Apparatus and method for adaptive channel information feedback in wireless communication system

Abstract

The objective of the present disclosure is to adaptively feed back channel information in a wireless communication system, and an operation method of a user equipment may comprise the steps of: receiving configuration information related to channel state information (CSI) feedback; receiving reference signals on the basis of the configuration information; generating CSI feedback information on the basis of the reference signals; and transmitting the CSI feedback information.

Description

Device and method for adaptive channel information feedback in wireless communication systems

[0001] The following description relates to a wireless communication system, and to a device and method for adaptively feeding back channel information in a wireless communication system.

[0002] Wireless access systems are widely deployed to provide various types of communication services, such as voice and data. Typically, wireless access systems are multiple access systems that support communications with multiple users by sharing available system resources (e.g., bandwidth, transmission power). Examples of multiple access systems include code division multiple access (CDMA), frequency division multiple access (FDMA), time division multiple access (TDMA), orthogonal frequency division multiple access (OFDMA), and single-carrier frequency division multiple access (SC-FDMA).

[0003] In particular, as numerous communication devices demand greater communication capacity, enhanced mobile broadband (eMBB) communication technologies are being proposed, improving upon existing radio access technology (RAT). Furthermore, massive machine type communications (mMTC), which connects numerous devices and objects to provide diverse services anytime and anywhere, as well as communication systems that consider reliability and latency-sensitive services / user equipment (UE), are being proposed. Various technological configurations are being proposed for these purposes.

[0004] The present disclosure can provide a device and method for effectively feeding back channel state information (CSI) in a wireless communication system.

[0005] The present disclosure may provide a device and method for adaptively controlling a feedback rate of CSI in a wireless communication system.

[0006] The present disclosure may provide a device and method for performing CSI feedback using temporal correlation in a wireless communication system.

[0007] The present disclosure can provide a device and method for generating side information necessary to restore information representing a channel and channel information in a wireless communication system.

[0008] The present disclosure may provide a device and method for encoding and decoding channel information based on statistical information about the channel information in a wireless communication system.

[0009] The present disclosure may provide a device and method for generating prior information for channel information in a wireless communication system.

[0010] The present disclosure may provide a device and method for determining Doppler information of a channel in a data-driven manner in a wireless communication system.

[0011] The present disclosure may provide a device and method for providing a framework for feeding back a channel instance expressed in an image format in a wireless communication system.

[0012] The present disclosure may provide a device and method for providing a framework for encoding and decoding channel information measured using a reference signal in a wireless communication system using the temporal correlation of the channel.

[0013] The present disclosure may provide a device and method for providing a framework for feeding back channel information measured using a reference signal in a wireless communication system at a variable transmission rate based on statistical information related to the channel.

[0014] The technical objectives to be achieved in the present disclosure are not limited to those mentioned above, and other technical tasks not mentioned can be considered by a person having ordinary skill in the technical field to which the technical configuration of the present disclosure is applied from the embodiments of the present disclosure described below.

[0015] As one embodiment of the present disclosure, a method of operating a terminal in a wireless communication system may include a step of receiving configuration information related to channel state information (CSI) feedback, a step of receiving reference signals based on the configuration information, a step of generating CSI feedback information based on the reference signals, and a step of transmitting the CSI feedback information. The CSI feedback information may include channel information determined based on the reference signals, information related to a temporal correlation of the channel information, and information for fine tuning of the channel information and information related to the temporal correlation.

[0016] As one embodiment of the present disclosure, a method of operating a base station in a wireless communication system may include a step of transmitting configuration information related to channel state information (CSI) feedback, a step of transmitting reference signals based on the configuration information, a step of receiving CSI feedback information corresponding to the reference signals, and a step of restoring channel information based on the CSI feedback information. The CSI feedback information may include channel information determined based on the reference signals, information related to a temporal correlation of the channel information, and information for fine tuning of the channel information and information related to the temporal correlation.

[0017] As one embodiment of the present disclosure, a terminal in a wireless communication system includes a transceiver and a processor connected to the transceiver, wherein the processor receives configuration information related to channel state information (CSI) feedback, receives reference signals based on the configuration information, generates CSI feedback information based on the reference signals, and controls transmission of the CSI feedback information, wherein the CSI feedback information may include channel information determined based on the reference signals, information related to temporal correlation of the channel information, and information for fine tuning of the channel information and information related to the temporal correlation.

[0018] As one embodiment of the present disclosure, a base station in a wireless communication system includes a transceiver and a processor connected to the transceiver, wherein the processor transmits configuration information related to channel state information (CSI) feedback, transmits reference signals based on the configuration information, receives CSI feedback information corresponding to the reference signals, and controls to restore channel information based on the CSI feedback information, wherein the CSI feedback information may include channel information determined based on the reference signals, information related to a temporal correlation of the channel information, and information for fine tuning of the channel information and information related to the temporal correlation.

[0019] As one embodiment of the present disclosure, a communication device includes at least one processor, and at least one computer memory coupled to the at least one processor and storing instructions that direct operations when executed by the at least one processor, wherein the operations may include: receiving configuration information related to channel state information (CSI) feedback, receiving reference signals based on the configuration information, generating CSI feedback information based on the reference signals, and transmitting the CSI feedback information. The CSI feedback information may include channel information determined based on the reference signals, information related to a temporal correlation of the channel information, and information for fine tuning of the channel information and the information related to the temporal correlation.

[0020] As one embodiment of the present disclosure, a non-transitory computer-readable medium storing at least one instruction includes at least one instruction executable by a processor, wherein the at least one instruction controls a device to receive configuration information related to channel state information (CSI) feedback, receive reference signals based on the configuration information, generate CSI feedback information based on the reference signals, and transmit the CSI feedback information, wherein the CSI feedback information may include channel information determined based on the reference signals, information related to a temporal correlation of the channel information, and information for fine tuning of the channel information and the information related to the temporal correlation.

[0021] The above-described aspects of the present disclosure are only some of the preferred embodiments of the present disclosure, and various embodiments reflecting the technical features of the present disclosure can be derived and understood by a person having ordinary skill in the art based on the detailed description of the present disclosure to be described below.

[0022] The following effects may be achieved by embodiments based on the present disclosure.

[0023] According to the present disclosure, channel information can be effectively fed back.

[0024] The effects that can be obtained from the embodiments of the present disclosure are not limited to the effects mentioned above, and other effects not mentioned can be clearly derived and understood by those skilled in the art to which the technical configuration of the present disclosure is applied, from the description of the embodiments of the present disclosure below. In other words, unintended effects resulting from implementing the configuration described in the present disclosure can also be derived from the embodiments of the present disclosure by those skilled in the art.

[0025] The accompanying drawings are intended to aid understanding of the present disclosure and, together with detailed descriptions, may provide embodiments of the present disclosure. However, the technical features of the present disclosure are not limited to specific drawings, and the features disclosed in each drawing may be combined with each other to form new embodiments. Reference numerals in each drawing may indicate structural elements.

[0026] Figure 1 illustrates an example of a communication system applicable to the present disclosure.

[0027] FIG. 2 illustrates an example of a wireless device applicable to the present disclosure.

[0028] FIG. 3 illustrates a method for processing a transmission signal applicable to the present disclosure.

[0029] Figure 4 illustrates a communication procedure between a terminal and a base station applicable to the present disclosure.

[0030] FIG. 5 illustrates an example of a communication structure that can be provided in a 6G (6th generation) system applicable to the present disclosure.

[0031] Figure 6 illustrates an electromagnetic spectrum applicable to the present disclosure.

[0032] FIG. 7 illustrates a THz wireless communication transceiver applicable to the present disclosure.

[0033] Figure 8 illustrates a THz signal generation method applicable to the present disclosure.

[0034] FIG. 9 illustrates a wireless communication transceiver applicable to the present disclosure.

[0035] Figure 10 illustrates a transmitter structure applicable to the present disclosure.

[0036] Figure 11 illustrates a modulator structure applicable to the present disclosure.

[0037] Figure 12 illustrates the structure of a perceptron included in an artificial neural network applicable to the present disclosure.

[0038] Figure 13 illustrates an artificial neural network structure applicable to the present disclosure.

[0039] Figure 14 illustrates an example of a functional framework for application of artificial intelligence technology applicable to the present disclosure.

[0040] Figure 15 illustrates an example of a procedure for utilizing an artificial intelligence model applicable to the present disclosure.

[0041] Figure 16 illustrates another example of a procedure for utilizing an artificial intelligence model applicable to the present disclosure.

[0042] Figure 17 illustrates another example of a procedure for utilizing an artificial intelligence model applicable to the present disclosure.

[0043] Figure 18 illustrates an AI technology-based communication procedure applicable to the present disclosure.

[0044] Figure 19 illustrates an example of a neural network structure for channel state information (CSI) feedback.

[0045] Figure 20 illustrates an example of the processing of a CSI matrix in a neural network for channel state information feedback.

[0046] Figure 21 illustrates another example of a neural network structure for channel state information feedback.

[0047] Figure 22 illustrates the concept of CSI feedback supporting variable feedback transmission rate.

[0048] Figure 23 illustrates an example of distortion imbalance between channel state information matrix instances.

[0049] Figure 24 illustrates an example of a procedure for channel state information feedback supporting variable transmission rates.

[0050] FIG. 25a illustrates an example of the structure of a transmitting device according to one embodiment of the present disclosure.

[0051] FIG. 25b illustrates an example of the structure of a receiving device according to one embodiment of the present disclosure.

[0052] FIG. 26a illustrates an example of the structure of a transmitting device using Doppler estimation and entropy coding according to one embodiment of the present disclosure.

[0053] FIG. 26b illustrates a more detailed example of the structure of a receiving device using Doppler estimation and entropy coding according to one embodiment of the present disclosure.

[0054] FIG. 26b illustrates a more detailed example of the structure of a receiving device using Doppler estimation and entropy coding according to one embodiment of the present disclosure.

[0055] FIG. 26c illustrates an example of a conceptual structure of a transmitting device and a receiving device according to one embodiment of the present disclosure.

[0056] FIG. 27a illustrates an example of an entropy bottleneck layer for entropy encoding / decoding according to one embodiment of the present disclosure.

[0057] FIG. 27b illustrates an example structure for training a bottleneck layer according to one embodiment of the present disclosure.

[0058] FIG. 28a illustrates an example of an entropy model according to one embodiment of the present disclosure.

[0059] FIG. 28b illustrates an example of an entropy model operating based on a channel latent vector according to one embodiment of the present disclosure.

[0060] FIG. 28c illustrates an example of an entropy model operating based on a Doppler potential vector according to one embodiment of the present disclosure.

[0061] Figure 29 shows an example of the elements of the code space, the quantization results of those elements, and the probability distribution of those elements combined with noise.

[0062] FIG. 30 illustrates an example of a CSI feedback procedure according to one embodiment of the present disclosure.

[0063] Figure 31 illustrates an example of a rate-distortion trade-off in a compression model.

[0064] FIG. 32 illustrates an example of a procedure for transmitting channel state information (CSI) according to one embodiment of the present disclosure.

[0065] FIG. 33 illustrates an example of a procedure for generating CSI feedback information according to one embodiment of the present disclosure.

[0066] FIG. 34 illustrates an example of a procedure for obtaining CSI including fine-tuning according to one embodiment of the present disclosure.

[0067] The following embodiments combine the components and features of the present disclosure in a predetermined form. Each component or feature may be considered optional unless explicitly stated otherwise. Each component or feature may be implemented without being combined with other components or features. Furthermore, some components and / or features may be combined to form embodiments of the present disclosure. The order of operations described in the embodiments of the present disclosure may be changed. Some components or features of one embodiment may be included in another embodiment or may be replaced with corresponding components or features of another embodiment.

[0068] In the description of the drawings, procedures or steps that may obscure the gist of the present disclosure are not described, and procedures or steps that can be understood by a person skilled in the art are also not described.

[0069] Throughout the specification, when a part is said to "comprising" or "including" a component, this does not mean that other components may be included, but rather that other components may be excluded, unless otherwise specifically stated. In addition, terms such as "...part," "...unit," and "module" described in the specification mean a unit that processes at least one function or operation, which may be implemented by hardware, software, or a combination of hardware and software. In addition, the words "a" or "an," "one," "the," and similar related words may be used in the context of describing the present disclosure (especially in the context of the claims below) to include both the singular and the plural, unless otherwise indicated herein or clearly contradicted by context.

[0070] Embodiments of the present disclosure described herein focus on the data transmission and reception relationship between a base station and a mobile station. Here, the base station is understood as a terminal node of a network that directly communicates with the mobile station. Certain operations described herein as being performed by the base station may, in some cases, be performed by an upper node of the base station.

[0071] That is, in a network consisting of multiple network nodes including a base station, various operations performed for communication with a mobile station may be performed by the base station or other network nodes other than the base station. In this case, the term 'base station' may be replaced by terms such as fixed station, Node B, eNB (eNode B), gNB (gNode B), ng-eNB, advanced base station (ABS), or access point.

[0072] Additionally, in the embodiments of the present disclosure, the term terminal may be replaced with terms such as user equipment (UE), mobile station (MS), subscriber station (SS), mobile subscriber station (MSS), mobile terminal, or advanced mobile station (AMS).

[0073] Additionally, a transmitter refers to a fixed and / or mobile node that provides data or voice services, and a receiver refers to a fixed and / or mobile node that receives data or voice services. Therefore, for uplink, a mobile station can be the transmitter, and a base station can be the receiver. Similarly, for downlink, a mobile station can be the receiver, and a base station can be the transmitter.

[0074] Embodiments of the present disclosure may be supported by standard documents disclosed in at least one of wireless access systems, such as IEEE 802.xx system, 3rd Generation Partnership Project (3GPP) system, 3GPP Long Term Evolution (LTE) system, 3GPP 5th generation (5G) NR (New Radio) system and 3GPP2 system, and in particular, embodiments of the present disclosure may be supported by 3GPP TS (technical specification) 38.211, 3GPP TS 38.212, 3GPP TS 38.213, 3GPP TS 38.321 and 3GPP TS 38.331 documents.

[0075] Furthermore, the embodiments of the present disclosure can be applied to other wireless access systems and are not limited to the systems described above. For example, they can be applied to systems implemented after the 3GPP 5G NR system and are not limited to a specific system.

[0076] That is, obvious steps or parts not described in the embodiments of the present disclosure can be explained by referring to the above documents. In addition, all terms disclosed in this document can be explained by the above standard documents.

[0077] Hereinafter, preferred embodiments according to the present disclosure will be described in detail with reference to the accompanying drawings. The detailed description set forth below, together with the accompanying drawings, is intended to illustrate exemplary embodiments of the present disclosure and is not intended to represent the only embodiments in which the technical configurations of the present disclosure may be implemented.

[0078] Additionally, specific terms used in the embodiments of the present disclosure are provided to aid in understanding of the present disclosure, and the use of such specific terms may be changed to other forms without departing from the technical spirit of the present disclosure.

[0079] The following technology can be applied to various wireless access systems such as CDMA (code division multiple access), FDMA (frequency division multiple access), TDMA (time division multiple access), OFDMA (orthogonal frequency division multiple access), and SC-FDMA (single carrier frequency division multiple access).

[0080] For clarity, the following description is based on 3GPP communication systems (e.g., LTE, NR, etc.), but the technical spirit of the present disclosure is not limited thereto. LTE may refer to technology after 3GPP TS 36.xxx Release 8. Specifically, LTE technology after 3GPP TS 36.xxx Release 10 may be referred to as LTE-A, and LTE technology after 3GPP TS 36.xxx Release 13 may be referred to as LTE-A pro. 3GPP NR may refer to technology after TS 38.xxx Release 15. 3GPP 6G may refer to technology after TS Release 17 and / or Release 18. "xxx" refers to a standard document detail number. LTE / NR / 6G may be collectively referred to as a 3GPP system.

[0081] For background information, terms, abbreviations, etc. used in this disclosure, reference may be made to standard documents published prior to this disclosure. For example, reference may be made to standard documents 36.xxx and 38.xxx.

[0082] Communication system applicable to the present disclosure

[0083] Although not limited thereto, the various descriptions, functions, procedures, proposals, methods and / or operational flowcharts of the present disclosure disclosed in this document may be applied to various fields requiring wireless communication / connectivity (e.g., 5G) between devices.

[0084] Hereinafter, more specific examples will be provided with reference to the drawings. In the drawings / descriptions below, the same drawing reference numerals may represent identical or corresponding hardware blocks, software blocks, or functional blocks, unless otherwise described.

[0085] Figure 1 illustrates an example of a communication system applied to the present disclosure.

[0086] Referring to FIG. 1, a communication system (100) applied to the present disclosure includes a wireless device, a base station, and a network. Here, the wireless device refers to a device that performs communication using a wireless access technology (e.g., LTE, LTE-A, LTE-A pro, NR, 5G, 5G-A, 6G) and may be referred to as a communication / wireless / 5G device. Although not limited thereto, the wireless device may include a robot (100a), a vehicle (100b-1, 100b-2), an XR (extended reality) device (100c), a hand-held device (100d), a home appliance (100e), an IoT (Internet of Things) device (100f), and an AI (artificial intelligence) device / server (100g). For example, the vehicle may include a vehicle equipped with a wireless communication function, an autonomous vehicle, a vehicle capable of performing vehicle-to-vehicle communication, etc. Here, the vehicles (100b-1, 100b-2) may include unmanned aerial vehicles (UAVs) (e.g., drones). The XR devices (100c) include augmented reality (AR) / virtual reality (VR) / mixed reality (MR) devices, and may be implemented in the form of head-mounted devices (HMDs), head-up displays (HUDs) installed in vehicles, televisions, smartphones, computers, wearable devices, home appliances, digital signage, vehicles, robots, etc. The portable devices (100d) may include smartphones, smart pads, wearable devices (e.g., smartwatches, smart glasses), computers (e.g., laptops, etc.), etc. The home appliances (100e) may include TVs, refrigerators, washing machines, etc. The IoT devices (100f) may include sensors, smart meters, etc.For example, the base station (120) and the network (130) may also be implemented as wireless devices, and a specific wireless device (120a) may act as a base station / network node to other wireless devices.

[0087] Wireless devices (100a to 100f) can be connected to a network (130) via a base station (120). AI technology can be applied to the wireless devices (100a to 100f), and the wireless devices (100a to 100f) can be connected to an AI server (100g) via a network (130). The network (130) can be configured using a 3G network, a 4G (e.g., LTE) network, a 5G (e.g., NR), or a 6G network. The wireless devices (100a to 100f) can communicate with each other via the base station (120) / network (130), but can also communicate directly (e.g., sidelink communication) without going through the base station (120) / network (130). For example, vehicles (100b-1, 100b-2) can communicate directly (e.g., V2V (vehicle to vehicle) / V2X (vehicle to everything) communication). Additionally, an IoT device (100f) (e.g., a sensor) can communicate directly with another IoT device (e.g., a sensor) or another wireless device (100a to 100f).

[0088] Wireless communication / connection (150a, 150b, 150c) can be established between wireless devices (100a to 100f) / base stations (120), and base stations (120) / base stations (120). Here, the wireless communication / connection can be established through various wireless access technologies such as uplink / downlink communication (150a), sidelink communication (150b) (or D2D communication), and base station-to-base station communication (150c) (e.g., relay, IAB (integrated access backhaul)). Through the wireless communication / connection (150a, 150b, 150c), the wireless device and base station / wireless device, and base stations and base stations can transmit / receive wireless signals to / from each other. For example, the wireless communication / connection (150a, 150b, 150c) can transmit / receive signals through various physical channels. To this end, based on various proposals of the present disclosure, at least some of various configuration information setting processes for transmitting / receiving wireless signals, various signal processing processes (e.g., channel encoding / decoding, modulation / demodulation, resource mapping / demapping, etc.), resource allocation processes, etc. may be performed.

[0089] Devices applicable to the present disclosure

[0090] FIG. 2 illustrates an example of a wireless device applicable to the present disclosure.

[0091] Referring to FIG. 2, the wireless device (200) can transmit and receive wireless signals via various wireless access technologies (e.g., LTE, LTE-A, LTE-A pro, NR, 5G, 5G-A, 6G). The wireless device (200) includes at least one processor (202) and at least one memory (204), and may additionally include at least one transceiver (206) and / or at least one antenna (208).

[0092] The processor (202) controls the memory (204) and / or the transceiver (206), and may be configured to implement the descriptions, functions, procedures, proposals, methods, and / or operational flowcharts disclosed in this document. For example, the processor (202) may process information in the memory (204) to generate first information / signal, and then transmit a wireless signal including the first information / signal via the transceiver (206). In addition, the processor (202) may receive a wireless signal including second information / signal via the transceiver (206), and then store information obtained from signal processing of the second information / signal in the memory (204). The memory (204) may be connected to the processor (202) and may store various information related to the operation of the processor (202). For example, the memory (204) may store software code including instructions for performing some or all of the processes controlled by the processor (202), or for performing the descriptions, functions, procedures, proposals, methods, and / or operational flowcharts disclosed herein. Here, the processor (202) and the memory (204) may be part of a communication modem / circuit / chip designed to implement wireless communication technology. The transceiver (206) may be connected to the processor (202) and may transmit and / or receive wireless signals via at least one antenna (208). The transceiver (206) may include a transmitter and / or a receiver. The transceiver (206) may be used interchangeably with an RF (radio frequency) unit. In the present disclosure, a wireless device may also mean a communication modem / circuit / chip.

[0093] Hereinafter, the hardware elements of the wireless device (200) will be described in more detail. Although not limited thereto, at least one protocol layer may be implemented by at least one processor (202). For example, at least one processor (202) may implement at least one layer (e.g., a functional layer such as physical (PHY), media access control (MAC), radio link control (RLC), packet data convergence protocol (PDCP), radio resource control (RRC), and service data adaptation protocol (SDAP)). At least one processor (202) may generate at least one Protocol Data Unit (PDU) and / or at least one Service Data Unit (SDU) according to the descriptions, functions, procedures, proposals, methods, and / or operation flowcharts disclosed in this document. At least one processor (202) may generate a message, control information, data, or information according to the descriptions, functions, procedures, proposals, methods, and / or operation flowcharts disclosed in this document. At least one processor (202) can generate a signal (e.g., a baseband signal) including a PDU, an SDU, a message, control information, data or information according to the functions, procedures, proposals and / or methods disclosed in this document, and provide the signal to at least one transceiver (206). At least one processor (202) can receive a signal (e.g., a baseband signal) from at least one transceiver (206) and obtain the PDU, SDU, message, control information, data or information according to the descriptions, functions, procedures, proposals, methods and / or operational flowcharts disclosed in this document.

[0094] At least one processor (202) may be referred to as a controller, a microcontroller, a microprocessor, or a microcomputer. The at least one processor (202) may be implemented by hardware, firmware, software, or a combination thereof. For example, at least one application specific integrated circuit (ASIC), at least one digital signal processor (DSP), at least one digital signal processing device (DSPD), at least one programmable logic device (PLD), or at least one field programmable gate array (FPGA) may be included in the at least one processor (202). The descriptions, functions, procedures, proposals, methods, and / or operation flowcharts disclosed in this document may be implemented using firmware or software, and the firmware or software may be implemented to include modules, procedures, functions, etc. The descriptions, functions, procedures, proposals, methods, and / or operation flowcharts disclosed in this document may be included in the at least one processor (202), or may be stored in at least one memory (204) and driven by the at least one processor (202). The descriptions, functions, procedures, suggestions, methods and / or flowcharts disclosed in this document may be implemented using firmware or software in the form of code, instructions and / or sets of instructions.

[0095] At least one memory (204) can be connected to at least one processor (202) and can store various forms of data, signals, messages, information, programs, codes, instructions and / or commands. The at least one memory (204) can be configured as a read only memory (ROM), a random access memory (RAM), an erasable programmable read only memory (EPROM), a flash memory, a hard drive, a register, a cache memory, a computer readable storage medium and / or a combination thereof. The at least one memory (204) can be located internally and / or externally to the at least one processor (202). In addition, the at least one memory (204) can be connected to the at least one processor (202) via various technologies such as a wired or wireless connection.

[0096] At least one transceiver (206) can transmit user data, control information, wireless signals / channels, etc., mentioned in the methods and / or flowcharts of this document to at least one other device. At least one transceiver (206) can receive user data, control information, wireless signals / channels, etc. mentioned in the descriptions, functions, procedures, proposals, methods and / or flowcharts disclosed in this document from at least one other device. For example, at least one transceiver (206) can be connected to at least one processor (202) and can transmit and receive wireless signals. For example, at least one processor (202) can control at least one transceiver (206) to transmit user data, control information, or wireless signals to at least one other device. Furthermore, at least one processor (202) can control at least one transceiver (206) to receive user data, control information, or wireless signals from at least one other device. In addition, at least one transceiver (206) may be connected to at least one antenna (208), and at least one transceiver (206) may be configured to transmit and receive user data, control information, wireless signals / channels, etc. mentioned in the descriptions, functions, procedures, proposals, methods and / or operation flowcharts disclosed in this document through at least one antenna (208). In this document, at least one antenna may be a plurality of physical antennas or a plurality of logical antennas (e.g., antenna ports). At least one transceiver (206) may convert the received wireless signals / channels, etc. from RF band signals to baseband signals in order to process the received user data, control information, wireless signals / channels, etc. using at least one processor (202). At least one transceiver (206) may convert the processed user data, control information, wireless signals / channels, etc. from baseband signals to RF band signals using at least one processor (202).For this purpose, at least one transceiver (206) may include an (analog) oscillator and / or filter.

[0097] The components of the wireless device described with reference to FIG. 2 may be referred to by different terms in terms of functionality. For example, the processor (202) may be referred to as a control unit, the transceiver (206) as a communication unit, and the memory (204) as a storage unit. In some cases, the communication unit may be used to mean at least a portion of the processor (202) and the transceiver (206).

[0098] The structure of the wireless device described with reference to FIG. 2 can be understood as the structure of at least a portion of various devices. For example, the structure of the wireless device illustrated in FIG. 2 can be at least a portion of various devices described with reference to FIG. 1 (e.g., a robot (100a), a vehicle (100b-1, 100b-2), an XR device (100c), a portable device (100d), a home appliance (100e), an IoT device (100f), an AI device / server (100g)). Furthermore, according to various embodiments, in addition to the components illustrated in FIG. 2, the device may further include other components.

[0099] For example, the device may be a portable device such as a smartphone, a smart pad, a wearable device (e.g., a smart watch, smart glasses), or a portable computer (e.g., a laptop, etc.). In this case, the device may further include at least one of a power supply unit that supplies power and includes a wired / wireless charging circuit, a battery, etc., an interface unit that includes at least one port for connection with another device (e.g., an audio input / output port, a video input / output port), and an input / output unit for inputting and outputting image information / signals, audio information / signals, data, and / or information input from a user.

[0100] For example, the device may be a mobile device such as a mobile robot, a vehicle, a train, an aerial vehicle (AV), a ship, etc. In this case, the device may further include at least one of a driving unit including at least one of an engine, a motor, a power train, wheels, brakes, and a steering unit of the device, a power supply unit including a wired / wireless charging circuit, a battery, etc. that supplies power, a sensor unit that senses status information, environmental information, and user information of the device or its surroundings, an autonomous driving unit that performs functions such as path maintenance, speed control, and destination setting, and a position measurement unit that obtains location information of the mobile device through a global positioning system (GPS) and various sensors.

[0101] For example, the device may be an XR device such as an HMD, a head-up display (HUD) installed in a vehicle, a television, a smartphone, a computer, a wearable device, a home appliance, a digital signage, a vehicle, a robot, etc. In this case, the device may further include at least one of a power supply unit that supplies power and includes a wired / wireless charging circuit, a battery, etc., an input / output unit that obtains control information, data, etc. from the outside and outputs the generated XR object, and a sensor unit that senses status information, environmental information, and user information of the device or the surroundings of the device.

[0102] For example, the device may be a robot that can be classified into industrial, medical, household, military, etc. types depending on the purpose or field of use. In this case, the device may further include at least one of a sensor unit that senses status information, environmental information, and user information of the device or its surroundings, and a driving unit that performs various physical actions, such as moving the robot joints.

[0103] For example, the device may be an AI device such as a TV, a projector, a smartphone, a PC, a laptop, a digital broadcasting terminal, a tablet PC, a wearable device, a set-top box (STB), a radio, a washing machine, a refrigerator, digital signage, a robot, a vehicle, etc. In this case, the device may further include at least one of an input unit that acquires various types of data from the outside, an output unit that generates output related to sight, hearing, or touch, a sensor unit that senses status information, environmental information, and user information of the device or its surroundings, and a training unit that trains a model composed of an artificial neural network using learning data.

[0104] The structure of the wireless device illustrated in FIG. 2 may be understood as a part of a RAN node (e.g., a base station, DU, RU, RRH, etc.). That is, the device illustrated in FIG. 2 may be a RAN node. In this case, the device may further include a wired transceiver for front haul and / or back haul communications. However, if the front haul and / or back haul communications are based on wireless communications, at least one transceiver (206) illustrated in FIG. 2 may be used for front haul and / or back haul communications, and a wired transceiver may not be included.

[0105] FIG. 3 illustrates a method for processing a transmission signal applicable to the present disclosure. For example, the transmission signal may be processed by a signal processing circuit. At this time, the signal processing circuit (300) may include a scrambler (310), a modulator (320), a layer mapper (330), a precoder (340), a resource mapper (350), and a signal generator (360). At this time, as an example, the operations / functions of FIG. 3 may be performed in the processor (202) and / or the transceiver (206) of FIG. 2. Furthermore, as an example, the hardware elements of FIG. 3 may be implemented in the processor (202) and / or the transceiver (206) of FIG. 2. As an example, blocks 310 to 360 may be implemented in the processor (202) of FIG. 2. Additionally, blocks 310 to 350 may be implemented in the processor (202) of FIG. 2, and block 360 may be implemented in the transceiver (206) of FIG. 2, and are not limited to the above-described embodiment.

[0106] The codeword can be converted into a wireless signal through the signal processing circuit (300) of FIG. 3. Here, the codeword is an encoded bit sequence of an information block. The information block may include a transport block (e.g., an UL-SCH transport block, a DL-SCH transport block). The wireless signal may be transmitted through various physical channels (e.g., a PUSCH, a PDSCH). Specifically, the codeword can be converted into a bit sequence scrambled by a scrambler (310). The scramble sequence used for scrambling is generated based on an initialization value, and the initialization value may include ID information of the wireless device, etc. The scrambled bit sequence can be modulated into a modulation symbol sequence by a modulator (320). The modulation scheme may include pi / 2-binary phase shift keying (pi / 2-BPSK), m-phase shift keying (m-PSK), m-quadrature amplitude modulation (m-QAM), etc.

[0107] A complex modulation symbol sequence can be mapped to at least one transmission layer by a layer mapper (330). The modulation symbols of each transmission layer can be mapped to corresponding antenna port(s) by a precoder (340). The output z of the precoder (340) can be obtained by multiplying the output y of the layer mapper (330) by an N×M precoding matrix W. Here, N is the number of antenna ports, and M is the number of transmission layers. Here, the precoder (340) can perform precoding after performing transform precoding (e.g., discrete Fourier transform (DFT) transform) on the complex modulation symbols. Additionally, the precoder (340) can perform precoding without performing transform precoding.

[0108] The resource mapper (350) can map modulation symbols of each antenna port to time-frequency resources. The time-frequency resources can include multiple symbols (e.g., CP-OFDMA symbols, DFT-s-OFDMA symbols) in the time domain and multiple subcarriers in the frequency domain. The signal generator (360) generates a wireless signal from the mapped modulation symbols, and the generated wireless signal can be transmitted to another device through each antenna. To this end, the signal generator (360) can include an inverse fast Fourier transform (IFFT) module, a cyclic prefix (CP) inserter, a digital-to-analog converter (DAC), a frequency uplink converter, and the like.

[0109] The signal processing process for a received signal in a wireless device may be configured in reverse order of the signal processing process (310 to 360) of FIG. 3. For example, a wireless device (e.g., 200 of FIG. 2) may receive a wireless signal from the outside through an antenna port / transceiver. The received wireless signal may be converted into a baseband signal through a signal restorer. For this purpose, the signal restorer may include a frequency downlink converter, an analog-to-digital converter (ADC), a CP remover, and a fast Fourier transform (FFT) module. Thereafter, the baseband signal may be restored to a codeword through a resource demapper process, a postcoding process, a demodulation process, and a descrambling process. The codeword may be restored to the original information block through decoding. Therefore, a signal processing circuit (not shown) for a received signal may include a signal restorer, a resource demapper, a postcoder, a demodulator, a descrambler, and a decoder.

[0110] Figure 4 illustrates a communication procedure between a terminal and a base station applicable to the present disclosure. Figure 4 illustrates operations of a terminal (410) and a base station (420) transmitting and / or receiving data and operations performed prior thereto.

[0111] Referring to FIG. 4, in step 401, the terminal (410) and the base station (420) perform synchronization. For example, the terminal (410) performs an initial cell search operation. Specifically, the terminal (410) can detect at least one synchronization signal transmitted from the base station (420) according to a predefined rule. Here, the synchronization signal can include multiple synchronization signals classified according to structure or purpose (e.g., primary synchronization signal, secondary synchronization signal). Through this, the terminal (410) can check the boundary of the frame, subframe, slot, and / or symbol of the base station (420) and obtain information about the base station (420) (e.g., cell identifier).

[0112] In step 403, the terminal (410) obtains system information transmitted from the base station (420). The system information is information related to the properties, characteristics, and / or capabilities of the base station (420) required to access the base station (420) and use the service, and may be classified according to the content (e.g., whether it is essential for access), transmission structure (e.g., channel used, whether provided on-demand), etc., and may be classified into, for example, a master information block (MIB) and a system information block (SIB). If necessary, the terminal (410) may transmit a signal requesting system information before receiving the system information. However, the request and provision of the system information may be performed after the random access procedure described below.

[0113] In step 405, the terminal (410) and the base station (420) perform a random access procedure. The terminal (410) may transmit and / or receive at least one message (e.g., a random access preamble, a random access response (RAR) message, etc.) for the random access procedure based on information related to the random access channel of the base station (420) obtained through system information (e.g., channel position, channel structure, supported preamble structure, etc.). For example, the terminal (410) may transmit a preamble (e.g., MSG1) through the random access channel, receive an RAR message (e.g., MSG2), transmit a message (e.g., MSG3) including information related to the terminal (410) (e.g., identification information) to the base station (420) using scheduling information included in the RAR message, and receive a message (e.g., MSG4) for contention resolution and / or connection establishment. As another example, MSG1 and MSG3 may be sent and received as one message, or MSG2 and MSG4 may be sent and received as one message.

[0114] In step 407, the terminal (410) and the base station (420) perform signaling of control information. Here, the control information may be defined in various layers, such as a layer that controls a connection (e.g., a radio resource control (RRC) layer), a layer that handles mapping between logical channels and transport channels (e.g., a media access control (MAC) layer), and a layer that handles physical channels (e.g., a physical (PHY) layer). For example, the terminal (410) and the base station (420) may perform at least one of signaling for establishing a connection, signaling for determining settings related to communication, and signaling for indicating allocated resources.

[0115] In step 409, the terminal (410) and the base station (420) transmit and / or receive data. In other words, the terminal (410) and the base station (420) can process, transmit, and / or receive data based on the signaling of the control information. For example, when transmitting data, the terminal (410) or the base station (420) can perform at least one of channel encoding, rate matching, scrambling, constellation mapping, layer mapping, waveform modulation, antenna mapping, and resource mapping on the information bits. Conversely, when receiving data, the terminal (410) or the base station (420) can perform at least one of signal extraction from resources, waveform demodulation for each antenna, signal arrangement considering layer mapping, constellation demapping, descrambling, and channel decoding.

[0116] 6G communication systems and core implementation technologies of 6G systems

[0117] The 5G system defines various operating bands within FR1 (frequency range 1), which covers 410 MHz to 7125 MHz, and FR2 (frequency range 2), which covers 24,250 MHz to 71,000 MHz. Various frequencies are being discussed as operating bands for the subsequent 6G system, and the use of higher frequencies than 5G systems is also being considered for wider bandwidth and higher transmission speeds. One such band is the THz (terahertz) frequency band, which covers approximately 100 GHz to 10 THz. The THz frequency band is a band that has both the transparency of radio waves and the straightness of light waves, and communications using the THz frequency band are expected to play a transitional role from existing radio-centered communications to lightwave-based communications.

[0118] 6G systems utilizing the THz frequency band have the following goals: i) very high data rates per device, ii) a very large number of connected devices, iii) global connectivity, iv) very low latency, v) reduced energy consumption of battery-free IoT devices, vi) ultra-reliable connectivity, and vii) connected intelligence with machine learning capabilities. The vision of the 6G system can be divided into four aspects: “intelligent connectivity,” “deep connectivity,” “holographic connectivity,” and “ubiquitous connectivity,” and the 6G system can be designed to satisfy the requirements as shown in [Table 1] below.

[0119] Per device peak data rate1 TbpsE2E latency1 msMaximum spectral efficiency100 bps / HzMobility supportup to 1000 km / hrSatellite integrationFullyAIFullyAutonomous vehicleFullyXRFullyHaptic CommunicationFully

[0120] At this time, the 6G system may have key factors such as enhanced mobile broadband (eMBB), ultra-reliable low latency communications (URLLC), massive machine type communications (mMTC), AI integrated communication, tactile internet, high throughput, high network capacity, high energy efficiency, low backhaul and access network congestion, and enhanced data security. Fig. 5 illustrates an example of a communication structure that can be provided in a 6G system applicable to the present disclosure. Referring to Fig. 5, the 6G system is expected to have simultaneous wireless communication connectivity that is 50 times higher than that of a 5G wireless communication system. URLLC, a key feature of 5G, is expected to become even more crucial in 6G communications by providing end-to-end latency of less than 1 ms. Furthermore, 6G systems will boast significantly higher volumetric spectral efficiency than the commonly used area spectral efficiency. 6G systems can offer extremely long battery life and advanced battery technologies for energy harvesting, potentially eliminating the need for separate charging for mobile devices in 6G systems.As core implementation technologies of the 6G system, technologies such as artificial intelligence (AI), THz (terahertz) communication, optical wireless technology, FSO backhaul network, massive MIMO technology, blockchain, 3D networking, quantum communication, unmanned aerial vehicles, cell-free communication, wireless information and energy transfer (WIET), integration of sensing and communication, integration of access backhaul networks, holographic beamforming, big data analysis, and large intelligent surface (LIS) can be adopted.

[0121] For example, THz communication is a communication that utilizes a spectrum in a frequency band between 0.1 THz and 10 THz with a corresponding wavelength in the range of 0.03 mm to 3 mm as shown in Fig. 6, and can be implemented using circuit elements having a structure as shown in Fig. 7. In addition, optical wireless technology is a technology that generates and modulates THz signals using optical elements, and can be implemented based on devices having structures as shown in Figs. 8, 9, 10, and 11.

[0122] In addition, AI can be implemented based on various models such as neural networks and machine learning (machine models). For example, an AI model of a neural network structure can be based on the structure of a perceptron as shown in Fig. 12. Referring to Fig. 12, an artificial neural network can be composed of multiple perceptrons. According to the structure of the perceptron, when an input vector x={x1, x2, …, xd} is input, each component is multiplied by a weight {W1, W2, …, Wd}, and all the results are added, and then an activation function σ(·) is applied. A large artificial neural network structure can be formed by extending the simplified perceptron structure illustrated in Fig. 12, and the input vector can be applied to perceptrons of different dimensions. When perceptrons are stacked, a neural network having an input layer, a hidden layer, and an output layer as shown in Fig. 13 can be configured.

[0123] AI technology utilizing the structure of the neural network as described above can be operated based on a functional framework as shown in FIG. 14 below. FIG. 14 illustrates an example of a functional framework for application of AI technology applicable to the present disclosure. First, the data collection block (1410) performs data preparation on input data collected from objects (e.g., UE, RAN node, network node, etc.) to generate training data (1411) and / or inference data (1411) including processed input data. The model training block (1420) performs training on an AI model using the training data (1411) and provides information on the trained model to the model inference block (1430). The model inference block (1430) generates an output (1416) by performing inference and / or prediction using the inference data (1411). Additionally, the model inference block (1430) can provide model performance feedback (1414) to the model training block (1420). Here, the output (1416) refers to the inference output of the AI ​​model generated by the model inference block (1430), and the details of the inference output may vary depending on the use case. The actor block (1440) triggers or performs a specified task / action based on the output (1416). The actor block (1440) can trigger a task / action for another object (e.g., at least one UE, at least one RAN node, at least one network node, etc.) or for itself. Any one of the functions illustrated in FIG. 14 described above may be performed by two or more entities including the RAN, the network node, the network operator's OAM, or the UE in collaboration. This may be referred to as a split AI operation.

[0124] FIG. 15 illustrates an example of a procedure for utilizing an AI model applicable to the present disclosure. FIG. 15 illustrates a case where a model training function (e.g., a function of a model training block (1420)) is included in a network node, and a model inference function (e.g., a function of a model inference block (1430)) is included in a RAN node. Referring to FIG. 15 , in step 1, RAN node 1 and RAN node 2 transmit input data (e.g., training data) for training an AI model to the network node. Here, RAN node 1 and RAN node 2 may also transmit data collected from the UE (e.g., measurements of the UE related to RSRP, RSRQ, SINR of the serving cell and neighboring cells, the UE's position, velocity, etc.) to the network node. In step 2, the network node trains the AI ​​model using the received training data. In step 3, the network node distributes / updates the AI ​​model to RAN node 1 and / or RAN node 2. RAN node 1 and / or RAN node 2 may continue model training based on the received AI model. In this procedure, it is assumed that the AI ​​model is deployed / updated only to RAN node 1. In step 4, RAN node 1 receives input data (e.g., inference data) for AI model inference from the UE and RAN node 2. In step 5, RAN node 1 performs inference based on the AI ​​model using the received inference data to generate output data (e.g., a prediction or a decision). In step 6, if applicable, RAN node 1 may send model performance feedback to network nodes. In step 7, RAN node 1, RAN node 2, and UE (or 'RAN node 1 and UE', or 'RAN node 1 and RAN node 2') perform actions based on the output data. For example, in case of load balancing operation, the UE may move from RAN node 1 to RAN node 2.In step 8, RAN node 1 and RAN node 2 transmit feedback information to the network nodes.

[0125] FIG. 16 illustrates another example of a procedure for utilizing an AI model applicable to the present disclosure. FIG. 16 illustrates a case where a model training function (e.g., a function of a model training block (1420)) and a model inference function (e.g., a function of a model inference block (1430)) are included in a RAN node. Referring to FIG. 16, in step 1, a UE and a RAN node 2 transmit input data (e.g., training data) for training an AI model to a RAN node 1. In step 2, RAN node 1 trains an AI model using the received training data. In step 3, RAN node 1 receives input data (e.g., inference data) for AI model-based inference from the UE and RAN node 2. In step 4, RAN node 1 performs AI model-based inference using the received inference data to generate output data (e.g., a prediction or decision). In step 5, RAN node 1, RAN node 2, and UE (or 'RAN node 1 and UE', or 'RAN node 1 and RAN node 2') perform actions based on the output data. For example, in case of load balancing operation, the UE may move from RAN node 1 to RAN node 2. In step 6, RAN node 2 transmits feedback information to RAN node 1.

[0126] FIG. 17 illustrates another example of a procedure for utilizing an AI model applicable to the present disclosure. FIG. 17 illustrates a case where a model training function (e.g., a function of a model training block (1420)) is included in a RAN node, and a model inference function (e.g., a function of a model inference block (1430)) is included in a UE. Referring to FIG. 17, in step 1, a UE transmits input data (e.g., training data) for training an AI model to a RAN node. Here, the RAN node may collect data from various UEs and / or other RAN nodes. In step 2, the RAN node trains an AI model using the received training data. In step 3, the RAN node distributes / updates the AI ​​model to the UE. The UE may continue model training based on the received AI model. In step 4, input data (e.g., inference data) for inference based on the AI ​​model is received from the UE, the RAN node, and / or other UEs. In step 5, the UE generates output data (e.g., predictions or decisions) by performing AI model-based inference using the received inference data. In step 6, if applicable, the UE may transmit model performance feedback to the RAN node. In step 7, the UE and the RAN node perform actions based on the output data. In step 8, the UE transmits feedback information to the RAN node.

[0127] According to the aforementioned framework and procedures, an AI model can be trained and utilized in a wireless communication system. In the aforementioned framework and procedures, various data, such as input data, training data, and inference data, are introduced. The specific content of the aforementioned data may vary depending on the task for which the AI ​​model is utilized. For example, information used in various embodiments of the present disclosure described below may be included in the aforementioned data.

[0128] Figure 18 illustrates an AI technology-based communication procedure applicable to the present disclosure. The detailed procedures illustrated in Figure 18 can be combined with various embodiments of the present disclosure described below. For example, data generated according to various embodiments of the present disclosure can be used for operations (e.g., configuration, training, inference, and / or data transmission / reception) in at least one of the detailed procedures illustrated in Figure 18. As another example, the results of the inference illustrated in Figure 18 can be used to transmit and / or receive data according to various embodiments of the present disclosure.

[0129] Referring to FIG. 18, in step 1801, at least one of a UE (1810), a RAN node (1820), and a network node (1830) performs an initial access procedure. For example, in this step, at least one of an initial cell search operation, a system information acquisition operation, a random access operation, and a registration operation may be performed. In step 1803, at least one of the UE (1810), the RAN node (1820), and the network node (1830) performs a configuration procedure. Through the configuration procedure, parameters, resources, connections, and / or entities necessary for performing subsequent procedures in layers between the UE (1810) and the RAN node (1820) and / or in at least one layer between the UE (1810) and the network node (1830) may be determined and / or created. In this case, the configuration procedure may be performed based on information, status, and / or characteristics of an AI model used for subsequent training and inference.

[0130] At step 1805, at least one of the UE (1810), the RAN node (1820), and the network node (1830) performs a model training procedure. At least one of the UE (1810), the RAN node (1820), and the network node (1830) may collect training data and perform learning using the training data. For example, the model training procedure may be performed as described with reference to FIG. 15, FIG. 16, or FIG. 17. If an offline-trained model is used, this step may be omitted.

[0131] At step 1807, at least one of the UE (1810), the RAN node (1820), and the network node (1830) performs a task using the trained model. That is, the task may be performed based on the results of inference and / or prediction using the trained model. For example, the task may be a procedure belonging to a communication protocol, and may be a preparatory operation for subsequent data transmission and / or reception, or may be related to data transmission and / or reception, or may be related to data processing (e.g., encoding, decoding, etc.).

[0132] At step 1809, at least one of the UE (1810), the RAN node (1820), and the network node (1830) transmits and / or receives data. At this time, the result of the task performed at step 1807 may be used. In some cases, the task performed at step 1807 may include transmitting and / or receiving data, in which case this step may be omitted as it is part of step 1807.

[0133] Specific embodiments of the present disclosure

[0134] The present disclosure relates to a technique for adaptive channel information feedback in a wireless communication system. Specifically, the present disclosure proposes a technique for generating and transmitting CSI feedback that reflects temporal correlation while supporting variable rates in a neural network-based CSI feedback framework.

[0135] In this disclosure, lowercase or uppercase italic letters represent scalars. Lowercase boldface letters and uppercase boldface letters represent vectors and matrices, respectively. Calligraphic letters represent sets. For example, , , and stands for scalars, vectors, matrices and sets. represents the set of complex numbers, represents an m×n dimensional complex space.

[0136] In this disclosure, an artificial neural network that compresses and reconstructs CSI based on deep learning (DL) is referred to as a "CSI network." Recently, various advancements have been made in the architecture of CSI networks.

[0137] Figure 19 illustrates an example neural network structure for CSI feedback. Figure 19 illustrates CsiNet, which is an example of a CSI network structure. Referring to Figure 19, the CSI network can be viewed as consisting of a CSI encoder (1910) and a CSI decoder (1920). For example, in the downlink, where data transmission is performed from a base station to a UE, the base station may operate as a transmitter, and the UE may operate as a receiver. In the downlink, the CSI encoder may be operated by the UE, which is a receiver, and the CSI decoder may be operated by the base station, which is a transmitter. In the present disclosure, for convenience of explanation, the case of downlink communication is assumed, but the various embodiments described below are not limited to the downlink and can also be applied to other links, such as uplink and sidelink.

[0138] A CSI encoder included in a UE can compress information about a channel state. The compressed information, which is the output of the CSI encoder, is transmitted to a base station via uplink feedback. The base station inputs the received compressed information into a CSI decoder, and the CSI decoder can restore information about the channel state of the UE. In the present disclosure, for convenience of description, the compressed information, which is the output of the CSI encoder and the input of the CSI decoder, may be referred to as a CSI feedback signal, CSI feedback information, or other terms having equivalent technical meanings. In the present disclosure, the CSI feedback signal may have the form of a bit stream. Here, the bit stream means a sequence of binary digits or bits of 0 or 1, rather than a vector of floating point numbers.

[0139] The temporal correlation of a time-varying wireless channel is a first-order Markov process, which can be modeled as shown in [Mathematical Equation 1] below.

[0140]

[0141] In [Equation 1], is the time A channel instance representing channel information in is a coefficient between 0 and 1 representing the temporal correlation between two consecutive channels, refers to a term representing randomness. For example, can be a zero-mean and unit-variance complex Gaussian matrix. As an example, the channel matrix in the spatial-frequency domain can be considered. The CSI matrix in the spatial-frequency domain can be expressed as follows.

[0142] In the present disclosure, the number of transmitting antennas of the base station is The number of receiving antennas of the UE is assumed to be 1. However, the various embodiments described below are not limited to a single receiving antenna, and can be extended to a multi-antenna case as well. In addition, in the following description, An OFDM system using orthogonal subcarriers is considered.

[0143] UE The signal received through the th subcarrier can be expressed as shown in [Mathematical Formula 2] below.

[0144]

[0145] In [Equation 2], is the instantaneous channel vector in the frequency domain, is a precoding vector, is a data symbol transmitted in the downlink, is AWGN (additive white Gaussian noise), is the subcarrier index, means the number of subcarriers.

[0146] The channel vector for the th subcarrier is is estimated by the UE and can be fed back to the base station. Considering all subcarriers in total, Only when the CSI matrix, which can be expressed as , is properly fed back from the UE to the base station, can the base station correctly determine the precoding vectors. Here, can be understood as a CSI matrix in the spatial-frequency domain.

[0147] As an example, a channel matrix in the angular-delay domain, rather than a channel matrix in the space-frequency domain, can be used. The CSI matrix in the space-frequency domain is can be processed as shown in Fig. 20 below. Fig. 20 illustrates an example of a CSI matrix processing process in a neural network for channel state information feedback. Fig. 20 represents the matrix in a reversed order compared to the general way of representing the matrix. Referring to Fig. 20, preprocessing can be performed in which a two-dimensional (2D) discrete Fourier transform (DFT) and truncation along the delay axis in the angular-delay domain, and separation into real and imaginary parts are sequentially performed. Through this, a channel matrix in the angular-delay domain can be obtained.

[0148] That is, in order to utilize the CSI network, preprocessing including the following three steps can be performed, as shown in Fig. 20.

[0149] (1) 2D-DFT

[0150] CSI matrix in the angular-delay domain is the CSI matrix in the spatial-frequency domain can be obtained from. The relation is Here, and are two DFT matrices.

[0151] (2) Truncation with respect to delay-axis

[0152] Since the time delay between multipath arrivals lies within a limited period, the time delays for all subcarriers fall within a specific period. Therefore, the CSI matrix in each delay domain is only the first It has large values ​​only in the rows and close to 0 in the rest. Therefore, the CSI matrix in each delay domain The beginning of If we take only the rows of the dog, is obtained.

[0153] (3) Split into real and imaginary parts

[0154] truncated CSI matrix Each element of the matrix is ​​composed of complex numbers, but general neural networks have difficulty handling complex numbers. Therefore, for the convenience of processing in neural networks, two matrices are created by dividing the real and imaginary parts of each element, and the two matrices are stacked in the third dimension. A tensor having the size of can be constructed.

[0155]

[0156] The above-described preprocessing process and the types of channel instances illustrated are only examples, and various embodiments of the present disclosure are not limited thereto. In the description of the above-described preprocessing process and FIG. 20, depending on which preprocessing process was performed or in which domain it is expressed, the channel matrix , , It is expressed as a distinction like this, but in the explanation below, time Channel instances representing channel information in are domain and preprocessed regardless of whether they are can be expressed as

[0157] Figure 21 illustrates another example of a neural network structure for channel state information feedback. Figure 21 illustrates the structure of CsiNet-LSTM (long short term memory). The neural network illustrated in Figure 21 introduces an LSTM structure on the decoder side (2110). LSTM is a type of recurrent neural network (RNN) that can be incorporated into the decoder structure to utilize the temporal correlation of the wireless channel for more effective CSI reconstruction. The role of the LSTM module (2112) is to utilize information extracted from the CSI of the previous time for the reconstruction of the CSI of the current time. By utilizing LSTM not only on the decoder side but also on the encoder side, it is possible to utilize the temporal correlation of the channel for compressing time-varying channels. In other words, the temporal correlation of the time-varying channel can be utilized for CSI compression and reconstruction.

[0158] Considering temporal correlation, time-varying channels can be treated like sequence data. For example, sequence data can be likened to video. However, conventional CSI network structures that consider temporal correlation are trained and define the loss function as the mean square error (MSE) from a distortion perspective, rather than being sufficiently studied or optimized from a rate-distortion performance perspective. Conventional CSI network structures that consider temporal correlation suffer from the problem of fixed code rates or code lengths.

[0159] Meanwhile, to address the problem that most existing deep learning-based CSI feedback techniques have fixed code lengths and support variable feedback rates, a framework utilizing an accumulable CSI feedback signal, as illustrated in Figure 22, can be applied. Figure 22 illustrates the concept of CSI feedback supporting a variable feedback rate.

[0160] Referring to FIG. 22, different CSI feedback bitstreams can be combined before being input to the decoder neural network (2220) of the CSI network. Accordingly, the input dimension of the decoder neural network (2220) can be maintained, the structure of the decoder neural network (2220) can be maintained as is, and further, the model parameter set of the decoder neural network (2220) can also be maintained as is.

[0161] Regardless of the number of CSI feedback bit strings that are combined before being input to the decoder neural network (2220), the same decoder neural network (2220) model can always be used. The number of feedback bits increases in proportion to the number of CSI feedback bit strings that are added and input to the decoder neural network (2220). For example, if the length of the CSI feedback bit string that can be input to the decoder neural network (2220) alone is 256 bits, the number of feedback bits will increase to 512, 768, and 1024 as the number of CSI feedback bit strings becomes 2, 3, and 4, respectively.

[0162] Meanwhile, even if the same decoder neural network (2220) model is used, the CSI reconstruction performance may improve as the number of CSI feedback bit-strings input to the decoder neural network (2220) increases. Fig. 21 metaphorically expresses the improvement in CSI reconstruction performance as the number of CSI feedback bit-strings increases through the change in resolution of the reconstructed Lenna image. Specifically, the second image (2292) reconstructed based on two CSI feedback bit-strings (2201, 2202) has a higher resolution than the first image (2291) reconstructed based on one CSI feedback bit-string (2201). Similarly, it is confirmed that the resolution gradually increases in the order of the third image (2293) reconstructed based on three CSI feedback bit-strings (2201 to 2203) and the fourth image (2294) reconstructed based on four CSI feedback bit-strings (2201 to 2204).

[0163] However, in Fig. 22, what is expressed as an image of Lena is purely a metaphor, and information such as the CSI matrix that can actually be restored from the CSI network is not recognized by the human eye like an image. Expressing the increase in CSI restoration performance as an increase in image resolution is also merely a metaphor, and should not be understood as meaning that the CSI restoration performance improves in terms of image resolution. This is because, due to the nature of deep learning, it is difficult to precisely explain what the different CSI feedback bit strings input and added to the decoder neural network (2220) mean and what role they play as signals.

[0164] In FIG. 22, a plurality of CSI feedback bit streams (2201 to 2204) having different roles are combined before being input to the decoder neural network (2220). The first CSI feedback bit stream (2201) may be a signal capable of CSI restoration even when input to the decoder neural network (2220) alone. The second CSI feedback bit stream (2202) may be a signal capable of CSI restoration only when input to the decoder neural network (2220) together with the first CSI feedback bit stream (2201). In this way, the CSI feedback bit streams generated by the encoder neural network (2210) may be classified into a CSI feedback bit stream that enables CSI restoration even when input to the decoder neural network (2220) alone and a CSI feedback bit stream that enables CSI restoration only when combined. At this time, the former CSI feedback bit string may be referred to as an 'independent CSI bit string', and the latter CSI feedback bit string may be referred to as a 'dependent CSI bit string'.

[0165] In the expressions intended to combine bit strings included in a CSI feedback signal, such as “added before being input to a decoder neural network,” “added and input to a decoder neural network,” or “added and input to a decoder neural network,” used in the present disclosure, the operation of “added” can be understood as not only summation, but also one of a weighted sum, a weighted average, or various numerical processing that can be derived therefrom.

[0166] As described above, when neural network structures and signaling procedures that can operate different feedback transmission rates through the same neural network model, different from the general CSI network, are applied, adaptive feedback transmission rate operation according to the downlink channel environment and uplink resource status becomes possible.

[0167] There are several studies to support variable transmission rates. Figure 23 illustrates an example of distortion imbalance between channel state information matrix instances. Referring to Figure 23, it can be seen that even when using the same number of CSI feedback bits, the distortion between the reconstructed CSI matrices and the original CSI matrices can vary depending on the CSI pattern. Noting this distortion imbalance, a variable transmission rate CSI feedback framework, as shown in Figure 24, has been proposed. Figure 24 illustrates an example of a procedure for channel state information feedback supporting a variable transmission rate. Referring to Figure 24, an estimator for predicting distortion is configured as a neural network, an appropriate codeword length is selected based on the distortion values ​​predicted by the estimator, and the channel can be compressed / decompressed using an encoder and decoder with the selected codeword length. However, the aforementioned techniques do not consider the temporal correlation of the channel.

[0168] The present disclosure proposes a CSI feedback framework that considers the temporal correlation of time-varying channels and supports variable feedback transmission rates from a rate-distortion performance perspective. In other words, the present disclosure describes various embodiments that support variable feedback transmission rates based on the temporal correlation of time-varying channels.

[0169] FIG. 25A illustrates an example of the structure of a transmitter according to one embodiment of the present disclosure. FIG. 25A illustrates the functional structure of the transmitter. The transmitter can perform functions such as those shown in FIG. 25A based on components such as those shown in FIG. 2.

[0170] Referring to FIG. 25A, the transmitting device includes a CSI encoder (2510), a variable transmission rate and temporal correlation processing unit (2520), and a transmitter (2530). The CSI encoder (2510) and the variable transmission rate and temporal correlation processing unit (2520) may be processors designed to implement the functions described below, or may be processors executing instructions or programs for the functions described below.

[0171] The CSI encoder (2510) encodes channel information. The CSI encoder (2510) can compress channel information generated based on a reference signal. According to one embodiment, the CSI encoder (2510) may have a neural network structure including multiple layers. That is, the CSI encoder (2510) can generate a latent vector representing a channel instance using the encoder neural network.

[0172] The variable transmission rate and temporal correlation processing unit (2520) generates feedback information including channel information by converting encoded channel information into a bit string. Specifically, the variable transmission rate and temporal correlation processing unit (2520) can estimate the temporal correlation of the channel and convert a latent vector representing a channel instance into a bit string based on the temporal correlation. In addition, the variable transmission rate and temporal correlation processing unit (2520) can support a variable transmission rate by adaptively adjusting the size of the bit string representing the channel information. To reflect the temporal correlation, the variable transmission rate and temporal correlation processing unit (2520) can estimate the Doppler characteristic of the channel based on channel instances determined at adjacent time points and encode a latent vector representing the channel instance based on the Doppler characteristic. At this time, since the feedback information is generated based on the temporal correlation, the receiving device requires information related to the temporal correlation to restore the channel information. Accordingly, the feedback information may further include information related to the temporal correlation in addition to the channel information. According to one embodiment, the variable transmission rate and temporal correlation processing unit (2520) may perform entropy encoding using an entropy model based on the Doppler characteristics of the channel.

[0173] The transmitter (2530) transmits feedback information. The transmitter (2530) may perform physical layer processing, such as modulation and resource mapping, on the feedback information and then transmit it via a wireless channel. Here, the feedback information may include channel information and information related to the temporal correlation of the channel.

[0174] FIG. 25b illustrates an example of the structure of a receiving device according to one embodiment of the present disclosure. FIG. 25b illustrates the functional structure of the receiving device. The receiving device can perform functions such as those shown in FIG. 25b based on components such as those shown in FIG. 2.

[0175] Referring to FIG. 25b, the receiving device includes a receiver (2560), a variable transmission rate and temporal correlation processing unit (2570), and a CSI decoder (2580). The variable transmission rate and temporal correlation processing unit (2570) and the CSI decoder (2580) may be processors designed to implement the functions described below, or may be processors executing instructions or programs for the functions described below.

[0176] The receiver (2560) receives feedback information. The transmitter (2560) can obtain feedback information by performing physical layer processing, such as resource demapping and demodulation, on a signal received via a wireless channel. Here, the feedback information may include channel information and information related to the temporal correlation of the channel.

[0177] The variable rate and temporal correlation processing unit (2570) can determine a latent vector representing a channel instance based on feedback information. Specifically, the variable rate and temporal correlation processing unit (2570) can verify the temporal correlation of the channel based on the feedback information, and obtain a latent vector representing the channel instance from a bit string representing channel information based on the temporal correlation. To consider the temporal correlation, the variable rate and temporal correlation processing unit (2570) can verify a Doppler characteristic based on the feedback information, and decode the bit string representing the channel information based on at least one of the Doppler characteristic and a previously restored channel instance. According to one embodiment, the variable rate and temporal correlation processing unit (2570) can perform entropy decoding using an entropy model based on the Doppler characteristic of the channel.

[0178] The CSI decoder (2580) recovers channel information by decoding a latent vector representing a channel instance. In one embodiment, the CSI decoder (2580) may have a neural network structure including multiple layers.

[0179] FIG. 26a illustrates an example of the structure of a transmitting device using Doppler estimation and entropy coding according to one embodiment of the present disclosure. FIG. 26a is an example of the detailed structure of FIG. 25a.

[0180] Referring to FIG. 26A, the transmitting device includes a CSI encoder (2510), a variable transmission rate and temporal correlation processing unit (2520), and a transmitter (2530). Here, the variable transmission rate and temporal correlation processing unit (2520) includes a first quantization and entropy encoder (2602), a first entropy decoding and dequantizer (2604), a CSI decoder (2606), a Doppler estimator (2608), a Doppler encoder (2610), a second quantization and entropy encoder (2612), a second entropy decoding and dequantizer (2614), a Doppler decoder (2616), a context generator (2618), a first entropy model (2620), and a second entropy model (2622).

[0181] Channel information is measured based on a reference signal and provided to a CSI encoder (2510) and a Doppler estimator (2608). The channel information may include channel instances expressed in an image format in the angle-delay domain or the space-frequency domain. The channel information input to the CSI encoder (2510) is intended to be transmitted in the current feedback occasion, and the channel information input to the Doppler estimator (2608) is intended to determine the temporal correlation for the channel information to be transmitted in the current feedback occasion.

[0182] The first quantization and entropy encoder (2602) performs quantization and entropy encoding on a latent vector (hereinafter referred to as a "channel latent vector") representing a channel instance generated by the CSI encoder (2510). Accordingly, the first quantization and entropy encoder (2602) generates a bitstream corresponding to the channel instance.

[0183] The first entropy decoding and dequantization unit (2604) performs entropy decoding and dequantization on the bitstream generated by the first quantization and entropy encoder (2602), thereby restoring the channel latent vector. The CSI decoder (2606) restores channel information (e.g., channel instances) based on the channel latent vector. To generate information related to temporal correlation, channel instances from previous time points may be used. Accordingly, the first entropy decoding and dequantization unit (2604) and the CSI decoder (2606) restore the channel information. Although not illustrated in FIG. 26A, a component that adds noise corresponding to channel noise may be further included between the first quantization and entropy encoder (2602) and the first entropy decoding and dequantization unit (2604).

[0184] The Doppler estimator (2608) generates Doppler information by estimating the Doppler characteristics of the channel based on the current channel information and the previous channel information. Doppler characteristics are related to temporal correlation. The current channel information includes the channel instance to be fed back. The previous channel information, as a previously fed back channel instance, can be obtained from the CSI decoder (2606).

[0185] The Doppler encoder (2610) encodes Doppler information to generate a latent vector (hereinafter referred to as a "Doppler latent vector") representing the Doppler information. The second quantization and entropy encoder (2612) performs quantization and entropy coding on the Doppler latent vector to generate a bit string representing temporal correlation. The bit string representing the temporal correlation is provided to the transmitter (2530) to be included in the feedback information.

[0186] The second entropy decoding and dequantization unit (2614) performs entropy decoding and dequantization on a bit string expressing temporal correlation, thereby restoring a Doppler latent vector. The Doppler decoder (2616) restores Doppler information based on the Doppler latent vector. Although not illustrated in FIG. 26a, a component that adds noise corresponding to channel noise may be further included between the second quantization and entropy encoder (2612) and the second entropy decoding and dequantization unit (2614).

[0187] The context generator (2618) generates context information for the first entropy model (2620) based on the Doppler information recovered by the Doppler decoder (2616). In addition, the context information generated by the context generator (2618) may be provided to the CSI encoder (2510) and the CSI decoder (2606). For example, the context information may include a channel instance predicted from a previous channel instance based on the Doppler information. The context information may be generated in at least one of various domains (e.g., each-delay domain, feature domain, etc.). The context generator (2618) may have a neural network structure including a plurality of layers.

[0188] The first entropy model (2620) generates an entropy model for the entropy encoding operation of the first quantization and entropy encoder (2602) and the entropy decoding operation of the first entropy decoding and dequantizer (2604). The first entropy model (2620) generates at least one prior (e.g., a temporal prior, a hierarchical prior, etc.) based on context information and a channel latent vector, and generates statistical parameters (e.g., a mean, a standard deviation, etc.) for the generated at least one prior. In addition, the first entropy model (2620) can provide information determined in the process of generating at least one prior (e.g., a result of encoding and quantizing a hierarchical prior for a channel latent vector) to the transmitter (2530), thereby providing the information to the receiving device. Through this, at least one prior required for entropy decoding can be determined in the receiving device.

[0189] The second entropy model (2622) generates an entropy model for the entropy encoding operation of the second quantization and entropy encoder (2612) and the entropy decoding operation of the second entropy decoding and dequantizer (2614). The second entropy model (2622) generates at least one prior (e.g., a hierarchical prior, etc.) based on the Doppler latent vector, and generates statistical parameters (e.g., a mean, a standard deviation, etc.) for the generated at least one prior. In addition, the second entropy model (2622) can transmit information determined in the process of generating at least one prior (e.g., a result of encoding and quantizing the hierarchical prior for the Doppler latent vector) to the transmitter (2530) and thereby transmit the information to the receiving device. Through this, at least one prior required for entropy decoding can be determined in the receiving device.

[0190] FIG. 26b illustrates a more detailed example of the structure of a receiving device utilizing Doppler estimation and entropy coding according to one embodiment of the present disclosure. FIG. 26b is an example of the detailed structure of FIG. 25b.

[0191] Referring to FIG. 26b, the receiving device includes a receiver (2560), a variable transmission rate and temporal correlation processing unit (2570), and a CSI decoder (2580). Here, the variable transmission rate and temporal correlation processing unit (2570) includes a first entropy decoding and dequantization unit (2662), a second entropy decoding and dequantization unit (2664), a Doppler decoder (2666), a context generator (2668), a first entropy model (2670), and a second entropy model (2672).

[0192] The receiver (2560) receives feedback information. Then, the receiver (2560) provides channel information included in the feedback information to a first entropy decoding and dequantization unit (2662) and a first entropy model (2670), and provides information related to temporal correlation included in the feedback information to a second entropy decoding and dequantization unit (2664) and a second entropy model (2672).

[0193] The first entropy decoding and dequantization unit (2662) performs entropy decoding and dequantization on a bit string representing channel information included in the feedback information, thereby restoring a channel latent vector. The channel latent vector is provided to the CSI decoder (2580) and used to restore channel information (e.g., channel instances).

[0194] The second entropy decoding and dequantization unit (2664) performs entropy decoding and dequantization on a bit string expressing temporal correlation, thereby restoring the Doppler latent vector. The Doppler decoder (2666) restores Doppler information based on the Doppler latent vector.

[0195] The context generator (2668) generates context information for the first entropy model (2670) based on the Doppler information restored by the Doppler decoder (2666). In addition, the context information generated by the context generator (2668) may be provided to the CSI decoder (2580). The context generator (2668) may have a neural network structure including multiple layers.

[0196] The first entropy model (2670) generates an entropy model for the entropy decoding operation of the first entropy decoding and dequantizer (2662). The first entropy model (2670) generates at least one prior (e.g., a temporal prior, a hierarchical prior, etc.) based on context information and a channel latent vector, and generates statistical parameters (e.g., a mean, a standard deviation, etc.) for the generated at least one prior. To this end, the first entropy model (2670) can perform entropy decoding and hierarchical prior decoding on information received from a transmitting device.

[0197] The second entropy model (2672) generates an entropy model for the entropy decoding operation of the second entropy decoding and dequantization unit (2664). The second entropy model (2672) generates at least one prior (e.g., a hierarchical prior, etc.) based on the Doppler latent vector, and generates statistical parameters (e.g., a mean, a standard deviation, etc.) for the generated at least one prior.

[0198] According to the structure of the transmitting device illustrated in Fig. 26a and the structure of the receiving device illustrated in Fig. 26c, there are functional blocks that perform similar functions in the transmitting device and the receiving device. If the functional blocks that perform similar functions are expressed as a single block, the transmitting device and the receiving device can be expressed as in Fig. 26c. Fig. 26c illustrates an example of the conceptual structure of the transmitting device and the receiving device according to one embodiment of the present disclosure. All blocks illustrated in Fig. 26 can be understood as a transmitting device, and the portion included in the dotted box can be understood as a receiving device.

[0199] Referring to Figure 26c, the encoder includes a Doppler estimator. The Doppler information is the output of the Doppler estimator. is a Doppler latent vector through a Doppler encoder neural network. is encoded as a Doppler potential vector can be converted into a bit string through an entropy bottleneck layer. The quantized Doppler potential vector is the reconstructed Doppler information through the Doppler decoder neural network. can be decoded. The Doppler-related information and the Doppler potential vector expressed in the present disclosure are intermediate information or signals obtained in a data-driven manner from the channel information of the current time and the channel information of the previous time, and it is not easy to clearly identify or accurately explain the physical meaning of the signal due to the characteristics of neural networks and deep learning. However, since the above-mentioned information or signals include information related to the Doppler shift, the present disclosure refers to them as Doppler-related information and the Doppler potential vector. However, in the present disclosure, the Doppler potential vector, which is a signal transmitted and / or received between a terminal and a base station, may actually include information related to temporal correlation other than the Doppler shift. Unlike traditional Doppler estimators based on mathematical model-driven methods, Doppler-related information, which is the output of a Doppler estimator that can be implemented as a data-driven neural network according to various embodiments of the present disclosure, can be extracted so that higher-dimensional, rich temporal correlation information that can be expressed in the feature domain can be better reflected.

[0200] In the present disclosure, the Doppler estimator takes as input a channel instance decoded at a previous time and a channel instance measured at a current time, expressed in each-delay domain, and has as output Doppler-related information extracted from each-delay domain or a specific feature domain, which is different from a Doppler estimator that generally takes as input a received signal and outputs the frequency shift itself. However, if a traditional Doppler estimator is utilized, and an appropriate modification is applied, the traditional Doppler estimator can also be used as a Doppler estimator according to an embodiment of the present disclosure. Thereafter, the restored Doppler information and the CSI matrix decoded in the previous time From context is obtained. If the CSI matrix is ​​generated in the angular-delay domain, is the predicted CSI matrix can be treated as . In this case, the residual - can be encoded to be fed back. On the other hand, If it is generated in the feature domain, Is can be used as a condition related to encoding (e.g., conditional coding). In this case, the context of the feature domain The higher dimension of It can provide richer and more correlated conditions for encoding. and From context The obtained operation can be performed by the context generator of Fig. 26. The context generator can be implemented based on a neural network. To implement the context generator, it is also possible to benchmark techniques used in video compression (e.g., warping functions). is input as a condition to the contextual encoder / decoder neural network, and the CSI instance to the potential vector or encoded as From can be used to decode.

[0201] In one embodiment, the system according to various embodiments may utilize an entropy bottleneck layer. The entropy bottleneck layer is an example of the first quantization and entropy encoder (2602), the first entropy model (2620), and the first entropy decoding and dequantizer (2604) of FIG. 26A, or an example of the second quantization and entropy encoder (2612), the second entropy model (2622), and the second entropy decoding and dequantizer (2614). The entropy bottleneck layer is a neural network designed to jointly optimize transmission rate and distortion. The entropy bottleneck layer has been introduced for deep learning-based CSI feedback, and the entropy bottleneck layer may be adopted for CSI compression / reconstruction according to various embodiments. An example of the structure of the entropy bottleneck layer is shown in Fig. 27a.

[0202] FIG. 27A illustrates an example of an entropy bottleneck layer for entropy encoding / decoding according to an embodiment of the present disclosure. Referring to FIG. 27A, the entropy bottleneck layer includes a quantizer (2712), an entropy encoder (2714), an entropy decoder (2722), a dequantizer (2724), and an entropy model (2726).

[0203] Referring to Fig. 27a, a quantizer (2712) quantizes an input vector, and an entropy encoder (2714) entropy-codes the quantized vector. This generates a bit string representing the input vector. The bit string passes through a channel and is input to an entropy decoder (2722), which entropy-decodes the bit string. A dequantizer (2724) dequantizes the decoded information. An entropy model (2726) provides probability distribution information for entropy encoding / decoding.

[0204] Channel matrix A compressed latent vector through an encoder neural network can be generated. The entropy bottleneck layer is Take as input, is an approximation of or Prints out. Here, is the output during training. It is the value with uniform noise added to it, is the output of inference. It can be obtained by performing quantization in Fig. 27a. Is can be obtained by performing dequantization again. In the case of the quantization operation performed during inference, since the gradient is 0 in almost all areas and the differentiation is undefined at the remaining points, problems may occur in the backpropagation operation, which is the process of training the neural network.

[0205] To solve the problem, the entropy bottleneck layer can be transformed into a structure as in Fig. 27b during training. Fig. 27b illustrates an example of a structure for training the bottleneck layer according to an embodiment of the present disclosure. Referring to Fig. 27b, the entropy bottleneck layer can be transformed into a structure as in Fig. 27b during training. , and outputs a probability distribution is stored as an entropy model. The stored entropy model is the inference time It is used to perform entropy coding / decoding. PMF (probability mass function) required for entropy coding / decoding is saved can be obtained from the compressed latent vector during inference through the entropy bottleneck layer. can be expressed as a bit stream.

[0206] An entropy model may have a structure as shown in FIG. 28A. FIG. 28A illustrates an example of an entropy model according to an embodiment of the present disclosure. Referring to FIG. 28A, the entropy model includes a temporal prior encoder (2812), a hierarchical prior encoder (2814), and a prior fusion unit (2816). Additionally, the entropy model may further include a spatial prior encoder (2818).

[0207] The temporal prior encoder (2812) generates temporal priors based on context information. The context information is generated based on the temporal correlation of the channel, for example, the Doppler characteristic. Furthermore, the temporal prior encoder (2812) generates information related to the prior distribution of the context information. The temporal prior encoder (2812) may have a neural network structure.

[0208] The hierarchical prior encoder (2814) generates a hierarchical prior based on a latent vector representing a channel (hereinafter, referred to as a “channel latent vector”). Specifically, the hierarchical prior encoder (2814) may include a prior distribution related to at least one parameter of a prior distribution for the latent vector, for example, a prior distribution related to a mean and / or a standard deviation. To this end, according to one embodiment, the hierarchical prior encoder (2814) may include at least one of a hierarchical prior encoder, a quantizer, an entropy encoder, an entropy decoder, and a hierarchical prior decoder. For example, the hierarchical prior encoder (2814) included in a transmitting device may include a hierarchical prior encoder, a quantizer, an entropy encoder, an entropy decoder, and a hierarchical prior decoder, and the hierarchical prior encoder (2814) included in a receiving device may include an entropy decoder and a hierarchical prior decoder. For the operation of the hierarchical prior encoder (2814) of the receiving device, the transmitting device may transmit a value including the output of the quantizer as side information to the receiving device. Specifically, the information transmitted as side information may include the output of the hierarchical prior encoder that has undergone quantization and entropy coding. Here, the hierarchical prior encoder and the hierarchical prior decoder may have a neural network structure.

[0209] The prior fusion unit (2816) combines temporal priors and hierarchical priors. That is, the prior fusion unit (2816) can generate parameters (e.g., a mean vector, a standard deviation vector, etc.) of the prior distribution of the channel latent vector as distribution information used for entropy encoding / decoding. Here, the prior fusion unit (2816) can have a neural network structure.

[0210] Additionally, a spatial prior encoder (2818) may be further included. The spatial prior encoder (2818) may be used in conjunction with the temporal prior encoder (2812) or the hierarchical prior encoder (2814) to perform a corresponding role, and may generate a spatial prior based on the quantization result for the latent vector. When the spatial prior encoder (2818) is further included, a prior fusion unit (2816) fuses the temporal prior, the hierarchical prior, and the spatial prior.

[0211] The structure of the entropy model is described with reference to FIG. 28a. Meanwhile, referring to FIGS. 26a and 26b, each of the transmitting device and the receiving device includes two entropy models. According to various embodiments, each of the two entropy models may have the structure illustrated in FIG. 28a, or one of the entropy models may have a structure different from the structure illustrated in FIG. 28a. For example, one of the entropy models may be composed solely of the hierarchical prior encoder (2814) illustrated in FIG. 28a.

[0212] According to one embodiment, the first entropy model (2620) or the first entropy model (2670) may have a structure as shown in FIG. 28b. FIG. 28b illustrates an example of an entropy model that operates based on a channel latent vector according to one embodiment of the present disclosure. In FIG. 28b, Q performs quantization, AE / AD (arithmetic encoding / arithmetic decoding) performs entropy coding / decoding, and HPE / HPD (hierarchical prior encoder / decoder) encodes / decodes hierarchical priors. Here, HPE and HPD can be implemented as neural networks. Context, which is information corresponding to temporal correlation is encoded as a temporal prior through a temporal prior encoder neural network. The temporal prior is used for prior fusion together with the hierarchical prior. Prior fusion can also be implemented as a neural network. By fusing the temporal prior and hierarchical prior through a prior fusion neural network, Parameters of the prior distribution (e.g., mean vector) and standard deviation vector ) can be obtained. In prior fusion, in addition to hierarchical priors and temporal priors, a third prior (e.g., spatial correlation) can also be a target of fusion.

[0213] According to one embodiment, the second entropy model (2622) or the first entropy model (2672) may have a structure as shown in FIG. 28C. FIG. 28C illustrates an example of an entropy model operating based on a Doppler latent vector according to one embodiment of the present disclosure. In FIG. 28C, Q performs quantization, EC / ED performs entropy coding / decoding, and HPE / HPD encodes / decodes hierarchical priors. Here, HPE and HPD may be implemented as neural networks. Hierarchical Priors is the Doppler potential vector Information about the prior distribution (e.g., the mean vector) and standard deviation vector ) may be included. Therefore, If you decode it through HPD, and This can be obtained. and can be utilized for entropy coding / decoding of Doppler potential vectors.

[0214] As described above, systems according to various embodiments can utilize temporal priors and contexts. The temporal priors can be extracted as follows. First, the channel matrix is assumed to be expressed in the angular-delay domain. It is known that Doppler terms of different paths in the angular-delay domain are less entangled with each other than in the space-frequency domain. However, various embodiments of the present disclosure are not limited to a channel matrix in a specific domain such as the space-frequency domain or the angular-delay domain. For example, the channel matrix may be expressed in a 3D tensor format in a 2D space-1D frequency domain, or in a tensor format in a 3D angular-delay domain consisting of an elevation angle, an azimuth angle, and a tap of delay. Alternatively, the channel matrix according to various embodiments of the present disclosure may be expressed in other domains.

[0215] The Doppler effect can be extracted in the angular-delay domain. For convenience of explanation, is represented as a matrix, but for planar arrays, to estimate the Doppler effect, it is represented as a 3D angular-domain tensor where each component corresponds to the elevation angle, azimuth angle, and tap of delay. can be used. Conceptually, information related to Doppler is and corresponds to the difference between the two. Therefore, as a conceptually simple method, Doppler information To extract , a subtraction operation can be performed on two consecutive channel matrices, i.e., the current channel matrix Previous channel matrix in A subtraction operation can be performed to subtract the previous CSI matrix. Instead, the previously decoded CSI matrix Because you can use it, at By subtracting, we can obtain Doppler-related information. That is, may be. However, in other embodiments, more complex operations and processes other than simple subtraction may be required, and for example, neural networks may be utilized. A Doppler estimator according to one embodiment of the present disclosure may be used to estimate the current CSI matrix. and the CSI matrix decoded in the previous time Takes as input, and Doppler information It may include a neural network that outputs a Doppler estimator. The Doppler estimator can be implemented based on a conventional channel prediction neural network. In addition, non-deep learning-based channel prediction approaches (e.g., Prony-based, autoregressive, matrix pencil, etc.) can be considered as a means for implementing the Doppler estimator.

[0216] As described above, according to the framework for the CSI feedback method according to one embodiment, the transmitting device includes a Doppler estimator. Doppler information, which is the output of the Doppler estimator is a Doppler latent vector through a Doppler encoder neural network. is encoded as a Doppler potential vector can be converted into a bit string through an entropy bottleneck layer. The quantized Doppler potential vector is the reconstructed Doppler information through the Doppler decoder neural network. can be decoded as

[0217] The Doppler-related information and Doppler potential vector expressed in the present disclosure are intermediate information or signals obtained in a data-driven manner from channel information of the current time and channel information of the previous time. Due to the characteristics of neural networks and deep learning, it is not easy to clearly identify or accurately describe the physical meaning of the signal. However, since the aforementioned information or signals include information related to the Doppler shift, the present disclosure refers to them as Doppler-related information and Doppler potential vector. However, in the present disclosure, the Doppler potential vector, which is a signal transmitted and / or received between a terminal and a base station, may actually include information related to temporal correlation other than the Doppler shift. Unlike traditional Doppler estimators based on mathematical model-driven methods, Doppler-related information, which is the output of a Doppler estimator that can be implemented as a data-driven neural network according to various embodiments of the present disclosure, can be extracted so that higher-dimensional, rich temporal correlation information that can be expressed in the feature domain can be better reflected.

[0218] In the present disclosure, the Doppler estimator takes as input a previously decoded channel instance and a current measured channel instance, expressed in the respective delay domain, and outputs Doppler-related information extracted from the respective delay domain or a specific feature domain. This differs from a Doppler estimator that typically takes as input a received signal and outputs the frequency shift itself. However, if a traditional Doppler estimator is utilized, and appropriate modifications are applied, the traditional Doppler estimator can also be used as the Doppler estimator according to an embodiment of the present disclosure.

[0219] Afterwards, the restored Doppler information and the CSI matrix decoded in the previous time From context is obtained. Context If the context is generated in the angular-delay domain, as is the case with the CSI matrix, is the predicted CSI matrix can be treated as . In this case, the residual can be decoded to provide feedback. On the other hand, the context If the context is generated from the feature domain, Is can be used as a condition related to encoding (i.e., conditional coding). In this case, the context of the feature domain The higher dimension of It can provide richer and more correlated conditions for encoding. and From context The operation of obtaining can be performed by the context generator (2618) of Fig. 26a. The context generator can be implemented based on a neural network. In order to implement the context generator (2618), it is also possible to benchmark techniques used in video compression (e.g., warping function). Context is input as a condition to the contextual encoder / decoder neural network, and the CSI instance to the potential vector or encoded as From can be used to decode.

[0220] As described above, according to various embodiments, the context is a CSI instance Channel potential vector can be used to encode or decode. Furthermore, the context is the latent vector It can also be used to represent a bit string. Specifically, a context generated through a context generator is the latent vector This is information required for the entropy model of the entropy bottleneck layer for transmission and reception. An example of the entropy model according to one embodiment of the present disclosure is as described above with reference to Figure 28a. That is, the channel potential vector An example of an entropy model required for entropy coding (e.g., arithmetic coding) for transmission and reception is as described above with reference to FIG. 28a.

[0221] Context, which is information corresponding to temporal correlation is encoded as a temporal prior through a temporal prior encoder neural network. The temporal prior is used for prior fusion together with the hierarchical prior. Prior fusion can also be implemented as a neural network. By fusing the temporal prior and hierarchical prior through the prior fusion neural network, Parameters of the prior distribution (e.g., mean vector) and standard deviation vector ) can be obtained. In prior fusion, in addition to hierarchical priors and temporal priors, a third prior (e.g., spatial correlation) can also be a target of fusion.

[0222] Obtaining an entropy model

[0223] Entropy model in this disclosure or is a parameter or can be obtained from. This can be explained through the graph of Figure 34 below. For convenience of explanation, go Explain the reasons that can be given from From The method of obtaining it is the same, so it is omitted.

[0224] Referring to Figure 29, , and Each one is , and Each random It means the th element, , and Each one is , and It means the probability distribution for each. Is This is the quantization result value. If the quantization bin size is 1 and the median of each bin is used as the representative value for quantization, then if the median of each bin is set as an integer, =round( ) can be expressed as an arbitrary integer. For each quantization bin, the interval corresponding to About ( ) is an integer In matches the value. Here, is an integral variable.

[0225] Is It can be obtained by adding uniform noise. At this time, the interval length of the uniform distribution is 1, like the quantization bin size. Since the probability distribution of a random variable defined as the sum of two random variables is expressed as a convolution of each of the two random variables, is a uniform distribution with a width of 1 and is obtained by convolution operation. A uniform distribution with an interval length of 1 is the same as a constant function with a function value of 1 within an interval of width 1 due to the characteristics of PDF. Therefore, a uniform distribution with a width of 1 and Performing a convolution operation on is to move a boxcar function of unit width with function value 1 over the overlapping interval. is equivalent to integrating . Therefore, the moving integration interval is exactly If located in, Is About ( ) is calculated as a definite integral, and this value is an integer. In Matches a value, i.e. any integer About Silver integer In It is equal to the value.

[0226] The concepts of prior, hyperprior, and context

[0227] In general, a prior is a concept that contrasts with a posterior probability. The posterior probability is proportional to the prior multiplied by the likelihood. The prior probability refers to the probability before a specific event occurs, meaning it is the probability before any observational information about that specific event is reflected.

[0228] In this disclosure, the prior is a probabilistic model for the quantized representation required for entropy coding, i.e., an entropy model. or corresponds to. The reason why the probability model, i.e. the probability distribution, is called a prior is because it is still a channel instance. or Doppler information Because it is not given. In contrast, the channel instance or Doppler information is the conditional probability given or corresponds to the posterior probability. Furthermore, in this disclosure,

[0229] or Each of them or Because we assume that we get it from the latent representation or Prior probability distribution for and Or and All of them can be treated as fryers.

[0230] Priors for parameters of an entropy model are called hyperpriors. Since the entropy model is called a prior, the term "hyperprior" can be used in the sense of "prior for priors." That is, a hyperprior means information about parameters of the entropy model (e.g., standard deviation, mean, etc.). The hierarchical priors and temporal priors in the present disclosure can be understood as a type of hyperprior.

[0231] As described above, systems according to various embodiments may utilize hyperprior or hierarchical priors. Latent representation The concept of hyperprior can be applied to model the dependency between elements. Hyperprior is Prior distribution for Prior distribution for the parameters (e.g. mean vector μ and standard deviation vector σ) It means. At this time, is the latent vector Another latent vector obtained by encoding has information about the hyperprior. by inputting it into another encoder neural network. get , is input into another entropy bottleneck layer and is printed A bit string can be generated for HyperFryer Because the structure for obtaining it is hierarchical, may be referred to as a hierarchical prior. In the present disclosure, Since other priors (e.g. temporal priors) are also used, to avoid confusion, the following is referred to as a hierarchical prior.

[0232] Additionally, in the present disclosure, context refers to information that can be provided as a conditional input to a function block to conditionally operate the function block. For example, in the case of a CSI encoder, the same channel instance Even if it is input, depending on what value the context is given, the output of the CSI encoder may vary. That is, the operation of a function block may vary depending on the context. In particular, according to various embodiments of the present disclosure, high-dimensional context in the feature domain may be extracted through learning so that the context may include rich information that may assist in the operation of the function block.

[0233] In other words, context can be used not only to encode or decode CSI instances into latent vectors, but also to represent latent vectors as bit streams. Specifically, the context generated by the context generator is information necessary for the operation of the entropy model in the entropy bottleneck layer for transmitting and receiving latent vectors.

[0234] FIG. 30 illustrates an example of a CSI feedback procedure according to one embodiment of the present disclosure. FIG. 30 illustrates an example of CSI feedback. The CSI feedback procedure according to various embodiments does not necessarily include all of the illustrated operations and may include only a portion of the procedures described below. For example, the order of signaling may vary, some signaling may be omitted, or additional signaling may be performed in addition to the described ones. In FIG. 30 , the terminal (3010) is an encoder-side device, including an encoder neural network, and the base station (3020) is a decoder-side device, including a decoder neural network. Additionally, the terminal (3010) may further include a decoder neural network.

[0235] Referring to FIG. 30, in steps S3001-1 to S3001-N, the base station (3020) transmits reference signals, for example, CSI-RSs, to the terminal (3010). The CSI-RSs may be transmitted periodically, aperiodicly, or semi-persistently. Although not illustrated in FIG. 30, the base station (3020) may transmit CSI-RSs after transmitting configuration information for the CSI-RSs to the terminal (3010).

[0236] In step S3003, the terminal (3010) determines a channel instance based on the CSI-RS. And, in step S3005, the terminal (3010) measures the channel potential vector and Doppler potential vector In step S3007, the base station (3020) estimates the channel instance. Restores.

[0237] In step S3009, the terminal (3010) determines a channel instance based on the CSI-RS. And, in step S3011, the terminal (3010) measures context information and channel instances Based on the channel latent vector Outputs the channel latent vector Hierarchical fryer for can be obtained. In step S3013, the terminal (3010) obtains Doppler information Based on the Doppler potential vector Outputs the Doppler potential vector. Hierarchical fryer for can be obtained.

[0238] At step S3015, the terminal (3010) quantizes and entropy-codes the channel latent vector and hierarchical fryer In step S3017, the terminal (3010) transmits a quantized and entropy-coded Doppler potential vector and hierarchical fryer Sends.

[0239] That is, the terminal (3010) is a channel potential vector and Doppler potential vector After quantizing and entropy coding, it is transmitted to the base station (3020), and also and Hierarchical priors for and After quantizing and entropy-coding, the terminal (3010) transmits the data to the base station (3020). That is, the terminal (3010) transmits the data to the base station (3020). and As a side note about and In this process, the channel potential vector and Hierarchical priors for , and the Doppler potential vector and Hierarchical priors for Each is converted into a sequence of symbols, such as a bitstream, by entropy coding. For convenience of explanation, the present disclosure considers the bitstream as a sequence of symbols, but binary transformation is only an example, and embodiments of the present disclosure are not limited to binary transformation.

[0240] At step S3019, the base station (3020) restores the Doppler potential vector Based on Doppler information In step S3021, the base station (3020) restores the restored Doppler information. and restored channel instances Based on context In step S3023, the base station (3020) generates a context and restored channel latent vectors Channel instances based on Restores.

[0241] According to the procedure according to the embodiment described above, the information signaled from the terminal to the base station is a channel potential vector and Hierarchical priors for , Doppler potential vector and Hierarchical priors for As explained with reference to Figure 30, the base station transmits a pilot or reference signal such as CSI-RS, and the terminal uses the pilot to determine the current time. Channel information The terminal measures the current CSI matrix measured Compressed through a CSI encoder, and the channel latent vector At this time, the CSI encoder obtains the channel matrix based on the context vector. compresses. Here, the context vector on the encoder side is the context vector on the decoder side. There may be differences.

[0242] For the procedure illustrated in Figure 30, the feedback rate can be adaptively adjusted based on channel variation. However, in actual wireless communication systems, adjustments to the channel feedback rate may be required due to factors other than channel variation, such as overhead control or traffic fluctuations. Furthermore, when a neural network model or parameter set is determined through end-to-end learning via offline training, the model or parameter set may be optimal in an average sense, but may be suboptimal for a specific channel instance.

[0243] Accordingly, the present disclosure proposes an additional fine-tuning technique for rate-distortion optimization specific to a channel instance. That is, the present disclosure proposes a hyperprior-based channel feedback framework operating technique that enables adjustment of the feedback transmission rate due to various factors such as channel changes, while also enabling fine-tuning specific to a channel instance. Here, the fine-tuning according to various embodiments can be understood as tuning the input / output signals of the neural network model, rather than tuning the neural network model or parameters. In other words, the proposed technique relates to optimization of the signal transmitted from the encoder side to the decoder side (hereinafter, the "transmission signal"). Therefore, the fine-tuning according to various embodiments can be referred to or understood as "pre- / post-filtering" or "pre- / post-processing" of the input / output signals.

[0244] Component-wise Mathematical Description for Problem Formulation

[0245] The fine-tuning technique described below can be applied to the aforementioned CSI feedback framework. However, the proposed technique can also be applied to CSI feedback frameworks or algorithms with different structures. That is, the inner function according to the embodiments described below can be utilized in all CSI feedback methods that use hierarchical priors as side latent information. Therefore, the proposed technique described below can be utilized in neural network structures for all CSI compression / reconstruction based on various hierarchical priors. For example, in addition to the hierarchical prior for the channel latent described in the present disclosure and the hierarchical prior for the Doppler latent, which corresponds to temporal correlation information, hierarchical priors for spatial correlation, frequency correlation, angular correlation, delay correlation, etc. can be considered. Furthermore, the outer function according to the embodiments described below can be utilized for operation of a neural network structure for various CSI compression / reconstruction optimized from a rate-distortion perspective.

[0246] Hereinafter, the present disclosure mathematically expresses each component of a hyperprior-based channel feedback framework for problem formulation. The components of the hyperprior-based channel feedback framework are as follows [Table 2]. For convenience of explanation, some components may be described as a single component. Furthermore, for convenience of explanation, encoder-side information and decoder-side information may be described without distinction. For example, and Without distinction can be expressed as . That is, it is assumed that the signal is decoded without error even when passing through a wireless channel. For convenience of explanation, the subscript indicating time is omitted, and instead the symbol ' is used to represent the previous time. ' is used. For example, is the previously decoded CSI matrix. can be understood as

[0247] Component DescriptionChannel Encoder : Channel encoder represented by contextual CSI encoder neural network in Fig. 26c Is A neural network having a parameter set or model, with a channel instance As input, context Treat as conditional input, and channel potential information Outputs the channel decoder : Channel encoder represented as a contextual CSI decoder neural network in Fig. 26c. Is A neural network having a parameter set or model, wherein the quantized channel latent information As input, context Treat as a conditional input and restore the channel instance Outputs a channel hierarchical prior encoder : Corresponding to HPE in Fig. 28b, channel potential information Hierarchical fryer from Encodes the channel hierarchical prior encoder Is A neural network with a parameter set or model. Entropy model for channel latent information. It corresponds to Fig. 28b and is a component expressed as an entropy model in Fig. 26c. Channel potential information The entropy model for has a set of parameters, and a hierarchical prior and context Channel potential information in a given situation It means the probability distribution (e.g., PMF (probability mass function)) for the parameter set. Here, refers to a parameter set or model for the HPD, temporal prior encoder, and prior fusion neural network of Fig. 28b. That is, can be understood as a parameter set or model for all three components. Entropy model for channel hierarchical prior Hierarchical fryer The entropy model for has a set of parameters. That is, Is is a set of parameters for the probability distribution of Doppler estimator and Doppler encoder. : One component including a two-component Doppler estimator and a Doppler encoder neural network in Fig. 26c. Is A neural network having a parameter set or model, with a channel instance and previously decoded channel instances Process as input, Doppler potential information Outputs . Here, the parameter set refers to a parameter set or model for the Doppler estimator and Doppler encoder neural network of Fig. 26c. That is, is a parameter set or model for both components. For convenience, in this disclosure, can be simply referred to as a Doppler encoder. That is, in the present disclosure, the Doppler encoder can be understood as including a Doppler estimator, a Doppler decoder and a context generator. : One component including the Doppler decoder neural network and context generator in Fig. 26c. Is A neural network having a set of parameters or a model, wherein the quantized Doppler potential information and previously decoded channel instances Process as input, and context Outputs . Here, the parameter set refers to a parameter set or model for the Doppler decoder neural network and context generator of Fig. 26c. That is, is a parameter set or model for both components. For convenience, in this disclosure, can be simply referred to as a Doppler decoder. That is, in the present disclosure, the Doppler decoder can be understood as including a context generator. Doppler hierarchical prior encoder : Corresponds to HPE in Fig. 28c, and Doppler potential information Hierarchical fryer from Encodes Doppler hierarchical prior encoder Is A neural network with a parameter set or model. Entropy model for Doppler latent information Corresponds to Fig. 28c, and is omitted as it is not explicitly shown in Fig. 26c. Doppler potential information The entropy model for has a set of parameters, and a hierarchical prior Doppler potential information in a given situation It means the probability distribution (e.g. PMF) for the parameter set refers to a parameter set or model for HPD of Fig. 28c. That is, is a parameter set or model of a hierarchical prior decoder neural network. Entropy model for Doppler hierarchical prior Hierarchical fryer The entropy model for has a set of parameters. That is, Is is a set of parameters for the probability distribution.

[0248] outer function

[0249] External functions correspond to procedures performed before the models used by the terminal and base station are determined. In contrast, internal functions, described later, correspond to procedures performed after the model is determined and before model transfer occurs. External functions are procedures related to how and on what basis the model ID is determined when model transfer is performed by signaling the model ID.

[0250] In order to optimize a hyper-priority-based channel feedback framework including at least one of the components illustrated in [Table 2] in terms of rate-distortion performance, a loss function can be defined as in the following [Mathematical Formula 3].

[0251]

[0252] In [Equation 3], is the loss function, , , , are transmission signals, is context information, Is A set of parameters for the entropy model, Is A set of parameters for the entropy model, is a channel instance, is the decoded channel, is the probability distribution of A given B and C, is distortion loss, represents a weighting coefficient for distortion loss. For example, the distortion loss can be MSE (mean square error).

[0253] In a loss function such as [Equation 3], is a positive value set prior to optimization to optimize the aforementioned parameter sets. The larger the value is set, the more optimized it is to reduce distortion rather than reduce the transmission rate. That is, If the value is set to a large value, a channel feedback model that can achieve higher restoration performance through a higher feedback transmission rate on average will be obtained. On the other hand, As the value is set smaller, a channel feedback model with reduced radio resource overhead for channel feedback will be obtained by reducing the feedback transmission rate, even if the average restoration performance is not better. Therefore, may be referred to as a 'rate-distortion trade-off hyperparameter', a 'rate-distortion trade-off parameter', a 'rate-distortion adjustment parameter', a 'rate-distortion control parameter', or other terms having an equivalent technical meaning. Generally, in the context of deep learning, 'parameter' means the weights and biases of a neural network that are learned through training, and 'hyperparameter' means values ​​that are set prior to training and used to control the learning process, but are not updated during the training process.

[0254] In a typical neural image compression model, When the loss function is given in the form of hyperparameters A conceptual representation of a typical rate-distortion tradeoff based on values ​​is shown in Figure 31. Figure 31 illustrates an example of a rate-distortion tradeoff in a compression model. In Figure 31, R represents rate, and D represents distortion.

[0255] Given a training set of channel instances, neural network parameter sets / models for the aforementioned components (e.g., at least one of the components illustrated in Table 2) can be obtained through learning to minimize a loss function.

[0256]

[0257] In [Equation 4], is a set of parameters of the framework, is the loss function, is a channel instance, is the training set, stands for the expectation operator.

[0258] Here, the loss function In If you set the values ​​differently, the parameter set corresponding to the optimized model can be determined differently, i.e. the parameter set is the loss function Depends on.

[0259] In the following description, through offline training, It is assumed that the corresponding model / parameter set has already been determined and deployed in the terminal and base station. In addition, it is assumed that there are multiple pairs of models when the encoder-side neural network model and the decoder-side neural network model are paired with each other. Therefore, the present disclosure is an ID of a neural network model for a hyperpriority-based channel feedback algorithm. We propose a method to perform model transition by having the terminal and the base station signal each other the values ​​corresponding to the predefined For the values, the terminal and base station are each We have a model for the values, each A model transition can be requested by specifying an ID corresponding to a value as a model ID.

[0260] The model ID may be determined or changed based on at least one factor. In one embodiment, the model ID may be changed based on the amount of traffic. For example, if the amount of traffic that the base station wants to transmit to the terminal increases, the model ID corresponding to the currently used model may be changed. Higher than the value The use of a model corresponding to the value may be required. In this case, the base station is expected to have channel restoration performance that can satisfy the downlink traffic demand. Determine the value, and the determined A change to a model corresponding to the value can be instructed or notified to the terminal. Alternatively, the terminal is expected to have channel restoration performance that can satisfy the required downlink traffic. Determine the value, and the determined You can request the base station to change to a model corresponding to the value.

[0261] In one embodiment, the model ID can be changed for overhead control. For example, the base station may expect a feedback transmission rate that can satisfy the constraints of the uplink resources allocated to the terminal. Determine the value, and the determined A change to a model corresponding to the value can be instructed or notified to the terminal. Alternatively, a feedback transmission rate is expected to satisfy the uplink resource constraints allocated to the terminal. Determine the value, and the determined You can request the base station to change to a model corresponding to the value.

[0262] According to various embodiments, hyperparameters on both the encoder-side and decoder-side A model ID according to a value is predefined in the form of a table, and a terminal and a base station can perform AI / ML-based CSI feedback using the predefined table. In the present disclosure, the model ID may be referred to as a term such as a model pairing ID in that it can be applied to a two-sided model. That is, an encoder-side AI / ML model and a decoder-side AI / ML model can be defined as being paired with each other based on the model pairing ID or model ID. [Table 3] below Here is an example of a table that can determine a model ID or model pairing ID based on a value. [Table 3] is an example, and the specific contents of the table may vary.

[0263] Rate-distortion trade-off hyperparameter Representation:normalized as Model(pairing) ID0.99900.510.2520.130.014

[0264] Hyperparameters according to various embodiments In addition to the model ID / model pairing ID, the value of may also be related to the payload size for CSI feedback. That is, The maximum payload size can be determined based on the value. A larger value can determine a larger maximum payload size. Conversely, The smaller the value, the smaller the maximum payload size can be determined. Information related to the CSI payload size determined based on the value can be transmitted from the base station to the terminal as CSI configuration information prior to the CSI feedback operation. That is, in the CSI-related configuration information transmitted by the base station and received by the terminal Information related to the maximum CSI payload size determined based on the value may be included.

[0265] According to the 3GPP NR standard document TS 38.214, a CSI report of a terminal may consist of Part 1 and Part 2. Part 1 is used to identify the number of information bits of Part 2, and Part 1 may be completely transmitted before Part 2 is transmitted. Therefore, Part 1 may have a fixed payload size. Hyperparameters according to various embodiments Information related to can be reported to the base station by the terminal, and hyperparameters Information related to may be included in Part 1 CSI reports.

[0266] Hyperparameters Or, if the corresponding model ID / model pairing ID is determined by the base station, the terminal Before making any decisions about values, hyperparameters Capability information about the terminal can be transmitted to the base station. According to one embodiment, the hyperparameters transmitted by the terminal to the base station The ability information related to hyperparameters The maximum supported value of the hyperparameter Candidates for hyperparameters It can include at least one of the ranges of values, i.e., the terminal supports The terminal may report information about the maximum value among the values ​​to the base station. Alternatively, the terminal may report information about the maximum value among the values ​​that the terminal supports or that the terminal prefers. Information about the candidates or ranges of values ​​can be transmitted to the base station.

[0267] inner function

[0268] Once the model is determined by the external functions described above, the following procedure, i.e., the internal function, can be performed for each channel instance based on the determined model. Given a training set for channel instances, a neural network parameter set / model for components (e.g., at least one of the components illustrated in [Table 2]) can be obtained through learning by minimizing a loss function.

[0269]

[0270] In [Equation 5], is a set of parameters of the framework, is the loss function, is a channel instance, is the training set, stands for the expectation operator.

[0271] Here, the parameter set corresponding to the optimized model is the loss function In It can be determined differently depending on the value. The internal function is It can be performed in a situation where the value is determined. That is, the internal function is performed after the model ID is determined. In addition, through offline training, The corresponding model / parameter set is determined, and the internal function is performed in a situation where it is deployed in the terminal and base station. That is, it is assumed that inference or testing is performed in a state that is sufficiently close to optimal from a learning perspective.

[0272] A given model or parameter set may be optimal in an average sense, but may be suboptimal for a specific channel instance. Therefore, there is room for further optimization in terms of throughput-distortion by performing fine-tuning in a channel-instance-specific manner during inference or testing operations. This channel-instance-specific fine-tuning is performed on the neural network model, i.e., With a fixed set of parameters, it is passed from the encoder side to the decoder side. , , , This is an operation that optimizes or quasi-optimizes information such as . It is expected that this fine-tuning will provide a transmission rate saving effect of 5 to 15%. The present disclosure is a channel feedback algorithm based on hyper-priority signals transmitted from the encoder side to the decoder side. , , , We propose a procedure for fine-tuning. In this disclosure, , , , may be collectively referred to as feedback information, transmission signals or other terms having equivalent technical meaning. , , , Each of these may be referred to as channel latent information, channel side latent information, Doppler latent information, Doppler side latent information, or other terms having equivalent technical meanings. , , , is described with reference to Figure 30. , , , It can be understood as the result of applying processing (e.g. quantization) for morphological and / or representational transformation.

[0273] With the parameter set / model of the neural network fixed, for optimization specific to a channel instance from a rate-distortion perspective, an optimization problem can be defined as follows [Mathematical Formula 6] based on a loss function identical to the loss function defined for optimization of the neural network model.

[0274]

[0275] In [Equation 6], , , , are optimized transmission signals, , , , are transmission signals, is the loss function, means a channel instance.

[0276] Here, can be understood as an unconstrained multi-objectives optimization problem, where the objectives are the minimum reconstruction error and the minimum bitlength for each of the four transmission signals. That is, Each of the five terms corresponds to a goal. The optimal solution to a multi-objective problem can be given as "Pareto Optimal" if no other goal can be improved without degrading one of the other goals.

[0277] It is known that the necessary and sufficient condition for a solution to a multi-objective optimization problem to be Pareto optimal is that it satisfies the Karush-Kuhn-Tucker (KKT) condition. In general, Consider a situation where multi-objective optimization is given, where: is the i-th goal to be minimized, And, At this time, the solution to the multi-objective optimization problem The necessary and sufficient condition for Pareto optimality is to satisfy the KKT condition as shown in [Mathematical Formula 7] below.

[0278]

[0279] In [Equation 7], is the ith goal, is the variable of interest, is the coefficient of the i-th objective, Is It means partial differentiation with respect to . Therefore, is the i-th loss function It means the gradient for .

[0280] To apply the general theory described above to the CSI framework, the loss function can be reorganized as shown in [Mathematical Formula 8] below.

[0281]

[0282] In [Equation 8], is the loss function, , , , are transmission signals, is context information, Is A set of parameters for the entropy model, Is A set of parameters for the entropy model, is a channel instance, is the decoded channel, is the probability distribution of A given B and C, is distortion loss, represents the weighting coefficient for distortion loss.

[0283] Each of the five terms that constitute the loss function of [Equation 8] can be defined as a target, as in [Equation 9] below.

[0284]

[0285] In [Equation 9], is the ith goal, , , , are transmission signals,, is context information, Is A set of parameters for the entropy model, Is A set of parameters for the entropy model, is a channel instance, is the decoded channel, Is A set of parameters, is a parameter set of the channel encoder, is the probability distribution of A given B and C, is the probability distribution of A given B and C, means distortion loss.

[0286] Here, the coefficients are And, To satisfy the condition , can be set as follows. Therefore, the loss function can be re-expressed as in [Mathematical Formula 10] below.

[0287]

[0288] In [Equation 10], is the loss function, is the weighting coefficient for distortion loss, , , , are transmission signals, is the ith goal, is the coefficient of the ith objective, is the training set, is the expectation operator, means a channel instance.

[0289] Using a loss function such as [Equation 9], the KKT condition for each transmission signal is expressed as [Equation 11] below.

[0290]

[0291] In [Equation 11], is the loss function, is the weighting coefficient for distortion loss, , , , are transmission signals, is the ith goal, is the coefficient of the ith objective, is the training set, is the expectation operator, means a channel instance.

[0292] Although the conditions may not be 100% valid for each channel instance, they are valid on average for the channel instances within the training set. In this disclosure, it is assumed that the conditions are also valid to some extent for any channel instance within the test data set for inference.

[0293] The decoding order of the transmission signals is the reverse of the encoding order. , , , It is in order. Therefore, Let us first look at the KKT conditions for the case where the neural network models for the hyperprior-based channel feedback algorithm are optimal, i.e., when learning is complete. and This is confirmed to be (in expectation) canceled out. In fact, during inference, for any channel instance, the two gradients have a correlation coefficient close to -1.

[0294] In the gradient-based optimization process, Doppler potential information To reduce the bit length for the Doppler side potential information Is can be updated by moving in the negative or opposite direction by a step size. However, Before decoding, the base station can't get it. Instead, has a strong negative correlation with Is Since the base station can obtain the Doppler potential information even before decoding it, the present disclosure To reduce the bit length for We propose to utilize . If we express this in a formula, it is as follows [Mathematical Formula 12].

[0295]

[0296] In [Equation 12], Doppler side potential information, one of the transmission signals, is the optimal step size, Is Partial differentiation for , Is A set of entropy model parameters for, means the probability distribution of A given B and C.

[0297] Doppler side potential information representing hierarchical priors for Doppler potential information using the same procedure as [Mathematical Formula 12] It can be called a 'latent shift' in the sense that it moves the optimal step size. can be obtained at the encoder side through brute force or other optimization techniques as shown in [Mathematical Formula 13] below.

[0298]

[0299] In [Equation 13], is the optimal step size, is the step size, Doppler potential information, one of the transmission signals, Doppler side potential information, one of the transmission signals, Is Partial differentiation for , Is A set of parameters for the entropy model, is the Doppler potential information A set of parameters for the entropy model, means the probability distribution of A given B and C.

[0300] On the encoder side To minimize the bit length, obtained Optimal step size for is signaled to the base station. After the information is properly converted into a bit string, it can be transmitted to the base station. In addition, Doppler potential information is provided at the encoder side. In encoding, the Doppler side latent information with latent shift is performed The entropy model given by can be used.

[0301] Next, referring to [Equation 11], The KKT condition for is more complex than the KKT conditions for other transmission signals, which consist of two terms, because it contains three terms. Therefore, Referring to the way the KKT conditions are handled for the Doppler side potential information As with the potential shift procedure performed on the Doppler potential information A method that does not perform the potential shift procedure for the Doppler potential information may be considered. However, in a different way, After some fine-tuning has been done on can be signaled, After it is decoded can be decoded. That is, according to another embodiment, If the KKT conditions for are defined differently, A potential shift can be performed on .

[0302] Channel-side latent information in [Equation 11] Looking at the KKT conditions for the hyperpriority-based channel feedback algorithm, the neural network models are optimal, i.e., when learning is complete. and This can be expected to cancel out on average. In fact, during inference, for any channel instance, the two gradients have a correlation coefficient close to -1.

[0303] In the gradient-based optimization process, channel potential information To reduce the bit length for channel side potential information Is can be updated by moving in the negative or opposite direction by a step size. However, Before decoding, the base station can't get it. Instead, has a strong negative correlation with Is Since the base station can obtain the channel potential information even before decoding, the present disclosure To reduce the bit length for We propose a latent shift that utilizes . This can be expressed in a formula as follows [Mathematical Formula 14].

[0304]

[0305] In [Equation 14], is one of the transmission signals, channel side potential information, is the optimal step size, Is Partial differentiation for , Is A set of parameters for the entropy model, means the probability distribution of A given B and C.

[0306] The reason why the procedure as in [Equation 14] is called latent shift is because the channel side latent information represents the hierarchical prior for the channel latent information. Because it moves. At this time, the optimal step size can be obtained at the encoder side through brute force or other optimization techniques as shown in [Mathematical Formula 15] below.

[0307]

[0308] In [Equation 15], is the optimal step size, is the step size, is the channel potential information, is channel side potential information, Is Partial differentiation for , Is A set of parameters for the entropy model, is the Doppler potential information, Is A set of parameters for the entropy model, Is A set of parameters for the entropy model, is the previously decoded channel instance, means the probability distribution of A given B and C.

[0309] On the encoder side To minimize the bit length, obtained Optimal step size for is signaled to the base station. After the information is properly converted into a bit string, it can be transmitted to the base station. Furthermore, the channel potential information can be obtained from the encoder side. In encoding, the channel side latent information on which the latent shift is performed The entropy model given by can be used.

[0310] Finally, in [Equation 11], the channel potential information Looking at the KKT conditions for the hyperprior-based channel feedback algorithm, the neural network models are optimal, i.e., when learning is complete, and the channel latent information Entropy model for of Slope for and channel potential information of restoration errors Weighted gradient for This can be expected to cancel out on average. In fact, during inference, for any channel instance, the two gradients have a correlation coefficient close to -1.

[0311] In the gradient-based optimization process, the restoration error To reduce channel potential information Is can be updated by moving in the negative or opposite direction by a step size. However, from the standpoint of the base station corresponding to the decoder side, the actual (true) channel instance Included is information that cannot be obtained. Instead, has a strong negative correlation with Is , , , and can also be obtained from the base station after decoding. Therefore, the present disclosure is to reduce the restoration error. We propose a latent shift that utilizes . This can be expressed in a formula as follows [Mathematical Formula 16].

[0312]

[0313] In [Equation 16], is the channel potential information, is the optimal step size, Is Partial differentiation for , is channel side potential information, is the Doppler potential information, Is A set of parameters for the entropy model, Is A set of parameters for the entropy model, is the previously decoded channel instance, means the probability distribution of A given B and C.

[0314] The reason why the procedure as in [Equation 16] is called latent shift is because the channel latent information Because it moves. At this time, the optimal step size can be obtained at the encoder side through brute force or other optimization techniques as shown in [Mathematical Formula 17] below.

[0315]

[0316] In [Equation 17], is the optimal step size, is the step size, is a channel instance, ( ) is a channel decoder, is the channel potential information, Is Partial differentiation for , is channel side potential information, is the Doppler potential information, Is A set of parameters for the entropy model, Is A set of parameters for the entropy model, is the previously decoded channel instance, ( ) is a Doppler decoder and context generator, Is ( ) parameter set, is a parameter set of the channel encoder, means the probability distribution of A given B and C.

[0317] Obtained to minimize the restoration error on the encoder side, Optimal step size for is signaled to the base station. After the information is properly converted into a bit string, it can be transmitted to the base station.

[0318] As with the various embodiments described above, fine-tuning of the representation values ​​of information transmitted for CSI feedback can be performed. The present disclosure below describes operations and procedures performed by a base station and a terminal according to the embodiments described above. In the following description, the terminal is described as an encoder-side device, and the base station is described as a decoder-side device. However, the proposed technology can be applied to various combinations of devices, such as terminals and terminals, base stations and base stations, terminals and other devices, and base stations and other devices, in addition to terminals and base stations, where channel feedback can be performed.

[0319] FIG. 32 illustrates an example of a procedure for transmitting CSI according to one embodiment of the present disclosure. FIG. 32 also illustrates an operation method of a terminal.

[0320] Referring to FIG. 32, in step S3201, the terminal receives configuration information related to CSI feedback. The configuration information may include at least one of information related to reference signals transmitted for channel measurement (e.g., resources, sequences, etc.), information related to channel measurement operations, and information related to feedback (e.g., formats, resources, feedback counts, cycles, etc.). According to various embodiments, the configuration information may be generated based on temporal correlation and may indicate CSI feedback having a variable transmission rate. Specifically, according to various embodiments, the configuration information may indicate feedback of values ​​representing a channel, for example, latent vectors, and information related to temporal correlation (e.g., Doppler information). Additionally, the configuration information may indicate feedback of values ​​for fine-tuning the values ​​representing the channel.

[0321] In step S3203, the terminal receives reference signals. The terminal receives the reference signals based on the configuration information. That is, the terminal can receive reference signals based on the sequence indicated by the configuration information through the resources indicated by the configuration information. Through this, the terminal can obtain reception values ​​or measurement values ​​for the reference signals. The reception values ​​or measurement values ​​for the reference signals can be understood as measurement information for the channel, i.e., channel information. Here, the channel information can be expressed in the time-frequency domain or in the angular-delay domain.

[0322] In step S3205, the terminal generates CSI feedback information. According to various embodiments, the CSI feedback information may be generated by at least one encoding operation on channel information expressed in an image format. Here, at least one encoding operation includes an inference operation using a trained neural network and is performed based on information related to the temporal correlation of the channel (e.g., Doppler information), thereby generating a value indicating the channel (hereinafter, "channel value"). In addition, at least one encoding operation may include entropy coding.

[0323] Since the channel value is generated based on the temporal correlation, information related to the temporal correlation is required to restore the channel information from the channel value. Therefore, according to one embodiment, the CSI feedback information may further include a value indicating the temporal correlation (hereinafter referred to as the "temporal correlation value") in addition to the channel value. Therefore, according to various embodiments, the CSI feedback information may include the temporal correlation value in addition to the channel value, and further, according to various embodiments, the CSI feedback information may include a side value corresponding to the channel value and a side value corresponding to the temporal correlation value. Here, the side value includes information about the hyperprior for the channel value or the temporal correlation value. In the present disclosure, the channel value, the temporal correlation value, and the side values ​​may be referred to as expression values ​​since they represent a transmission signal for transmitting channel information.

[0324] In step S3207, the terminal performs an operation for fine-tuning the CSI feedback information. According to various embodiments, the terminal may determine information (hereinafter, “fine-tuning assistance information”) for adjusting representation values ​​including at least one of a channel value, a temporal correlation value, and at least one side value from the perspective of a rate-distortion trade-off. For example, the terminal may determine a step size value for fine-tuning as fine-tuning assistance information for at least one of the values ​​included in the CSI feedback information generated in step S3205. In addition, the terminal may include the fine-tuning assistance information in the CSI feedback information. That is, the operation for fine-tuning the CSI feedback information includes an operation for determining fine-tuning assistance information for representation values ​​used to restore a channel instance (e.g., at least one of a channel value, a temporal correlation value, a side value corresponding to the channel value, and a side value corresponding to the temporal correlation value) and an operation for including the determined fine-tuning assistance information in the CSI feedback information.

[0325] In step S3209, the terminal transmits CSI feedback information. The terminal may transmit the CSI feedback information based on the configuration information received in step S3201. The CSI feedback information includes various feedback values, including the values ​​generated in step S3205 and fine-tuning assistance information (e.g., step size values) generated in step S3207. According to various embodiments, the various feedback values ​​may be transmitted through a single message or through different messages. Here, the different messages may be transmitted at different times, at different cycles, and according to different mechanisms.

[0326] As described with reference to FIG. 32, CSI feedback information may be transmitted. Although not illustrated in FIG. 32, at least one additional operation may be performed to determine a model to be used in the CSI feedback framework. For example, the model may be determined based on the aforementioned rate-distortion tradeoff hyperparameters. For example, the terminal may exchange information related to the model through a capability negotiation procedure and perform the procedure of FIG. 32 using the determined model. The determined model may be indicated through the capability negotiation procedure or another procedure. For example, the determined model may be indicated through the configuration information of step S3201 of FIG. 32.

[0327] FIG. 33 illustrates an example of a procedure for obtaining CSI according to one embodiment of the present disclosure. FIG. 33 illustrates a method performed by a base station.

[0328] Referring to FIG. 33, in step S3301, the base station transmits configuration information related to CSI feedback. The configuration information may include at least one of information related to reference signals transmitted for channel measurement (e.g., resources, sequences, etc.), information related to channel measurement operations, and information related to feedback (e.g., formats, resources, feedback counts, cycles, etc.). According to various embodiments, the configuration information may be generated based on temporal correlation and may indicate CSI feedback having a variable transmission rate. Specifically, according to various embodiments, the configuration information may indicate feedback of values ​​representing a channel, for example, latent vectors, and information related to temporal correlation (e.g., Doppler information). Additionally, the configuration information may indicate feedback of values ​​for fine-tuning the values ​​representing the channel.

[0329] In step S3303, the base station transmits reference signals. The base station transmits reference signals based on the configuration information. That is, the base station can transmit reference signals based on the sequence indicated by the configuration information through the resources indicated by the configuration information.

[0330] In step S3305, the base station receives CSI feedback information. That is, the base station receives CSI feedback information generated based on the transmitted reference signals. According to various embodiments, the CSI feedback information may include a value indicating a channel generated by at least one encoding operation for channel information expressed in an image format, i.e., a channel value. Here, the channel value may be generated by an encoding operation performed based on information related to the temporal correlation of the channel (e.g., Doppler information). Furthermore, according to one embodiment, the CSI feedback information may further include a value indicating the temporal correlation, i.e., a temporal correlation value, in addition to the channel value. Furthermore, the CSI feedback information may include a side value corresponding to the channel value and a side value corresponding to the temporal correlation value. Here, the side value includes information about a hyperprior for the channel value or the temporal correlation value. In addition, the CSI feedback information may further include fine-tuning auxiliary information for fine-tuning the expression values ​​such as the channel value, the temporal correlation value, and the side value. According to various embodiments, various feedback values, including expression values ​​and fine-tuning assistance information (e.g., step size values), may be received via a single message or via different messages. Here, the different messages may be received at different times, at different cycles, and using different mechanisms.

[0331] In step S3307, the base station performs an operation for fine-tuning the CSI feedback information. Specifically, the base station modifies at least one of the representation values ​​included in the CSI feedback information using the fine-tuning assistance information included in the CSI feedback information. The fine-tuning assistance information includes a step size value for a potential shift for each representation value. Accordingly, the base station can modify at least one of the received representation values ​​based on at least one step size value.

[0332] In step S3309, the base station acquires channel information. In other words, the base station restores the channel information based on at least one of a channel value and / or a temporal correlation value modified based on the fine-tuning assistance information. Specifically, according to one embodiment, the base station acquires Doppler information of the channel based on the modified temporal correlation value, acquires a channel potential vector based on the Doppler information and the modified channel value, and then determines the channel information based on the channel potential vector. To this end, the base station may perform at least one decoding operation, and the decoding operation may include an inference operation using a trained neural network.

[0333] As described with reference to FIG. 33, channel information can be recovered from CSI feedback information. Although not illustrated in FIG. 33, at least one additional operation may be performed to determine a model to be used in the CSI feedback framework. For example, the model may be determined based on the aforementioned rate-distortion tradeoff hyperparameters. For example, the base station may exchange information related to the model through a capability negotiation procedure and perform the procedure of FIG. 33 using the determined model. The determined model may be indicated through the capability negotiation procedure or another procedure. For example, the determined model may be indicated through the configuration information of step S3301 of FIG. 33.

[0334] FIG. 34 illustrates an example of a procedure for obtaining CSI, including fine-tuning, according to one embodiment of the present disclosure. FIG. 34 illustrates an example of CSI feedback. The CSI feedback procedure according to various embodiments does not necessarily include all of the illustrated operations and may include only a portion of the procedures described below. For example, the order of signaling may vary, some signaling may be omitted, or additional signaling may be performed beyond what is described. In FIG. 34 , the terminal (3410) is an encoder-side device, including an encoder neural network, and the base station (3420) is a decoder-side device, including a decoder neural network. Furthermore, the terminal (3410) may further include a decoder neural network.

[0335] Referring to Figure 34, in step S3401, the terminal (3410) sends a Transmits UE capability information. For example, UE capability information may be a hyperparameter Includes information related to, and specifically, information that can be supported by the terminal (3410). It can indicate at least one of the maximum value, candidate values, and range of values.

[0336] In step S3403, the base station (3420) determines a model for CSI feedback. For example, the base station (3420) may determine the model based on at least one of the downlink traffic load to the terminal (3410) and the uplink resources.

[0337] In step S3405, the base station (3420) transmits the determined model ID or model pairing ID to the terminal (3410). Accordingly, the terminal (3410) can confirm the determined model. Furthermore, the terminal (3410) can configure the CSI encoder.

[0338] In step S3407, the base station (3420) transmits a CSI-RS to the terminal (3410). Prior to this, the base station (3420) may transmit configuration information related to the CSI-RS to the terminal (3410).

[0339] In step S3409, the terminal (3410) transmits Doppler side potential information and a related step size value to the base station (3420). To this end, the terminal (3410) can determine an optimal step size for the Doppler side potential information. Furthermore, the base station (3420) can fine-tune the Doppler side potential information using the step size value. Specifically, the base station (3420) can determine an adjustment amount based on the step size value and modify the value of the Doppler side potential information based on the adjustment amount. In step 3411, the terminal (3410) transmits the Doppler side potential information to the base station (3420).

[0340] In step S3413, the terminal (3410) transmits channel-side potential information and a related step size value to the base station (3420). To this end, the terminal (3410) can determine an optimal step size for the channel-side potential information. Furthermore, the base station (3420) can fine-tune the channel-side potential information using the step size value. Specifically, the base station (3420) can determine an adjustment amount based on the step size value and modify the value of the channel-side potential information based on the adjustment amount.

[0341] In step S3415, the terminal (3410) transmits channel potential information and a related step size value to the base station (3420). To this end, the terminal (3410) can determine an optimal step size for the channel potential information. Furthermore, the base station (3420) can perform fine-tuning on the channel potential information using the step size value. Specifically, the base station (3420) can determine an adjustment amount based on the step size value and modify the value of the channel potential information based on the adjustment amount.

[0342] In step S3417, the base station (3420) restores the channel instance. Specifically, the base station (3420) can restore channel information, i.e., the channel instance, using the modified Doppler side potential information, the received Doppler side potential information, the modified channel side potential information, and the modified channel potential information.

[0343] The procedure described with reference to Figure 34 includes external and internal functions. In Figure 34, dashed arrows represent signaling for external functions, and solid arrows represent signaling for internal functions. In other words, multiple internal function operations can be performed for a single external function operation.

[0344] According to the procedure illustrated in FIG. 34, in step 3411, the terminal (3410) transmits Doppler potential information to the base station (3420). However, according to another embodiment, the terminal (3410) may transmit a step size value related to the Doppler potential information in addition to the Doppler potential information. That is, although the embodiment of FIG. 34 excludes the potential shift for the Doppler potential information, according to another embodiment, the potential shift for the Doppler potential information may be performed. Furthermore, according to another embodiment, the potential shift for at least one of the Doppler side potential information, the channel side potential information, and the channel side potential information may be excluded, and in this case, the signaling operation of the step size value related to the corresponding information may be excluded.

[0345] Optimal step size information (e.g. , , ) can be appropriately converted into a bit string to be signaled via digital communication. At this time, the step size information can be quantized using a predefined index according to the range of values. The range of the step size can be set to a linear scale or a log scale. An example of a quantization rule for the optimal step size is as follows [Table 5].

[0346] range of step size(quantization interval)representative value......(0.00001, 0.0001]0.000316(0.0001, 0.001]0.00316(0.001, 0.01]0.0316(0.01, 0.1]0.316......

[0347] The examples in [Table 4] are merely examples, and the quantization rules may be defined differently according to various embodiments, and the scope of the rights is not limited to the format of [Table 4]. For example, [Table 4] uses a logarithmic scale for interval setting, but a linear scale may be used. Quantization rules such as [Table 4] must be agreed upon in advance between the terminal and the base station, and the quantization rules for the optimal step size are , , and It can be defined differently depending on the situation.

[0348] According to the various embodiments described above, it is possible to adjust the channel feedback model to account for downlink traffic load and / or uplink resource conditions. Furthermore, by fine-tuning the signal transmitted from the encoder to the decoder from a rate-distortion perspective, it is possible to further reduce the transmission rate and improve channel restoration performance.

[0349] It is clear that the examples of the proposed methods described above can also be considered as a type of proposed methods, as they can be included as one of the implementation methods of the present disclosure. Furthermore, the proposed methods described above can be implemented independently, but they can also be implemented in the form of a combination (or merge) of some of the proposed methods. Information regarding the applicability of the proposed methods (or information regarding the rules of the proposed methods) can be defined by a rule such that the base station notifies the terminal of the application of the proposed methods through a predefined signal (e.g., a physical layer signal or a higher layer signal).

[0350] The present disclosure may be embodied in other specific forms without departing from the technical ideas and essential features described herein. Therefore, the above detailed description should not be construed as limiting in all respects but rather as illustrative. The scope of the present disclosure should be determined by a reasonable interpretation of the appended claims, and all modifications within the equivalent scope of the present disclosure are intended to be included within the scope of the present disclosure. Furthermore, claims that are not explicitly cited in the claims may be combined to form an embodiment or incorporated into a new claim through a post-filing amendment.

[0351] Embodiments of the present disclosure can be applied to various wireless access systems. Examples of various wireless access systems include the 3rd Generation Partnership Project (3GPP) or 3GPP2 systems.

[0352] The embodiments of the present disclosure can be applied not only to the various wireless access systems described above, but also to all technical fields that utilize these various wireless access systems. Furthermore, the proposed method can also be applied to mmWave and THz communication systems utilizing ultra-high frequency bands.

[0353] Additionally, embodiments of the present disclosure can be applied to various applications such as autonomous vehicles and drones.

Claims

1. A method performed by a terminal in a wireless communication system, A step of receiving configuration information related to CSI (channel state information) feedback; A step of receiving reference signals based on the above setting information; A step of generating CSI feedback information based on the above reference signals; and Comprising a step of transmitting the above CSI feedback information, A method wherein the CSI feedback information includes channel information determined based on the reference signals, information related to temporal correlation of the channel information, and information for fine tuning of the channel information and information related to temporal correlation.

2. In claim 1, Further comprising a step of transmitting capability information related to the above CSI feedback, A method wherein the above capability information includes information for determining a rate-distortion trade-off hyperparameter included in a loss function of a model for the CSI feedback.

3. In claim 1, A method wherein the above configuration information includes model identification information corresponding to a transmission rate-distortion trade-off hyperparameter included in a loss function of a model for the CSI feedback, or information related to a maximum payload size determined based on the transmission rate-distortion trade-off hyperparameter.

4. In claim 1, The above CSI feedback information includes Part 1 CSI and Part 2 CSI, The above Part 1 CSI method includes information related to the rate-distortion tradeoff hyperparameter.

5. In claim 1, A method wherein the information for the fine-tuning includes information related to a step size for a latent shift of at least one of the channel information, the temporal correlation information, and side information for the channel information.

6. In claim 1, The step of transmitting the above CSI feedback information is: A method comprising the steps of sequentially transmitting Doppler side potential information, Doppler potential information, channel side potential information, and channel potential information.

7. In claim 1, A method wherein the channel information includes the result of entropy coding for a latent vector of a first channel instance measured using the reference signals.

8. In claim 7, A method in which the above CSI feedback information includes hierarchical prior information determined based on the latent vector of the first channel instance as side information.

9. In claim 7, A method in which the above entropy coding is performed using an entropy model determined based on a hierarchical prior related to a channel determined based on a latent vector of the first channel instance and a temporal prior determined based on Doppler information related to the channel.

10. In claim 9, A method wherein the temporal prior is determined based on context information, which is determined based on the channel instance reconstructed from the Doppler information and the latent vector of the second channel instance measured at a previous point in time.

11. In claim 1, A method wherein information related to the above temporal correlation comprises the result of entropy coding for a latent vector of Doppler information estimated based on the first channel instance.

12. In claim 11, A method further comprising the step of estimating the Doppler information based on the channel instance restored from the latent vector of the second channel instance measured at a previous point in time and the first channel instance measured at the current point in time.

13. In claim 11, A method wherein the CSI feedback information includes hierarchical prior information related to a latent vector of the Doppler information.

14. In claim 1, The step of generating the above CSI feedback information is: A step for estimating Doppler information for a channel; A step of generating a latent vector of the above Doppler information; A step of generating context information based on a latent vector of the above Doppler information; A step of generating a latent vector of a first channel instance measured using the above reference signals, based on the context information; and A method comprising the steps of performing quantization and entropy coding on a latent vector of the first channel instance based on the context information.

15. In claim 14, A method wherein the first channel instance comprises information represented in one of the following forms: an image format in the angular-delay domain, a 3D tensor format in the 2D space-1D frequency domain, or a tensor format in the 3D angular-delay domain consisting of an elevation angle, an azimuth angle, and taps of delay.

16. In a method of operating a base station in a wireless communication system, A step of transmitting configuration information related to CSI (channel state information) feedback; A step of transmitting reference signals based on the above setting information; A step of receiving CSI feedback information corresponding to the above reference signals; and A step of restoring channel information based on the above CSI feedback information is included. A method wherein the CSI feedback information includes channel information determined based on the reference signals, information related to temporal correlation of the channel information, and information for fine tuning of the channel information and information related to temporal correlation.

17. In claim 16, A method wherein the information for the fine-tuning includes information related to a step size for a latent shift of at least one of the channel information, the temporal correlation information, and side information for the channel information.

18. In claim 17, The steps to restore the above channel information are: A method comprising the step of performing a latent shift for at least one of the channel information, the temporal correlation information, and side information for the channel information based on information related to the step size.

19. In a wireless communication system, at a terminal, Transmitter and receiver; and comprising a processor connected to the above transceiver, The above processor, Receive configuration information related to CSI (channel state information) feedback, Receive reference signals based on the above setting information, Generate CSI feedback information based on the above reference signals, Controls to transmit the above CSI feedback information, A terminal wherein the CSI feedback information includes channel information determined based on the reference signals, information related to temporal correlation of the channel information, and information for fine tuning of the channel information and information related to temporal correlation.

20. In a base station in a wireless communication system, Transmitter and receiver; and comprising a processor connected to the above transceiver, The above processor, Transmit configuration information related to CSI (channel state information) feedback, Transmit reference signals based on the above setting information, Receive CSI feedback information corresponding to the above reference signals, Control to restore channel information based on the above CSI feedback information, A base station, wherein the CSI feedback information includes channel information determined based on the reference signals, information related to temporal correlation of the channel information, and information for fine tuning of the channel information and information related to temporal correlation.

21. In communication devices, At least one processor; At least one computer memory coupled to said at least one processor and storing instructions that direct operations when executed by said at least one processor, The above actions are, A step of receiving configuration information related to CSI (channel state information) feedback; A step of receiving reference signals based on the above setting information; A step of generating CSI feedback information based on the above reference signals; and Comprising a step of transmitting the above CSI feedback information, A communication device, wherein the CSI feedback information includes channel information determined based on the reference signals, information related to temporal correlation of the channel information, and information for fine tuning of the channel information and information related to temporal correlation.

22. In a non-transitory computer-readable medium storing at least one instruction, comprising at least one instruction executable by the processor; At least one of the above commands causes the device to: Receive configuration information related to CSI (channel state information) feedback, Receive reference signals based on the above setting information, Generate CSI feedback information based on the above reference signals, Controls to transmit the above CSI feedback information, A non-transitory computer-readable medium comprising: channel information determined based on the reference signals; information related to temporal correlation of the channel information; and information for fine tuning of the channel information and information related to temporal correlation.

Images