Method and device for transmitting and receiving signals using artificial intelligence in a wireless communication system

Abstract

The present disclosure relates to a 5G communication system or a 6G communication system for supporting a higher data rate than a 4G communication system such as Long Term Evolution (LTE). A method for transmitting or receiving a signal by a user equipment (UE) in a mobile communication system is provided. The method may include: identifying a neural network model for transmitting first information to a base station (BS); learning connection weights of the neural network model using the first information; transmitting second information for updating weights of a second part of the neural network corresponding to the base station based on a result of the learning to the base station; and updating weights of the first part of the neural network corresponding to the UE based on the result of the learning.

Description

Technical Field

[0001] The present disclosure relates to a method and apparatus for transmitting or receiving a signal including channel information using deep learning and artificial intelligence in a wireless communication system. Background Art

[0002] As wireless communications have evolved over generations, technologies such as voice calls, multimedia services, and data services have been developed primarily for services targeting people. With the commercialization of 5G (fifth generation) communication systems, the number of interconnected devices is expected to grow exponentially. These devices will increasingly be connected to communication networks. Examples of the Internet of Things can include vehicles, robots, drones, home appliances, displays, smart sensors connected to various infrastructures, construction machinery, and factory instruments. Mobile devices are expected to evolve in various forms (such as augmented reality glasses, virtual reality headsets, and holographic devices). In order to provide various services by connecting hundreds of billions of devices and things in the 6G (sixth generation) era, efforts have been made to develop improved 6G communication systems. For these reasons, 6G communication systems are referred to as super 5G systems.

[0003] The 6G communication system, which is expected to be commercialized around 2030, will have a peak data rate of tera-level bps and a radio latency of less than 100 μs, making it 50 times faster than the 5G communication system and having a radio latency of 1 / 10 of that of the 5G communication system.

[0004] In order to achieve such high data rates and ultra-low latency, the implementation of 6G communication systems in the terahertz band (e.g., the 95 GHz to 3 THz band) has been considered. It is expected that since the path loss and atmospheric absorption in the terahertz band are more severe than those in the millimeter wave (mmWave) band introduced in 5G, technologies that can ensure the signal transmission distance (i.e., coverage) will become more critical. As the main technology to ensure coverage, it is necessary to develop radio frequency (RF) elements, antennas, and new waveforms with better coverage than orthogonal frequency division multiplexing (OFDM), beamforming and massive multiple input multiple output (MIMO), full-dimensional MIMO (FD-MIMO), array antennas, and multi-antenna transmission technologies such as massive antennas. In addition, new technologies to improve the coverage of terahertz band signals, such as metamaterial-based lenses and antennas, orbital angular momentum (OAM), and reconfigurable smart surfaces (RIS), have been discussed.

[0005] Furthermore, to improve spectrum efficiency and overall network performance, the following technologies have been developed for 6G communication systems: full-duplex technology for enabling uplink and downlink transmissions to simultaneously use the same frequency resources; network technologies that utilize satellites, high-altitude platform stations (HAPS), and the like in an integrated manner; an improved network structure for supporting mobile base stations and the like and achieving network operation optimization and automation; dynamic spectrum sharing technology for conflict avoidance based on spectrum usage prediction; the use of artificial intelligence (AI) in wireless communications to improve overall network operations by leveraging AI from the design phase of 6G development and internalizing end-to-end AI support functions; and next-generation distributed computing technologies that overcome the computing power limitations of UEs through ultra-high-performance communication and computing resources accessible on the network, such as mobile edge computing (MEC) and the cloud. Furthermore, efforts are continuing to strengthen connectivity between devices, optimize networks, promote the softwareization of network entities, and increase the openness of wireless communications by designing new protocols to be used in 6G communication systems, developing mechanisms for implementing hardware-based security environments and secure data usage, and developing technologies for maintaining privacy.

[0006] Research and development of 6G communication systems for hyperconnectivity, including both human-to-machine (P2M) and machine-to-machine (M2M), are expected to bring about the next hyperconnected experience. Specifically, services such as truly immersive extended reality (XR), high-fidelity mobile holograms, and digital replicas are expected to be provided through 6G communication systems. Furthermore, services such as remote surgery for enhanced safety and reliability, industrial automation, and emergency response will be provided through 6G communication systems, enabling these technologies to be applied in various fields such as industry, healthcare, automobiles, and home appliances.

[0007] In connection with communication between a user equipment (UE) and a base station (BS) in a communication system, research on a signal transmission or reception method using an artificial neural network is ongoing.

[0008] The above information is presented as background information only to assist in understanding the present disclosure. No determination has been made, and no assertion is made, as to whether any of the above might be applicable as prior art to the present disclosure. Summary of the Invention

[0009] Technical issues

[0010] An aspect of the present disclosure is to provide a method and apparatus for accurately transmitting or receiving information using a limited number of bits during communication performed between a User Equipment (UE) and a Base Station (BS) in a communication system.

[0011] Solution to the problem

[0012] According to one aspect of the present disclosure, a method for sending or receiving a signal by a user equipment (UE) in a mobile communication system may include: identifying a neural network model for sending first information to a base station (BS); using the first information, learning connection weights of the neural network model; based on the learning result, sending second information to the base station for updating the weights of a second part of the neural network corresponding to the base station; and based on the learning result, updating the weights of the first part of the neural network corresponding to the UE.

[0013] According to one aspect of the present disclosure, a method for sending or receiving a signal by a BS in a mobile communication system may include: identifying a neural network model for receiving first information from a UE; receiving second information from the UE for updating the weights of a second part of the neural network corresponding to the BS; and updating the weights of the second part of the neural network based on the second information.

[0014] According to one aspect of the present disclosure, a UE that sends or receives signals in a mobile communication system may include: a transceiver configured to send or receive signals; and a controller configured to: identify a neural network model for sending first information to a BS, learn connection weights of the neural network model using the first information, send second information for updating the weights of a second part of the neural network corresponding to the BS based on a result of the learning to the BS, and update the weights of a first part of the neural network corresponding to the UE based on the result of the learning.

[0015] According to one aspect of the present disclosure, a BS that sends or receives signals in a mobile communication system may include: a transceiver configured to send or receive signals; and a controller configured to: identify a neural network model for receiving first information from a UE, receive second information from the UE for updating the weights of a second part of the neural network corresponding to the BS, and update the weights of the second part of the neural network based on the second information.

[0016] Advantageous Effects of the Invention

[0017] According to various embodiments, information to be sent or received may be accurately conveyed via an artificial neural network using a limited number of bits.

[0018] According to various embodiments, performance degradation of artificial neural networks may be mitigated, and overhead associated with learning may be reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] For a more complete understanding of the present disclosure and its advantages, reference is now made to the following description taken in conjunction with the accompanying drawings, wherein like reference numerals represent like parts, and wherein:

[0020] Figure 1A diagram illustrating a basic structure of a time-frequency domain as a radio resource area for transmitting data or a control channel in an LTE system is shown;

[0021] Figure 2 A diagram illustrating the PDCCH as a physical downlink channel on which DCI for LTE is transmitted is shown;

[0022] Figure 3 A diagram illustrating an example of a basic unit of time and frequency resources configured for a downlink control channel that can be used in 5G is shown;

[0023] Figure 4 A diagram illustrating an example of a control region (control resource set (CORESET)) in which a downlink control channel is transmitted in a 5G wireless communication system is shown;

[0024] Figure 5 A diagram illustrating a case where 14 OFDM symbols are configured to be used as a single time slot (or subframe) in the downlink, a PDCCH is transmitted in the first two OFDM symbols, and a DMRS is transmitted in the third symbol;

[0025] Figure 6 shows the configuration of an autoencoder according to an embodiment;

[0026] Figure 7 shows a diagram illustrating an autoencoder based downlink channel feedback scheme 700 according to an embodiment;

[0027] Figure 8 shows a block diagram illustrating the operations of a user equipment (UE) and a base station (BS) according to a first embodiment;

[0028] Figure 9 shows a flowchart illustrating operations of the UE and the BS according to the first embodiment;

[0029] Figure 10 shows a block diagram illustrating operations of a UE and a BS according to a second embodiment;

[0030] Figure 11 shows a flowchart illustrating operations of the UE and the BS according to the second embodiment;

[0031] Figure 12 shows a block diagram illustrating operations of a UE and a BS in Mode 1 according to the third embodiment;

[0032] Figure 13 shows a flowchart illustrating operations of the UE and the BS in Mode 1 according to the third embodiment;

[0033] Figure 14shows a block diagram illustrating operations of a UE and a BS in Mode 2 according to the third embodiment;

[0034] Figure 15 shows a flowchart illustrating operations of the UE and the BS in Mode 2 according to the third embodiment;

[0035] Figure 16 shows a block diagram illustrating operations of a UE and a BS in Mode 3 according to a third embodiment;

[0036] Figure 17 shows a flowchart illustrating operations of the UE and the BS in Mode 3 according to the third embodiment;

[0037] Figure 18 shows a block diagram illustrating the structure of a BS according to an embodiment; and

[0038] Figure 19 A block diagram illustrating the structure of a UE according to an embodiment is shown. DETAILED DESCRIPTION

[0039] Before proceeding to the following detailed description, it may be helpful to set forth definitions of certain words and phrases used throughout this patent document: the terms "include" and "comprising" and their derivatives mean inclusion without limitation; the term "or" is inclusive, meaning and / or; the phrases "associated with..." and "associated with..." and their derivatives may mean including, being included, interconnected with..., containing, being contained within, connected to or connected with..., coupled to or coupled with..., communicable with..., cooperating with..., interwoven, juxtaposed, proximate, coupled to or coupled with..., having, having the property of..., etc.; the term "controller" refers to any device, system, or portion thereof that controls at least one operation, which device may be implemented in hardware, firmware, or software, or some combination of at least two of hardware, firmware, or software. It should be noted that the functionality associated with any particular controller may be centralized or distributed, whether locally or remotely.

[0040] In addition, the various functions described below can be implemented or supported by one or more computer programs, each of which is formed of computer-readable program code and contained in a computer-readable medium. The terms "application" and "program" refer to one or more computer programs, software components, instruction sets, processes, functions, objects, classes, instances, related data, or parts thereof suitable for implementation with suitable computer-readable program code. The phrase "computer-readable program code" includes any type of computer code, including source code, object code, and executable code. The phrase "computer-readable medium" includes any type of medium that can be accessed by a computer, such as read-only memory (ROM), random access memory (RAM), hard drive, compact disk (CD), digital video disc (DVD), or any other type of memory. "Non-transitory" computer-readable media excludes wired, wireless, optical, or other communication links that transmit temporary electrical or other signals. Non-transitory computer-readable media include media that can permanently store data and media that can store data and overwrite it later, such as rewritable optical disks or erasable memory devices.

[0041] Definitions for certain words and phrases are provided throughout this patent document, those of ordinary skill in the art should understand that in many, if not most instances, such definitions apply to prior, as well as future uses of such defined words and phrases.

[0042] Discussed below Figures 1 to 19 The various embodiments used to describe the principles of the present disclosure in this patent document are merely exemplary and should not be construed in any way to limit the scope of the present disclosure. Those skilled in the art will understand that the principles of the present disclosure can be implemented in any suitably arranged system or device.

[0043] Wireless communication systems have evolved into broadband wireless communication systems that provide high-speed and high-quality packet data services (in addition to the voice-based services initially provided, similar to communication standards such as 3GPP's High Speed ​​Packet Access (HSPA), Long Term Evolution (LTE) or Evolved Universal Terrestrial Radio Access (E-UTRA), LTE-Advanced (LTE-A), LTE-Pro, 3GPP2's High Speed ​​Packet Data (HRPD), Ultra Mobile Broadband (UMB), and IEEE's 802.16e).

[0044] As a representative example of a broadband wireless communication system, the LTE system adopts an orthogonal frequency division multiplexing (OFDM) scheme for the downlink (DL) and a single carrier frequency division multiple access (SC-FDMA) scheme for the uplink (UL). The uplink refers to a wireless link through which a terminal (user equipment (UE) or mobile station (MS)) sends data or a control signal to a base station (eNode B or base station (BS)). The downlink refers to a wireless link through which a BS sends data or a control signal to a UE. The above-mentioned multiple access scheme can allocate or manage time-frequency resources, through which data or control information is carried for each user, so that they do not overlap with each other, that is, they have orthogonality, thereby distinguishing the data or control information of each user.

[0045] Beyond LTE communication systems (i.e., 5G communication systems) need to freely support a variety of requirements from users, service providers, and others. Therefore, they need to support services that simultaneously meet these requirements. Services considered for 5G communication systems may include enhanced mobile broadband (eMBB), massive machine-type communications (mMTC), and ultra-reliable low-latency communications (URLLC).

[0046] eMBB is designed to provide data transmission rates that exceed the data transmission rates supported by conventional LTE, LTE-A, or LTE-pro. For example, in a 5G communication system, from the perspective of a single BS, eMBB needs to provide a peak data rate of up to 20 Gbps in the downlink and a peak data rate of up to 10 Gbps in the uplink. In addition, the 5G communication system needs to provide enhanced user-perceived UE data rates as well as peak data rates. To meet these requirements, it is expected that various transmission or reception technologies will be improved, including advanced multiple-input multiple-output (MIMO) transmission technology. In addition, LTE uses a maximum transmission bandwidth of 20 MHz in the 2 GHz band currently used by LTE to transmit signals. However, the 5G communication system uses a frequency bandwidth wider than 20 MHz in a frequency band ranging from 3 to 6 GHz or in a frequency band greater than or equal to 6 GHz, so that the data transmission rate required by the 5G communication system can be met.

[0047] In addition, in 5G communication systems, in order to support application services such as the Internet of Things (IoT), mMTC is being considered. In order to effectively provide IoT, mMTC needs to support a large number of UE access within a cell, increase the coverage area of ​​UE, increase battery life expectancy, reduce UE costs, etc. The Internet of Things (IoT) provides communication functions by attaching to various sensors and various devices, so a large number of UEs (for example, 1,000,000 UEs / km) need to be supported within a cell. 2). In addition, due to the characteristics of the service, mMTC-enabled UEs are likely to be located in shadowed areas not covered by cells, such as basements of buildings, and may require a wider coverage area than other services of the 5G communication system. mMTC-enabled UEs need to be manufactured as inexpensive UEs, and the UE's battery may not be frequently replaced. Therefore, a long battery life, such as 10 to 15 years, may be required.

[0048] URLLC is a cellular-based wireless communication service for mission-critical purposes. For example, remote control services for robots or machines, industrial automation services, drone services, remote healthcare services, emergency alert services, etc. can be considered. Therefore, the communication provided by URLLC may need to provide very low latency and very high reliability. For example, services supporting URLLC need to exhibit an air interface latency of less than 0.5 milliseconds and need to meet a requirement of less than or equal to 10 -5 Therefore, for services supporting URLLC, the 5G system needs to provide a smaller transmission time interval (TTI) than other services, and at the same time needs to allocate wide resources in the frequency band to ensure the reliability of the communication link.

[0049] The three services in 5G (i.e., eMBB, URLLC, and mMTC) can be multiplexed and transmitted in a single system. In this case, in order to meet the different requirements of the services, different transmission or reception schemes and transmission or reception parameters can be used between the services.

[0050] Hereinafter, frame structures of LTE and LTE-A systems will be described in detail with reference to the accompanying drawings.

[0051] Figure 1 A diagram illustrating a basic structure of a time-frequency domain as a radio resource region for transmitting data or a control channel in an LTE system is shown.

[0052] refer to Figure 1, the horizontal axis is the time domain, and the vertical axis is the frequency domain. In the time domain, the minimum transmission unit is the OFDM symbol. A time slot 102 includes Nsymb OFDM symbols 101, and a subframe 103 includes two time slots. The length of a time slot is 0.5 milliseconds, and the length of a subframe is 1.0 milliseconds (ms). A radio frame 104 is a time domain unit including 10 subframes. In the frequency domain, the minimum transmission unit is a subcarrier. The entire system transmission bandwidth includes a total of NBW subcarriers 105. In the time-frequency domain, the basic resource unit is a resource element (RE) 106, and RE is represented by an OFDM symbol index and a subcarrier index. A resource block (RB) (or physical resource block (PRB)) 107 is defined by Nsymb consecutive OFDM symbols 102 in the time domain and NRB consecutive subcarriers 108 in the frequency domain. Therefore, one RB 108 includes Nsymb×NRB REs 106. Generally, the minimum transmission unit of data is an RB. In the LTE system, Nsymb=7 and NRB=12, and NBW and NRB may be proportional to the bandwidth of the system transmission band.

[0053] Subsequently, downlink control information (DCI) in LTE and LTE-A systems will be described in detail.

[0054] In the LTE system, scheduling information associated with downlink data or uplink data is transmitted from the base station to the user equipment (UE) via DCI. DCI operations can be performed by defining various formats and applying corresponding DCI formats, depending on whether the scheduling information is associated with uplink data or downlink data, whether the control information size is the small compact DCI size, whether spatial multiplexing using multiple antennas is applied, whether the DCI is used for power control, and so on. For example, DCI format 1, which is scheduling control information associated with downlink data, can be configured to include at least the following control information.

[0055] - Resource Allocation Type 0 / 1 Flag: Indicates whether the resource allocation scheme is Type 0 or Type 1. Type 0 applies a bitmap scheme and allocates resources in units of resource block groups (RBGs). In LTE systems, the basic scheduling unit is a resource block (RB), which is represented by time and frequency domain resources. An RBG includes multiple RBs and serves as the basic scheduling unit in the Type 0 scheme. Type 1 allows allocation of predetermined RBs within an RBG.

[0056] - Resource Block Allocation: Indicates the RBs allocated for data transmission. The resources to be indicated are determined based on the system bandwidth and resource allocation scheme.

[0057] - Modulation and Coding Scheme (MCS): Indicates the modulation scheme used for data transmission and the size of the transport block, which is the data to be transmitted.

[0058] -HARQ process number: indicates the HARQ process number.

[0059] - New data indicator: indicates HARQ initial transmission or HARQ retransmission.

[0060] - Redundancy version: indicates the redundancy version of HARQ.

[0061] - Transmit Power Control (TPC) Command for Physical Uplink Control Channel (PUCCH): indicates a transmit power control command for PUCCH, which is an uplink control channel.

[0062] After channel coding and modulation processes, the DCI is transmitted through a Physical Downlink Control Channel (PDCCH).

[0063] A cyclic redundancy check (CRC) is added to the payload of the DCI message, and the CRC can be scrambled with a radio network temporary identifier (RNTI) corresponding to the UE identity. Different RNTIs can be used depending on the purpose of the DCI message (e.g., UE-specific data transmission, power control command, random access response, etc.). That is, the RNTI is not sent explicitly, but is sent by being included in the CRC calculation process. If the UE receives a DCI message sent on the PDCCH, the UE can use the allocated RNTI to identify the CRC. If the CRC recognition result is correct, the UE can recognize that the corresponding message was sent for the UE.

[0064] Figure 2 A diagram illustrating PDCCH 201 as a physical downlink channel on which DCI of LTE is transmitted is shown.

[0065] refer to Figure 2, PDCCH 201 is time-multiplexed with PDSCH 202, which is a data transmission channel, and is sent across the entire system bandwidth. The area of ​​PDCCH 201 is represented as multiple OFDM symbols, which are indicated to the UE using a control format indicator (CFI) sent via the physical control format indicator channel (PCFICH). By assigning PDCCH 201 to the OFDM symbol present in the front of the subframe, the UE is able to decode the downlink scheduling assignment as quickly as possible. Therefore, the downlink shared channel (DL-SCH) decoding delay, that is, the total downlink transmission delay, can be reduced. A single PDCCH delivers a single DCI message and can schedule multiple UEs simultaneously in the downlink and uplink, so the transmission of multiple PDCCHs can be performed in parallel in each cell. The cell-specific reference signal (CRS) 203 is used as a reference signal for decoding PDCCH 201. CRS 203 is sent for each subframe across the entire frequency band, and the scrambling and resource mapping can be different for each cell identity (ID). CRS 203 is a reference signal used by all UEs, so UE-specific beamforming may not be applied. Therefore, the multi-antenna transmission scheme for PDCCH in LTE may be limited to an open-loop transmit diversity scheme. The UE implicitly obtains the number of CRS ports by decoding the Physical Broadcast Channel (PBCH).

[0066] Resource allocation for PDCCH 201 is performed in units of control channel elements (CCEs), with a single CCE comprising 9 resource element groups (REGs), i.e., a total of 36 resource elements (REs). The number of CCEs required for a predetermined PDCCH 201 may be 1, 2, 4, or 8, which is determined based on the channel coding rate of the DCI message payload. As described above, link adaptation for PDCCH 201 may be achieved using different numbers of CCEs. The UE needs to detect the signal without knowing the information associated with the PDCCH 201. In LTE, a search space indicating a set of CCEs is defined for blind decoding. The search space includes multiple sets according to each CCE aggregation level (AL), which is not explicitly signaled but is implicitly defined by a subframe number and a function associated with the UE identity. In each subframe, the UE decodes PDCCH 201 for all possible resource candidates that can be selected from the CCEs in the configured search space, and processes the information declared to be valid for the corresponding UE via a CRC check.

[0067] The search space can be categorized into a UE-specific search space and a common search space. A group of UEs or all UEs can search the common search space of PDCCH 201 to receive cell-common control information, such as dynamic scheduling or paging messages associated with system information. For example, scheduling allocation information for the DL-SCH, which transmits the system information block (SIB)-1 including cell operator information, can be received by searching the common search space of PDCCH 201.

[0068] In LTE, the entire PDCCH region is configured as a set of CCEs in a logical region, and there is a search space including the set of CCEs. The search space can be classified into a common search space and a UE-specific search space, and the search space for LTE PDCCH can be defined as follows.

[0069]

[0070]

[0071] According to the above definition of the search space of the PDCCH, the UE-specific search space is not explicitly signaled, but is implicitly defined by the subframe number and the function associated with the UE identity. In other words, the fact that the UE-specific search space changes depending on the subframe number means that the UE-specific search space may change over time. Through the above description, the problem that a predetermined UE cannot use the search space due to other UEs (blocking problem) can be overcome. Since all CCEs searched by the UE are currently used by other scheduled UEs within the subframe, the UE may not be scheduled in the same subframe. However, since the search space changes over time, this problem does not occur in subsequent subframes. For example, although the UE-specific search spaces of UE#1 and UE#2 partially overlap in the predetermined subframe, since the UE-specific search space is different for each subframe, it can be expected that the overlap is different in subsequent subframes.

[0072] According to the above definition of the PDCCH search space, because a group of UEs or all UEs need to receive the PDCCH, the common search space is defined as a set of pre-agreed CCEs. In other words, the common search space does not change depending on the UE identity, subframe number, etc. Although the common search space exists for the transmission of various system messages, it can be used to transmit control information for a single UE. As described above, the common search space can be used as a solution to the phenomenon in which a UE is not scheduled due to a lack of available resources in the UE-specific search space.

[0073] A search space is a set of candidate control channels, including CCEs that the UE should attempt to decode at a given aggregation level. There are various aggregation levels for bundling one, two, four, and eight CCEs into a single bundle, so the UE has multiple search spaces. In the LTE PDCCH, the number of PDCCH candidates that the UE will monitor in the search space and that are defined based on the aggregation level is defined in the following table.

[0074] [Table 1]

[0075]

[0076] According to Table 1, in the case of the UE-specific search space, aggregation levels {1, 2, 4, 8} are supported, and in this case, there are {6, 6, 2, 2} PDCCH candidates, respectively. In the case of the common search space 302, aggregation levels {4, 8} are supported, and in this case, there are {4, 2} PDCCH candidates, respectively. The common search space only supports aggregation levels {4, 8} to improve coverage characteristics, because system messages generally need to reach the edge of the cell.

[0077] The DCI transmitted in the common search space is defined only for predetermined DCI formats, such as 0 / 1A / 3 / 3A / 1C, corresponding to purposes such as power control for UE groups or system messages. In the common search space, DCI formats involving spatial multiplexing are not supported. The downlink DCI format that should be decoded in the UE-specific search space may change depending on the transmission mode configured for the corresponding UE. The transmission mode is configured via RRC signaling, so the subframe number is not accurately defined in association with whether the corresponding configuration is valid for the corresponding UE. Therefore, regardless of the transmission mode, the UE always performs decoding for DCI format 1A, thereby operating in a manner without losing communication.

[0078] In the above description, the method of transmitting or receiving a downlink control channel and downlink control information and a search space in conventional LTE and LTE-A has been described.

[0079] Hereinafter, a downlink control channel in the 5G communication system currently under discussion will be described in detail with reference to the accompanying drawings.

[0080] Figure 3 A diagram illustrating an example of a basic unit of time and frequency resources configured for a downlink control channel that can be used in 5G is shown. Figure 3, the basic unit (REG) of time and frequency resources configured for the control channel includes one OFDM symbol 301 on the time axis and 12 subcarriers 302 (i.e., 1 RB) on the frequency axis. When configuring the basic unit of the control channel, by assuming 1 OFDM symbol 301 as the basic time axis unit, the data channel and the control channel can be time-division multiplexed within a single subframe. By placing the control channel before the data channel, the processing time perceived by the user can be reduced, and therefore, the delay requirement can be easily met. The basic frequency axis unit of the control channel is set to 1 RB 302, so that frequency multiplexing between the control channel and the data channel can be effectively performed.

[0081] By joining Figure 3 The REG 303 shown can be configured in various sizes to configure the control channel region. For example, when CCE 304 is the basic unit for allocating downlink control channels in 5G, one CCE 304 can include multiple REGs 303. Figure 3 A description is provided for REG 303 of a plurality of REs. If REG 303 includes 12 REs and one CCE 304 includes six REGs 303, this means that one CCE 304 includes 72 REs. If a downlink control region is configured, the corresponding region includes multiple CCEs 304, and depending on the aggregation level (AL), a predetermined control channel can be transmitted by mapping a single CCE or multiple CCEs 304 in the control region. CCEs 304 in the control region can be distinguished by numbering, and the numbers can be allocated according to a logical mapping scheme.

[0082] Figure 3 The basic unit of the downlink control channel (ie, REG 303) may include the RE to which the DCI is mapped and the area to which the demodulation reference signal (DMRS) 305 is mapped. The demodulation reference signal 305 is a reference signal for decoding the DCI. Figure 3 As shown, DMRS 305 may be transmitted in 6 REs within one REG 303. For reference, DMRS 305 is transmitted using the same precoding as the control signal mapped to REG 303, so the UE can decode the control information without information associated with the precoding used by the base station.

[0083] Figure 4 A diagram illustrating an example of a control region (control resource set (CORESET)) in which a downlink control channel is transmitted in a 5G wireless communication system is shown. Figure 4 In the example, it is assumed that two control regions (control region #1 401 and control region #2 402) are configured within a system bandwidth 410 on the frequency axis and one time slot 420 on the time axis (e.g., Figure 4 The example assumes that 1 time slot includes 7 OFDM symbols). The control region 401 or 402 can be configured based on a predetermined subband 403 of the entire system bandwidth 410 on the frequency axis. On the time axis, the control region can be configured based on one or more OFDM symbols (which can be defined as the control region length (control resource set duration 404)). Figure 4 In the example of , control region #1 401 is configured based on a control region length of 2 symbols, and control region #2 is configured based on a control region length of 1 symbol.

[0084] As described above, the control region in 5G can be configured via higher-layer signaling (e.g., system information, master information block (MIB), RRC signaling) from the base station to the user equipment terminal. Configuring the control region for the user equipment terminal involves providing information related to the location of the control region, subband, resource allocation of the control region, and control region length. For example, the following information may be included.

[0085] [Table 2]

[0086]

[0087] In addition to the above configuration information, various types of information required to send downlink control channels may be configured for the UE.

[0088] Next, the downlink control information (DCI) in 5G will be described in detail.

[0089] In the 5G system, scheduling information associated with uplink data (physical uplink shared channel (PUSCH)) or downlink data (physical downlink shared channel (PDSCH)) can be transmitted from the BS to the UE via DCI. The UE can monitor the DCI format for fallback and the DCI format for non-fallback associated with the PUSCH or PDSCH. The fallback DCI format can be implemented as a fixed field between the BS and the UE, and the non-fallback DCI format can include a configurable field.

[0090] The fallback DCI for scheduling PUSCH may include the following information.

[0091] [Table 3]

[0092]

[0093]

[0094] The non-fallback DCI for scheduling PUSCH may include the following information.

[0095] [Table 4]

[0096]

[0097]

[0098] The fallback DCI for scheduling PDSCH may include the following information.

[0099] [Table 5-1]

[0100]

[0101] The non-fallback DCI for scheduling PDSCH may include the following information.

[0102] [Table 5-2]

[0103]

[0104]

[0105] After the channel coding and modulation process, the DCI can be sent via the Physical Downlink Control Channel (PDCCH). A cyclic redundancy check (CRC) is added to the payload of the DCI message, and the CRC can be scrambled with a radio network temporary identifier (RNTI) corresponding to the UE identity. Different RNTIs can be used depending on the purpose of the DCI message (e.g., UE-specific data transmission, power control commands, random access responses, etc.). That is, the RNTI is not sent explicitly, but is sent by being included in the CRC calculation process. If the UE receives a DCI message sent on the PDCCH, the UE can use the allocated RNTI to identify the CRC. If the CRC recognition result is correct, the UE can recognize that the corresponding message has been sent for the UE.

[0106] For example, DCI scheduling a PDSCH associated with system information (SI) may be scrambled with the SI-RNTI. DCI scheduling a PDSCH associated with a random access response (RAR) message may be scrambled with the RA-RNTI. DCI scheduling a PDSCH associated with a paging message may be scrambled with the P-RNTI. DCI reporting a slot format indicator (SFI) may be scrambled with the SFI-RNTI. DCI reporting transmit power control (TPC) may be scrambled with the TPC-RNTI. DCI scheduling a UE-specific PDSCH or PUSCH may be scrambled with the cell RNTI (C-RNTI).

[0107] If a data channel (ie, PUSCH or PDSCH) is scheduled for a predetermined UE via the PDCCH, the data may be transmitted or received together with the DMRS within the corresponding scheduled resource region. Figure 5A diagram illustrating a case where 14 OFDM symbols are configured to be used as a single time slot (or subframe) in the downlink, PDCCHs are transmitted in the first two OFDM symbols, and DMRSs are transmitted in the third symbol. Figure 5 In the case of the LTE / LTE-A system, in a predetermined RB in which the PDSCH is scheduled, the PDSCH is transmitted by mapping data to the REs through which the DMRS is not transmitted in the third symbol and the REs from the fourth to the last symbol. Figure 5 The subcarrier spacing Δf represented in is 15 kHz, and in the case of a 5G system, the subcarrier spacing Δf can be one of {15, 30, 60, 120, 240, 480} kHz.

[0108] As mentioned above, the BS needs to send a reference signal in order to measure the downlink channel state in the cellular system. In the case of the 3GPP's Advanced Long Term Evolution (LTE-A) system, the UE can use the CRS or CSI-RS sent by the BS to measure the channel state between the BS and the UE. Various factors may need to be considered to measure the channel state, and the amount of interference in the downlink may be one of the factors. The amount of interference in the downlink may include interference signals generated by antennas belonging to adjacent BSs, thermal noise, etc., and the amount of interference is important when the UE determines the channel state in the downlink. For example, when a BS with a single transmit antenna sends a signal to a UE with a single receive antenna, the UE needs to determine Es / Io based on the reference signal received from the BS by determining the energy per symbol that can be received in the downlink and the amount of interference received simultaneously in the part where the corresponding symbol is received. The determined Ex / Io can be converted into a data transmission rate or a value corresponding thereto, can be sent to the BS in the form of a channel quality indicator (CQI), and can be used when the BS determines the data transmission rate to be used for transmission to the UE.

[0109] In LTE-A systems, the UE feeds back information related to the downlink channel state to the base station, which then uses this information for downlink scheduling. Specifically, the UE measures the reference signal transmitted by the base station in the downlink and feeds back information extracted from the measured reference signal to the base station in a format defined in the LTE / LTE-A standards. As mentioned above, the information fed back by the UE in LTE / LTE-A is called channel state information, and this channel state information can include the following three pieces of information:

[0110] - Rank Indicator (RI): Indicates the number of spatial layers that the UE can receive under the current channel state.

[0111] - Precoding Matrix Indicator (PMI): an indicator associated with the precoding matrix preferred by the UE under the current channel state.

[0112] - Channel Quality Indicator (CQI): Indicates the maximum data rate that the UE can receive under the current channel status.

[0113] The CQI may be replaced by the signal to interference plus noise ratio (SINR), the maximum error correction code rate and modulation scheme, the data per frequency efficiency, etc. (which may be exploited in a manner similar to the maximum data transmission rate).

[0114] RI, PMI, and CQI are interrelated. For example, the precoding matrix supported in LTE / LTE-A can be defined differently for each rank. Therefore, the PMI value X when the RI is 1 and the PMI value X when the RI is 2 can be interpreted as being different. In addition, the UE determines the CQI based on the assumption that the PMI reported by the UE to the BS and X are applied in the BS. That is, when the rank is RI_X and the PMI is PMI_Y, reporting RI_X, PMI_Y, and CQI_Z to the BS can be a report that the corresponding UE is able to perform reception at a data transmission rate corresponding to CQI_Z. As described above, the UE calculates the CQI based on the assumption of the transmission scheme to be performed for the BS, so when the UE actually performs transmission using the corresponding transmission scheme, the UE can obtain optimal performance.

[0115] In LTE / LTE-A, RI, PMI, and CQI as channel state information fed back by the UE can be fed back periodically or aperiodically. In the case where the BS expects to obtain the channel state information of a predetermined UE aperiodically, the BS can configure the UE to perform aperiodic feedback (or aperiodic channel state information report) using an aperiodic feedback indicator (or channel state information request field or channel state information request information) included in the downlink control information (DCI). In addition, if the UE receives an indicator configured for aperiodic feedback in the nth subframe, the UE can perform uplink transmission by including the aperiodic feedback information (or channel state information) in the data transmission in the n+kth subframe. Here, k is a parameter defined in the 3GPP LTE Release 11 standard, which is 4 in frequency division duplex (FDD), and can be defined as shown in Table 6 in the case of time division duplex (TDD).

[0116] [Table 6] k for each subframe number n in TDD UL / DL configuration

[0117]

[0118] In the case where aperiodic feedback is configured, the feedback information (or channel state information) may include RI, PMI, and CQI, and depending on the feedback configuration (or channel state report configuration), RI and PMI may not be fed back.

[0119] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. In addition, although embodiments of the present disclosure are described with reference to LTE or LTE-A systems, embodiments of the present disclosure may be applicable to other communication systems having similar technical backgrounds or using similar channel types. For example, 5G mobile communication technology (5G, New Radio (NR)) developed after LTE-A may be included. In addition, those skilled in the art may modify the embodiments of the present disclosure without departing from the scope of the present disclosure, and the embodiments of the present disclosure may be applied to other communication systems.

[0120] In the following description of the present disclosure, when known functions or configurations incorporated herein may make the subject matter of the present disclosure unclear, detailed descriptions of these functions or configurations will be omitted. The terms described below are defined in consideration of the functions in the present disclosure and may vary depending on the user, the user's intention, or habits. Therefore, the definition of the terms should be based on the content of the entire specification.

[0121] The present disclosure provides a method for efficiently learning and updating the weights of an autoencoder neural network (DNN) when an autoencoder, which is a type of deep neural network (DNN), is used in signal transmission or reception between a UE and a base station. The training method in the present disclosure is called "shadow training."

[0122] Figure 6 6. The configuration of an autoencoder according to an embodiment of the present disclosure is shown. The autoencoder is a neural network (NN) trained in a manner in which the output 602 and the input 601 are the same. According to an embodiment, a signal to be transmitted may be input 601 to the autoencoder NN, and the input signal may be calculated via the trained autoencoder NN and may be output 602. Hereinafter, for the convenience of describing the embodiments of the present disclosure, a downlink or uplink channel matrix is ​​described as input data 601 or output data 602. However, this is merely an example, and the description does not limit the scope of the present disclosure, and any data may be input or output, just like a common neural network.

[0123] The autoencoder NN can be configured to include an input layer, an output layer, and one or more hidden layers, and the autoencoder NN can be defined based on the number of layers, the number of nodes in each layer, and the connection weights between the nodes. A connection weight is a value indicating the relationship between nodes and can be, for example, a real number. The value of each node can be calculated based on the value of another node connected to the corresponding node in the NN and the connection weights associated with the other nodes. In the following, "weight" refers to the connection weight unless otherwise specified.

[0124] According to an embodiment, the autoencoder NN may include a TxNN 603 including an input layer and an RxNN 604 including an output layer. TxNN 603 may include an input layer and at least one hidden layer, and RxNN 604 may include an output layer and at least one hidden layer. The number of hidden layers included in each of RxNN 604 and TxNN 603 may not always be the same, and the number of hidden layers included in each of RxNN 604 and TxNN 603 may be different from each other.

[0125] According to an embodiment, the autoencoder NN may be configured such that the number of nodes in the input layer and the number of nodes in the output layer are the same, and the number of nodes in the hidden layer is less than the number of nodes in the input layer and the number of nodes in the output layer. Therefore, the fact that the autoencoder NN is used can be understood to mean that the input value is encoded in the Tx NN 603 and decoded in the Rx NN 604.

[0126] For ease of description, this description will refer to downlink channel state feedback as an example of signal transmission between a UE and a base station using an autoencoder NN. However, this is merely an example, and the description does not limit the scope of the present disclosure. Obviously, the technical concepts of the present disclosure can be applied to any signal transmission or reception process in a communication system, such as data transmission or control signal transmission.

[0127] Figure 7 FIG. 7 is a diagram illustrating a downlink channel feedback scheme 700 based on an autoencoder according to an embodiment. According to the feedback scheme, a Tx NN 705 of an autoencoder NN learned via deep learning may be arranged in a UE 701, and an Rx NN 708 may be arranged in a BS 702. The UE may perform preprocessing 704 on the estimated downlink channel matrix H 703 to generate a new matrix Tx NN 705 can convert the pre-processed output matrix The codeword vector is converted into a signal in a form that can be transmitted via the CSI transmitter 706 and can be fed back to the BS (CSI report). Feedback can be performed periodically or aperiodically via the PUCCH or PUSCH. The BS can obtain the codeword vector received via the CSI receiver 707 by decoding it using the Rx NN 708. BS can perform post-processing 709 To obtain the downlink channel matrix H. This scheme can use deep learning to efficiently encode the information of the channel matrix, so it can send more accurate channel information using a limited number of feedback bits.

[0128] There are two examples of methods for training an autoencoder NN for a channel feedback scheme. First, an offline training method can be used, which pre-learns the weights of the autoencoder NN under predetermined assumptions and uses fixed weights in real-world situations. Second, an online training method can be used, which utilizes the weights of the autoencoder NN by learning and updating the weights of the autoencoder NN in real time. However, in the case of the offline training method, the performance of the pre-trained autoencoder may deteriorate if the channel environment changes. In addition, the online training method may involve high real-time learning complexity and high weight feedback overhead.

[0129] In the present disclosure, as a method for training an autoencoder NN, a shadow training scheme is provided, which uses an autoencoder NN for training that is different from the autoencoder NN actually used by the UE or BS for signal transmission to perform training, and shares the training results, so that the autoencoder NN can be efficiently trained and updated.

[0130] <First embodiment>

[0131] According to the first embodiment, the UE learns connection weights of the autoencoder NN through shadow training and transmits information associated with the learned weights to the BS, thereby updating the weights of the autoencoder NN.

[0132] Figure 8 A diagram illustrating operations of a UE and a BS according to a first embodiment is shown. Figure 9 A diagram illustrating operations of a UE and a BS according to a first embodiment is shown.

[0133] UE 801 and BS 802 can share a set of autoencoder NNs, which includes a limited number of predefined / trained elements. The set can include one or more autoencoder NNs as elements, and each autoencoder has a structure corresponding to the number of layers and the number of nodes per layer determined based on the physical conditions between the UE and BS. Here, the physical conditions may include operating frequency, UE bandwidth, UE / BS antenna configuration, etc. Based on the assumption of the predetermined conditions, the initial connection weights of each autoencoder NN in the set can be learned in advance.

[0134] After the initial access 901 and 907, each of the UE 801 and the BS 802 may obtain information associated with the physical situation of the other, and may select an element from the set of autoencoder NNs in operations 902 and 908. That is, the UE and the BS may select an autoencoder NN having an appropriate structure from among a plurality of predefined / trained autoencoder NNs based on the obtained information associated with the physical situation after the initial access.

[0135] In operation 903, the UE 801 may arrange the Tx NN 805 of the selected autoencoder NN on the UE side, and in operation 909, the BS 802 may arrange the Rx NN 808 of the selected NN on the BS side. In this case, the BS may arrange a plurality of Rx NNs that are different for each UE existing in the cell on the base station side. The UE 801 and the BS 802 may perform the operation using the Tx NN 805 and the Rx NN 808 arranged therein, respectively. Figure 7 The downlink channel feedback based on the auto-encoder 700 has been referred to. Figure 7 The autoencoder-based downlink channel feedback 700 is described, and thus a detailed description thereof will be omitted here.

[0136] In addition to the Tx NN 805 arranged in the UE, the UE 801 may also prepare the entire autoencoder NN including both the Tx NN and the Rx NN for shadow training. According to an embodiment, the UE may use the downlink channel matrix H 803 (which is estimated using the signal received from the BS) as learning data, and may continuously perform shadow training 810 associated with the entire autoencoder NN in operation 904. The signal received from the BS may include, for example, CRS, CSI-RS, synchronization signal, DMRS, etc., but is not limited thereto. According to another embodiment, the UE 801 may receive a channel matrix estimated using a reference signal sent by the UE to the BS 802 from the BS 802, and may perform shadow training using the estimated channel matrix. The reference signal sent by the UE 801 to the BS 802 may include, for example, SRS and DMRS, but is not limited thereto. That is, UE 801 can use the channel matrix estimated by the UE based on the signal (CRS, CSI-RS, synchronization signal, DMRS, ...) received from BS 802 or the channel matrix estimated by the BS and received from the BS to continuously update the connection weights of the autoencoder NN prepared for shadow training.

[0137] In operation 905, the UE 801 may transmit the weight of the Rx NN (weight report) among the weights of the new entire autoencoder NN (Tx NN and Rx NN) updated through the shadow training 810 to the BS 802. Depending on the embodiment, the weight of the Rx NN may be transmitted to the BS via an uplink channel (e.g., PUCCH or PUSCH).

[0138] According to an embodiment, UE 801 may transmit the weights of the Rx NN to the BS, which are updated via periodic weight reporting. In this case, the period for transmitting the weights of the Rx NN may be set taking into account the amount of time required for UE 801 to perform shadow training. Preferably, this period may be set to be longer than the channel state information reporting period (CSI reporting period).

[0139] According to an embodiment, UE 801 may send the weight of Rx NN to BS 802, which is updated via non-periodic weight reporting. UE 801 may perform weight reporting in response to a non-periodic weight reporting request from BS 802, or may perform weight reporting if the performance of the autoencoder NN performing shadow training meets a predetermined condition. Here, the performance of the autoencoder NN may be defined using an indicator indicating the degree of difference between the input and output of the autoencoder NN. For example, the performance of the autoencoder NN may be defined using a mean square error (MSE) value between the input and output. In addition, the predetermined condition may be, for example, a condition that the performance of the autoencoder NN is greater than or equal to a predetermined reference, i.e., the difference between the input and output is less than a predetermined reference. Here, the predetermined reference may be a value predetermined and input to the UE, or may be a value arbitrarily set by the UE.

[0140] The weight report transmitted by the UE 801 may include a request to update the Rx NN arranged in the BS 802. The BS 802 receiving the weight report may transmit an ACK signal to the UE 801 confirming receipt of the weight report, and may update the weight of the currently arranged Rx NN 808 to the newly received weight in operation 911. Upon receiving the ACK signal, the UE 801 may update the weight of the arranged Tx NN 805 to the newly learned weight in operation 906.

[0141] UE 801 and BS 802 may use Tx NN 805 and Rx NN 808 to perform Figure 7 FIG. 7 is an autoencoder-based downlink channel feedback 700 , wherein the weights of the Tx NN 805 and the Rx NN 808 are updated.

[0142] <Second embodiment>

[0143] According to the second embodiment, the BS learns connection weights of the autoencoder NN through shadow training and transmits information associated with the learned weights to the UE in order to update the weights of the autoencoder NN.

[0144] Figure 10 A diagram illustrating operations of a UE and a BS according to a second embodiment is shown. Figure 11A diagram illustrating operations of a UE and a BS according to a second embodiment is shown.

[0145] UE 1001 and BS 1002 can share a set of autoencoder NNs, which includes a limited number of predefined / trained elements. The set can be a set including one or more autoencoder NNs as elements, and each autoencoder has a structure corresponding to the number of layers and the number of nodes per layer determined based on the physical conditions between the UE and BS. Here, the physical conditions may include operating frequency, UE bandwidth, UE / BS antenna configuration, etc. Based on the assumption of the predetermined conditions, the initial connection weights of each autoencoder NN in the set can be learned in advance.

[0146] After initial access 1101 and 1107, each of the UE 1001 and the BS 1002 may obtain information associated with the physical situation of the other, and may select an element from a set of autoencoder NNs in operations 1102 and 1108. That is, the UE and the BS may select an autoencoder NN having an appropriate structure from among a plurality of predefined / trained autoencoder NNs based on the obtained information associated with the physical situation after the initial access.

[0147] In operation 1109, the UE 1001 may arrange the Tx NN 1005 of the selected autoencoder NN on the UE side, and in operation 1103, the BS 1002 may arrange the Rx NN 1008 of the selected NN on the BS side. In this case, the BS may arrange a plurality of Rx NNs different for each UE existing in the cell on the BS side. The UE 1001 and the BS 1002 may perform the Tx NN 1005 and the Rx NN 1008 arranged therein, respectively. Figure 7 The downlink channel feedback based on the auto-encoder 700 has been referred to. Figure 7 The autoencoder-based downlink channel feedback 700 is described, and thus a detailed description thereof will be omitted here.

[0148] In addition to the Rx NN 1008 deployed in the BS, BS 1002 may prepare the entire autoencoder NN, including both the Tx NN and the Rx NN, for shadow training. According to an embodiment, the UE may estimate the downlink channel matrix H 1003 using signals received from the BS and may transmit it to the BS via autoencoder-based downlink channel feedback 700. In this case, the signals received from the BS may include, for example, but are not limited to, CRS, CSI-RS, synchronization signals, DMRS, etc. In operation 1104, the BS may continuously perform shadow training 1010 associated with the entire autoencoder NN using the received downlink channel matrix H 1009 as learning data. According to another embodiment, BS 1002 may estimate the channel matrix using reference signals received from UE 1001 and may perform shadow training using the estimated channel matrix as learning data. The reference signals transmitted by UE 1001 to BS 1002 may include, for example, SRS, DMRS, etc., but are not limited to these. That is, BS 1002 can continuously update the connection weights of the autoencoder NN prepared for shadow training using the channel matrix estimated by the BS based on the signal (SRS, DMRS, . . . ) received from UE 1001 or the channel matrix estimated by the UE and received from the UE.

[0149] In operation 1105, the BS 1002 may transmit the weight of the Tx NN (weight report) among the weights of the new entire autoencoder NN (Tx NN and Rx NN) updated via the shadow training 1010 to the UE 1001. Depending on the embodiment, the weight of the Tx NN may be transmitted to the UE 1001 via a downlink channel (e.g., PDCCH or PDSCH).

[0150] According to an embodiment, BS 1002 may transmit the Tx NN weights to UE 1001, which are updated via periodic weight reporting. In this case, the period for transmitting the Tx NN weights may be set taking into account the amount of time required for BS 1002 to perform shadow training. Preferably, this period may be set to be longer than the channel state information reporting period (CSI reporting period).

[0151] According to an embodiment, BS 1002 may send the weight of Tx NN to UE 1001, which is updated via non-periodic weight reporting. BS 1002 may perform weight reporting in response to a non-periodic weight reporting request from UE 1001, or may perform weight reporting if the performance of the autoencoder NN performing shadow training meets a predetermined condition. Here, the performance of the autoencoder NN may be defined using an indicator indicating the degree of difference between the input and output of the autoencoder NN. For example, the performance of the autoencoder NN may be defined using a mean square error (MSE) value between the input and output. In addition, the predetermined condition may be, for example, a condition that the performance of the autoencoder NN is greater than or equal to a predetermined reference, i.e., the difference between the input and output is less than a predetermined reference. Here, the predetermined reference may be a value predetermined and input to the BS, or may be a value arbitrarily set by the UE.

[0152] The weight report transmitted by the BS 1002 may include a request to update the Tx NN arranged in the UE 1001. The UE 1001 that receives the weight report may transmit an ACK signal to the BS 1002 to confirm receipt of the weight report, and may update the weight of the currently arranged Tx NN 1005 to the newly received weight in operation 1111. Upon receiving the ACK signal, the BS may update the weight of the arranged Rx NN 1008 to the newly learned weight in operation 1106.

[0153] UE 1001 and BS 1002 may use Tx NN 1005 and Rx NN 1008 to perform Figure 7 An autoencoder based downlink channel feedback 700 is shown, where the weights of the Tx NN 1005 and the Rx NN 1008 are updated.

[0154] <Third embodiment>

[0155] According to the third embodiment, channel feedback can be performed by switching modes between mode 1200 (mode 1) in which the UE uses a scheme of directly feeding back the channel matrix to perform shadow training, mode 1400 (mode 2) based on an autoencoder, and mode 1600 (mode 3) in which the base station uses a scheme of directly feeding back the channel matrix to perform shadow training.

[0156] In the following, reference will be made to Figures 12 to 17 A method of transmitting or receiving a signal based on an autoencoder NN according to a third embodiment is described.

[0157] Figure 12 A diagram illustrating operations of a UE and a BS in Mode 1 according to the third embodiment is shown. Figure 13A diagram illustrating operations of a UE and a BS in Mode 1 according to the third embodiment is shown.

[0158] Specifically, Figure 12 and Figure 13 A scheme in which the UE in mode 1 directly feeds back the channel matrix to perform shadow training is shown.

[0159] UE 1201 and BS 1202 can share a set of autoencoder NNs, which includes a limited number of predefined / trained elements. The set can be a set including one or more autoencoder NNs as elements, and each autoencoder has a structure corresponding to the number of layers and the number of nodes per layer determined based on the physical conditions between the UE and BS. Here, the physical conditions may include operating frequency, UE bandwidth, UE / BS antenna configuration, etc. Based on the assumption of the predetermined conditions, the initial connection weights of each autoencoder NN in the set can be learned in advance.

[0160] After the initial access 1301 and 1307, each of the UE 1201 and the BS 1202 may obtain information associated with the physical situation of the other and may select an element from a set of autoencoder NNs in operations 1302 and 1308. That is, the UE 1201 and the BS 1202 may select an autoencoder NN having an appropriate structure from among a plurality of predefined / trained autoencoder NNs based on the obtained information associated with the physical situation after the initial access.

[0161] UE 1201 may prepare the entire autoencoder NN 1207 including both the Tx NN and the Rx NN of the autoencoder NN selected for shadow training. According to an embodiment, the UE may use the downlink channel matrix H 1203 (which is estimated using the signal received from the BS) as learning data, and may continuously perform shadow training 1207 associated with the entire autoencoder NN in operation 1303. The signal received by the UE from the BS may include, for example, a CRS, a CSI-RS, a synchronization signal, a DMRS, etc., but is not limited thereto. According to another embodiment, UE 1201 may receive a channel matrix estimated using a reference signal transmitted by the UE to BS 1202 from BS 1202, and may perform shadow training using the estimated channel matrix as learning data. The reference signal transmitted by UE 1201 to BS 1202 may include, for example, an SRS and a DMRS, but is not limited thereto. That is, UE 1201 can use the channel matrix estimated by the UE based on the signal (CRS, CSI-RS, synchronization signal, DMRS, ...) received from BS 1202 or the channel matrix estimated by the BS and received from the BS to continuously update the connection weights of the autoencoder NN prepared for shadow training.

[0162] In operation 1304, UE 1201 may transmit the weight of the Rx NN (weight report) among the weights of the new entire autoencoder NN (Tx NN and Rx NN) updated via shadow training 1207 to BS 1202. Depending on the embodiment, the weight of the Rx NN may be transmitted to BS 1202 via an uplink channel (e.g., PUCCH or PUSCH). Depending on the embodiment, the weight report transmitted by UE 1201 may include a request for BS 1202 to switch to Mode 2.

[0163] According to an embodiment, UE 1201 may transmit the weight of the Rx NN updated via periodic weight reporting to BS 1202. In this case, the period for transmitting the weight of the Rx NN may be set in consideration of the amount of time required for UE 1201 to perform shadow training. Preferably, this period may be set to be longer than the channel state information reporting period (CSI reporting period).

[0164] According to an embodiment, UE 1201 may transmit the weights of the Rx NN to BS 1202, which are updated via aperiodic weight reporting. UE 1201 may perform weight reporting in response to an aperiodic weight reporting request from BS 1202, or may perform weight reporting if the performance of the autoencoder NN performing shadow training meets a predetermined condition. Here, the performance of the autoencoder NN may be defined using an indicator indicating the degree of difference between the input and output of the autoencoder NN. For example, the performance of the autoencoder NN may be defined using a mean square error (MSE) value between the input and output. Furthermore, the predetermined condition may be, for example, a condition that the performance of the autoencoder NN is greater than or equal to a predetermined reference, i.e., the difference between the input and output is less than a predetermined reference. Here, the predetermined reference may be a value predetermined and input to the UE, or may be a value arbitrarily set by the UE. That is, if the result of the shadow training determines that the performance of the autoencoder NN is greater than or equal to a predetermined level, UE 1201 may request BS 1202 via the weight reporting to switch to Mode 2, so that signal transmission or reception is performed via the trained autoencoder NN. The conditions used as criteria for the aperiodic request of Mode 1 may be set to be the same as or different from the conditions used in Mode 2 or Mode 3 .

[0165] The weight report transmitted by UE 1201 may include a request for BS 1202 to update the Rx NN. In operation, BS 1202, having received the weight report, may transmit an ACK signal to UE 1201 to confirm receipt of the weight report, and may update the retained Rx NN weights to the newly received weights in operation 1310. Upon receiving the ACK signal, UE 1201 may update the retained Tx NN weights to the newly learned weights in operation 1305.

[0166] The UE 1201 and the BS 1202 may update the weights of the Tx NN and the Rx NN, respectively, and may switch to Mode 2 in operations 1306 and 1311 .

[0167] Figure 14 A diagram illustrating operations of a UE and a BS in Mode 2 according to the third embodiment is shown. Figure 15 A diagram illustrating operations of a UE and a BS in Mode 2 according to the third embodiment is shown.

[0168] Specifically, Figure 14 and Figure 15 A scheme is shown in which the UE performs a shadow test 1410 when performing the autoencoder based channel feedback scheme 700 according to Mode 2.

[0169] UE 1401 may arrange Tx NN 1405 updated according to the operation performed based on Mode 1 or Mode 3 on the UE side, and BS 1402 may arrange Rx NN 1408 updated according to the operation performed based on Mode 3 on the BS side. In this case, the BS may arrange multiple Rx NNs different for each UE existing in the cell on the BS side. The UE and the BS may perform the operation using the Tx NN 1005 and Rx NN 1008 arranged separately. Figure 7 The downlink channel feedback based on the auto-encoder 700 has been referred to. Figure 7 The autoencoder-based downlink channel feedback 700 is described, and a detailed description thereof will be omitted here.

[0170] According to an embodiment, in operation 1501, UE 1401 may use the estimated channel matrix to continuously test the performance of the currently used autoencoder NN. In the present disclosure, this is referred to as "shadow testing" 1410. Subsequently, UE 1401 may request BS 1402 to switch to Mode 1 or Mode 3. This request may be sent to the BS via an uplink channel (e.g., PUCCH or PUSCH).

[0171] According to an embodiment, the UE 1401 may periodically request the BS 1402 to switch to Mode 1 or Mode 3. In this case, the period at which the UE 1401 requests the BS 1402 to switch modes may be randomly determined.

[0172] According to an embodiment, UE 1401 may aperiodically request BS 1402 to switch to Mode 1 or Mode 3. In response to the request from BS 1402, UE 1401 may request switching to Mode 1 or Mode 3, or if the performance of the autoencoder NN measured via shadow testing satisfies a predetermined condition, then in operation 1502, UE 1401 may request switching to Mode 1 or Mode 3. Here, the performance of the autoencoder NN may be defined using an indicator indicating the degree of difference between the input and output of the autoencoder NN. For example, the performance of the autoencoder NN may be defined using a mean squared error (MSE) value between the input and output. Furthermore, the predetermined condition may be, for example, a condition that the performance of the autoencoder NN is less than a predetermined reference, i.e., the difference between the input and output exceeds the predetermined reference. That is, if the results of continuous shadow testing indicate that the performance of the autoencoder NN for signal transmission or reception is below a predetermined level, UE 1401 may request BS 1402 to switch to Mode 1 or Mode 3 for shadow training. The conditions used as criteria for the aperiodic request for Mode 2 may be set to be the same as or different from the conditions used in Mode 1 or Mode 3.

[0173] In operation 1505, the BS 1402 may switch to Mode 1 or Mode 3 after transmitting an ACK signal to the UE confirming reception of the mode switching request 1504. Also, upon receiving the ACK signal, the UE 1401 may switch to Mode 1 or Mode 3 in operation 1503.

[0174] Figure 16 A diagram showing operations of a UE and a BS in Mode 3 according to the third embodiment. Figure 17 A diagram showing operations of a UE and a BS in Mode 3 according to the third embodiment.

[0175] Specifically, Figure 16 and Figure 17 A scheme (Mode 3) is shown in which a BS according to Mode 3 performs shadow training using a scheme of directly feeding back a channel matrix.

[0176] UE 1601 and BS 1602 can share a set of autoencoder NNs, which includes a limited number of predefined / trained elements. The set can be a set including one or more autoencoder NNs as elements, and each autoencoder has a structure corresponding to the number of layers and the number of nodes per layer determined based on the physical conditions between the UE and BS. Here, the physical conditions may include operating frequency, UE bandwidth, UE / BS antenna configuration, etc. Based on the assumption of the predetermined conditions, the initial connection weights of each autoencoder NN in the set can be learned in advance.

[0177] After initial access 1701 and 1707, each of the UE 1601 and the BS 1602 may obtain information associated with the physical situation of the other, and may select an element from a set of autoencoder NNs in operations 1702 and 1708. That is, the UE 1601 and the BS 1602 may select an autoencoder NN having an appropriate structure from among a plurality of predefined / trained autoencoder NNs based on the obtained information associated with the physical situation after the initial access.

[0178] BS 1602 may prepare the entire autoencoder NN 1607, including both the Tx NN and the Rx NN of the autoencoder NN, selected for shadow training. According to an embodiment, the UE may estimate the downlink channel matrix H 1603 using the signal received from the BS and may transmit it to BS 1602. In operation 1703, BS 1602 may continuously perform shadow training 1607 associated with the entire autoencoder NN using the received downlink channel matrix H 1606 as learning data. In this case, the signal received by the UE from the BS may include, for example, a CRS, a CSI-RS, a synchronization signal, a DMRS, etc., but is not limited thereto. According to another embodiment, BS 1602 may estimate the channel matrix using the reference signal received from UE 1601 and may perform shadow training using the estimated channel matrix as learning data. The reference signal transmitted by UE 1601 to BS 1602 may include, for example, an SRS and a DMRS, but is not limited thereto. That is, BS 1602 may continuously update the connection weights of the autoencoder NN prepared for shadow training using a channel matrix estimated by the BS based on a signal (SRS, DMRS, ...) received from UE 1601 or a channel matrix estimated by the UE and received from the UE.

[0179] In operation 1704, BS 1602 may transmit to the UE the weight of the Tx NN (weight report) among the weights of the new entire autoencoder NN (Tx NN and Rx NN) updated via shadow training 1607. According to an embodiment, the weight of the Tx NN may be transmitted to UE 1601 via a downlink channel (e.g., PDCCH or PDSCH). According to an embodiment, the weight report transmitted by BS 1602 may include a request for UE 1601 to switch to Mode 2.

[0180] According to an embodiment, BS 1602 may transmit the Tx NN weights to UE 1601, which are updated via periodic weight reporting. In this case, the period for transmitting the Tx NN weights may be set taking into account the amount of time required for BS 1602 to perform shadow training. Preferably, this period may be set to be longer than the channel state information reporting period (CSI reporting period).

[0181] According to an embodiment, BS 1601 may transmit the weights of the Tx NN to UE 1601, which are updated via aperiodic weight reporting. BS 1602 may perform weight reporting in response to an aperiodic weight reporting request from UE 1601, or may perform weight reporting if the performance of the autoencoder NN performing shadow training meets a predetermined condition. Here, the performance of the autoencoder NN may be defined using an indicator indicating the degree of difference between the input and output of the autoencoder NN. For example, the performance of the autoencoder NN may be defined using a mean squared error (MSE) value between the input and output. Furthermore, the predetermined condition may be, for example, a condition that the performance of the autoencoder NN is greater than or equal to a predetermined reference, i.e., the difference between the input and output is less than a predetermined reference. Here, the predetermined reference may be a value predetermined and input to the BS, or may be a value arbitrarily set by the BS. That is, if the results of shadow training indicate that the performance of the autoencoder NN is greater than or equal to a predetermined level, BS 1602 may request UE 1601 via weight reporting to switch to Mode 2, so that signal transmission or reception is performed via the trained autoencoder NN. The conditions used as criteria for the aperiodic request of Mode 3 may be set to be the same as or different from the conditions used in Mode 1 or Mode 2 .

[0182] The weight report transmitted by BS 1602 may include a request for UE 1601 to update the Tx NN. UE 1601, having received the weight report, may transmit an ACK signal to BS 1602 to confirm receipt of the weight report and may update the weight of the retained Tx NN to the newly received weight in operation 1710. Upon receiving the ACK signal, BS 1602 may update the weight of the retained Rx NN to the newly learned weight in operation 1705.

[0183] The UE 1601 and the BS 1602 may update the weights of the Tx NN and the Rx NN, respectively, and may switch to Mode 2 in operations 1706 and 1711 .

[0184] In order to implement the above-mentioned embodiments of the present disclosure, Figure 18 and Figure 19 The transmitter, receiver, and controller of each of the base station and the user equipment (UE) are shown in FIG. Transmission or reception methods of the base station and the user equipment (UE) are disclosed, and are used to apply the method of transmitting or receiving uplink / downlink control channels and data channels in a communication system according to the embodiments. To this end, the transmitter, receiver, and processor of each of the base station and the user equipment (UE) need to operate according to each embodiment.

[0185] Figure 18 1 is a block diagram illustrating the structure of a BS according to an embodiment. Figure 18As shown, the BS of the present disclosure may include a BS processor 1801, a BS receiver 1802, and a BS transmitter 1803. The BS processor 1801 may control a series of processes so that the BS operates according to the above-mentioned embodiments. For example, the BS processor 1801 may use OFDM signals, RS and data channel resource mapping and their transmission or reception, etc. to control downlink control channel allocation and transmission. In an embodiment, the BS receiver 1802 and the BS transmitter 1803 are generally referred to as a transceiver. The transceiver can perform signal transmission or reception with the UE. The signal may include control information and data. To this end, the transceiver may include an RF transmitter that up-converts and amplifies the frequency of the transmitted signal, an RF receiver that low-noise amplifies the received signal and down-converts the frequency, etc. In addition, the transceiver can output signals received via a wireless channel to the BS processor 1801, and can transmit signals output from the BS processor 1801 via a wireless channel.

[0186] Figure 19 1 shows a block diagram illustrating the structure of a UE according to an embodiment. Figure 19 As shown, the UE of the present disclosure may include a UE processor 1901, a UE receiver 1902, and a UE transmitter 1903. The UE processor 1901 can control a series of processes so that the UE operates according to the above-mentioned embodiments. For example, the UE processor 1901 can control the reception of downlink control channels using OFDM signals, RS, and data channel transmission or reception. In an embodiment, the UE receiver 1902 and the UE transmitter 1903 are generally referred to as a transceiver. The transceiver can perform signal transmission or reception with the base station. The signal may include control information and data. To this end, the transceiver may include an RF transmitter that up-converts and amplifies the frequency of the transmitted signal, an RF receiver that low-noise amplifies the received signal and down-converts the frequency, etc. In addition, the transceiver can output signals received via a wireless channel to the UE processor 1901, and can transmit signals output from the UE processor 1901 via a wireless channel.

[0187] The embodiments of the present disclosure described and illustrated in the specification and drawings have been presented to easily explain the technical content of the present disclosure and to help understand the present disclosure, and are not intended to limit the scope of the present disclosure. That is, for those skilled in the art, other modifications and changes that can be made to the present disclosure based on the technical spirit of the present disclosure will be obvious. In addition, the above-mentioned various embodiments can be used in combination as needed.

[0188] Although the present disclosure has been described with various embodiments, various changes and modifications may occur to those skilled in the art. The present disclosure is intended to encompass such changes and modifications as fall within the scope of the appended claims.

Claims

1. A method performed by a user equipment (UE) in a mobile communication system, the method comprising: Determine a first neural network for sending information about a channel to a base station BS, where the first neural network includes a transmit Tx neural network corresponding to the UE and a receive Rx neural network corresponding to the BS; learning weights of the first neural network in a second neural network of the UE based on a channel matrix estimated based on a signal transmitted on a channel between the UE and the BS; Sending weight information for updating the weights of the Rx neural network to the BS, wherein the weight information is generated based on the learning result; receiving an ACK signal from the BS indicating that the weight information is received by the BS; as well as After receiving the ACK signal from the BS, the weights of the Tx neural network are updated based on the learning results. The weight information is periodically sent to the BS at a period longer than the channel state information CSI reporting period.

2. The method according to claim 1, wherein The first neural network includes at least one hidden layer, and the number of nodes in the hidden layer is less than the number of nodes in the input layer and the number of nodes in the output layer.

3. The method according to claim 1, wherein Determining the first neural network includes: A first neural network is identified based on at least one of an operating frequency, a UE bandwidth, and antenna configurations of the UE and the BS.

4. A method performed by a base station BS in a mobile communication system, the method comprising: Determine a first neural network for receiving information about a channel from a user equipment (UE), the first neural network including a transmit Tx neural network corresponding to the UE and a receive Rx neural network corresponding to the BS; receiving, from the UE, weight information for updating weights of the Rx neural network, the weight information being generated based on a result of learning weights of the first neural network, wherein the weights of the first neural network are learned in a second neural network of the UE based on a channel matrix estimated based on a signal transmitted on a channel between the UE and the BS; Sending an ACK signal to the UE indicating that the weight information is received by the BS; as well as After the ACK signal is sent to the UE, the weights of the Rx neural network are updated based on the weight information. The weight information is periodically received from the UE at a period longer than a channel state information CSI reporting period.

5. The method according to claim 4, wherein The first neural network includes at least one hidden layer, and the number of nodes in the hidden layer is less than the number of nodes in the input layer and the number of nodes in the output layer.

6. The method according to claim 4, wherein: Determining the first neural network includes: The first neural network is determined based on at least one of an operating frequency, a UE bandwidth, and antenna configurations of the UE and the BS.

7. A user equipment (UE) in a mobile communication system, the UE comprising: a transceiver configured to transmit or receive signals; as well as The controller is configured as: Determine a first neural network for sending information about a channel to a base station BS, where the first neural network includes a transmit Tx neural network corresponding to the UE and a receive Rx neural network corresponding to the BS; learning weights of the first neural network in a second neural network of the UE based on a channel matrix estimated based on a signal transmitted on a channel between the UE and the BS; Sending weight information for updating the weights of the Rx neural network to the BS, wherein the weight information is generated based on the learning result; receiving an ACK signal from the BS indicating that the weight information is received by the BS; as well as After receiving the ACK signal from the BS, the weights of the Tx neural network are updated based on the learning results. The weight information is periodically sent to the BS at a period longer than a channel state information CSI reporting period.

8. The UE according to claim 7, in, The first neural network includes at least one hidden layer, the number of nodes in the hidden layer is less than the number of nodes in the input layer and the number of nodes in the output layer, and The first neural network is determined based on at least one of an operating frequency of the UE and the BS, a UE bandwidth, and an antenna configuration.

9. A base station BS in a mobile communication system, the BS comprising: a transceiver configured to transmit or receive signals; as well as The controller is configured as: Determine a first neural network for receiving information about a channel from a user equipment (UE), the first neural network including a transmit Tx neural network corresponding to the UE and a receive Rx neural network corresponding to the BS; receiving, from the UE, weight information for updating weights of the Rx neural network, wherein the weight information is generated based on a result of learning weights of the first neural network, and wherein the weights of the first neural network are learned in a second neural network of the UE based on a channel matrix estimated based on a signal transmitted on a channel between the UE and the base station; Sending an ACK signal to the UE indicating that the weight information is received by the BS; as well as After the ACK signal is sent to the UE, the weights of the Rx neural network are updated based on the weight information. The weight information is periodically received from the UE at a period longer than a channel state information CSI reporting period.

10. The BS according to claim 9, in, The first neural network includes at least one hidden layer, the number of nodes in the hidden layer is less than the number of nodes in the input layer and the number of nodes in the output layer, and The first neural network is determined based on at least one of an operating frequency of the UE and the BS, a UE bandwidth, and an antenna configuration.

Images