addition, the communication system 100 may further include a core network. When the communication system 100 supports 4G communication, the core network may include a serving-gateway (S-GW), a packet data network (PDN)-gateway (P-GW), a mobility management entity (MME), and the like. there is. When the communication system 100 supports 5G communication, the core network may include a user plane function (UPF), a session management function (SMF), an access and mobility management function (AMF), and the like.

Meanwhile, a plurality of communication nodes 110-1, 110-2, 110-3, 120-1, 120-2, 130-1, 130-2, 130-3, 130-1 constituting the communication system 100 4, 130-5, 130-6) may each have the following structure.

2 is a block diagram illustrating an embodiment of a communication node constituting a communication system.

Referring to FIG. 2 , a communication node 200 may include at least one processor 220, a memory 220, and a transceiver 230 connected to a network to perform communication. In addition, the communication node 200 may further include an input interface device 240, an output interface device 250, a storage device 260, and the like. Each component included in the communication node 200 may be connected by a bus 270 to communicate with each other.

However, each component included in the communication node 200 may be connected through an individual interface or an individual bus centered on the processor 220 instead of the common bus 270 . For example, the processor 220 may be connected to at least one of the memory 220, the transmission/reception device 230, the input interface device 240, the output interface device 250, and the storage device 260 through a dedicated interface. .

The processor 220 may execute a program command stored in at least one of the memory 220 and the storage device 260 . The processor 220 may mean a central processing unit (CPU), a graphics processing unit (GPU), or a dedicated processor on which methods according to embodiments of the present disclosure are performed. Each of the memory 220 and the storage device 260 may include at least one of a volatile storage medium and a non-volatile storage medium. For example, the memory 220 may include at least one of a read only memory (ROM) and a random access memory (RAM).

Referring back to FIG. 1, the communication system 100 includes a plurality of base stations (110-1, 110-2, 110-3, 120-1, 120-2), a plurality of terminals 130- 1, 130-2, 130-3, 130-4, 130-5, 130-6). Base stations 110-1, 110-2, 110-3, 120-1, and 120-2 and terminals 130-1, 130-2, 130-3, 130-4, 130-5, and 130-6 The inclusive communication system 100 may be referred to as an “access network”. Each of the first base station 110-1, the second base station 110-2, and the third base station 110-3 may form a macro cell. Each of the fourth base station 120-1 and the fifth base station 120-2 may form a small cell. The fourth base station 120-1, the third terminal 130-3, and the fourth terminal 130-4 may belong to the cell coverage of the first base station 110-1. The second terminal 130-2, the fourth terminal 130-4, and the fifth terminal 130-5 may belong to the cell coverage of the second base station 110-2. The fifth base station 120-2, the fourth terminal 130-4, the fifth terminal 130-5, and the sixth terminal 130-6 may belong to the cell coverage of the third base station 110-3. there is. The first terminal 130-1 may belong to the cell coverage of the fourth base station 120-1. The sixth terminal 130-6 may belong to the cell coverage of the fifth base station 120-2.

Here, each of the plurality of base stations 110-1, 110-2, 110-3, 120-1, and 120-2 is a NodeB, an evolved NodeB, a base transceiver station (BTS), Radio base station, radio transceiver, access point, access node, RSU (road side unit), RRH (radio remote head), TP (transmission point), TRP ( transmission and reception point), eNB, gNB, etc.

Each of the plurality of terminals 130-1, 130-2, 130-3, 130-4, 130-5, and 130-6 is a UE (user equipment), terminal, access terminal, mobile Mobile terminal, station, subscriber station, mobile station, portable subscriber station, node, device, IoT (Internet of Thing) It may be referred to as a device, a mounted device (mounted module/device/terminal or on board device/terminal, etc.), and the like.

Meanwhile, each of the plurality of base stations 110-1, 110-2, 110-3, 120-1, and 120-2 may operate in different frequency bands or may operate in the same frequency band. Each of the plurality of base stations 110-1, 110-2, 110-3, 120-1, and 120-2 may be connected to each other through an ideal backhaul link or a non-ideal backhaul link, and , information can be exchanged with each other through an ideal backhaul link or a non-ideal backhaul link. Each of the plurality of base stations 110-1, 110-2, 110-3, 120-1, and 120-2 may be connected to the core network through an ideal backhaul link or a non-ideal backhaul link. Each of the plurality of base stations 110-1, 110-2, 110-3, 120-1, and 120-2 transmits a signal received from the core network to a corresponding terminal 130-1, 130-2, 130-3, and 130 -4, 130-5, 130-6), and signals received from corresponding terminals 130-1, 130-2, 130-3, 130-4, 130-5, 130-6 are transmitted to the core network can be sent to

In addition, each of the plurality of base stations 110-1, 110-2, 110-3, 120-1, and 120-2 transmits MIMO (eg, single user (SU)-MIMO, multi-user (MU)- MIMO, massive MIMO, etc.), CoMP (coordinated multipoint) transmission, CA (carrier aggregation) transmission, transmission in an unlicensed band, device to device communication (D2D) (or ProSe ( proximity services)) may be supported. Here, each of the plurality of terminals 130-1, 130-2, 130-3, 130-4, 130-5, and 130-6 is a base station 110-1, 110-2, 110-3, 120-1 , 120-2) and operations supported by the base stations 110-1, 110-2, 110-3, 120-1, and 120-2 may be performed. For example, the second base station 110-2 can transmit a signal to the fourth terminal 130-4 based on the SU-MIMO scheme, and the fourth terminal 130-4 uses the SU-MIMO scheme. A signal may be received from the second base station 110-2. Alternatively, the second base station 110-2 may transmit a signal to the fourth terminal 130-4 and the fifth terminal 130-5 based on the MU-MIMO scheme, and the fourth terminal 130-4 And each of the fifth terminal 130-5 may receive a signal from the second base station 110-2 by the MU-MIMO method.

Each of the first base station 110-1, the second base station 110-2, and the third base station 110-3 may transmit a signal to the fourth terminal 130-4 based on the CoMP scheme, and The terminal 130-4 may receive signals from the first base station 110-1, the second base station 110-2, and the third base station 110-3 by CoMP. Each of the plurality of base stations 110-1, 110-2, 110-3, 120-1, and 120-2 includes a terminal 130-1, 130-2, 130-3, and 130-4 belonging to its own cell coverage. , 130-5, 130-6) and a CA method. Each of the first base station 110-1, the second base station 110-2, and the third base station 110-3 controls D2D between the fourth terminal 130-4 and the fifth terminal 130-5. and each of the fourth terminal 130-4 and the fifth terminal 130-5 may perform D2D under the control of the second base station 110-2 and the third base station 110-3, respectively. .

Next, methods for transmitting and receiving signals in a communication system will be described. Here, even when a method performed in a first communication node among communication nodes (for example, transmission or reception of a signal) is described, the second communication node corresponding thereto is a method performed in the first communication node and a method corresponding thereto. (eg, receiving or transmitting a signal). For example, when an operation of a receiving node is described, a corresponding transmitting node may perform an operation corresponding to that of the receiving node. Conversely, when the operation of the transmitting node is described, the corresponding receiving node may perform an operation corresponding to that of the transmitting node.

In one embodiment of the communication system, the transmitter inputs a transport block (TB) composed of information (or message) bits to be transmitted to a channel coding block, and performs a channel coding process to obtain a given code rate (or data symbol modulation). It outputs coded bits suitable for order and modulation and coding scheme (MCS) level related to channel coding rate. The channel coding block calculates and adds cyclic redundancy check (CRC) bits of the transport block, (if the size of the transport block exceeds a certain size), code block segmentation (ie, CRC bits are added). dividing a transport block into multiple code blocks), calculating and adding CRC bits for each code block, channel encoding for each code block, rate matching (RM) for each code block (CB), and HARQ ( It may consist of all or part of hybrid automatic repeat request processing, bit interleaving, and code block concatenation (ie, concatenation of coded bits of multiple code blocks). In the above, if the transport block is not divided into multiple code blocks, calculating and adding CRC bits for each code block is unnecessary, and the remaining function blocks are performed on the transport block itself to which the CRC bit is added, and the last code block concatenation may be unnecessary In the case of rate matching, the code rate can be adjusted by performing puncturing and/or shortening and/or repetition on coded bits, which are outputs of channel encoding. . As channel encoding, various channel encoding methods such as turbo code, low density parity check (LDPC) code, and polar code can be applied. Encoded bits output from the channel coding block may be input to a modulation/constellation mapper after scrambling (bit level scrambling), and then data symbols may be output through a data modulation process.

In the following, each set of data symbols output through a channel coding process and a data modulation process for each transport block is referred to as a codeword. That is, one transport block and one codeword are regarded as one-to-one correspondence. Additionally, the transmitter may perform the following additional procedure on data symbols (or symbols spread to resources on a second 2D domain (or area) described below) for multi-antenna transmission. The transmitter may map data symbols belonging to the codeword(s) (or symbols spread to resources on a second 2D domain described below) to layer(s). Accordingly, when a plurality of transport blocks are transmitted, corresponding codewords may be mapped to different layer(s). Data symbols belonging to different codewords (or symbols spread to resources on a second 2D domain described below) may be restricted to be mapped only to different layers. Data symbols mapped to each layer (or symbols spread to resources on a second 2D domain described below) may be spread to each antenna port through multi-antenna preprocessing.

The transmitter converts the output signals into analog signals through a digital-to-analog converter (DAC) after a two-dimensional (partial) spread multicarrier modulation process described below in the baseband, and then passes through an analog/RF stage to a receiver through an antenna. can be sent to The receiver converts the RF band signal received from the transmitter through the reception antenna to a digital signal through an analog/RF stage and an analog-to-digital converter (ADC), and then converts the baseband to a two-dimensional (partial) transmission described below. A spread multicarrier demodulation process may be performed.

The receiver may decode data symbols detected after the demodulation process through the following process. The detected data symbols may be input to a modulation/constellation demapper to output a log-likelihood ratio (LLR) for coded bits. After descrambling the LLR of the coded bits output in this way, input them to a channel decoding block and undergo a channel decoding process to output the information bits of the corresponding transport block and the result of whether or not the transport block passed the CRC. . If the CRC for the corresponding transport block passes, it can be regarded as decoding success, and otherwise, as decoding failure. When a codeword is composed of a plurality of code blocks, the channel decoding block performs code block segmentation/deconcatenation into a plurality of code blocks, performs bit deinterleaving for each code block, and rate dematching for each code block ( rate dematching) and LLR combining, channel decoding for each code block, CRC check for each code block, code block concatenation/desegmentation (ie, decoded bits for each code block) concatenation), and transport block CRC check. If the codeword received above is not composed of a plurality of code blocks, code block division is unnecessary and the remaining functional blocks are performed on the codeword itself, CRC check for each code block and code block concatenation are unnecessary, and channel Whether or not the CRC passes can be checked from information bits and CRC bits of the corresponding transport block, which are decoding outputs.

In the case of downlink (DL) in the following, the transmitter may be a base station or repeater or transmission point (TP) or transmission and reception point (TRP), and the receiver may be a device or mobile or user terminal (user equipment: UE). In the case of uplink (UL), the transmitter may be a terminal or a user terminal, and the transmitter may be a base station, a relay, a transmission point, or a transmission/reception point. Unless otherwise specified, transmitters and receivers do not differentiate between link directions. In the case of sidelink (SL), a transmitter and a receiver may be different terminals.

On the other hand, the discrete Fourier transform (DFT), inverse DFT (IDFT), Walsh-Hadamard transform (WHT), and inverse WHT (IWHT) referred to in this specification are normalized so that the magnitudes of the input signal and the output signal do not change. may have been applied.

3 is a block diagram for explaining an embodiment of a communication system including a transmitting communication node and a receiving communication node.

Referring to FIG. 3 , a communication system 300 may include a plurality of communication nodes. Some of the communication nodes of the communication system 300 may operate as transmitting communication nodes that perform radio signal transmission operations. A transmitting communication node may also be referred to as a 'transmitting node'. Meanwhile, some of the communication nodes of the communication system 300 may operate as transmission communication nodes that perform radio signal reception operations. The receiving communication node may also be referred to as a 'receiving node'. In one embodiment of the communication system 300, each of the communication nodes may operate as a transmitting node and/or a receiving node. 3 can be seen as illustrating an embodiment of a communication system 300 including a communication node operating as a transmission node 310 and a communication node operating as a reception node 320 at a specific point in time. However, this is only an example for convenience of description, and the embodiment of the communication system is not limited thereto.

The transmitting node 310 may modulate a symbol or signal to be transmitted to the receiving node 320. The transmitting node 310 may transmit the modulated signal through an antenna or an emitter. A signal transmitted from the transmission node 310 may be carried on a transmission waveform and transmitted over a channel or medium. The reception node 320 may receive a signal transmitted through a channel or medium through an antenna or a collector. The receiving node 320 demodulates the received signal based on the received waveform, thereby restoring the original signal that the transmitting node 310 intends to transmit. Here, the original signal to be transmitted from the transmitting node 310 to the receiving node 320 may include various types of symbols or signals such as data symbols, pilot symbols, and reference signals. Embodiments of a modulation and demodulation method and apparatus in a wireless communication system will be described in this specification by taking operations in which the transmission node 310 and the reception node 320 mutually transmit and receive an original signal that is a data symbol as an example. However, this is only an example for convenience of description and the embodiment of the communication system is not limited thereto. For example, embodiments of the communication system described with reference to FIG. 3 may be equally or similarly applied to operations for transmitting and receiving various types of original signals such as pilot symbols and reference signals.

Specifically, the communication system 300 includes a transmission node 310 for transmitting a signal, a reception node 320 for receiving a signal from the transmission node 310, and a signal between the transmission node 310 and the reception node 320. It may include a channel 330 through which it can be transmitted. The transmitting node 310 may include an encoding unit 312 and a modulating unit 314, and the like, and the receiving node may include a demodulation unit 322 and a decoding unit 324 and the like. Here, the encoder 312, the modulator 314, the demodulator 322, and the decoder 324 are an encoder 312, a modulator 314, and a demodulator 322, respectively. ) and a decoder 324.

Each of the transmitting node 310 and the receiving node 320 is a base station (eg, the base stations 110-1, 110-2, 110-3, 120-1, and 120-2 shown in FIG. 1) or a terminal ( For example, it may correspond to terminals 130-1, 130-2, 130-3, 130-4, 130-5, and 130-6 shown in FIG. When the transmitting node 310 is a base station, the receiving node 320 may be a terminal. Alternatively, when the transmitting node 310 is a terminal, the receiving node 320 may be a base station or another terminal. Each of the transmitting node 310 and the receiving node 320 may be configured identically or similarly to the communication node 200 shown in FIG. 2 .

The transmitting node 310 may perform an encoding process and a modulation process before transmitting the transport block to the receiving node 320. The encoder 312 may perform an encoding process on the transport block. Specifically, the encoder 312 may encode the transport block into a codeword including a plurality of code blocks. At this time, the encoder 312 may divide the transport block into small blocks. The encoder 312 may configure a block set by combining the divided blocks into an arbitrary number. The number of blocks may be determined by a channel environment of a communication network, performance information of the transmitting node 310 and the receiving node 320, and requirements of an application program. The encoder 312 may perform encoding in units of block sets. The encoder 312 may transmit the encoded codeword to the modulator 314.

The modulator 314 may generate modulation symbols by modulating the codeword. The modulator 314 may modulate the codeword into symbols using various modulation methods. Modulation may mean converting signal information into an appropriate form such as the strength or displacement, frequency or phase of a signal (information) in accordance with the channel characteristics of a transmission medium. Modulation may be a process of converting a signal containing data into a waveform suitable for a transmission channel.

The transmission node 310 maps modulated symbols to time/frequency resources, and transmits a signal generated based on the mapped symbols to a channel 330 formed between the transmission node 310 and the reception node 320. It can be transmitted to the receiving node 320 through. Specifically, signals propagated through the antenna of the transmitting node 310 may be transmitted to the antenna of the receiving node 320 through the channel 330 . At this time, noise may occur on the channel.

A signal received through the antenna of the reception node 320 may be transmitted to the demodulator 322 . The demodulator 322 may demodulate a signal according to a demodulation method determined according to a channel environment. The demodulator 322 may demodulate the signal to generate a codeword and transmit the demodulated codeword to the decoder 324 . The decoder 324 receiving the demodulated codeword may decode the codeword to obtain an output transport block.

In one embodiment of the communication system, an orthogonal frequency division multiplexing (OFDM) method or an orthogonal frequency division multiple access (OFDMA) method may be used for modulation and demodulation. The OFDM scheme (or OFDMA scheme) may be characterized in that data symbols are mapped to time-frequency two-dimensional resources and transmitted. The time-frequency two-dimensional resources may have intervals of mutual OFDM symbol intervals on the time axis and intervals of mutual subcarrier intervals on the frequency axis. The OFDM scheme may have advantages of relatively excellent frequency efficiency and simple implementation. However, the OFDM scheme may have a problem in that communication performance is greatly deteriorated due to inter-carrier interference (ICI) that may occur in a high-speed moving environment due to a high Doppler shift or the like. When the sub-carrier spacing (SPS) is increased to alleviate the problem of the OFDM scheme, the ratio of the guard interval is increased, which may cause transmission efficiency to deteriorate.

In order to improve the problems of the OFDM scheme, an orthogonal time frequency space (OTFS) scheme (or an OTFS transmission scheme) is being studied. In the OTFS scheme, unlike the OFDM scheme, resources in the delay-Doppler domain can be used in a process of transmitting and receiving data symbols. The OTFS scheme can have a long channel coherence time in the delay-Doppler domain, so channel feedback can not be required with high frequency, and high transmission efficiency can be achieved. However, when radio resources are limited or non-ideal pulses are used in a communication environment, performance of the OTFS scheme may deteriorate. A technique for improving the performance of the OTFS method by efficiently mitigating this problem may be required.

4 is a block diagram for explaining an embodiment of a modulator of a transmission node in a communication system.

Referring to FIG. 4 , a transmission node in a communication system may include a modulator 400 . Here, the modulator may be the same as or similar to the modulator 314 of the transmission node 310 described with reference to FIG. 3 . The modulator 400 may receive a codeword generated through a process such as encoding in a transport block or data symbols within the codeword. The modulator 400 of the transmission node may modulate the input codeword or data symbols within the codeword to output transmission signals. In one embodiment of the communication system, the modulator 400 may include components such as a first mapper 410, a preprocessor 420, a second mapper 430, and a multicarrier modulator 440. . However, this is only an example for convenience of description, and the embodiment of the communication system is not limited thereto.

A first mapper 410 of the modulator 400 may map input data symbols to resources on a first 2-dimensional (2D) domain. Next, the pre-processor 420 of the modulator 400 performs pre-processing for spreading data symbols mapped to resources on the first 2D domain to resources on the second 2D domain. processing or precoding). Next, the second mapper 430 of the modulator 400 may map the preprocessed data symbols to resources on the second 2D domain. Next, the multicarrier modulator 440 of the modulator 400 may perform multicarrier modulation for each resource in the second 2D domain on data symbols mapped to resources in the second 2D domain after preprocessing. there is. Here, multicarrier modulation for each resource on the second 2D domain of data symbols may mean that multicarrier modulation is performed for each OFDM symbol. Specifically, the multicarrier modulator 440 may generate transmission signals by modulating data symbols mapped to resources on the second 2D domain after preprocessing into individual waveforms for each corresponding resource on the second 2D domain. . The transmission signals generated by the multicarrier modulator 440 may be transmitted to a receiving node.

During the modulation process in the modulator 400, data symbols may be 2D spread or 2D partial spread for resources in the 2D domain. Here, '2D partial spreading' may mean that data symbols are spread into a specific spreading resource group or spreading resource block in the 2D domain. Here, a specific spreading resource group or spreading resource block may include some resources or all resources in the 2D domain. Hereinafter, in this specification, a 'resource block' may be an expression referring to a 'diffusion resource block' or a 'diffusion resource group'.

A delay-Doppler or Doppler-delay domain as an embodiment of the first 2D domain and a frequency (or subcarrier)-time (or multicarrier symbol) as an embodiment of the second 2D domain ) (frequency-time or subcarrier-MC symbol) or time (or multi-carrier symbol)-frequency (or multi-carrier) (time-frequency or MC symbol-subcarrier) domains may be considered.

In the case of multiple-input and multiple-output (MIMO) or multiple-input and single-output (MISO) transmission using multiple antenna ports, preprocessing for multiple-antenna transmission is provided. It may be performed between the first mapper 410 and/or the preprocessor 420 and the second mapper 430. In the former case, the operation of the first mapper 410 can be performed for each antenna port, which is an output of preprocessing for multi-antenna transmission. In the latter case, the operations of the first mapper 410 and the pre-processing unit 420 are performed for each layer, which is the input of pre-processing for multi-antenna transmission, and after pre-processing for multi-antenna transmission is performed, the output of pre-processing for multi-antenna transmission is performed. Operations of the second mapper 430 and the modulator 440 may be performed for each in-antenna port.

In one embodiment of the communication system, a modulation process of data symbols constituting one codeword may be performed as follows. Data symbols within one codeword may be mapped to N _G 2D spreading resource blocks (including a case where N _G is 1) on the first 2D domain. To spread data symbols mapped to each 2D spreading resource block in the first 2D domain to resources in each corresponding 2D spreading resource block in the second 2D domain, each spreading resource in the first 2D domain Pre-processing of corresponding data symbols for each block (as an example of pre-processing, discrete symplectic Fourier transform (DSFT) (or discrete Fourier transform (DFT) for the first dimension and inverse DFT (IDFT) for the second dimension) alone or with this Another example of preprocessing is the Walsh-Hadamard transform (WHT) for the first dimension and the inverse WHT (IWHT) for the second dimension alone or together. (which may be configured including transmission windowing following this) may be performed. Data symbols preprocessed for each spreading resource block in the first 2D domain may be mapped to resources in each corresponding spreading resource block in the second 2D domain. Multicarrier modulation may be performed on data symbols spread over resources on the second 2D domain.

The first 2-dimensional domain may be defined as a delay-Doppler domain or a Doppler-delay domain or other 2-dimensional domain types. For convenience, the following 2-dimensional delay-Doppler domain (the first dimension in the first 2-dimensional domain corresponds to the delay domain) and the second dimension corresponds to the Doppler domain).

The second 2-dimensional domain may be defined as a frequency-time domain or a time-frequency domain or other 2-dimensional domain types, and in the following, for convenience, the frequency-time domain (the first dimension in the second 2-dimensional domain corresponds to the frequency domain) and the second dimension corresponds to the time domain).

Different spreading resource blocks on the first 2D domain may have different independent resource grids. Different spreading resource blocks on the second 2D domain may share a common lattice. Each spreading resource block on the first 2D domain may correspond to different spreading resource blocks on the second 2D domain. The size of each spreading resource block in the first 2D domain may be the same as the size of each corresponding spreading resource block in the second 2D domain. Spreading resource blocks on the first 2-dimensional domain may be sequentially assigned indexes, and data symbols may be sequentially mapped in the order of spreading resource block indexes. When the spreading resource block size is K , the first K data symbols are sequentially mapped to the 0th spreading resource block, the next K data symbols are sequentially mapped to the next 1st spreading resource block, and the next spreading resource blocks It is also possible to proceed in the order of spreading resource block indexes in this way. These mapping rules can be defined in advance.

When data symbols are mapped within each spreading resource block on the first 2-dimensional domain, the delay axis may be mapped first and then the Doppler axis may be mapped. Alternatively, you can map to the Doppler axis first and then to the delay axis. These mapping rules may be defined in advance, or rules to be applied to the terminal by the base station may be set.

When mapping the spreading resource block on the first 2D domain to resources on the second 2D domain, it may be first mapped to the frequency axis and then mapped to the time axis. (This may have the advantage of having a relatively small processing delay.) Alternatively, when mapping the spreading resource block on the first 2D domain to the resources on the second 2D domain, it is first mapped on the time axis and then the frequency Axis can be mapped. These mapping rules may be defined in advance, or rules to be applied to the terminal by the base station may be set.

Modulations such as CP-OFDM, W-OFDM (or PS-OFDM), and F/SBF-OFDM can be applied as multi-carrier modulation. Data symbols or spread data symbols may be multiplexed with the reference signals on the first 2D domain to the second 2D domain and transmitted.

Specifically, a codeword or data symbols within the codeword may be input to the modulator 400 of the transmitting node. The modulator 400 may map data symbols in the codeword to resources on the first 2D domain. Such a mapping operation may be performed by the first mapper 410 . The modulator 400 may perform preprocessing to spread data symbols mapped to resources on the first 2D domain to resources on the second 2D domain. Such a pre-processing operation may be performed by the pre-processing unit 420 . The modulator 400 may map the preprocessed data symbols to resources on the second 2D domain. Such a mapping operation may be performed by the second mapper 430 . The modulator 400 may multicarrier-modulate data symbols mapped to resources on the second 2D domain for each resource or resource block on the second 2D domain. Such a modulation operation may be performed by the multicarrier modulators 400 and 440 .

Specifically, the modulator 400 may map only data symbols to resources on the first 2D domain, or may further map additional signals to resources on the first 2D domain in addition to the data symbols. For example, for some of the resources on the first 2D domain, one or more reference signals (RSs) or one or more symbols corresponding to one or more RSs may be mapped instead of data symbols. Alternatively, one or more demodulation reference signals (DMRS) or one or more symbols corresponding to one or more DMRSs may be mapped to some of the resources in the first 2D domain instead of data symbols. Meanwhile, some of the resources on the first 2D domain may be emptied without mapping in consideration of interference between resources on the first 2D domain. Alternatively, guard resources may be allocated to some of the resources on the first 2D domain. A mapping operation of resources on the first 2D domain by the first mapper 410 will be described in detail with reference to FIG. 6 below.

5 is a conceptual diagram for explaining embodiments of a data symbol and reference signal (RS) mapping method for resources on a first two-dimensional (2D) domain in a communication system.

Referring to FIG. 5 , in one embodiment of a communication system, a first communication node may modulate a signal for transmission to a second communication node. Here, the first communication node may be the same as or similar to the transmission node 310 described with reference to FIG. 3 . The second communication node may be the same as or similar to the reception node 320 described with reference to FIG. 3 . The first communication node may modulate the codeword output as a result of operation on the transport block by the encoding unit through the modulation unit. Here, the modulator may be the same as or similar to the modulator 400 described with reference to FIG. 4 . Here, the modulator of the first communication node may map data symbols constituting a codeword to resources on the first 2D domain in a process of performing a modulation operation. The modulator of the first communication node may map one or more RSs in addition to data symbols to resources on the first 2D domain.

The first example of FIG. 5 may refer to a mapping scheme in which one RS is mapped to a centrally located resource and data symbols are mapped to the remaining resources. The second example of FIG. 5 may refer to a mapping scheme in which one RS is mapped to a centrally located resource, 8 null resources surrounding it are not mapped, and data symbols are mapped to the remaining resources. The third example of FIG. 5 may refer to a mapping scheme in which RS is mapped to nine centrally located resources and data symbols are mapped to the remaining resources.

When symbols mapped or spread to different resource blocks on the second 2D domain are transmitted (described in detail below), channels experienced within each resource block on the second 2D domain are different, so that each resource on the second 2D domain Reference signal(s) may need to be transmitted to resource(s) on the first 2D domain corresponding to each block.

Resources on the first 2D domain that are spread to different resource blocks on the second 2D domain can be separated and despread on the second 2D domain, so that symbols spread to different resource blocks on the second 2D domain are the first It may be advantageous to allocate resources on one 2D domain contiguously, whereas data symbols transmitted on resources on a first 2D domain spread over the same resource block on a second 2D domain have a channel spread on the first 2D domain. Since they may interfere with each other, it may be advantageous to allocate them as far apart as possible. For data symbols spread and transmitted within the same resource block on the second 2D domain to a specific receiver, resources on the first 2D domain It may be desirable to map in an interleaved pattern to .

6A and 6B are conceptual diagrams for explaining an embodiment of a relationship between data symbols, resources on a first 2D domain, and resources on a second 2D domain in a communication system.

6A shows an example in which 12 data symbols in one codeword are mapped to 12 resources on a first 2D domain and each of the 4 data symbols is spread into the same resource block on a second 2D domain. there is. It can be seen that each of the 4 data symbols spread into each same resource block on the second 2D domain is mapped in an interleaved pattern to the resource grid on the first 2D domain. On the other hand, referring to FIG. 6B, symbols spread into different resource blocks in the second 2D domain will be considered to be distributed and mapped to independently separated lattices (ie, three lattices) in the first 2D domain. can

For multi-access of different users, all resources on the first 2D domain may be divided and data symbols for different users may be mapped without overlapping. Additionally or alternatively, for multi-access of different users, data symbols for different users may be mapped without overlapping by dividing all resources on the second 2D domain. The latter case will be described in detail in the operation of the second mapper 130 below. In the case of non-orthogonal multiple access, resources on the first 2D domain and/or resources on the second 2D domain may be mapped to different users by partially or entirely overlapping them.

7 is a block diagram for explaining an embodiment of a demodulation unit of a receiving node in a communication system.

Referring to FIG. 7 , a receiving node in a communication system may include a demodulation unit 700 . Here, the demodulation unit may be the same as or similar to the demodulation unit 322 of the receiving node 310 described with reference to FIG. 3 . The demodulator 700 may receive information on reception signals in which a radio signal modulated and transmitted from a transmission node of a communication system is received via a radio channel. The demodulator 700 may modulate the received signals received through the radio channel and output demodulated data symbols. The demodulated data symbols may constitute a demodulated codeword that is a restoration result of a codeword intended to be transmitted by a transmitting node. In one embodiment of the communication system, the demodulator 700 may include components such as a multicarrier demodulator 710, a first demapper 720, a postprocessor 740, and a second demapper 750. can The demodulator 700 may further include a first channel equalizer 730 and a second channel equalizer 760 . However, this is only an example for convenience of explanation, and embodiments of the present disclosure are not limited thereto.

The multicarrier demodulator 710 may receive information on received signals received through a radio channel. The multicarrier demodulation unit 710 may perform multicarrier demodulation on the received signals for each resource in the second 2D domain. Here, multicarrier demodulation for each resource on the second 2D domain of the received signals may mean that multicarrier demodulation is performed for each OFDM symbol. The multicarrier demodulation unit 710 may output information on multicarrier demodulated signals. The multicarrier demodulated signals output from the multicarrier demodulator 710 can be viewed as being mapped to resources within a total of _NG resource blocks on the second 2D domain.

Signals output from the multicarrier demodulator 710 may be input to the first demapper 720 . The first demapper 720 may perform demapping on demodulated signals mapped to resources within a total of N _G resource blocks on the second 2D domain. The demapped signals may be input to the signal post-processing unit 740 . Alternatively, the demapped signals may be input to the post-processing unit 740 after undergoing a channel equalization operation in the first channel equalizer 730 .

The post-processing unit 740 may perform a post-processing operation for each resource block on signals demapped from a total of _NG resource blocks on the second 2D domain or signals that have undergone a channel equalization operation after demapping. . Signals that have undergone post-processing in the post-processing unit 740 can be regarded as the same as or similar to states mapped to resource blocks on the first 2D domain. Signals that have undergone post-processing in the post-processing unit 740 may be input to the second demapper 750 .

Post-processing in the post-processing unit 740 is receive windowing (RX windowing) followed by inverse DSFT (IDSFT) (or IDFT for the first dimension and DFT for the second dimension). Alternatively, it may be configured with only inverse DSFT (IDSFT) without receiving windowing. As another example, post-processing may consist of receive windowing followed by IWHT for the first dimension and WHT for the second dimension or IWHT for the first dimension and WHT for the second dimension without receive windowing.

8 is a conceptual diagram for explaining an embodiment of a machine learning structure.

Referring to FIG. 8 , a communication system may include a plurality of communication nodes. Here, the plurality of communication nodes may be the same as or similar to any one of the transmitting communication node 310 and the receiving communication node 320 described with reference to FIG. 3 . In the communication system, machine learning control based on the machine learning control device may be performed. Here, the machine learning control device is included in at least some of the communication nodes constituting the communication system, or is connected to at least some of the communication nodes constituting the communication system to perform operations for machine learning control. can do. The machine learning control device may control machine learning (or machine learning operation) for generating a learning model for efficiently determining the size of a window to be applied to communication between transmission and reception nodes. The machine learning control device may include a machine learning structure for performing a machine learning operation. Hereinafter, in describing an embodiment of the machine learning structure with reference to FIG. 8 , contents overlapping those described with reference to FIGS. 1 to 7 may be omitted.

The machine learning control device may perform a machine learning control operation based on a machine learning structure identical or similar to that shown in FIG. 8 or a machine learning structure including a structure identical or similar to that shown in FIG. 8 . For example, the machine learning control device may obtain a learning model for performing an operation to reduce the influence of mutual interference between communication nodes through a machine learning structure. The memory and/or storage device of the machine learning control device may include program instructions for performing machine learning according to a predetermined machine learning structure. The machine learning control device may include a machine learning unit for performing machine learning according to a predetermined machine learning structure. A machine learning control device including a machine learning unit or a machine learning unit may be referred to as a 'machine learning device'.

The machine learning control device can obtain a learning model for efficiently performing machine learning control through machine learning according to a structure such as an artificial neural network (ANN) or a deep neural network (DNN). . For example, it can be seen that FIG. 7 shows a DNN structure composed of multi-layer multi-nodes among machine learning structures. However, this is only one example for convenience of explanation, and the embodiment of the machine learning structure is not limited thereto. For example, in one embodiment of the communication system, an ANN structure, a recurrent neural network (RNN) structure, a neuron structure composed of a single node, a perceptron structure composed of a single node, and a knowledge-based system ) structure, a structure applying reasoning techniques such as Bayesian, and a deep neural network structure, various machine learning structures can be applied to the machine learning unit. A machine learning structure selected from among various machine learning structures according to a predetermined criterion may be applied to the machine learning unit. For example, a machine learning structure selected according to various conditions, such as development and/or production cost of a communication system and/or individual device, performance requirements, and processor capability, may be applied to the machine learning unit.

In one embodiment of the communication system, the plurality of layers constituting the artificial neural network may include an input layer, a hidden layer, an output layer, and the like. The input layer may be a layer into which a data set or data group to be learned is input. An input layer may include one or more input nodes. Some or all of the entries constituting the data set may be input to each of the at least one input node constituting the input layer. A data set input to one or more input nodes constituting the input layer may be data that has undergone data preprocessing in advance. The output layer may refer to a layer in which data or signals input to the artificial neural network are output through operation in the artificial neural network. An output layer may include at least one or more output nodes.

At least one hidden layer may be disposed between the input layer and the output layer. An artificial neural network having two or more hidden layers may be referred to as a deep neural network (DNN). That is, DNN may mean a neural network structure in which a plurality of hidden layers are disposed between an input layer and an output layer in a neural network structure composed of an input layer, a hidden layer, and an output layer. A machine learning method based on a DNN structure may be referred to as deep learning. A hidden layer may be connected to an input layer, an output layer, or another hidden layer through weight vectors.

In one embodiment of the communication system, the machine learning device may perform a learning operation of updating weight vectors of an artificial neural network. The machine learning device may include a multi-layer perceptron classifier. The learning operation of the artificial neural network may be performed by a multi-layer perceptron classifier included in the machine learning device. The multilayer perceptron classifier may train an artificial neural network through a preset learning algorithm. The learning algorithm may include machine learning algorithms such as a supervised learning (SL) algorithm and an unsupervised learning (UL) algorithm.

In one embodiment of the communication system, the machine learning device may perform a series of calculations and obtain an output value through a feed-forward operation in an artificial neural network structure. The machine learning device may obtain error information based on an output value and a preset reference value. The machine learning apparatus may perform a learning operation of correcting weight vectors between layers of an artificial neural network by back-propagating the calculated error information. The machine learning device may modify weight vectors between layers of the artificial neural network through a preset optimization algorithm. For example, the optimization algorithm may include gradient descent, alternating gradient descent, stochastic gradient descent, or an adam-optimizer algorithm. The machine learning device may repeatedly perform the learning operation as many times as the number of epochs corresponding to the preset learning number. As the number of epochs increases, prediction performance or accuracy of a model obtained through machine learning may be improved. On the other hand, as the number of epochs increases, the amount of calculation in the machine learning process increases, the calculation load increases, and learning efficiency and the like may decrease. The number of epochs may be set to a value determined by a person skilled in the art to be suitable for improving the performance of the machine learning device.

In one embodiment of the communication system, the machine learning control device may perform machine learning based on a predetermined neural network structure for a machine learning control operation. Here, the total number of layers of the neural network structure may be L, and L may be a natural number of 3 or more. When the neural network corresponds to a DNN, L may be a natural number of 4 or greater. Each of the layers may be expressed as an l (lowercase English alphabet L)-th layer from the input layer to the output layer. Each of the layers may be expressed as an l(0, 1, ..., L-1) th layer, and among them, layers from l=1 th to l=L-2 th may be hidden layers. For example, the DNN structure may include three hidden layers and may be composed of 32, 64, and 32 hidden nodes, respectively. However, this is only one example for convenience of explanation, and the embodiment of the communication system is not limited thereto and may encompass various embodiments of machine learning or artificial neural network technology.

Input data may be input to the input layer of the neural network structure. The input data may be referred to as I (capital child of the English alphabet). As input data I input to the input layer passes through each layer and is calculated through successive functions, output data may be output from the output layer. Output data can be referred to as O (capital letter O in the English alphabet). The output data O may be equal to Equation 1, for example.

In Equation 1, W may be at least one weight coefficient or weight parameter set between nodes of each layer. Alternatively, W may be weight vector data including at least one weight coefficient or weight parameter set between nodes of each layer. Meanwhile, f ^(l) may be a function established between the 1-th layer and the 1-1 th layer. For example, f ^(l) may correspond to a function such as a sigmoid function or a rectified linear unit (ReLu). The sigmoid function may refer to a function used for calculation between layers such as output mapping in a machine learning structure. For example, the sigmoid function can be expressed as Equation 2.

The ReLu function may refer to a function used as an activation function in an operation between layers such as an input layer and/or a hidden layer. For example, the ReLu function can be expressed as Equation 3.

However, functions such as Equation 2 and Equation 3 are merely examples presented to aid understanding, and embodiments of the machine learning structure are not limited thereto and may encompass various types of neural network embodiments.

In one embodiment of the communication system, the neural network structure of the machine learning control device may receive input data I and output output data O to generate a learning model. Here, the operation of generating the learning model may be performed one or more times before actually performing data communication between the transmitting and receiving nodes or at a specific time point between performing data communication between the transmitting and receiving nodes.

The input data I may include information related to a radio channel between transmitting and receiving nodes. Input data I may be generated by pre-processing radio channel information or radio channel matrix information between transmitting and receiving nodes. Here, information of a radio channel (or information of a radio channel matrix) may correspond to channel information in a delay-Doppler domain acquired by a machine learning control device (or a transmission node including a machine learning control device) from a reception node. However, this is only an example for convenience of description, and the embodiment of the communication system is not limited thereto.

For example, input data I may be generated by transforming information of a 2D matrix about a radio channel between transmitting and receiving nodes into a form of 1D information (eg, 1D vector). That is, input data I composed of a plurality of pieces of information may be input in a vector form. The neural network structure may output output data O by performing an operation based on a weight coefficient (or weight data) W in a vector form with respect to input data I input in a vector form. For example, in some or all of the neural network structures configured through neurons or perceptrons in one embodiment of the communication system, an output value may be calculated using a sigmoid function in the form of Equation 4.

In Equation 4, I may be vector data corresponding to information input as input data I. For example, I may be vector data composed of information about a radio channel matrix or information about elements constituting a radio channel matrix. Alternatively, I may be vector data composed of values obtained by processing or scaling information such as information about the radio channel matrix or information about elements constituting the radio channel matrix to a value between 0 and 1. W may correspond to at least one weight coefficient configured in a vector form. Meanwhile, y may correspond to a value obtained as an operation result or output data O. The output data O may correspond to information of a window (for example, information of a transmission window used for transmission windowing of a transmission node). Alternatively, the output data O may correspond to a value for calculating window information. In other words, the information of the window corresponding to the radio channel information input to the input data I can be obtained through an operation (for example, post-processing) on the output data O output from the neural network structure.

In one embodiment of the communication system, the learning model to be acquired through iterative learning using the machine learning structure by the machine learning control device is, when input data about delay-Doppler channel information between transmitting and receiving nodes is input, the corresponding delay-Doppler channel It may correspond to a learning model that adaptively (or environment-adaptively) outputs information of a window (or transmission window) suitable for information. In other words, the machine learning control device may acquire a first learning model through repeated learning using a machine learning structure, and the first learning model is input when input data about delay-Doppler channel information between transmitting and receiving nodes is input, corresponding It may correspond to a learning model that adaptively outputs information of a first window suitable for delayed-Doppler channel information.

In a machine learning method using an artificial neural network having a structure such as DNN, an output value may be evaluated according to a regression learning model and each parameter may be updated. For the evaluation of the output value, a loss function may be defined based on the relationship between the output value and the correct value. For example, in an embodiment of the artificial neural network, a loss function may be defined as Equation 5 or Equation 6.

In Equation 5, K may mean the number of learning samples, x may correspond to the correct answer value (or correct answer vector) to be learned through the artificial neural network, and x _o is output through the output layer of the artificial neural network. It may correspond to an output value (or an output vector). In Equation 6, x may correspond to the correct answer value (or correct answer vector) to be learned through the artificial neural network, and x _o may correspond to the output value (or output vector) output through the output layer of the artificial neural network. . However, the loss functions in the form of Equations 5 and 6 are only examples to aid understanding, and embodiments of the machine learning structure are not limited thereto and may include various types of loss function embodiments. For example, in Equation 5 or 6, x and x _o correspond to a specific value obtained through operation (pre-processing, post-processing, normalization, etc.) on the correct answer value (or correct answer vector) and output value (or output vector), respectively. may apply. Alternatively, x and x _o may correspond to specific values used to obtain a correct answer value (or correct answer vector) and an output value (or output vector), respectively. In other words, correct answer values (or correct answer vectors) and output values (or output vectors) may be obtained through operations on x and x _o (pre-processing, post-processing, normalization, etc.).

In Equations 5 and 6, the loss function may have a smaller value as the difference between x _o and x is smaller. In one embodiment of the machine learning control device, iterative learning and parameter update may be performed in a direction in which a loss function is minimized. Accordingly, x _o obtained through the output of the artificial neural network structure may approach x . That is, calculation performance or prediction performance of a learning model or a prediction model obtained through an artificial neural network structure may be improved. For example, in a learning process for generating a learning model in the machine learning control device, the artificial neural network structure may output an output value related to a window corresponding thereto when an input value related to a wireless channel is input. Here, comparison between x _o (or output value) and x (or correct value) may be performed. Accordingly, iterative learning and parameter update may be performed in a direction in which a loss function defined based on x _o and x is minimized.

9 is a flowchart for explaining an embodiment of a machine learning-based communication method.

Referring to FIG. 9 , a communication system may include a plurality of communication nodes. At least some of the plurality of communication nodes included in the communication system may be the same as or similar to the transmission node described with reference to FIG. 3 or 4 and the reception node described with reference to FIG. 4 or 7. At least some of the plurality of communication nodes may support transmission and reception of radio signals of the OTFS method. The communication system may include a first machine learning device. The first machine learning device may be the same as or similar to the machine learning control device or machine learning device described with reference to FIG. 8 . The first machine learning device may generate a learning model for adaptively supporting a radio signal transmission/reception operation. Hereinafter, in describing an embodiment of a machine learning-based communication method with reference to FIG. 9 , contents overlapping those described with reference to FIGS. 1 to 8 may be omitted.

According to an embodiment of a machine learning-based communication method, first learning is performed through a procedure for generating virtual channel information (hereinafter, step S900), a procedure for performing a machine learning operation (hereinafter, step S910), and repetition of a machine learning operation. A procedure for acquiring a model (hereinafter, step S920), a procedure for performing real-time wireless signal transmission and reception in an environment-adaptive manner using the first learning model (hereinafter, step S930), and the like may be performed. In FIG. 9, steps S900 to S930 are expressed as being performed time-sequentially in one communication system (or by one subject), but this is only an example for convenience of description and is an embodiment of a machine learning-based communication method. is not limited to For example, at least some of the procedures (steps S900 to S930) shown in FIG. 9 may be performed by the same or different entities. The first machine learning device may be a subject separated from communication nodes of the communication system. In this case, the operation of generating the first learning model according to steps S900 to S920 may be performed by the first machine learning device before communication nodes are produced or while communication nodes are produced. The generated first learning model can be loaded into production communication nodes, and can be used for real-time communication in the loaded communication nodes. Alternatively, the first machine learning device may be included in at least some of the communication nodes or configured to be connected to some of the communication nodes. In this case, the first machine learning device may generate or update the first machine learning model by performing a machine learning operation on a regular or irregular basis. The first machine learning device may provide information of the generated or updated first machine learning model to at least some of the communication nodes. Communication nodes that have acquired the information of the first machine learning model may perform real-time communication using the first machine learning model.

Specifically, the first machine learning device may generate channel information that can be generated in the first 2D domain (S900). Here, an embodiment of the machine learning method will be described taking a case where the first 2D domain corresponds to the delay-frequency domain as an example. However, this is only an example for convenience of explanation, and the embodiment of the machine learning method is not limited thereto.

와 같이 표현될 수 있고, 도플러 성분은

과 같이 표현될 수 있다. 또는, 제1 2D 도메인에서 도플러 성분은

와 같이 표현될 수 있고, 지연 성분은

과 같이 표현될 수도 있다.The first 2D domain may correspond to a delay-Doppler domain. Here, the first 2D domain may have a MХN (or NХM) lattice structure. M and N may be natural numbers corresponding to the number of delay components or Doppler components, respectively. The delay component in the first 2D domain is

It can be expressed as, and the Doppler component is

can be expressed as Alternatively, the Doppler component in the first 2D domain is

It can be expressed as, and the delay component is

It can also be expressed as

일 수 있다). 가상 채널 정보 H _v 를 구성하는 엘리먼트들은, h_ij와 같이 표현될 수 있다. 제1 기계학습 장치는 S900 단계에서 생성된 가상 채널 정보 H _v 에 기초한 가상의 무선 송수신 결과에 기초하여, 기계학습 동작을 수행할 수 있다(S910). S910 단계에 따른 기계학습 동작과 관련하여는, 이하 도 10을 참조하여 보다 구체적으로 설명한다.In step S900, the first machine learning apparatus may randomly generate a power-delay-Doppler profile within a possible power-delay-Doppler frequency range. Alternatively, a predefined power-delay-Doppler profile may be provided instead of randomly generated. Here, the power-delay-Doppler profile may mean information composed of average channel power, delay, and Doppler frequency values for each channel path having a non-zero average channel power. The first machine learning device may generate virtual wireless channels (or virtual channel information H _v ) based on the generated (or provided) power-delay-Doppler profile. Virtual channel information H _v can be expressed as a MХN-sized matrix (that is,

can be). Elements constituting the virtual channel information H _v may be expressed as h _ij . The first machine learning device may perform a machine learning operation based on a virtual wireless transmission/reception result based on the virtual channel information H _v generated in step S900 (S910). The machine learning operation according to step S910 will be described in more detail with reference to FIG. 10 below.

10 is a flowchart for explaining an embodiment of a machine learning method.

Referring to FIG. 10 , the first machine learning device may be the same as or similar to the first machine learning device described with reference to FIG. 9 . The first machine learning device may generate a learning model for adaptively supporting a radio signal transmission/reception operation. Hereinafter, in describing an embodiment of the machine learning method with reference to FIG. 10 , contents overlapping those described with reference to FIGS. 1 to 9 may be omitted.

에 해당할 수 있다. 정규화된 가상 채널 정보

를 구성하는 엘리먼트들은

와 같이 표현될 수 있다. 가상 채널 정보 H _v 를 구성하는 엘리먼트들 h_ij, 및 정규화된 가상 채널 정보

를 구성하는 엘리먼트들

은, 수학식 7과 동일 또는 유사한 관계를 가질 수 있다.The first machine learning device may perform virtual wireless signal transmission and reception (or simulation of wireless signal transmission and reception) to update the machine learning structure. The first machine learning device may input an input value generated by preprocessing the virtual channel information H _v generated in step S900 described with reference to FIG. 9 to the first machine learning structure, and in the first machine learning structure An output value may be obtained (S1020). In step S1020, the input value generated by preprocessing the virtual channel information H _v is the normalized virtual channel information.

may correspond to Normalized Virtual Channel Information

elements that make up

can be expressed as Elements constituting virtual channel information H _v h _ij , and normalized virtual channel information

elements that make up

may have the same or similar relationship as Equation 7.

는 가상 채널 정보 H _v 를 구성하는 엘리먼트들 h_ij의 크기(또는 절대값), 또는 각 엘리먼트들에 대한 경로 이득에 해당할 수 있다.In Equation 7, P may be the number of non-zero elements or MN among elements constituting the virtual channel information H _v . Referring to Equation 7,

may correspond to the size (or absolute value) of the elements h _ij constituting the virtual channel information H _v or the path gain for each element.

는, 가상 채널 정보 H _v 를 구성하는 엘리먼트들 h_ij 각각에 대하여 수학식 7에 따른 연산(이를테면, 정규화 연산)을 수행함으로써 생성될 수 있다. 제1 기계학습 장치는 수학식 7에 따른 연산을 수행하여 생성된 정규화된 가상 채널 정보

에 대응되는 입력값을, 제1 기계학습 구조에 입력할 수 있다. 제1 기계학습 장치는 제1 기계학습 구조에서의 연산을 거쳐서 출력된 출력값 u를 획득할 수 있다. 여기서 출력값 u는, 제1 2D 도메인에서의 무선 신호 전송을 위해 사용되는 제1 윈도우 U에 대응되는 값일 수 있다. 여기서, 제1 윈도우 U는 도 4를 참조하여 설명한 송신 노드의 송신 윈도잉에 사용되는 송신 윈도우 U에 해당할 수 있다. 제1 기계학습 장치는, S1020 단계에서 획득한 출력값 u에 대한 연산을 통하여 제1 윈도우 U를 구성할 수 있다(S1030). S1030 단계에서 출력값 u에 대하여 수행되는 연산은, 이를테면 절대값 연산, 정규화 연산 등을 포함할 수 있다.Normalized Virtual Channel Information

may be generated by performing an operation (eg, normalization operation) according to Equation 7 for each of the elements h _ij constituting the virtual channel information H _v . The first machine learning device normalized virtual channel information generated by performing an operation according to Equation 7

An input value corresponding to may be input to the first machine learning structure. The first machine learning device may obtain an output value u output through an operation in the first machine learning structure. Here, the output value u may be a value corresponding to the first window U used for radio signal transmission in the first 2D domain. Here, the first window U may correspond to the transmission window U used for transmission windowing of the transmission node described with reference to FIG. 4 . The first machine learning device may configure a first window U through an operation on the output value u obtained in step S1020 (S1030). The operation performed on the output value u in step S1030 may include, for example, an absolute value operation and a normalization operation.

The first machine learning device may generate a virtual transmission signal x _u based on the first window U (S1040). Specifically, the first machine learning device may randomly generate the same number (MN) of data symbols as the size (MХN) of the first 2D domain. A set of data symbols generated in this way can be expressed as transmission symbol information x . The transmission symbol information x may be vector data having a size of MNХ1. The first machine learning apparatus performs a first transform operation (eg, an inverse symplectic Fourier transform (ISFT) operation) on the generated transmission symbol information x to obtain a signal x' in the time-frequency domain. The first machine learning apparatus may generate a virtual transmission signal x _u by applying a first window U to the obtained signal x' in the time-frequency domain. For example, the virtual transmission signal x _u may be generated identically or similarly to Equation 8.

항은 시간-주파수 영역의 신호 x'의 생성을 위한 제1 변환 연산에 대응될 수 있다.

항은 시간-주파수 영역의 신호 x'에 제1 윈도우 U를 적용하여 가상 송신 신호 x _u를 생성하는 연산(또는 제2 변환 연산)에 대응될 수 있다. S1040 단계에서 수학식 8 등에 기초하여 전송 심볼 정보 x로부터 가상 송신 신호 x _u를 생성하는 동작은, 송신 윈도잉(TX windowing) 동작에 대응될 수 있다.In Equation 8, U _diag may be a diagonal matrix having a size of MNХMN corresponding to the first window U (or generated based on the first window U ). For example, U _diag may be a diagonal matrix having time-frequency domain window values as diagonal components.

The term may correspond to a first transform operation for generating a signal x' in the time-frequency domain.

The term may correspond to an operation (or a second conversion operation) of generating a virtual transmission signal x _u by applying a first window U to a signal x' in the time-frequency domain. An operation of generating a virtual transmission signal x _u from transmission symbol information x based on Equation 8 or the like in step S1040 may correspond to a transmission windowing (TX windowing) operation.

에 기초하여 생성된 제1 채널 H, 및 소정의 수신 잡음 w를 적용함으로써, 가상 수신 신호 y를 구성할 수 있다. 이를테면, 가상 수신 신호 y는 수학식 9와 동일 또는 유사하게 구성될 수 있다.The first machine learning device may calculate a first loss function defined based on a virtual transmission/reception result of the virtual transmission signal x _u (S1050). Specifically, the first machine learning device provides normalized virtual channel information to the virtual transmission signal x _u

A virtual received signal y may be configured by applying the first channel H generated based on and a predetermined received noise w . For example, the virtual received signal y may be configured the same as or similar to Equation 9.

에 기초하여 생성될 수 있다. 구체적으로는, M×N 크기의 정규화된 가상 채널 정보

에 기초하여 생성된 제1 채널 H는 MN×MN 크기의 행렬(또는 행렬 정보, 행렬 데이터 등)에 해당할 수 있다. 제1 채널 H는 가상 채널 정보 H _v 또는 정규화된 가상 채널 정보

에 대응 가능한 순시 채널 중 하나일 수 있다. 제1 채널 H를 구성하는 각각의 열들은 정규화된 가상 채널 정보

에 대한 변환 연산, 벡터화(vectorization) 연산, 블록 순환 시프트(block circulant shift) 연산, 순환 시프트(circulant shift) 연산 중 적어도 일부에 기초하여 구성될 수 있다.In Equation 9, the first channel H is normalized virtual channel information

can be created based on Specifically, normalized virtual channel information of M×N size

The first channel H generated based on may correspond to a MN×MN size matrix (or matrix information, matrix data, etc.). The first channel H is virtual channel information H _v or normalized virtual channel information

It may be one of the instantaneous channels that can correspond to . Each column constituting the first channel H is normalized virtual channel information

It may be configured based on at least some of a conversion operation, a vectorization operation, a block circulant shift operation, and a circulant shift operation for .

를 벡터화한 결과로 구성될 수 있다. 제1 채널 H에서 제1 열을 제외한 나머지 열들은, 제1 열에 대한 소정의 시프트(shift 또는 천이) 연산에 기초하여 구성될 수 있다. 이를테면, 제1 채널 H에서 제1 열을 제외한 나머지 열들은, 제1 열에 대한 블록 순환 시프트(block circulant shift) 연산 및/또는 순환 시프트(circulant shift) 연산에 기초하여 생성될 수 있다.For example, the first column (ie, the first column) of the first channel H is normalized virtual channel information

It can be composed of the result of vectorizing . Columns other than the first column in the first channel H may be configured based on a predetermined shift (transition) operation for the first column. For example, columns other than the first column in the first channel H may be generated based on a block circulant shift operation and/or a circulant shift operation on the first column.

The first machine learning apparatus may calculate restored symbol information x _o based on the virtual received signal y generated as in Equation 9 based on the first channel H. Reconstructed symbol information x _o may be calculated based on the effective channel matrix H _u in which the influence of the first window U is reflected and the virtual received signal y . Here, the effective channel matrix H _u and reconstruction symbol information x _o reflecting the influence of the first window U may be calculated identically or similarly to Equation 10.

는 이를테면

와 같이 정의될 수 있다. 이 경우,

는 제1 2D 도메인(즉, 지연-도플러 영역)의 잡음(이를테면, 수신 잡음 w)의 분산 값에 해당할 수 있다. 수학식 10과 같이 계산된 복원 심볼 정보 x _o 는, S1040 단계에서 제1 윈도우 U에 기초하여 전송 심볼 정보 x로부터 생성된 가상 송신 신호 x _u에 대한 송수신 결과(또는, 가상 송신 신호 x _u의 수신 결과에 대한 복원 결과)에 해당하는 것으로 볼 수 있다. 다르게 표현하면, 복원 심볼 정보 x _o 는 제1 기계학습 장치가 제1 기계학습 구조에 의하여 생성한 제1 윈도우 U에 기초하여 수행된 가상의 무선 신호 송수신 결과에 해당하는 것으로 볼 수 있다. 제1 기계학습 장치는, 전송 심볼 정보 x 및 복원 심볼 정보 x _o 에 기초하여, 제1 손실함수를 계산할 수 있다. 제1 손실함수는 수학식 5 또는 수학식 6과 동일 또는 유사할 수 있다. 또는, 제1 손실함수는 수학식 11과 동일 또는 유사할 수 있다.In Equation 10,

is for example

can be defined as in this case,

may correspond to a variance value of noise (eg, received noise w ) in the first 2D domain (ie, delay-Doppler domain). The restored symbol information x _o calculated as in Equation 10 is the result of transmission and reception of the virtual transmission signal x _u generated from the transmission symbol information x based on the first window U in step S1040 (or the reception of the virtual transmission signal x _u result) can be seen as corresponding to the restoration result for the result. In other words, the restored symbol information x _o can be regarded as corresponding to a virtual radio signal transmission/reception result performed based on the first window U generated by the first machine learning apparatus by the first machine learning structure. The first machine learning apparatus may calculate a first loss function based on the transmitted symbol information x and the restored symbol information x _o . The first loss function may be the same as or similar to Equation 5 or Equation 6. Alternatively, the first loss function may be the same as or similar to Equation 11.

, 및 OOBE(out-of-band emission) 억제를 위한 제3 손실함수

의 조합으로 정의될 수 있다. 또한 c는 제3 손실함수 항에 대한 가중치를 결정하는 0 이상의 실수값일 수 있다. 여기서, 제2 손실함수 및 제3 손실함수는 수학식 12 및 수학식 13과 같이 정의될 수 있다.Referring to Equation 11, the first loss function is the second loss function for suppressing mean square error (MSE)

, and a third loss function for suppressing out-of-band emission (OOBE)

can be defined as a combination of In addition, c may be a real value of 0 or more that determines a weight for the third loss function term. Here, the second loss function and the third loss function may be defined as Equations 12 and 13.

The second loss function according to Equation 12 may be the same as or similar to the loss function according to Equation 6 described with reference to FIG. 8 . The third loss function according to Equation 13 may be defined as a ratio between the power of the main lobe and the power of the side lobe in the transmission signal x _u .

The first machine learning device may update the first machine learning structure based on the value of the first loss function calculated in step S1050 (S1060). The update of the first machine learning structure may be performed in a direction in which the value of the first loss function decreases. For example, the update of the first machine learning structure may be performed based on stochastic gradient descent (SGD), etc., but this is only an example for convenience of description and embodiments of the machine learning method are not limited thereto. don't

Referring back to FIG. 9 , the first machine learning apparatus may obtain a first learning model by repeatedly performing the machine learning operation according to step S910. In step S920, the first machine learning device may repeatedly perform the machine learning operation according to step S910 until a preset termination condition is satisfied. For example, the first machine learning device may perform iterative learning until a preset termination condition is satisfied. For example, iterative learning may be terminated after repetitions of a predetermined number of epochs are performed.

Alternatively, the iterative learning may be terminated when it is determined that the value of the first loss function converges to a certain level or more. For example, iterative learning may be terminated when the value of the first loss function becomes equal to or less than a predetermined threshold value. Iterative learning may be terminated when it is determined that the value of the first loss function is no longer reduced. The iterative learning may be terminated when the value of the first loss function does not decrease while repeating iterations by a predetermined first number of times.

The first machine learning device may obtain a first learning model based on the updated first machine learning structure until the iterative learning ends. Here, the first learning model is information of one or more artificial neural networks constituting the first machine learning structure (or information of parameters constituting the artificial neural networks) that is finally updated through iterative learning described with reference to steps S910 and S920. can correspond to When an input value related to radio channel information (eg, channel information on the first 2D domain) between transmitting and receiving nodes is input, the first learning model acquired in step S920 is input, the first window (eg, transmission used in the transmitting node) window) can be output.

The first machine learning device may provide the first learning model acquired in step S920 to at least some of the plurality of communication nodes included in the communication system. One or more communication nodes that have acquired the first learning model may perform a real-time radio signal transmission/reception procedure adaptively to the environment using the first learning model (S930).

For example, the first communication node may check real-time first 2D domain channel information. The first communication node may calculate an input value to be input to the first learning model based on the checked real-time first 2D domain channel information. Here, the calculation of the input value may be the same as or similar to the calculation of generating the input value of the first machine learning structure from the virtual channel information H _v in an embodiment of the machine learning method described with reference to FIG. 9 . The first communication node may input the calculated input value to the first learning model, and obtain an output value through an operation in the first learning model. The first communication node may calculate a first window in real time based on the output value. Here, the calculation of the real-time window based on the output value may be the same as or similar to the calculation of configuring the first window U through the calculation of the output value u in one embodiment of the machine learning method described with reference to FIG. 9 . The real-time window generated based on the first learning model may be considered to be environment-adaptively generated based on real-time first 2D domain channel information.

When the first communication node is a transmission node, the first communication node may perform transmission windowing by applying a real-time window generated based on the first learning model when transmitting a radio signal. When the first communication node is a receiving node, the first communication node generates a radio signal transmitted to the first communication node by another communication node (eg, a second communication node) in the communication system based on the first learning model. It can be estimated that it is transmitted based on the same or similar transmission window as the real-time window, and a radio signal can be received according to the estimation result.

The first learning model may be generated by repeatedly performing machine learning in a direction in which the first machine learning device reduces the value of the first loss function defined by Equation 11. In other words, the first learning model reduces the value of the first loss function defined as a combination of the values of the second loss function for MSE suppression and the third loss function for OOBE suppression. Based on repeated learning in the direction can be created by Due to this, the first learning model can adaptively generate a transmission window that suppresses the occurrence of OOBE and prevents signal distortion at the receiving end (or receiving node) side from occurring significantly.

According to an embodiment of the machine learning-based communication method and apparatus, the first machine learning apparatus may repeatedly perform a machine learning operation based on a result of transmitting and receiving a virtual wireless signal. The first machine learning device generates a first learning model for adaptively generating a window used for wireless signal transmission in a first two-dimensional domain (eg, a two-dimensional domain used in a transmission scheme such as the OTFS scheme) can create One or more communication nodes of the communication system may generate a window suitable for a real-time communication environment based on the first learning model generated by the first machine learning device, and perform real-time communication based on the created window. As a result, performance of transmitting and receiving radio signals can be efficiently improved.

Claims (1)

As a method of operating the first device,
performing a machine learning operation based on a result of transmitting and receiving a virtual wireless signal; and
And outputting a first learning model obtained by repeatedly performing the step of performing the machine learning operation,
The step of performing the machine learning operation,
obtaining an output value for a first window by inputting an input value generated based on virtual channel information in a first 2-dimensional domain to a first machine learning structure;
generating a virtual transmission signal corresponding to one or more transmission symbols based on an output value for the first window;
obtaining a virtual received signal and one or more reconstruction symbols corresponding to the virtual received signal, based on the virtual channel information and the virtual transmission signal; and
Based on a loss function defined based on the one or more transmission symbols and the one or more recovery symbols, updating the first machine learning structure.
Method of operation of the first device.

Images