Communication method and apparatus

Abstract

The present disclosure relates to a communication method and an apparatus. A terminal device obtains M first downlink channel subdata, each of which corresponds to one data space among the M data spaces. For an i-th first downlink channel subdata among the M first downlink channel subdata, the terminal device determines first information corresponding to the i-th first downlink channel subdata based on a first dictionary corresponding to the i-th data space, and M first information is determined in total. The terminal device transmits first indication information to indicate the M first information. Different data spaces can represent different channel environment information. The terminal device feeds back the first information corresponding to the different data spaces, so that the access network device can determine a correspondence relationship between the first information and the environment information. In this case, the first information fed back by the terminal device can reflect an actual communication environment, and the accuracy of the first information fed back by the terminal device is improved.

Description

[Technical Field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims priority to Chinese Patent Application No. 202111663303.8, entitled "Communication Method and Apparatus," filed with the State Intellectual Property Office of China on December 31, 2021, the entire contents of which are incorporated herein by reference.

[0002] The present disclosure relates to the field of communication technologies, and more particularly to communication methods and devices. [Background technology]

[0003] The 5th generation (5G) mobile communication system has higher requirements for system capacity and spectral efficiency. In a 5G communication system, the application of massive multiple-input multiple-output (massive-MIMO) technology plays an important role in improving the system's spectral efficiency. Base stations can use massive-MIMO technology to simultaneously provide high-quality services to a larger number of user equipment (UEs). The key step is for the base station to precode downlink data for multiple UEs. Precoding enables spatial multiplexing, reduces interference between UEs, increases the signal-to-interference-plus-noise ratio (SINR) at the receiver, and improves system throughput. To more accurately precode the UE's downlink data, the base station may acquire channel state information (CSI) for the downlink channel, reconstruct the downlink channel based on the CSI, and use the reconstructed downlink channel to determine a precoding matrix and perform precoding. Therefore, how to make the CSI fed back by the UE more accurate is a technical problem worth studying. Summary of the Invention

[0004] The present disclosure provides a communication method and apparatus for improving the accuracy of CSI fed back by a UE. [Means for solving the problem]

[0005] According to a first aspect, a first communication method is provided. The method can be performed on a terminal device side. The method can be performed using software, hardware, or a combination of software and hardware. For example, the method can be performed by a terminal device, a circuit system, or a larger device including a terminal device. The circuit system can perform the functions of the terminal device. The method includes the steps of: acquiring M pieces of first downlink channel subdata, each of which corresponds to a data space among the M data spaces, where M is an integer greater than 1; determining, for an ith first downlink channel subdata among the M pieces of first downlink channel subdata, first information corresponding to the ith first downlink channel subdata based on a first dictionary corresponding to the ith data space among the M data spaces, where a total of M pieces of first information are determined, where i is an integer from 1 to M, the ith first downlink channel subdata corresponds to the ith data space, the first dictionary includes a plurality of elements, and the first information corresponding to the ith first downlink channel subdata corresponds to P elements among the plurality of elements, where P is a positive integer; and transmitting first indication information, where the first indication information indicates the M pieces of first information.

[0006] In the present disclosure, each of the M first downlink channel subdata acquired by the terminal device may correspond to one data space among the M data spaces, and the first information corresponding to all the first downlink channel subdata may be determined based on dictionaries corresponding to different data spaces. Different data spaces may represent different location information, i.e., different data spaces may represent different channel environment information. The terminal device feeds back the first information corresponding to the different data spaces, thereby allowing the access network device to determine a correspondence between the first information and the environment information. In this case, the first information fed back by the terminal device may reflect the actual communication environment, and the accuracy of the first information fed back by the terminal device may be improved. The access network device may reconstruct an accurate downlink channel based on the first information fed back by the terminal device.

[0007] In an optional embodiment, the first instruction information indicates M identifiers of the first information, and the step of transmitting the first information includes the step of transmitting the M identifiers of the first information in a first order, where the first order is the arrangement order of the M data spaces. The first order specifies that the terminal device first transmits the identifier of the first information corresponding to a specific data space, and then transmits the identifier of the first information corresponding to a specific data space. The first order is known to the terminal device and the access network device. Therefore, after receiving the M identifiers of the first information, the access network device can further determine the correspondence between the data spaces and the identifiers of the first information to avoid correspondence errors.

[0008] In optional embodiments, the first order may be a predefined order, or second indication information may be received, with the second indication information indicating the first order, or the first order may be determined, and third indication information may be transmitted, with the third indication information indicating the first order. For example, the first order may be a predefined order in a protocol, and the terminal device and the access network device may determine the first order according to the protocol. Alternatively, the first order may be preconfigured in the terminal device and the access network device. Alternatively, the first order may be determined by the access network device. After determining the first order, the access network device may send second indication information to the terminal device, which allows the terminal device to determine the first order based on the second indication information. Alternatively, the first order may be determined by the terminal device. After determining the first order, the terminal device may send third indication information to the access network device, which allows the access network device to determine the first order based on the third indication information. It may be noted that the method of determining the first order is flexible.

[0009] In an optional embodiment, the M first downlink channel subdata are obtained based on the first downlink channel data, where the first downlink channel data is a preprocessing result, or the first downlink channel data includes F consecutive data columns in the preprocessing result, or the first downlink channel data is compressed information obtained by compressing the preprocessing result, and the preprocessing result is obtained by preprocessing the second downlink channel data. The preprocessing result of the second downlink channel data may be directly used as the first downlink channel data, and there is no need to perform excessive processing on the preprocessing result. This is simple. Alternatively, it is considered that in the frequency domain (delay domain), energy is generally concentrated mainly around delay=0, and energy in other domains can be essentially ignored. Therefore, the terminal device may select F consecutive columns on both sides of delay=0 as the first downlink channel data, and the coefficients of the remaining parts may be set to 0 by default. In this case, the complexity of processing the first downlink channel data may be reduced. Alternatively, the pre-processing result may be compressed to obtain the first downlink channel data, thereby reducing the complexity of processing the first downlink channel data. The process of pre-processing the downlink channel data may include, for example, performing space-frequency joint projection on the downlink channel data.

[0010] In optional embodiments, the division method of the M data spaces is predefined; fourth indication information is received, and the fourth indication information indicates the division method of the M data spaces; or the division method of the M data spaces is determined, and fifth indication information is sent, and the fifth indication information indicates the division method of the M data spaces. For example, if the division method of the M data spaces is predefined in a protocol, both the terminal device and the access network device may determine the division method of the M data spaces according to the protocol. Alternatively, the division method of the M data spaces is determined by the access network device. The access network device may send the fourth indication information to the terminal device, thereby allowing the terminal device to determine the division method of the M data spaces based on the fourth indication information. Alternatively, the division method of the M data spaces may be determined by the UE. The UE may send the fifth indication information to the access network device, thereby allowing the access network device to determine the division method of the M data spaces based on the fifth indication information. It may be known that the method of dividing the data spaces is flexible.

[0011] According to a second aspect, a second communication method is provided. The method may be executed on the access network device side. The method may be executed using software, hardware, or a combination of software and hardware. For example, the method may be executed by an access network device, a larger device including an access network device, or a circuit system. The circuit system may perform the functions of the access network device. Alternatively, the method may be executed by an access network device or a network element of the access network device with the assistance of an AI module independent of the access network device. This is not limited to this. For example, the access network device may be an access network device, such as a base station. The method includes the steps of: receiving first instruction information, where the first instruction information indicates M pieces of first information, where M is an integer greater than 1; reconstructing i-th second downlink channel subdata, for an i-th first information among the M pieces of first information, based on a first dictionary corresponding to an i-th data space among the M data spaces, where a total of M pieces of second downlink channel subdata are obtained, the i-th first information corresponds to the i-th data space, where i is an integer from 1 to M, the first dictionary includes a plurality of elements, and the first information corresponding to the i-th second downlink channel subdata corresponds to P elements among the plurality of elements; and reconstructing downlink channel information based on the M pieces of second downlink channel subdata.

[0012] In an optional embodiment, the step of receiving the first instruction information includes a step of receiving identifiers of M pieces of first information in a first order, the first order being the arrangement order of the M pieces of data space.

[0013] In optional embodiments, the first order is a predefined order, or second instruction information is sent and the second instruction information indicates the first order, or third instruction information is received and the third instruction information indicates the first order.

[0014] In an optional embodiment, the M data spaces correspond to M dictionaries, and each data space corresponds to one dictionary, or all of the M data spaces correspond to the same dictionary, or the number of dictionaries corresponding to the M data spaces is greater than 1 and less than M. In other words, the data spaces correspond one-to-one to the dictionaries, which can improve the accuracy of the first information determined based on the dictionaries; all data spaces can correspond uniformly to one dictionary, which can make the samples used for training to obtain the dictionaries richer, which can make the contents included in the dictionaries more detailed; or the number of dictionaries corresponding to the data spaces can be less than the number of data spaces; for example, one dictionary can correspond to multiple data spaces, which can reduce complexity to a certain extent.

[0015] In an optional embodiment, the step of reconstructing downlink channel information based on the M pieces of second downlink channel subdata includes the steps of obtaining compressed information based on the M pieces of second downlink channel subdata, and obtaining the downlink channel information based on the compressed information.

[0016] In optional embodiments, the division method of the M data spaces is predefined, or fourth instruction information is sent, and the fourth instruction information indicates the division method of the M data spaces, or fifth instruction information is received, and the fifth instruction information indicates the division method of the M data spaces.

[0017] For the technical effects provided by the second aspect or various optional implementations of the second aspect, please refer to the description of the technical effects of the first aspect or corresponding implementations.

[0018] According to a third aspect, a communication device is provided. The communication device may perform the method according to the first aspect. The communication device has the functionality of a terminal device. In an optional embodiment, the device may include modules corresponding to one another for performing the methods / operations / steps / actions described in the first aspect. The modules may be hardware circuits, software, or may be implemented by hardware circuits in combination with software. In an optional embodiment, the communication device includes a baseband device and a radio frequency device. In another optional embodiment, the communication device includes a processing unit (sometimes referred to as a processing module) and a transceiver unit (sometimes referred to as a transceiver module). The transceiver unit may perform transmitting and receiving functions. When the transceiver unit performs transmitting functions, the transceiver unit may be referred to as a transmitting unit (sometimes referred to as a transmitting module). When the transceiver unit performs receiving functions, the transceiver unit may be referred to as a receiving unit (sometimes referred to as a receiving module). The transmitting unit and the receiving unit may be the same functional module, and this functional module is referred to as the transceiver unit. This functional module may perform transmitting and receiving functions, or the transmitting unit and receiving unit may be different functional modules, and the transceiver unit is a collective term for these functional modules.

[0019] The processing unit is configured to obtain M pieces of first downlink channel subdata, each of which corresponds to a data space among the M data spaces, where M is an integer greater than 1. For an ith first downlink channel subdata among the M pieces of first downlink channel subdata, the processing unit is further configured to determine first information corresponding to the ith first downlink channel subdata based on a first dictionary corresponding to the ith data space among the M data spaces, where a total of M pieces of first information are determined, where i is an integer from 1 to M, the ith first downlink channel subdata corresponds to the ith data space, the first dictionary includes a plurality of elements, and the first information corresponding to the ith first downlink channel subdata corresponds to P elements among the plurality of elements, where P is a positive integer. The transceiver unit is configured to transmit first indication information, where the first indication information indicates the M pieces of first information.

[0020] In another example, a communication device includes a processor coupled to a memory and configured to execute instructions in the memory to perform a method according to the first aspect. Optionally, the communication device further includes other components, such as an antenna, an input / output module, and an interface. These components may be hardware, software, or a combination of software and hardware.

[0021] According to a fourth aspect, a communication device is provided. The communication device may perform the method according to the second aspect. The communication device has functionality of an access network device. The access network device is, for example, a base station or a baseband device in a base station. In an optional embodiment, the device may include modules corresponding to one another for performing the methods / operations / steps / actions described in the second aspect. The modules may be hardware circuits, or software, or may be implemented by hardware circuits in combination with software. In an optional embodiment, the communication device includes a baseband device and a radio frequency device. In another optional embodiment, the communication device includes a processing unit (sometimes referred to as a processing module) and a transceiver unit (sometimes referred to as a transceiver module). For an embodiment of the transceiver unit, please refer to the relevant description of the third aspect.

[0022] The transceiver unit is configured to receive first indication information, where the first indication information indicates M pieces of first information, where M is an integer greater than 1. For an ith first information among the M pieces of first information, the processing unit is configured to reconstruct an ith second downlink channel subdata based on a first dictionary corresponding to the ith data space among the M data spaces, so that a total of M pieces of second downlink channel subdata are obtained, where the ith first information corresponds to the ith data space, where i is an integer from 1 to M, the first dictionary includes a plurality of elements, and the first information corresponding to the ith second downlink channel subdata corresponds to P elements among the plurality of elements. The processing unit is further configured to reconstruct downlink channel information based on the M pieces of second downlink channel subdata.

[0023] In another example, a communication device includes a processor coupled to a memory and configured to execute instructions in the memory to perform a method according to the second aspect. Optionally, the communication device further includes other components, such as an antenna, an input / output module, and an interface. These components may be hardware, software, or a combination of software and hardware.

[0024] According to a fifth aspect, there is provided a computer-readable storage medium configured to store computer programs or instructions which, when run, perform the methods according to the first and / or second aspects.

[0025] According to a sixth aspect, there is provided a computer program product comprising instructions, which when run on a computer, perform the method according to the first aspect and / or the second aspect.

[0026] According to a seventh aspect, there is provided a chip system. The chip system includes a processor and may further include a memory, and the chip system is configured to perform the method according to the first aspect and / or the second aspect. The chip system may include a chip, or may include a chip and another individual component.

[0027] According to an eighth aspect, there is provided a communication system including a communication device according to the third aspect and a communication device according to the fourth aspect. [Brief explanation of the drawings]

[0028] [Figure 1] FIG. 1 is a diagram of a communication system. [Figure 2] 1 is a flowchart of a CSI feedback mechanism. [Figure 3] 1 is a diagram of an application scenario. [Figure 4A]A diagram of an application framework for AI in a communications system. [Figure 4B] A diagram of an application framework for AI in a communications system. [Figure 4C] A diagram of an application framework for AI in a communications system. [Figure 4D] A diagram of an application framework for AI in a communications system. [Figure 4E] A diagram of an application framework for AI in a communications system. [Figure 5] 1 is a flowchart of a communication method. [Figure 6] A diagram of a dictionary. [Figure 7] FIG. 10 is a diagram of a communication method used when both the UE and the access network device process compressed information. [Figure 8] 10 is a flowchart of another communication method. [Figure 9A] FIG. 1 is a diagram of the network training and inference stages. [Figure 9B] FIG. 1 is a diagram of the network training phase. [Figure 9C] FIG. 1 is a diagram of the network training phase. [Figure 9D] FIG. 1 is a diagram of the network training phase. [Figure 10] 10 is a flowchart of yet another communication method. [Figure 11] FIG. 1 is another diagram of the network training and inference stages. [Figure 12] FIG. 1 is a block diagram of a communication device. DETAILED DESCRIPTION OF THE INVENTION

[0029] To make the objectives, technical solutions, and advantages of the present disclosure clearer, the following further describes the present disclosure in detail with reference to the accompanying drawings.

[0030] The techniques provided in this disclosure may be applied to a communication system 10 shown in FIG. 1 . The communication system 10 includes one or more communication devices 30 (e.g., terminal devices). The one or more communication devices 30 are connected to one or more core network (CN) devices via one or more access network (radio access network) (RAN) devices 20 to facilitate communication between the communication devices. For example, the communication system 10 may be, but is not limited to, a communication system supporting the 4th generation (4G) (including long term evolution (LTE)) access technology, a communication system supporting 5G (sometimes referred to as new radio (NR)) access technology, a wireless fidelity (Wi-Fi) system, a cellular system associated with the 3rd generation partnership project (3GPP®), a communication system supporting a collection of multiple radio technologies, or a future-oriented evolution system.

[0031] The following provides a detailed description of the terminal device and the RAN in FIG. 1 separately.

[0032] 1. Terminal Device A terminal device may simply be referred to as a terminal. A terminal device may be a device with wireless transceiver functionality. A terminal device may be mobile or fixed. A terminal device may be located on land, including indoors or outdoors, or handheld or vehicle-mounted, on water (e.g., a ship), or in the air (e.g., an aircraft, balloon, or satellite). A terminal device may include a mobile phone, a tablet computer, a computer with wireless transceiver functionality, a virtual reality (VR) terminal device, an augmented reality (AR) terminal device, an industrial control wireless terminal device, a self-driving wireless terminal device, a remote medical wireless terminal device, a smart grid wireless terminal device, a transportation safety wireless terminal device, a smart city wireless terminal device, and / or a smart home wireless terminal device. Alternatively, the terminal device may be a cellular phone, a cordless phone, a session initiation protocol (SIP) phone, a wireless local loop (WLL) station, a personal digital assistant (PDA), a handheld or computing device with wireless communication capabilities, an in-vehicle device, a wearable device, a terminal device in the 5th generation (5G) network, or a terminal device in a further evolved public land mobile network (PLMN), etc. The terminal device may also be referred to as user equipment (UE).Optionally, a terminal device may communicate with multiple access network devices using different technologies. For example, a terminal device may communicate with an access network device supporting LTE, or may communicate with an access network device supporting 5G, or may implement dual connectivity to an access network device supporting LTE and an access network device supporting 5G. This is not a limitation of the present disclosure.

[0033] In the present disclosure, an apparatus configured to perform the functions of a terminal device may be a terminal device, or may be an apparatus capable of supporting a terminal device in performing its functions, such as a chip system, a hardware circuit, a software module, or a hardware circuit and a software module. The apparatus may be incorporated into a terminal device or adapted for use in a terminal device. In the technical solutions provided in the present disclosure, an example in which an apparatus configured to perform the functions of a terminal device is a terminal device and the terminal device is a UE is used to describe the technical solutions provided in the present disclosure.

[0034] In the present disclosure, a chip system may include a chip, or may include a chip and other separate components.

[0035] 2.RAN The RAN may include one or more RAN devices, such as RAN device 20. The interface between the RAN device and the terminal device may be a Uu interface (also called an air interface). In future communications, the names of these interfaces may remain unchanged or may be replaced by other names. This is not a limitation of the present disclosure.

[0036] A RAN device is a node or device that enables a terminal device to access a wireless network. RAN devices may also be referred to as network devices or base stations. Examples of RAN devices include, but are not limited to, a base station, a 5G generation NodeB (gNB), an evolved NodeB (eNB), a radio network controller (RNC), a NodeB (NB), a base station controller (BSC), a base transceiver station (BTS), a home base station (e.g., a home evolved NodeB or home NodeB (HNB)), a baseband unit (BBU), a transmitting and receiving point (TRP), a transmitting point (TP), and / or a mobile switching center. Alternatively, the access network device may be at least one of a central unit (CU), a distributed unit (DU), a central unit control plane (CU-CP) node, a central unit user plane (CU-UP) node, an integrated access and backhaul (IAB), or a radio controller in a cloud radio access network (CRAN) scenario, etc. Alternatively, the access network device may be a relay station, an access point, an in-vehicle device, a terminal device, a wearable device, an access network device in a 5G network, or an access network device in a future evolved public land mobile network (PLMN), etc.

[0037] In the present disclosure, an apparatus configured to perform the functions of an access network device may be an access network device, or may be an apparatus capable of supporting an access network device in performing its functions, such as a chip system, a hardware circuit, a software module, or a hardware circuit and a software module. The apparatus may be incorporated into an access network device or adapted for use in an access network device. In the technical solutions provided in the present disclosure, an example in which an apparatus configured to perform the functions of an access network device is an access network device, and the access network device is a base station, is used to describe the technical solutions provided in the present disclosure.

[0038] (1) Protocol layer structure Communications between the access network device and the terminal device conform to a specified protocol layer structure. The protocol layer structure may include a control plane protocol layer structure and a user plane protocol layer structure. For example, the control plane protocol layer structure may include at least one of the following: a radio resource control (RRC) layer, a packet data convergence protocol (PDCP) layer, a radio link control (RLC) layer, a media access control (MAC) layer, or a physical layer (PHY). For example, the user plane protocol layer structure may include at least one of the following: a service data adaptation protocol (SDAP) layer, a PDCP layer, an RLC layer, a MAC layer, and a physical layer.

[0039] The protocol layer structure between the access network device and the terminal device may be considered an access stratum (AS) structure. Optionally, a non-access stratum (NAS) may further exist above the AS and is used by the access network device to forward information from the core network device to the terminal device, or used by the access network device to forward information from the terminal device to the core network device. In this case, it may be considered that there is a logical interface between the terminal device and the core network device. Optionally, the access network device may forward information between the terminal device and the core network device in a transparent transmission manner. For example, NAS messages may be mapped to or included in RRC signaling as elements of the RRC signaling.

[0040] Optionally, the protocol layer structure between the access network device and the terminal device may further include an artificial intelligence (AI) layer configured to transmit data related to AI functions.

[0041] (2) Central Unit (CU) and Distributed Unit (DU) A RAN device may include a CU and a DU. This design may be referred to as CU and DU separation. Multiple DUs may be centrally controlled by one CU. For example, the interface between the CU and the DU may be referred to as an F1 interface. The control plane (CP) interface may be F1-C, and the user plane (UP) interface may be F1-U. The specific names of the interfaces are not limited in this disclosure. The CU and the DU may be classified according to the protocol layer of the wireless network. For example, the functions of the PDCP layer and protocol layers above the PDCP layer (such as the RRC layer and the SDAP layer) are configured in the CU, and the functions of the protocol layers below the PDCP layer (such as the RLC layer, the MAC layer, and the PHY layer) are configured in the DU. In another example, the functions of the protocol layers above the PDCP layer are configured in the CU, and the functions of the PDCP layer and protocol layers below the PDCP layer are configured in the DU.

[0042] The above-described division of processing functions of the CU and DU based on protocol layers is merely an example, and the division may alternatively be performed in other manners. For example, the CU or DU may be divided to have more protocol layer functions. In another example, the CU or DU may be divided to have some of the processing functions of the protocol layers. In one design, some of the RLC layer functions and functions of protocol layers above the RLC layer are configured in the CU, and the remaining RLC layer functions and functions of protocol layers below the RLC layer are configured in the DU. In another design, the division of functions of the CU or DU may alternatively be performed based on service type or other system requirements. For example, the division may be performed based on latency. Functions whose processing time must meet latency requirements are configured in the DU, and functions whose processing time does not need to meet latency requirements are configured in the CU.

[0043] Optionally, the CU may have one or more functions of a core network, for example, the CU may be located on the network side to facilitate centralized management.

[0044] Optionally, the radio unit (RU) of the DU is remotely located. The RU has radio frequency functions. For example, the DU and the RU may be separated at the PHY layer. For example, the DU may perform functions of an upper layer of the PHY layer, and the RU may perform functions of a lower layer of the PHY layer. When the PHY layer is used for transmission, the functions of the PHY layer may include at least one of the following functions: adding cyclic redundancy check (CRC) bits, channel coding, rate matching, scrambling, modulation, layer mapping, precoding, resource mapping, physical antenna mapping, or radio frequency transmission. When the PHY layer is used for reception, the functions of the PHY layer may include at least one of the following functions: CRC checking, channel decoding, de-rate matching, descrambling, demodulation, layer demapping, channel detection, resource demapping, physical antenna demapping, or radio frequency reception. The functions of the upper layer of the PHY layer may include some of the functions of the PHY layer, some of which are closer to the MAC layer. The functions of the lower layer of the PHY layer may include other parts of the functions of the PHY layer. For example, some of the functions are closer to radio frequency functions. For example, functions of an upper layer of the PHY layer may include CRC bit addition, channel coding, rate matching, scrambling, modulation, and layer mapping, while functions of a lower layer of the PHY layer may include precoding, resource mapping, physical antenna mapping, and radio frequency transmission functions. Alternatively, functions of an upper layer of the PHY layer may include CRC bit addition, channel coding, rate matching, scrambling, modulation, layer mapping, and precoding, while functions of a lower layer of the PHY layer may include resource mapping, physical antenna mapping, and radio frequency transmission functions. For example, functions of an upper layer of the PHY layer may include CRC checking, channel decoding, de-rate matching, decoding, demodulation, and layer demapping, while functions of a lower layer of the PHY layer may include channel detection, resource demapping, physical antenna demapping, and radio frequency reception functions.Alternatively, the functions of the upper layer of the PHY layer may include CRC checking, channel decoding, de-rate matching, decoding, demodulation, layer demapping, and channel detection, and the functions of the lower layer of the PHY layer may include resource demapping, physical antenna demapping, and radio frequency reception functions.

[0045] Optionally, the functions of the CU may be further divided. Specifically, the control plane and user plane of the CU are separated and implemented by different entities, i.e., a control plane CU entity (CU-CP entity) and a user plane CU entity (CU-UP entity). The CU-CP entity and the CU-UP entity may be separately coupled or connected to the DU to jointly perform the functions of the RAN device.

[0046] In the above architecture, signaling generated by the CU may be transmitted to the terminal device via the DU, or signaling generated by the terminal device may be transmitted to the CU via the DU. For example, signaling at the RRC layer or the PDCP layer may ultimately be processed as signaling at the physical layer and transmitted to the terminal device, or converted from signaling received at the physical layer. In this architecture, signaling at the RRC layer or the PDCP layer may be considered to be transmitted via the DU or via the DU and the RU.

[0047] Optionally, any one of the DU, CU, CU-CP, CU-UP, and RU may be a software module, a hardware structure, or a combination of a software module and a hardware structure. This is not limited thereto. Different entities may exist in different forms, this is not limited thereto. For example, the DU, CU, CU-CP, and CU-UP are software modules, and the RU is a hardware structure. These modules and methods performed by these modules also fall within the scope of protection of the present disclosure. For example, when the method of the present disclosure is performed by an access network device, the method may be specifically performed by at least one of the CU, CU-CP, CU-UP, DU, RU, or the near real-time RIC described below. The method performed by the module also falls within the scope of protection of the present disclosure.

[0048] It should be noted that, since the network devices in this disclosure are mainly access network devices, hereinafter, unless otherwise specified, "network device" may refer to "access network device."

[0049] It should be understood that the number of devices in the communication system shown in Figure 1 is used as an example only, and the present disclosure is not limited thereto. In actual applications, the communication system may further include more terminal devices and more RAN devices, or may further include other devices, such as core network devices and / or nodes configured to perform artificial intelligence functions.

[0050] The network architecture shown in FIG. 1 is applicable to communication systems of various radio access technologies (RATs), such as a 4G communication system, a 5G (also called new radio (NR)) communication system, a transition system between an LTE communication system and a 5G communication system, or a future communication system, such as a 6G communication system. The transition system may also be called a 4.5G communication system. The network architecture and service scenarios described in this disclosure are intended to more clearly explain the technical solutions in this disclosure and do not constitute limitations on the technical solutions provided in this disclosure. Those skilled in the art may know that with the development of network architectures and the emergence of new service scenarios, the technical solutions provided in this disclosure may also be applicable to similar technical problems.

[0051] In addition to communication between an access network device and a terminal device, the methods provided in the present disclosure can also be used for communication between other communication devices, such as communication between a macro base station and a micro base station in a wireless backhaul link, or communication between a first terminal device and a second terminal device in a sidelink (SL). This is not limited to this. The present disclosure will be described using communication between a network device and a terminal device as an example.

[0052] When transmitting data to a terminal device, the access network device may perform precoding based on channel state information (CSI) fed back by the terminal device. To facilitate understanding of the present disclosure, the following briefly describes some technical terms in the present disclosure.

[0053] 1. Precoding technology When channel state information is known, an access network device may process a signal to be transmitted using a precoding matrix that matches the channel conditions. The precoded signal to be transmitted can adapt to the channel using this technique, thereby improving the quality of the signal received by the terminal device (e.g., signal-to-interference plus noise ratio (SINR)) and improving system throughput. The precoding technique is used to enable a transmitting device (e.g., an access network device) and multiple receiving devices (e.g., terminal devices) to effectively perform transmission on the same time-frequency resource, i.e., effectively implementing multi-user multiple-input multiple-output (MU-MIMO). The precoding technique is used to enable a transmitting device (e.g., an access network device) and multiple receiving devices (e.g., terminal devices) to effectively perform multi-data stream transmission on the same time-frequency resource, i.e., effectively implementing single-user multiple-input multiple-output (SU-MIMO). It should be noted that the relevant description of the precoding technique is merely an example for easy understanding and is not intended to limit the scope of protection of the present disclosure. In a specific implementation process, the transmission device may alternatively perform precoding in another manner. For example, when channel information (e.g., but not limited to, a channel matrix) cannot be obtained, precoding is performed using a pre-configured precoding matrix or a weighting processing method. For brevity, the specific content thereof will not be described again in this specification.

[0054] 2. CSI feedback CSI feedback may also be referred to as a CSI report. In a wireless communication system, CSI feedback refers to a receiver (e.g., a terminal device) of data (e.g., but not limited to, data carried on a physical downlink shared channel (PDSCH)) reporting information used to describe channel attributes of a communication link to a transmitter (e.g., an access network device). For example, the CSI report includes one or more pieces of information such as a downlink channel matrix, a precoding matrix indicator (PMI), a rank indicator (RI), or a channel quality indicator (CQI). The foregoing contents included in the above-described CSI are merely examples for illustration and do not constitute any limitation on the present disclosure. The CSI may include one or more of the foregoing described contents, or may include information different from the foregoing described contents and used to represent the CSI. This is not a limitation of the present disclosure.

[0055] 3.Neural network (NN). Neural networks are a specific embodiment of machine learning technology. According to the universal approximation theorem, neural networks can theoretically approximate any continuous function, thereby giving them the ability to learn any mapping. In traditional communication systems, communication modules must be designed with extensive expertise. However, deep learning communication systems based on neural networks can automatically discover implicit pattern structures from multiple data sets, establish mapping relationships between data, and achieve better performance than traditional modeling methods.

[0056] For example, a deep neural network (DNN) is a neural network with multiple layers. Based on different network structures and / or usage scenarios, the DNN may include a multi-layer perceptron (MLP), a convolutional neural network (CNN), a recurrent neural network (RNN), etc. The specific form of the DNN is not limited in this disclosure.

[0057] 4. Auto-encoder (AE) network, or AE for short. An AE network may include an encoder and a corresponding decoder. For example, the encoder and / or decoder may be implemented using a neural network (such as a DNN). In this case, the encoder may be referred to as an encoder network, and the decoder may be referred to as a decoder network. For example, in an AE network, the encoder and the corresponding decoder may be obtained by joint training. The encoder and decoder obtained by training may be used to encode and decode information.

[0058] In this disclosure, unless otherwise specified, the number of nouns refers to "singular or plural," i.e., "one or more." "At least one" means one or more, "multiple" means two or more, and "and / or" is an associative relationship for describing associated objects, indicating that three relationships may exist. For example, A and / or B may indicate the following cases: only A is present, both A and B are present, and only B is present, where A and B may be singular or plural. When representing a feature, the symbol " / " may represent an "or" relationship between associated objects. For example, A / B indicates A or B. When representing an operation, the symbol " / " may also represent division. In addition, in this disclosure, the symbol "x" may also be replaced with the symbol "*."

[0059] In this disclosure, ordinal numbers such as "first" and "second" are used to distinguish between multiple objects and are not intended to limit the size, content, order, chronological order, application scenario, priority, importance, etc. of the multiple objects. For example, the first instruction information and the second instruction information may be the same instruction information or different instruction information. In addition, this type of name does not indicate different sizes, transmission modes, instruction contents, priority, application scenario, importance, etc. of the two instruction information.

[0060] In a possible implementation, the CSI feedback mechanism uses the procedure shown in FIG.

[0061] S21: The base station sends signaling, and in response, the UE receives signaling from the base station.

[0062] The signaling is used to configure channel measurement information, for example, the signaling notifies the UE of at least one of the following: time information for performing channel measurement, type of reference signal (RS) for performing channel measurement, time domain resource of the reference signal, frequency domain resource of the reference signal, and reporting condition of the number of measurements, etc.

[0063] S22: The base station transmits a reference signal to the UE, and in response, the UE receives the reference signal from the base station.

[0064] The UE measures the reference signal to obtain the CSI.

[0065] S23: The UE transmits the CSI to the base station, and in response, the base station receives the CSI from the UE.

[0066] S24: The base station sends data to the UE based on the CSI, and the UE correspondingly receives data from the base station.

[0067] The base station determines a precoding matrix based on the CSI and performs precoding on data to be transmitted to the UE using the precoding matrix. The data transmitted by the base station to the UE is carried on a downlink channel, for example, a PDSCH.

[0068] When the CSI fed back by the UE is more accurate, the information is richer, and the downlink channel reconstructed by the base station based on the CSI is more accurate. In this case, the precoding matrix determined by the base station is more accurate, the downlink spatial multiplexing performance is better, the UE's received signal-to-interference-and-noise ratio is higher, and the system throughput is higher. Meanwhile, as the size of the antenna array in a MIMO system continues to increase, the number of supported antenna ports also increases. Because the size of the complete downlink channel matrix is ​​directly proportional to the number of antenna ports, in a massive MIMO system, a large amount of feedback overhead is required to ensure high accuracy in the CSI fed back by the UE. This large feedback overhead reduces the resources available for data transmission, resulting in a reduced system capacity. Therefore, methods for reducing the CSI feedback overhead need to be researched to improve system capacity. Feedbacking CSI based on a bi-domain compressed codebook is a method that can effectively reduce feedback overhead.

[0069] The two-domain compression codebook is generally designed based on factors such as the expected antenna panel configuration and the number of subbands. However, in an actual communication environment, the channel environment is complex and variable, and the actual antenna panel configuration is diverse. Therefore, a codebook determined for a fixed antenna panel configuration and the number of subbands may not necessarily satisfy the actual communication environment, resulting in a decrease in the accuracy of the CSI fed back by the UE. Therefore, how to improve the accuracy of the CSI fed back by the UE is a technical problem worth studying.

[0070] In consideration of this, the present disclosure provides a technical solution. In the present disclosure, each of M first downlink channel subdata acquired by a UE may correspond to one data space among the M data spaces, and first information corresponding to all of the first downlink channel subdata may be determined based on dictionaries corresponding to different data spaces. Different data spaces may represent different location information, i.e., different data spaces may represent different channel environment information. The UE feeds back the first information corresponding to the different data spaces, so that the access network device can determine a correspondence between the first information and the environment information. In this case, the first information fed back by the UE can reflect the actual communication environment, and the accuracy of the first information fed back by the UE is improved. The access network device can reconstruct an accurate downlink channel based on the first information fed back by the UE.

[0071] 3 shows a communication network architecture in a communication system 10 provided in the present disclosure. Any of the embodiments provided hereinafter are applicable to this architecture. A network device included in FIG. 3 is, for example, an access network device 20 included in the communication system 10, and a terminal device included in FIG. 3 is, for example, a communication apparatus 30 included in the communication system 10. The network device can communicate with the terminal device.

[0072] The present disclosure may relate to machine learning technology. Machine learning technology is a specific embodiment of AI technology. For ease of understanding, the following describes AI technology. It can be understood that the description is not intended to limit the present disclosure.

[0073] AI is a technology that mimics the human brain to perform complex calculations, and is being increasingly applied as data storage and capabilities improve.

[0074] In this disclosure, an independent network element (e.g., referred to as an AI network element, AI node, or AI device) may be introduced into the communication system shown in FIG. 1 to perform AI functions. The AI ​​network element may be directly connected to the access network device or indirectly connected to the access network device via a third-party network element. Optionally, the third-party network element may be a core network element. Alternatively, an AI entity may be configured or disposed in another network element within the communication system to perform AI-related operations. The AI ​​entity may also be referred to as an AI module, AI unit, or by another name, and is primarily configured to perform some or all of the AI ​​functions. The specific name of the AI ​​entity is not limited in this disclosure. Optionally, the other network element may be an access network device, a core network device, a network management device (operation, administration, and maintenance, OAM), etc. In this case, the network element that performs the AI ​​function is a network element with built-in AI functions.

[0075] In this disclosure, AI functions may include at least one of the following: data collection, model training (or model learning), model information release, model inference (also referred to as model inference, inference, prediction, etc.), model monitoring or model checking, or inference result release, etc. AI functions may also be referred to as AI (related) operations or AI-related functions.

[0076] In the present disclosure, an OAM network element (a network management device of a core network device) is configured to operate, manage, and / or maintain a core network device, and / or an OAM network element (a network management device of an access network device) is configured to operate, manage, and / or maintain an access network device. For example, the present disclosure includes a first OAM network element and a second OAM network element, where the first OAM network element is a network management device of a core network device and the second OAM network element is a network management device of an access network device. Optionally, the first OAM network element and / or the second OAM network element include an AI entity. In another example, the present disclosure includes a third OAM network element, where the third OAM network element is a network management device of both a core network device and an access network device. Optionally, the third OAM network element includes an AI entity.

[0077] FIG. 4A is a diagram of a first application framework for AI in a communication system. A data source is used to store training data and inference data. A model training node performs training or update training on training data provided by the data source to obtain an AI model, and deploys the AI ​​model in a model inference node. The AI ​​model represents a mapping relationship between the model's input and output. Obtaining an AI model through learning by the model training node is equivalent to obtaining a mapping relationship between the model's input and output through learning by the model training node using training data. The model inference node uses the AI ​​model to perform inference based on inference data provided by the data source and obtain an inference result. This method can also be described as follows: the model inference node inputs inference data to the AI ​​model and obtains an output through the AI ​​model. The output is the inference result. The inference result may indicate configuration parameters used (acted upon) by a target of an action and / or an operation performed by the target of an action. The inference results may be uniformly planned by an actor entity and sent to one or more targets of the action (e.g., a core network element, a base station, or a UE) for action. Optionally, the model inference node may feed back the inference results of the model inference node to the model training node. This process may be referred to as model feedback. The fed back inference results are used by the model training node to update the AI ​​model, and the updated AI model is placed in the model inference node. Optionally, the target of the action may feed back network parameters collected by the target of the action to a data source. This process may be referred to as performance feedback, and the fed back network parameters may be used as training data or inference data.

[0078] For example, the AI ​​model includes a decoder network in the AE network. The decoder network is located on the access network device side. The inference result of the decoder network is used, for example, to reconstruct a downlink channel matrix. The AI ​​model includes an encoder network in the AE network. The encoder network is located on the UE side. The inference result of the encoder network is used, for example, to encode the downlink channel matrix.

[0079] The application framework shown in FIG. 4A may be deployed in the network elements shown in FIG. 1. For example, the application framework in FIG. 4A may be deployed in at least one of a terminal device, an access network device, a core network device (not shown), or an independently deployed AI network element (not shown) in FIG. 1. For example, the AI ​​network element (which may be considered a model training node) may analyze or train training data provided by the terminal device and / or the access network device to obtain a model. At least one of the terminal device, the access network device, or the core network device (which may be considered a model inference node) may perform inference using the model and inference data to obtain a model output. The inference data may be provided by the terminal device and / or the access network device. The model input includes the inference data, and the model output is an inference result corresponding to the model. At least one of the terminal device, the access network device, or the core network device (which may be considered a target of the action) may perform a corresponding operation based on the inference result. The model inference node and the target of the action may be the same or different. This is not a limitation.

[0080] With reference to FIGS. 4B to 4E, the following will use an example to describe a network architecture to which the method provided in the present disclosure can be applied.

[0081] As shown in FIG. 4B , in a first possible implementation, the access network device includes a near-real-time radio access network intelligent controller (RAN intelligent controller, RIC) module configured to perform model training and inference. For example, the near-real-time RIC may be configured to train an AI model and perform inference using the AI ​​model. For example, the near-real-time RIC may obtain network-side or terminal-side information from at least one of the CU, DU, or RU, which may be used as training data or inference data. Optionally, the near-real-time RIC may submit inference results to at least one of the CU, DU, RU, or terminal device. Optionally, the CU and the DU may exchange inference results. Optionally, the DU and the RU may exchange inference results. For example, the near-real-time RIC submits the inference results to the DU, and the DU forwards the inference results to the RU.

[0082] As shown in FIG. 4B , in a second possible implementation, the non-real-time RIC is located outside the access network device (optionally, the non-real-time RIC may be located in an OAM network element or a core network device) and is configured to perform model training and inference. For example, the non-real-time RIC is configured to train an AI model and perform inference using the model. For example, the non-real-time RIC may obtain network-side and / or terminal-side information from at least one of the CU, DU, or RU. This information may be used as training data or inference data, and the inference results may be submitted to at least one of the CU, DU, RU, or terminal device. Optionally, the CU and DU may exchange the inference results. Optionally, the DU and RU may exchange the inference results. For example, the non-real-time RIC submits the inference results to the DU, and the DU forwards the inference results to the RU.

[0083] As shown in FIG. 4B , in a third possible embodiment, the access network device includes a near-real-time RIC, and the non-real-time RIC is located outside the access network device (optionally, the non-real-time RIC may be located in an OAM network element or a core network device). Like the non-real-time RIC in the second possible embodiment, the non-real-time RIC may be configured to perform model training and inference, and / or like the near-real-time RIC in the first possible embodiment, the near-real-time RIC may be configured to perform model training and inference, and / or the non-real-time RIC may perform model training, and the near-real-time RIC may obtain AI model information from the non-real-time RIC, obtain network-side and / or terminal-side information from at least one of a CU, a DU, or a RU, and obtain an inference result based on the information and the AI ​​model information. Optionally, the near-real-time RIC may submit the inference result to at least one of a CU, a DU, a RU, or a terminal device. Optionally, the CU and the DU may exchange the inference result. Optionally, the DU and the RU may exchange the inference result. For example, the near real-time RIC submits inference results to the DU, and the DU forwards the inference results to the RU. For example, the near real-time RIC is configured to train model A and perform inference using model A. For example, the non-real-time RIC is configured to train model B and perform inference using model B. For example, the non-real-time RIC is configured to train model C and send information about model C to the near real-time RIC, and the near real-time RIC performs inference using model C.

[0084] 4C is an exemplary diagram of a network architecture to which the method according to the present disclosure can be applied. Compared with FIG. 4B, in FIG. 4C, the CU is separated into a CU-CP and a CU-UP.

[0085] 4D is an example diagram of a network architecture to which the method according to the present disclosure can be applied. As shown in FIG. 4D, optionally, an access network device includes one or more AI entities, and the functionality of the AI ​​entities is similar to that of a near-real-time RIC. Optionally, an OAM network element includes one or more AI entities, and the functionality of the AI ​​entities is similar to that of a non-real-time RIC. Optionally, a core network device includes one or more AI entities, and the functionality of the AI ​​entities is similar to that of a non-real-time RIC. When an OAM network element and a core network device each include an AI entity, the models obtained by training using the AI ​​entities of the OAM and core network devices are different, and / or the models used for inference are different.

[0086] In the present disclosure, differences between models include differences in at least one of the following: structural parameters of the models (e.g., at least one of the number of neural network layers, the neural network width, the connectivity between layers, the weights of neurons, the activation functions of neurons, or the biases in the activation functions); input parameters of the models (e.g., the types of input parameters and / or the dimensions of the input parameters); or output parameters of the models (e.g., the types of output parameters and / or the dimensions of the output parameters).

[0087] FIG. 4E is an exemplary diagram of a network architecture to which the method according to the present disclosure can be applied. Compared with FIG. 4D, in FIG. 4E, the access network device is separated into a CU and a DU. Optionally, the CU may include an AI entity, whose function is similar to that of a near-real-time RIC. Optionally, the DU may include an AI entity, whose function is similar to that of a near-real-time RIC. When the CU and the DU each include an AI entity, the models obtained by training using the AI ​​entities of the CU and the DU are different, and / or the models used for inference are different. Optionally, the CU in FIG. 4E may be further divided into a CU-CP and a CU-UP. Optionally, one or more AI models may be located in the CU-CP. Optionally, one or more AI models may be located in the CU-UP.

[0088] In Figure 4D or 4E, the OAM network elements of the access network devices and the OAM network elements of the core network devices are uniformly arranged. Alternatively, as described above, in Figure 4D or 4E, the OAM network elements of the access network devices and the OAM network elements of the core network devices may be separately arranged.

[0089] In the present disclosure, a model may obtain an output through inference, and the output includes one or more parameters. The learning or training processes of different models may be located on different devices or nodes, or on the same device or node. The inference processes of different models may be located on different devices or nodes, or on the same device or node.

[0090] Optionally, the AI ​​model includes a decoder network in the AE network. On the network side, the inference result of the decoder network is used, for example, to reconstruct a downlink channel matrix. Optionally, the AI ​​model includes an encoder network in the AE network, and model information of the encoder network may be transmitted to the UE for the UE to perform inference.

[0091] Note that in the framework of FIGS. 4A-4E, the AI ​​model may be referred to as a model or a network model for short, and may be considered as a mapping from input parameters (e.g., an input matrix) to output parameters (e.g., an output matrix). For example, with respect to a network-side decoder network, the input matrix may be a matrix determined based on received CSI. The training data may include a known input matrix, or may include a known input matrix and a corresponding output matrix, and is used to train the AI ​​model. The training data may be data from an access network device, a CU, a CU-CP, a CU-UP, a DU, a RU, a UE, and / or another entity, and / or data inferred using AI techniques. This is not limited to this. The inference data includes the input matrix and is used to infer an output matrix using the model. The inference data may be data from an access network device, a CU, a CU-CP, a CU-UP, a DU, a RU, a UE, and / or another entity. The inferred matrix may be considered policy information and is transmitted to a target for action. The inferred matrix may be transmitted to an access network device, such as a CU, CU-CP, CU-UP, DU, RU, or UE, for further processing, e.g., for reconstruction of the downlink channel matrix.

[0092] In the present disclosure, when the decoder network in the AE network is located on the network side, the decoder network may be located within an access network device (e.g., a base station) or outside the access network device, for example, in an OAM network element, an AI network element, or a core network device, or in an RU, DU, or near-real-time RIC. This is not limited. The inference result of the decoder network may be obtained by the access network device through inference, or may be transmitted to the access network device after a non-real-time RIC performs inference. For simplicity, the present disclosure will be described using an example in which the decoder network is located in the access network device.

[0093] In the present disclosure, when the encoder network in the AE network is located on the terminal side, the encoder network may be located in the UE, and the UE may perform inference using the encoder network.

[0094] The following describes the methods provided in the present disclosure with reference to the accompanying drawings. The steps or operations included in these methods are merely examples, and other operations or variations of various operations may also be performed in the present disclosure. In addition, steps may be performed in an order different from that presented in the present disclosure, and in some cases, not all operations need to be performed.

[0095] FIG. 5 is a flowchart of a communication method according to the present disclosure.

[0096] S501: A UE obtains M first downlink channel subdata, each of the first downlink channel subdata corresponding to one data space among the M data spaces, where M is an integer greater than 1.

[0097] The M first downlink channel subdata are obtained, for example, based on the first downlink channel data. For example, to obtain the M first downlink channel subdata, the UE may divide and allocate the first downlink channel data to M data spaces, or it may be understood that the UE may divide the first downlink channel data into M portions. Corresponding each first downlink channel subdata to one data space may also be understood as a one-to-one correspondence between the data space and the first downlink channel subdata. The first downlink channel data is, for example, original downlink channel data (also referred to as an original downlink channel matrix or downlink channel response). In other words, after obtaining the original downlink channel data, the UE may directly divide the original downlink channel data into M portions without performing any other processing on the original downlink channel data. In this case, the processing steps can be reduced. Alternatively, the first downlink channel data may be data obtained by pre-processing second downlink channel data, and the second downlink channel data is obtained based on the original downlink channel matrix. In this case, the original downlink channel data can be simplified by a pre-processing process to simplify the process of processing the first downlink channel data by the UE. Alternatively, the first downlink channel data can be data output by a neural network, for example, the contents of the original downlink channel matrix are invisible to the UE, and the UE directly obtains the first downlink channel data output by the neural network.

[0098] A pre-processing process is included when the first downlink channel data is obtained by pre-processing the second downlink channel data. The second downlink channel data is obtained based on the original downlink channel matrix. For example, the second downlink channel data is the original downlink channel matrix, or the second downlink channel data is an eigenvector obtained by processing the original downlink channel matrix. Different implementations of the second downlink channel data may have different pre-processing processes. These will be described below.

[0099] 1. The second downlink channel data is the original downlink channel matrix. For example, the original downlink channel matrix is ​​called the first downlink channel matrix.

[0100] For example, the dimension of the first downlink channel matrix is ​​[N tx ,N rx ,N RB ] and N tx represents the number of antennas or ports at the transmitter (e.g., access network device) of the downlink signal, and N rx represents the number of antennas or ports at the receiver (e.g., UE) of the downlink signal, and N RB represents the number of frequency domain units, for example, the number of resource blocks (RBs) or the number of subbands.

[0101] Optionally, the UE may further perform a dimension transformation operation on the first downlink channel matrix to obtain transformed data, i.e., to obtain a transformed first downlink channel matrix. The dimension of the transformed first downlink channel matrix is ​​[N tx *N rx ,N RB ] or [N tx ,N rx ,N RB For example, a matrix is ​​represented by H, where H is a complex matrix, i.e.

number

[0102] Optionally, a discrete Fourier transform (DFT) can be further used to generate two groups of DFT bases, namely, spatial domain bases

number

number

number

[0103] Complex matrix C complex After S is obtained, the pre-processing process of the second downlink channel data is completed. H is the Hermitian matrix of S, which is also called the self-conjugate matrix and can be obtained by performing a conjugate transpose on the matrix S. sb represents the number of frequency domain subbands, e.g., N sb =N rb / a, where a represents the frequency domain subband granularity or subband bandwidth, i.e., the number of RBs included in each subband. The common frequency domain subband granularity may be 1 RB, 2 RB, 4 RB, or 8 RB, etc. This is not limited here. For example, if the frequency domain subband granularity is 4 RBs, then N sb =N rb / 4. S represents the spatial domain basis, and the specific form of S is related to the antenna panel. Assuming that the antenna panel is dual-polarized, with a horizontal component of Nh and a vertical component of Nv, the expression form of S obtained is as follows:

number

[0104] F represents the frequency domain basis, and the representation of F is subband N sb For example, F may satisfy the following equation: F=DFT(N sb ) (Formula 3)

[0105] Optionally, an oversampling factor may be further added in the DFT process. For example, multiple groups of orthogonal space-domain bases {S1, S2, S3...} and multiple groups of orthogonal frequency-domain bases {F1, F2, F3...} may be generated in an oversampling manner, and S i and F j The groups are selected from these groups as the spatial domain basis and the frequency domain basis of the present disclosure. For example, a group having a correct projection direction may be selected from these groups. For example, the oversampling factors of the spatial domain and the frequency domain are 4, respectively.

[0106] Optionally, the first downlink channel data may be a complex matrix, for example, a complex matrix C obtained by pre-processing the second downlink channel data. complex is.

[0107] 2. The second downlink channel data is an eigenvector obtained by processing the first downlink channel matrix.

[0108] In this case, the first downlink channel matrix needs to be first processed to obtain eigenvectors, and then the eigenvectors are pre-processed to obtain the first downlink channel data. Alternatively, it may be understood that the process of processing the first downlink channel matrix to obtain eigenvectors and the process of pre-processing the eigenvectors to obtain the first downlink channel data can both be considered as pre-processing processes of the first downlink channel matrix.

[0109] For example, the dimension of the first downlink channel matrix is ​​[N tx ,N rx ,N RB ], and dimension reduction is performed by singular value decomposition (SVD) to obtain the eigensubspace matrix (or called eigensubspace for short) of the downlink channel [N tx ,N rx ,N RB ]-dimensional first downlink channel matrix. The dimension of the eigensubspace is [N tx ,N sb ]. When performing dimensionality reduction on the first downlink channel matrix by SVD, the UE may process different ranks of the first downlink channel matrix separately, and different ranks may also be understood as different streams or different layers. One piece of channel information (or one channel estimation result) may correspond to one or more layers. The following describes a process in which the UE processes the Lth layer of the first downlink channel matrix. There may be multiple methods, which are not limited.

[0110] Each subband in the Lth layer may include an RB, and the UE may calculate the equivalent downlink channel of one subband by referring to the downlink channel of the RB. The downlink channel corresponding to the kth RB in the Lth layer subband c is expressed as H k , and the equivalent downlink channel for subband c may be expressed as:

number

[0111] UE is

number

number

[0112] in particular,

number

number

number

number

[0113] Optionally, in addition, the eigenvectors

number

number

number

number

[0114] The obtained complex matrix C complex is a sparse representation of the eigensubspace of the original downlink channel, and the dimensions of the complex matrix match the dimensions of the eigenvectors that existed before the space-frequency joint projection, and N tx *N sbThe complex matrix C complex After S is obtained, the pre-processing process of the second downlink channel data is completed. H , N sb For a description of parameters such as , and the spatial domain basis S, see the previous discussion.

[0115] Optionally, an oversampling factor may be further added in the DFT process. For example, multiple groups of orthogonal space-domain bases {S1, S2, S3...} and multiple groups of orthogonal frequency-domain bases {F1, F2, F3...} may be generated in an oversampling manner, and S i and F j are selected from these groups as the spatial domain basis and the frequency domain basis of the present disclosure. For example, a group having a correct projection direction may be selected from these groups. For example, the oversampling factors of the spatial domain and the frequency domain are 4, respectively. The complex matrix C is calculated in one of two ways: complex After obtaining the complex matrix C complex Optionally, the UE may obtain the first downlink channel data based on a complex matrix C complex The method for obtaining the first downlink channel data based on complex directly as the first downlink channel data, that is, the first downlink channel data is a result of pre-processing the second downlink channel data.

[0116] Alternatively, if UE is a complex matrix C complex Another method for obtaining the first downlink channel data based on complexFor example, in the frequency domain (delay domain), energy is generally concentrated mainly around delay=0, and energy in other domains can be essentially ignored. Therefore, the UE can select F consecutive columns on both sides of delay=0 as the first downlink channel data, and the coefficients of the remaining part can be 0 by default. For example, the UE can select a complex matrix C as the first downlink channel data. complex F consecutive columns may be selected from the complex matrix C complex Columns not selected from may not be processed, in which case the energy distribution may be considered and processing overhead may be reduced.

[0117] For example, F is a positive integer, and the value of F may be predefined in a protocol, or different F may be determined based on different overhead. For example, a mapping relationship between overhead and F may be provided in a protocol, allowing the UE and the access network device to determine the same F based on their current overhead requirements. Alternatively, the value of F may be indicated by the access network device. For example, the access network device transmits information indicating the value of F to the UE. The UE can determine the value of F after receiving the information. Alternatively, the value of F may be determined by the UE. For example, the UE determines the value of F based on factors such as channel conditions and / or network topology to reduce the impact on air interface transmission. After determining the value of F, the UE may transmit information indicating the value of F to the access network device. The access network device can determine the value of F after receiving the information.

[0118] Alternatively, if UE is a complex matrix C complex Yet another method for obtaining the first downlink channel data based on complex For example, the UE may perform a compression process on the complex matrix Ccomplex may be input to the encoder network, which then outputs a complex matrix C complex The encoder network performs a compression process on the first downlink channel data, and outputs the compressed information. In this way, the first downlink channel data is obtained through compression. This can reduce the complexity of processing the first downlink channel data by the UE.

[0119] The aforementioned process is to acquire first downlink channel data. After acquiring the first downlink channel data, the UE may divide the first downlink channel data and allocate it to M data spaces to acquire M first downlink channel subdata. The first downlink channel subdata correspond one-to-one to the data spaces. For example, the ith first downlink channel subdata among the M first downlink channel subdata corresponds to the ith data space among the M data spaces, where i may be an integer from 1 to M.

[0120] In the present disclosure, M data spaces are included, and the M data spaces may correspond to dictionaries. For example, the M data spaces may correspond to N dictionaries, where N is an integer greater than or equal to 1 and less than or equal to M. Optionally, N=M, i.e., the data spaces correspond one-to-one to the dictionaries, and each data space corresponds to one dictionary, or N=1, i.e., all M data spaces correspond to the same dictionary, and a dictionary corresponds to each data space. Optionally, N=M / 2, and every two data spaces correspond to one dictionary. Other possible cases are not described one by one. Different data spaces may correspond to the same dictionary or different dictionaries. This is not limited. The use of dictionaries will be described in S502 below. In addition, the M data spaces (or the division method of the M data spaces) are also included in the dictionary training process. The dictionary training process will be described in subsequent embodiments. Accordingly, the division method of the M data spaces, etc. will also be described in subsequent embodiments.

[0121] A variable stored in a dictionary includes at least one of {data space index, element index, element}. That is, a variable stored in a dictionary may include one or more of a data space index, an element index, or an element. In addition, a dictionary may further include other information, or may not include other information. This is not limited. An index of a data space included in a dictionary is an index of a data space corresponding to the dictionary. For example, if a dictionary corresponds one-to-one to a data space, one dictionary corresponds to one data space, and the dictionary includes an index of a data space corresponding to the dictionary. Alternatively, if all M data spaces correspond to the same dictionary, the dictionary may correspond to M data spaces, and the dictionary may not include an index of a data space. An element may be, for example, a vector, and a dictionary may include multiple elements. Each element may have a corresponding index, that is, an element may correspond one-to-one to an index of the element. If N is greater than 1, the indexes of elements included in different dictionaries may be reused. For example, the index of an element in each dictionary may start from 1 or 0, i.e., elements included in different dictionaries are numbered independently. Alternatively, the index of an element included in different dictionaries may be different, i.e., elements included in different dictionaries are numbered together. For example, the index of an element in a first dictionary is 0 to d-1, and the index of an element in a second dictionary starts from d. Figure 6 is a diagram of N dictionaries. In Figure 6, N=M is ​​used as an example, i.e., M dictionaries are included in total. In Figure 6, 0 to 3 in each dictionary represent the index of an element. Here, the number of indexes of elements included in each dictionary is 4 as an example. This is not actually a limitation. In addition, the number of elements included in different dictionaries may be the same or different.

[0122] Optionally, the dictionary representation method further includes: when N is between 1 and M, the dictionary may include {dictionary index, element index, element}, and the correspondence between the dictionary index and the data space index may be interoperable between the access network device and the terminal device. When N=1, the dictionary index may be omitted. By default, all data space indexes correspond to the dictionary. When M>N>1, the correspondence between the dictionary index and the data space index may be a default rule predefined in the protocol. For example, when M=4 and N=2, in the rule, data space index 0 and data space index 2 correspond to dictionary index 0, and data space index 1 and data space index 3 correspond to dictionary index 1. Alternatively, when M>N>1, the access network device may indicate the correspondence between the dictionary index and the data space index to the UE. Alternatively, when M>N>1, the UE may report the correspondence between the dictionary index and the data space index to the access network device. If M=N, the dictionary indices may correspond one-to-one to the data space indices, or the dictionary may contain {data space index, element index, element}.

[0123] S502: The UE determines first information corresponding to the ith first downlink channel subdata among the M first downlink channel subdata based on the first dictionary corresponding to the ith data space among the M data spaces, where i is an integer from 1 to M. Thus, the UE determines a total of M pieces of first information.

[0124] The i-th first downlink channel subdata corresponds to the i-th data space among the M data spaces. For example, when the first downlink channel data is divided and assigned to the M data spaces, M pieces of first downlink channel subdata are obtained, and the i-th first downlink channel subdata is the portion of the first downlink channel data divided and assigned to the i-th data space. For example, N=M, and each data space has a dictionary corresponding to the data space. In this case, the first dictionary is, for example, the dictionary corresponding to the i-th data space among the M data spaces. That is, the UE may determine the first information corresponding to the i-th first downlink channel subdata based on the first dictionary corresponding to the i-th data space. When i is an integer from 1 to M, dictionaries corresponding to different data spaces may be referred to as first dictionaries, and the first dictionaries corresponding to different data spaces may be the same or different. Alternatively, when N=1 and one dictionary corresponds to the M data spaces, the first dictionary is a dictionary. For any data space among the M data spaces, a first dictionary may be used, and the UE may determine first information corresponding to the i-th first downlink channel subdata based on the first dictionary.

[0125] If N=1, the UE may determine that the dictionary is the first dictionary corresponding to the ith data space. Alternatively, if M=N and the dictionaries correspond one-to-one to the data spaces, the UE can determine the first dictionary corresponding to the ith data space. Alternatively, if M>N>1, the UE may determine the first dictionary corresponding to the ith data space based on a correspondence between dictionary indexes and data space indexes. For example, M=4 and N=2. The correspondence specifies that data space index 0 and data space index 2 correspond to dictionary index 0, and data space index 1 and data space index 3 correspond to dictionary index 1. i is equivalent to the data space index. In this case, the UE may determine the first dictionary corresponding to the ith data space based on the value of i and the correspondence. For example, if i=1, the UE may determine that the first dictionary corresponding to the first data space is the dictionary indicated by dictionary index 1.

[0126] From the above description of the dictionary, it can be known that the first dictionary may include multiple elements, and the UE may determine P elements corresponding to the i-th first downlink channel subdata from the multiple elements, where P is a positive integer. For example, among the multiple elements included in the first dictionary, the P elements most relevant to the i-th subdata are the P elements corresponding to the i-th first downlink channel subdata, and the P elements may be used as first information corresponding to the i-th first downlink channel subdata. If P is greater than 1, the P elements may form the first information in a first combination method. For example, the first combination method may be multiplying the P elements, or performing weighted addition on the P elements (e.g., averaging or performing weighted addition using other possible weight values), or connecting the P elements in series. The first combination method is not limited. For example, the first combination method may be predefined in a protocol, or determined by the access network device and notified to the UE, or determined by the UE and notified to the access network device. For the M first downlink channel subdata, the UE may determine first information corresponding to the M first downlink channel subdata, and the UE may determine a total of M pieces of first information, where the M pieces of first information are M elements.

[0127] S503: The UE transmits first indication information. For example, the UE transmits the first indication information to the access network device, and the access network device may receive the first indication information from the UE in response. The first indication information may indicate M pieces of first information, and the access network device may determine the M pieces of first information based on the first indication information.

[0128] Optionally, the first indication information may include M identifiers of the first information, thereby indicating the M pieces of first information. Each identifier of the first information may be, for example, an index of the first information in a corresponding dictionary. For example, the M pieces of first information may include first information corresponding to the i-th first downlink channel subdata, and the identifier of the first information may be an index of the first information in the first dictionary. After determining the M pieces of first information, the UE may determine M identifiers of the first information. For example, the UE may determine M identifiers in total, and the UE may transmit the M identifiers to the access network device. When the UE transmits the M identifiers of the first information, it may be considered that the UE transmits CSI, i.e., the M identifiers of the first information may be used as CSI, or the M identifiers of the first information may be used as PMI, or the M identifiers of the first information may perform a function similar to that of a PMI or a CSI identifier.

[0129] Alternatively, the first indication information may not include identifiers of the M pieces of first information and may indicate the first information in another manner. For example, there may be different combination relationships between dictionary elements, and each combination relationship may include one element in each of the N dictionaries. Each combination relationship may correspond to one piece of indication information. When the UE transmits specific indication information, it indicates that the combination relationship corresponding to the indication information is indicated. For example, if the first indication information corresponds to the combination relationship of the M pieces of first information, the first indication information transmitted by the UE may indicate the M pieces of first information.

[0130] Optionally, when transmitting the M first information identifiers, the UE may transmit the identifiers in a first order. The first order is the arrangement order of the M data spaces, i.e., the first order specifies that the UE first transmits the first information identifier corresponding to a specific data space, and then transmits the first information identifier corresponding to a specific data space. For example, M=4, the M data spaces are data space 1 to data space 4, and the first order is 2-1-4-3. When transmitting the M first information identifiers, the UE first transmits the first information identifier corresponding to data space 2, then transmits the first information identifier corresponding to data space 1, then transmits the first information identifier corresponding to data space 4, and finally transmits the first information identifier corresponding to data space 3. The first order is known to the UE and the access network device. Therefore, after receiving the M first information identifiers, the access network device can further determine the correspondence between the data spaces and the first information identifiers to avoid correspondence errors.

[0131] For example, the first order may be a predefined order in a protocol, and the UE and the access network device may determine the first order according to the protocol. Alternatively, the first order may be preconfigured in the UE and the access network device. Alternatively, the first order may be determined by the access network device. After determining the first order, the access network device may send second indication information to the UE, where the second indication information indicates the first order, and the UE may determine the first order based on the second indication information. Alternatively, the first order may be determined by the UE. After determining the first order, the UE may send third indication information to the access network device, where the third indication information indicates the first order, and the access network device may determine the first order based on the third indication information.

[0132] S504: For an i-th first information among the M pieces of first information, the access network device reconstructs an i-th second downlink channel subdata based on a first dictionary corresponding to the i-th data space among the M pieces of data spaces. When i is an integer from 1 to M, the access network device may obtain a total of M pieces of second downlink channel subdata.

[0133] For example, when an access network device receives M identifiers of first information in a first order, the access network device can determine a correspondence between a data space and an identifier of the first information, thereby determining the first information corresponding to the identifier of the first information based on the dictionary corresponding to the data space, and the first information determined by the access network device is considered as the second downlink channel subdata reconstructed by the access network device. For example, N=M, the data space corresponds one-to-one to the dictionary, and the dictionary corresponding to the i-th data space is, for example, the first dictionary. In this case, for the i-th first information identifier, the access network device can determine the i-th first information identifier in the first dictionary to determine the first information corresponding to the i-th first information identifier in the first dictionary, i.e., to reconstruct the second downlink channel subdata (i-th second downlink channel subdata) corresponding to the i-th first information. In another example, N=1, and all M data spaces correspond to the first dictionary. In this case, for the identifier of the i-th first information, the access network device may determine the identifier of the i-th first information in the first dictionary to determine the first information corresponding to the identifier of the i-th first information in the first dictionary, i.e., to reconstruct the second downlink channel subdata (the i-th second downlink channel subdata) corresponding to the i-th first information. In another example, M>N>1. In this case, for the identifier of the i-th first information, the access network device may determine the data space corresponding to the identifier of the i-th first information in a first order, for example, the i-th data space. The access network device may further determine the dictionary corresponding to the i-th data space, for example, the first dictionary, based on the correspondence between the dictionary index and the data space index.In this case, the access network device may determine the first information corresponding to the identifier of the i-th first information in the first dictionary, i.e., may reconstruct the second downlink channel subdata (i-th second downlink channel subdata) corresponding to the i-th first information.

[0134] In an ideal situation, the M second downlink channel subdata acquired by the access network device and the M first downlink channel subdata acquired by the UE may be the same data. For example, the i-th first downlink channel subdata and the i-th second downlink channel subdata are the same data. In practical application, there may be a discrepancy between the M second downlink channel subdata acquired by the access network device and the M first downlink channel subdata acquired by the UE. The process in which the UE acquires the first information based on a dictionary is equivalent to the process in which the M first downlink channel subdata are quantized. In other words, the UE transmits quantization information to the access network device, and the access network device reconstructs the M second downlink channel subdata based on the quantization information and the dictionary. The quantization and reconstruction process may have some loss. Therefore, there may be a certain discrepancy between the M second downlink channel subdata and the M first downlink channel subdata. For example, the i-th first downlink channel subdata and the i-th second downlink channel subdata may be different data, but the discrepancy between the M second downlink channel subdata and the M first downlink channel subdata may tend to decrease due to improvements in dictionary accuracy and transmission quality.

[0135] S505: The access network device reconstructs downlink channel information based on the M second downlink channel subdata. In other words, the access network device reconstructs a downlink channel matrix based on the M second downlink channel subdata, for example, reconstructs a first downlink channel matrix.

[0136] In S501, the UE receives a complex matrix C as first downlink channel data. complex or use the complex matrix C as the first downlink channel data complex When selecting F consecutive columns from, after obtaining M pieces of second downlink channel subdata, the access network device may concatenate the M pieces of second downlink channel subdata, and the obtained information is, for example, called an angle delay region coefficient, and the angle delay region coefficient is a matrix:

number

[0137] Alternatively, in S501, the UE calculates a complex matrix C complex to obtain compressed information and use the compressed information as the first downlink channel data, the M second downlink channel subdata obtained by the access network device are actually M compressed subinformation. Optionally, the access network device may reconstruct the M compressed subinformation to obtain K reconstructed information, where K is a positive integer, and K may be equal to or not equal to M. For example, if the UE obtains the compressed information through an encoder network, a decoder network corresponding to the encoder network may be configured on the access network device side. The access network device may input the M second downlink channel subdata to the decoder network, and the decoder network may output the K reconstructed information. The access network device may:

number

[0138] 7 is a diagram in which the UE uses the compressed information as the first downlink channel data, and the access network device needs to reconstruct the compressed information. In FIG. 7, for example, the UE inputs the second downlink channel data to the encoder network, the encoder network compresses the second downlink channel data, and the encoder network outputs the compressed information, which can be used as the first downlink channel data. Alternatively, in FIG. 7, the UE inputs the complex matrix C complex may be input to the encoder network, which then receives the complex matrix C complex The encoder network outputs compressed information, which may be used as the first downlink channel data. The UE divides the compressed information and allocates it to M data spaces. In FIG. 7, M=4 is used as an example, in which the UE obtains four first downlink channel subdata. The UE processes the four first downlink channel subdata based on four dictionaries to obtain identifiers of the four first information. In FIG. 7, circles represent dictionaries, and C M represents the number of elements in the Mth dictionary, and log2C M represents the number of transmission bits corresponding to the Mth dictionary. Optionally, in FIG. 7, the number of transmission bits corresponding to the dictionary may be obtained by rounding up. For example, in FIG. 7, log2C M teeth

number

number

[0139] Regardless of how the angular delay region coefficients are obtained, the access network device may

number

[0140] For example, the first downlink channel data is a first downlink channel matrix

number

number

number

number

[0141]

number

number

number

number

[0142] In another example, the first downlink channel data is represented by a first downlink channel matrix

number

[0047] In this case, the access network device may perform the following to obtain an eigensubspace of the reconstructed downlink channel:

number

number

number

[0143]

number

number

number

[0144] In the present disclosure, a UE may divide and allocate first downlink channel data into M data spaces and determine first information corresponding to each first downlink channel subdata based on dictionaries corresponding to different data spaces. Different data spaces may represent different location information, i.e., different channel environment information. The UE feeds back the first information corresponding to the different data spaces, allowing the access network device to determine a correspondence between the first information and the environment information. In this case, the first information fed back by the UE may reflect the actual communication environment, and the accuracy of the first information fed back by the UE may be improved. The access network device may reconstruct an accurate downlink channel based on the first information fed back by the UE.

[0145] The embodiment shown in FIG. 5 describes a network inference process. A dictionary is included in the network inference process, and the dictionary can be obtained by network training. There can be multiple methods for training to obtain the dictionary. For example, if the encoder network is not located on the UE side and the decoder network is not located on the access network device side, or if the encoder network is located on the UE side and the decoder network is located on the access network device side, the encoder network and the decoder network may or may not be trained with the dictionary. If only the dictionary needs to be obtained by training and the encoder / decoder network does not need to be obtained, reference can be made to another communication method described later in this disclosure. In this method, a network training process is described, and the dictionary can be obtained in the training process. FIG. 8 is a flowchart of the method.

[0146] S801: A first node acquires M pieces of third downlink channel subdata. Each piece of third downlink channel subdata corresponds to one data space among the M data spaces. The M data spaces in the present disclosure and the M data spaces in the embodiment shown in FIG. 5 may have the same characteristics.

[0147] The M third downlink channel subdata are obtained, for example, based on the third downlink channel data. For example, to obtain the M third downlink channel subdata, the UE or the first node may divide the third downlink channel data and allocate it to M data spaces, or it may be understood that the UE or the first node may divide the third downlink channel data into M portions. The third downlink channel data is, for example, the original downlink channel data. For example, the original downlink channel data in this embodiment is referred to as a third downlink channel matrix. Alternatively, the third downlink channel data may be data obtained by pre-processing the fourth downlink channel data, and the fourth downlink channel data is obtained based on the third downlink channel matrix. Alternatively, the third downlink channel data may be data output by a neural network. The third downlink channel matrix may be considered training data or a training sample. For example, the third downlink channel matrix may be considered to include one or more training data. For example, the third downlink channel matrix may actually include one or more downlink channel submatrices, and each downlink channel submatrix may be considered as training data. The third downlink channel submatrices here may be independent of each other and not included in a large matrix. In other words, the third downlink channel matrix may not be considered as a large matrix, and the third downlink channel matrix may be understood as a combination of one or more third downlink channel submatrices.

[0148] If the third downlink channel data is obtained by pre-processing the fourth downlink channel data, a pre-processing process is included. For the pre-processing process of the fourth downlink channel data, please refer to the description of the pre-processing process of the second downlink channel data in S501 of the embodiment shown in Figure 5.

[0149] In the present disclosure, the first node may be, for example, a UE or an access network device, or may be a third party device (e.g., an AI node). The training process may be an online training process or an offline training process.

[0150] After acquiring the third downlink channel data, the first node may divide and allocate the third downlink channel data to M data spaces to acquire M pieces of third downlink channel subdata. The third downlink channel subdata correspond one-to-one to the data spaces. For example, the ith third downlink channel subdata among the M pieces of third downlink channel subdata corresponds to the ith data space among the M data spaces, where i may be an integer from 1 to M.

[0151] To divide the third downlink channel data and allocate it to the M data spaces, the first node first needs to determine the M data spaces, i.e., determine how to divide the M data spaces. For example, the first node is a UE or an access network device. For example, if the division method of the M data spaces is predefined in a protocol, both the UE and the access network device may determine how to divide the M data spaces according to the protocol. Alternatively, the division method of the M data spaces may be determined by the access network device. The access network device may send fourth instruction information to the UE, where the fourth instruction information may indicate how to divide the M data spaces, and the UE may determine how to divide the M data spaces based on the fourth instruction information. Alternatively, the division method of the M data spaces may be determined by the UE. The UE may send fifth instruction information to the access network device, where the fifth instruction information may indicate how to divide the M data spaces, and the access network device may determine how to divide the M data spaces based on the fifth instruction information.

[0152] In the present disclosure, for example, the division method of M data spaces is specified in the protocol as follows, where M=4, and these four data spaces each include four portions of one data. The four portions are the real part of polarization 1 included in the data, the imaginary part of polarization 1 included in the data, the real part of polarization 2 included in the data, and the imaginary part of polarization 2 included in the data. For example, if third downlink channel data is divided and allocated to four data spaces, the four third downlink channel subdata obtained by the division respectively include the real part of polarization 1 included in the third downlink channel data, the imaginary part of polarization 1 included in the third downlink channel data, the real part of polarization 2 included in the third downlink channel data, and the imaginary part of polarization 2 included in the third downlink channel data. From the perspective of antenna configuration, the antenna element is dual-polarized, and polarization 1 and polarization 2 represent two polarization directions. The two polarization directions may be considered to be independent of each other. From the perspective of complex numbers, data includes a real part and an imaginary part, and the processing process of the real part and the processing process of the imaginary part are independent of each other. Therefore, the data space can be divided based on the antenna polarization direction and the real part and the imaginary part of the complex number, so that each data space can be processed independently. The data space is divided, and the size of each data space is 1 / M of the original data, and different data spaces can represent different environmental information. Optionally, the data space division method may alternatively be an unequal division method. This is not limited.

[0153] Optionally, in the network inference stage, the UE and the access network device may also determine a division method for the M data spaces, and the determination method is the same as that in the present disclosure. Alternatively, the first node may indicate the division method for the M data spaces to the UE and / or the access network device. In other words, in the embodiment shown in FIG. 5, to divide and allocate the first downlink channel data to the M data spaces, the UE also needs to first determine a division method for the M data spaces. In this case, the division method for the M data spaces provided in the present disclosure may be used. The division methods applied to the M data spaces in the network inference stage and the network training stage are the same.

[0154] S802: The first node performs clustering training to obtain N dictionaries.

[0155] As described in the embodiment shown in Figure 5, N may be equal to M, or may be equal to 1, or M > N > 1. In these solutions, the training process of the first node may be different. The following will describe them separately.

[0156] 1. N=M, i.e., the data space corresponds one-to-one to the dictionary.

[0157] The first node may perform clustering training based on the ith third downlink channel subdata among the M third downlink channel subdata to obtain a dictionary (e.g., a first dictionary) corresponding to the ith data space, where the ith third downlink channel subdata corresponds to the ith data space. In other words, the first node may perform training separately in each data space to obtain a dictionary corresponding to each data space, thereby obtaining M dictionaries in total. Clustering is the division of a dataset into different classes or clusters based on a specific criterion (e.g., distance) so that data objects in the same cluster are as similar as possible and data objects that are not in the same cluster are as different as possible. In other words, after clustering, data in the same class can be aggregated as much as possible, and data in different classes can be separated as much as possible. Each class of data has a class center value. When network model training is performed using a clustering method in the present disclosure, the elements included in the dictionary may also be referred to as class center values.

[0158] Training in a data space involves obtaining elements corresponding to the data space, which can be used as elements included in a dictionary corresponding to the data space.

[0159] The number of elements included in the dictionary corresponding to one data space may be related to the bit overhead corresponding to the data space. For example, the bit overhead is 48 bits, which is the total transmission overhead corresponding to M data spaces. For example, the bit overhead corresponding to all data spaces is equal. It is assumed that there are four data spaces in total, and the transmission overhead corresponding to each data space is 1 / M of the total bit overhead. In this case, the bit overhead corresponding to one data space is 12 bits. 12 bits can be used to store up to 2 12 The first dictionary can carry 2 identifiers. 12It must be below. It can be known that different bit overheads correspond to different numbers of elements. Optionally, the first node may separately train different dictionaries based on different bit overheads, and the bit overheads may correspond one-to-one to the dictionaries. In other words, the first node may train one or more dictionaries for one data space. When multiple dictionaries are trained, the multiple dictionaries may correspond to different bit overheads, so that the UE can select an appropriate dictionary based on the current bit overhead when performing network inference.

[0160] For example, in a network inference process, the bit overhead corresponding to a data space is generally determined on the network side, e.g., by an access network device. The access network device may determine the bit overhead based on real-time channel conditions between the access network device and the UE. When a UE performs the network inference process in the embodiment shown in FIG. 5, the access network device may first deliver information to the UE to determine the bit overhead of the current transmission. This information may indicate the total bit overhead corresponding to M data spaces, or may determine the bit overhead corresponding to one data space. The UE may select an appropriate dictionary based on the bit overhead indicated by the access network device. For example, if the access network device indicates that the total bit overhead is 48 bits and M=4, the UE may determine that the transmission overhead of each data space is 48 / 4=12 bits. For example, when a UE determines a dictionary corresponding to an i-th data space, if the i-th data space corresponds to multiple dictionaries (different dictionaries correspond to different bit overheads), the UE may select a dictionary corresponding to 12 bits from the dictionary to perform network inference.

[0161] Additionally, the first node may further determine the dimension of the elements included in the first dictionary based on the dimension of the i-th third downlink channel subdata. The dimension of the elements included in the first dictionary may also be considered the depth of the first dictionary and is related to the dimension of the downlink channel subdata used to train the first dictionary. Thus, the first node may determine the dimension of the elements included in the first dictionary based on the dimension of the i-th third downlink channel subdata. For example, the first node may convert the i-th third downlink channel subdata into a vector, and the length of the vector is the dimension of the elements included in the first dictionary. For example, the i-th third downlink channel subdata is the real part of polarization 1 included in the third downlink channel data, and this part is, for example, a matrix with dimensions [16, 13]. The first node may convert this matrix into a vector with a length of 16 x 13. In this case, the dimension of the elements included in the first dictionary is 16 x 13.

[0162] The first node may convert a matrix to a vector row by row, or may convert a matrix to a vector column by column. In the network inference process, the UE also needs to perform this conversion process. To enable the access network device to reconstruct accurate downlink channel information, the UE's conversion sequence needs to be known by both the UE and the access network device. For example, the UE's conversion sequence may be predefined in a protocol, or the UE's conversion sequence may be determined by the access network device and notified to the UE, or the UE's conversion sequence may be determined by the UE and notified to the access network device. The conversion sequence in the network inference process may be consistent with the conversion sequence in the network training process.

[0163] After determining the number of elements to be included in the first dictionary and the dimensions of the elements to be included in the first dictionary, the first node may perform clustering training based on the i-th third downlink channel subdata to obtain the first dictionary. For each data space, the first node may perform training in a similar manner to obtain M dictionaries.

[0164] 2. N=1, that is, M data spaces correspond to one dictionary. For example, this dictionary is called the first dictionary.

[0165] The first node may perform clustering training based on the M pieces of third downlink channel subdata to obtain a dictionary (e.g., a first dictionary) corresponding to the M pieces of data spaces. In other words, after obtaining the M pieces of third downlink channel subdata, the first node may perform joint training to obtain a dictionary, where the dictionary corresponds to all of the M pieces of data spaces. The M pieces of third downlink channel subdata are used to train the dictionary. In terms of the dictionary, this is equivalent to the sampled data (or training data) being increased by M-1 times, and the training data is richer, so that the elements included in the dictionary are richer and more detailed, which helps the access network device reconstruct more accurate downlink channel information.

[0166] The M third downlink channel subdata are used to perform training to obtain elements corresponding to the M data spaces, and these elements may be used as elements included in the dictionary obtained by training.

[0167] Optionally, the first node may separately train different dictionaries based on different bit overheads, and the bit overheads may correspond one-to-one to the dictionaries. In other words, the first node may train one or more dictionaries. When multiple dictionaries are trained, the multiple dictionaries may correspond to different bit overheads, so that the UE may select an appropriate dictionary based on the current bit overhead when performing network inference.

[0168] In addition, the first node may further determine the dimension of the elements included in the first dictionary based on the dimension of the third downlink channel data. For the determination method, see the above description. The first dictionary is, for example, a dictionary obtained by the first node through training.

[0169] After determining the number of elements included in the first dictionary and the dimensions of the elements included in the first dictionary, the first node may perform clustering training based on the M third downlink channel subdata to obtain the first dictionary.

[0170] Alternatively, when N=1, the first node may perform training separately in all data spaces. For example, the first node uses the training method used when N=M, but a condition needs to be added during training, that is, the dictionaries obtained by training in all data spaces are the same. In this case, the first node may obtain N dictionaries through training, but the N dictionaries are the same. This is equivalent to the first node still obtaining one dictionary.

[0171] 3. M>N>1, that is, M data spaces correspond to N dictionaries.

[0172] The first node may perform clustering training based on at least one third downlink channel subdata among the M third downlink channel subdata to obtain one dictionary (e.g., the first dictionary) among the N dictionaries. For example, if the third downlink channel subdata correspond one-to-one to the data space, the at least one third downlink channel subdata corresponds to at least one data space. One dictionary may correspond to one or more data spaces. For example, there is a correspondence between an index of the dictionary and an index of the data space. In this case, when training the dictionary, the first node performs training based on the third downlink channel subdata in the data space corresponding to the dictionary.

[0173] Optionally, the first node may separately train different dictionaries based on different bit overheads, and the bit overheads may correspond one-to-one to the dictionaries. In other words, one dictionary corresponds to one or more data spaces. For one or more data spaces, the first node may train one or more dictionaries based on different bit overheads. When multiple dictionaries are trained, the multiple dictionaries may correspond to different bit overheads, allowing the UE to select an appropriate dictionary based on the current bit overhead during network inference.

[0174] In addition, the first node may further determine the dimension of the elements included in the first dictionary based on the dimension of the third downlink channel data. For the determination method, see the above description. The first dictionary is, for example, a dictionary obtained by the first node through training.

[0175] After determining the number of elements included in the first dictionary and the dimensions of the elements included in the first dictionary, the first node may perform clustering training based on at least one third downlink channel subdata to obtain the first dictionary.

[0176] There may be multiple clustering training methods that can be applied in the present disclosure, such as the K-Means clustering method. Additionally, in the neural network training process, a loss function may be defined, which describes the gap or difference between an ideal target value and the output value of the neural network. In the clustering training process, a loss function may or may not be used. For example, the loss function may use the minimum distance between the clustering center and each of the multiple training samples as the target, or may use the training sample that is most related to the clustering center and is within the multiple training samples as the target. Additionally, the loss function may be another function, and the implementation of the loss function is not limited by the present disclosure. The dictionary training process is a process in which dictionary parameters are adjusted so that the value of the loss function is less than a threshold or so that the value of the loss function meets the target requirement. Adjusting the dictionary parameters may, for example, include adjusting the dictionary elements.

[0177] The contents contained in the dictionary have been explained in the embodiment shown in FIG. 5 and will not be explained in detail again.

[0178] In the above process, the first node obtains N dictionaries through training, so that the UE can use the N dictionaries in the network inference process in the embodiment shown in Figure 5, and the access network device can also use the N dictionaries during reconstructing downlink channel information. The data space and the N dictionaries are divided, so that environmental information corresponding to the downlink channel can be reflected, which helps the access network device reconstruct more accurate downlink channel information.

[0179] For ease of understanding, the following uses some accompanying drawings as examples to explain the network training process and the network inference process in the present disclosure.

[0180] FIG. 9A is a diagram of a training process and a network inference process according to the present disclosure. The training process in FIG. 9A and subsequent accompanying drawings uses an example in which a UE performs the training process. The process from the third downlink channel data q1 to q4, i.e., the process before the information is transmitted to the access network device, can be considered the training process. The entire process in FIG. 9A can also be considered a network inference process, and the network inference process can also be considered a data processing process. Of course, the data is not actually the training data used for training, but the data processing process is similar to the training data processing process. The training process includes obtaining a dictionary by performing clustering training on multiple training data. The processing process can be considered to represent the downlink channel data using the obtained dictionary.

[0181] In the training process, the training data is the original downlink channel data, and it is assumed that the original downlink channel data actually includes multiple training data (also called training samples). The UE processes each training data in the original downlink channel data to obtain an eigenvector, and the dimension of the eigenvector is [N tx =32,N sb=13]. The UE preprocesses the eigenvector to obtain sparse coefficients of the eigenvector, and the sparse coefficients of the eigenvector corresponding to the plurality of training data may be used as the third downlink channel data. Because real training is used for network training, the data input is divided into two parts, a real part and an imaginary part, and the dimension of the third downlink channel data is, for example, [E, 2, 32, 13]. The "E" in the dimension of the third downlink channel data is considered to be the amount of training data. In other words, in this case, the third downlink channel data may be considered to include E pieces of training data, where E is a positive integer. In the dimension of the third downlink channel data, "2" represents the real part and the imaginary part, and "32" represents the N tx and "13" represents N sb Represents.

[0182] The third downlink channel data is divided and allocated to M data spaces, i.e., the third downlink channel data is divided into M portions. In FIG. 9A, M=4 is used as an example, in which case, four third downlink channel subdata can be obtained by division. The four third downlink channel subdata are y1, y2, y3, and y4. The dimensions of y1, y2, y3, and y4 are [S, 16×13], respectively, where S represents the amount of training data corresponding to one third downlink channel subdata, and 16×13 is, for example, the dimension of the training target dictionary. q1 to q4 in FIG. 9A represent four training targets, i.e., N=M is ​​used as an example in FIG. 9A. The UE trains the four dictionaries using a clustering method. Optionally, in an offline training method, information about the four dictionaries may be agreed upon in a protocol or transmitted by the UE to an access network device. In the online training method, information about the four dictionaries may be sent by the UE to the access network device.

[0183] In the inference process, for example, the UE may obtain four pieces of first information based on four pieces of first downlink channel subdata and four dictionaries obtained by training, and one piece of first information is an element corresponding to one piece of first downlink channel subdata in the corresponding dictionary.

[0184] For example, the UE may transmit four identifiers of the first information to the access network device, where each identifier of the first information occupies X bits. After receiving the four identifiers of the first information, the access network device may reconstruct the four pieces of first information based on the four dictionaries. Then, the access network device performs processing such as concatenation on the four pieces of first information to reconstruct downlink channel information, i.e., to reconstruct a downlink channel matrix.

[0185] 9B is another diagram of the training process according to the present disclosure. The UE processes each training data in the original downlink channel data to obtain an eigenvector, and the dimension of the eigenvector is [N tx =32,N sb = 13]. The UE preprocesses the eigenvectors to obtain sparse coefficients of the eigenvectors, and the sparse coefficients of the eigenvectors corresponding to multiple training data included in the original downlink channel data can be used as the third downlink channel data. Because real-valued training is used for network training, the data input is divided into two parts, a real part and an imaginary part, and the dimension of the third downlink channel data is, for example, [E, 2, 32, 13]. Here, E is the amount of training data, and E is a positive integer.

[0186] The third downlink channel data is divided and allocated to M data spaces, i.e., the third downlink channel data is divided into M portions. In FIG. 9B, M=4 is used as an example, and in this case, the four third downlink channel subdata obtained by the division can be uniformly represented as y1, i.e., y1 can be considered to include four third downlink channel subdata. The dimension of y1 is [4×S, 16×13], where S represents the amount of training data corresponding to one third downlink channel subdata, and the amount of training data corresponding to four third downlink channel subdata is 4×S. In addition, 16×13 is, for example, the dimension of the training target dictionary. In FIG. 9B, q1 represents the training target dictionary to which the M data spaces uniformly correspond, i.e., N=1 is used as an example in FIG. 9B. The UE trains the dictionary using a clustering method.

[0187] In the network inference process, for each data space, the UE can independently find corresponding first information in the corresponding dictionary. As shown in FIG. 9B, if the M data spaces uniformly correspond to one dictionary, in the network inference process, for each data space, the UE can find corresponding first information in the dictionary obtained by training in FIG. 9B. FIG. 9B is still used as an example. In this case, the UE can determine four pieces of first information. For example, the UE can transmit identifiers of the four pieces of first information to the access network device. After receiving the identifiers of the four pieces of first information, the access network device can reconstruct the four pieces of first information based on the four dictionaries. Then, the access network device performs processing such as concatenation on the four pieces of first information to reconstruct downlink channel information, i.e., to reconstruct a downlink channel matrix.

[0188] 9A and 9B use an example in which the third downlink channel data is a pre-processing result of the original downlink channel data (or eigenvectors). From the description of the embodiment shown in FIG. 5, it can be known that the third downlink channel data may alternatively be F consecutive data strings extracted from the pre-processing result. In this case, the dimension of the elements included in the dictionary may change. For example, if the i-th third downlink channel subdata is one of M data obtained by dividing the F data strings extracted from the pre-processing result, the i-th third downlink channel subdata may be, for example, a matrix with dimensions [16, F], and the UE may convert this matrix into a vector with a length of 16×F. In this case, the dimension of the elements included in the first dictionary is 16×F. F is generally less than the number of subbands. In this case, the storage space occupied by the dictionary may be reduced.

[0189] 9C is yet another diagram of a training process according to the present disclosure. The UE processes each training data in the original downlink channel data to obtain an eigenvector, and the dimension of the eigenvector is [N tx =32,N sb= 13]. The UE preprocesses the eigenvectors to obtain sparse coefficients of the eigenvectors, and the sparse coefficients of the eigenvectors corresponding to multiple training data included in the original downlink channel data can be used as the third downlink channel data. Because real-valued training is used for network training, the data input is divided into two parts, a real part and an imaginary part, and the dimension of the third downlink channel data is, for example, [E, 2, 32, 13]. Here, E is the amount of training data, and E is a positive integer. To obtain M third downlink channel subdata, F consecutive data strings are extracted from the third downlink channel data, and the F consecutive data strings are divided and allocated to M data spaces. In FIG. 9C, M = 4 is used as an example, and the four third downlink channel subdata obtained by the division are y1, y2, y3, and y4. The dimensions of y1, y2, y3, and y4 are each [S, 16 × F]. 9C, q1 to q4 represent four dictionaries, that is, N=M is ​​used as an example in FIG. 9C. The UE trains the four dictionaries in a clustering method.

[0190] In the network inference process, for each data space, the UE can independently find the corresponding first information in the corresponding dictionary. Figure 9C is still used as an example. In this case, the UE can determine the four pieces of first information based on the four dictionaries obtained by the training of Figure 9C. For example, the UE transmits identifiers of the four pieces of first information to the access network device. After receiving the identifiers of the four pieces of first information, the access network device can reconstruct the four pieces of first information based on the four dictionaries. Then, the access network device performs processing such as concatenation on the four pieces of first information to reconstruct the downlink channel information, i.e., to reconstruct the downlink channel matrix.

[0191] 9D is another diagram of the training process according to the present disclosure. The UE processes each training data in the original downlink channel data to obtain an eigenvector, and the dimension of the eigenvector is [N tx =32,N sb = 13]. The UE preprocesses the eigenvector to obtain sparse coefficients of the eigenvector, and the sparse coefficients of the eigenvector corresponding to multiple training data included in the original downlink channel data can be used as the third downlink channel data. Because real training is used for network training, the data input is divided into two parts, a real part and an imaginary part, and the dimension of the third downlink channel data is, for example, [E, 2, 32, 13].

[0192] To obtain M third downlink channel subdata, F consecutive data strings are extracted from the third downlink channel data, and the F consecutive data strings are divided and allocated to M data spaces. In FIG. 9D, M=4 is used as an example. These four third downlink channel subdata can be uniformly represented as y1, and the dimension of y1 is [4×S, 16×13]. In FIG. 9B, q1 represents a dictionary to which the M data spaces uniformly correspond, that is, N=1 is used as an example in FIG. 9C. The UE trains the dictionary using a clustering method.

[0193] In the network inference process, for each data space, the UE can independently find corresponding first information in the corresponding dictionary. As shown in FIG. 9D, if the M data spaces uniformly correspond to one dictionary, in the network inference process, for each data space, the UE can find corresponding first information in the dictionary obtained by training in FIG. 9D. FIG. 9D is still used as an example. In this case, the UE can determine four pieces of first information. For example, the UE can transmit identifiers of the four pieces of first information to the access network device. After receiving the identifiers of the four pieces of first information, the access network device can reconstruct the four pieces of first information based on the four dictionaries. Then, the access network device performs processing such as concatenation on the four pieces of first information to reconstruct downlink channel information, i.e., to reconstruct a downlink channel matrix.

[0194] The network training process described in the embodiment shown in Figure 8 is a process of obtaining a dictionary by training. The above also describes that an encoder network may be configured on the UE side, and a decoder network corresponding to the encoder network may be configured on the access network device side. In this case, another network training process is a process of jointly training the encoder network, the decoder network, and the dictionary. The following describes yet another communication method in the present disclosure. In this method, a joint training process is described. Figure 10 is a flowchart of this method.

[0195] S1001: The second node acquires fifth downlink channel data.

[0196] The fifth downlink channel data may be, for example, the original downlink channel data. Alternatively, the fifth downlink channel data may be data obtained by preprocessing the original downlink channel data. Alternatively, the fifth downlink channel data may be data output by a neural network. The original downlink channel data may be considered training data or may be called training samples. In the process of training the dictionary, the second node needs to train the training samples. The original downlink channel data may include one or more training data.

[0197] If the fifth downlink channel data is obtained by pre-processing the original downlink channel data, a pre-processing process is included. For the pre-processing process of the original downlink channel data, please refer to the description of the pre-processing process of the second downlink channel data in S501 of the embodiment shown in Figure 5.

[0198] In the present disclosure, the second node may be, for example, a UE or an access network device, or a third-party device (e.g., an AI node). The training process may be an online training process or an offline training process. The second node and the first node in the embodiment shown in Figure 8 may be the same node or different nodes.

[0199] The second node may use the fifth downlink channel data to perform joint training on the encoder network, the dictionary, and the decoder network. The following describes the training process by performing S1002 to S1006.

[0200] S1002: The second node inputs the fifth downlink channel data to the encoder network to obtain sixth downlink channel data output by the encoder network.

[0201] The encoder network is an encoder network that needs to be trained. The second node inputs the fifth downlink channel data to the encoder network, and the encoder network may perform processing such as compression on the fifth downlink channel data. After processing, the encoder network outputs the sixth downlink channel data.

[0202] S1003: The second node acquires M sixth downlink channel subdata. Each sixth downlink channel subdata corresponds to one data space among the M data spaces. The M data spaces in the present disclosure and the M data spaces in the embodiment shown in FIG. 5 may have the same characteristics.

[0203] The M sixth downlink channel subdata are obtained based on the sixth downlink channel data. For example, the M sixth downlink channel subdata may be obtained by dividing the sixth downlink channel data and allocating it to M data spaces. For further details of S1001, please refer to S801 in the embodiment shown in Figure 8.

[0204] S1004: The second node obtains M pieces of third information based on the M pieces of sixth downlink channel subdata and the N training object dictionaries.

[0205] For example, in the process of training a dictionary, the second node may train a training target dictionary based on an ith data space among M data spaces, where i is an integer from 1 to M, and the second node may train M training target dictionaries. The second node trains a training target dictionary corresponding to a data space based on the ith data space among M data spaces. For example, in the training method, for the ith sixth downlink channel subdata among M sixth downlink channel subdata, if the second node obtains third information corresponding to the ith sixth downlink channel subdata based on the training target dictionary corresponding to the ith data space, the second node may obtain M pieces of third information in total. For example, the third information corresponding to the ith sixth downlink channel subdata is an element corresponding to the ith sixth downlink channel subdata in the training target dictionary corresponding to the ith data space.

[0206] Before training begins, an initial model may be set as a training dictionary, and multiple training rounds (here, a process of training using one training data may be considered as one training process) are performed on the initial model using original downlink channel data. After the training is completed, a dictionary to be used in the network inference stage may be obtained. Therefore, the training dictionary corresponding to the ith data space may be the initial model, or may be an intermediate model obtained by performing at least one training round on the initial model.

[0207] Optionally, in the process of training the dictionary, the second node may further train the training dictionary based on the M data spaces, and the second node may obtain the same M dictionaries or one dictionary through training. The second node trains the training dictionary based on the M data spaces. For example, in the training method, for an i-th sixth downlink channel subdata among the M sixth downlink channel subdata, if the second node obtains third information corresponding to the i-th sixth downlink channel subdata based on the training dictionary, the second node may obtain M pieces of third information in total. For example, the third information corresponding to the i-th sixth downlink channel subdata is an element in the training dictionary corresponding to the i-th sixth downlink channel subdata.

[0208] S1005: For the i-th third information among the M pieces of third information, the second node reconstructs the i-th fifth downlink channel subdata based on the training target dictionary corresponding to the i-th data space among the M data spaces. When i is an integer from 1 to M, the second node may obtain a total of M pieces of fifth downlink channel subdata.

[0209] In an ideal situation, the M fifth downlink channel subdata acquired by the second node and the M sixth downlink channel subdata acquired by the second node may be the same data. For example, the i-th sixth downlink channel subdata and the i-th fifth downlink channel subdata are the same data. In practical application, there may be a discrepancy between the M sixth downlink channel subdata and the M fifth downlink channel subdata. For details, please refer to S504 in the embodiment shown in FIG. 5.

[0210] For further details of S1005, please refer to S504 in the embodiment shown in FIG.

[0211] S1006: The second node inputs the M fifth downlink channel subdata to the decoder network to obtain L reconstructed information to be output by the decoder network, where L is a positive integer. Alternatively, the second node concatenates the M fifth downlink channel subdata and inputs the concatenated downlink channel subdata to the decoder network to obtain first reconstructed information to be output by the decoder network.

[0212] The decoder network is the decoder network that needs to be trained, and is also the decoder network corresponding to the encoder network of S1002.

[0213] For example, the original downlink channel data may include multiple training data, one of which may include training sub-data and a label. The second node may input the training sub-data to the encoder network to obtain encoded data. After the encoded data is processed by the decoder network, the decoder network may output an inference result (e.g., the L pieces of reconstructed information or the first reconstructed information described in this disclosure). The second node may calculate an error between the inference result and the label according to a loss function. Based on the error, the second node may optimize parameters of the encoder network and / or the decoder network according to a backpropagation optimization algorithm (or a model optimization algorithm, etc.). The encoder network and the decoder network are trained using a large amount of training data, whereby neural network training is completed after the difference between the output of the decoder network and the label becomes less than a preset value.

[0214] It should be noted that the above-described training process of the encoder network and the decoder network uses a supervised learning training method, i.e., a loss function is used to train the encoder network and the decoder network based on the training data and the label. Alternatively, the training process of the intelligent model may use unsupervised learning, in which an algorithm is used to learn the internal patterns of the training data to train the intelligent model based on the training data. Alternatively, the training process of the intelligent model may use reinforcement learning, in which an excitation signal that is fed back by the environment is obtained through interaction with the environment to learn a problem-solving policy and optimize the model. The present disclosure does not limit the model training method, model type, etc.

[0215] From the above description, it can be known that when training the encoder network and the decoder network, the second node may perform training according to a loss function. Optionally, the same loss function may be set for the M data spaces. In other words, for all data spaces within the M data spaces, the second node may perform joint training according to this loss function. For example, if the decoder network outputs L pieces of reconstructed information, the mean square error (MSE) between the fifth downlink channel data and data obtained by concatenating the L pieces of reconstructed information reconstructed by the decoder network may be used as the loss function, or the correlation between the third downlink channel data and data obtained by concatenating the L pieces of reconstructed information reconstructed by the decoder network may be used as the loss function. Alternatively, if the decoder network outputs the first reconstructed information, the MSE between the first reconstructed information and the fifth downlink channel data may be used as the loss function, or the correlation between the first reconstructed information and the third downlink channel data may be used as the loss function, etc.

[0216] Alternatively, different loss functions may be set for different data spaces. For example, if the decoder network outputs L pieces of reconstructed information, the MSE between the reconstructed information reconstructed by the decoder network and the data input to the encoder network may be used as the loss function corresponding to the data space. The reconstructed information reconstructed by the decoder network corresponding to the loss function is the reconstructed information reconstructed by the decoder network corresponding to the data space. The data input to the encoder network corresponding to the loss function is the data in the fifth downlink channel data input to the encoder network corresponding to the data space.

[0217] In the above process, the second node obtains N dictionaries and performs joint training on the encoder network, the decoder network, and the dictionaries to obtain an encoder network and a corresponding decoder network. In this case, the UE can use the N dictionaries and the encoder network in the network inference process in the embodiment shown in FIG. 5, and the access network device can also use the N dictionaries and the decoder network in the process of reconstructing downlink channel information. The data space and the N dictionaries are divided, which can reflect environmental information corresponding to the downlink channel, helping the access network device reconstruct more accurate downlink channel information. If the encoder network and the decoder network are not used in the embodiment shown in FIG. 5, the network training method provided in the embodiment shown in FIG. 8 can be used to train the dictionaries separately. If the encoder network and the decoder network need to be used in the embodiment shown in FIG. 5, the network training method provided in the embodiment shown in FIG. 10 can be used to obtain the encoder network, the decoder network, and the dictionary by joint training.

[0218] For example, FIG. 11 is a diagram of a training process and a network inference process according to the present disclosure. In the training process, for example, an encoder network, a decoder network, and a dictionary are obtained by joint training, and the encoder network, the decoder network, and the dictionary can be used in the network inference process. The process from the third downlink channel data q1 to q4, that is, the process before the information is transmitted to the access network device, can be considered a training process. The entire process in FIG. 11 can also be considered a network inference process, and the network inference process can also be considered a training data processing process. Of course, the data is not actually the training data used for training, but the data processing process is consistent with the training data processing process.

[0219] In the training process, it is assumed that the original downlink channel data may include multiple training data, and the UE processes each training data in the original downlink channel data to obtain an eigenvector, and the dimension of the eigenvector is [N tx =32,N sb =13]. The UE pre-processes the eigenvectors to obtain sparse coefficients of the eigenvectors. The UE compresses the sparse coefficients of the eigenvectors using an encoder network to obtain compressed information. The compressed information corresponding to the plurality of training data may be used as the third downlink channel data.

[0220] The UE divides the third downlink channel data into four data spaces to obtain four third downlink channel subdata, where the four third downlink channel subdata are y1, y2, y3, and y4. The dimensions of y1, y2, y3, and y4 are [S, 16×13], respectively, where S represents the amount of training data corresponding to one third downlink channel subdata, and 16×13 is, for example, the dimension of the training target dictionary. q1 to q4 in FIG. 11 represent four training target dictionaries, i.e., N=M is ​​used as an example in FIG. 11. The UE trains the four dictionaries using a clustering method.

[0221] In the inference process, for example, the UE may obtain four pieces of first information based on four training dictionaries and four pieces of first downlink channel subdata, where one piece of first information is an element in the corresponding dictionary corresponding to one piece of first downlink channel subdata.

[0222] For example, the UE may send four first information identifiers to the access network device, where each first information identifier occupies X bits. After receiving the four first information identifiers, the access network device may reconstruct the four compressed sub-information based on the four dictionaries. The access network device performs processing such as concatenation on the four compressed sub-information to obtain the reconstructed information output by the decoder network, and then inputs the processing result into the decoder network. After obtaining the reconstructed information output by the decoder network, the access network device may reconstruct downlink channel information based on the reconstructed information. The loss function in the embodiment shown in FIG. 10 may be applied in the training process, which improves the performance of the encoder / decoder network obtained by training.

[0223] The communication device provided in the present disclosure is described based on the above method embodiments.

[0224] It can be understood that to implement the functions in the above-mentioned manner, the access network device, the UE, etc. include corresponding hardware structures and / or software modules for performing the functions. Those skilled in the art should easily recognize that the present disclosure can be implemented by hardware or a combination of hardware and computer software with reference to the units and method steps in the examples described in the present disclosure. Whether the functions are performed by hardware or by hardware driven by computer software depends on the specific application scenario and design constraints of the technical solution.

[0225] The present disclosure provides a communications device. The communications device includes, for example, a processing unit and a transceiver unit (also referred to as a communications unit). The processing unit may be configured to implement the processing functions of the UE in the embodiment shown in FIG. 5, the embodiment shown in FIG. 8, or the embodiment shown in FIG. 10, and the transceiver unit may be configured to implement all or a portion of the transceiver functions of the UE in the embodiment shown in FIG. 5, the embodiment shown in FIG. 8, or the embodiment shown in FIG. 10. Alternatively, the processing unit may be configured to implement the processing functions implemented by the access network device in the embodiment shown in FIG. 5, the embodiment shown in FIG. 8, or the embodiment shown in FIG. 10, and the transceiver unit may be configured to implement all or a portion of the transceiver functions of the access network device in the embodiment shown in FIG. 5, the embodiment shown in FIG. 8, or the embodiment shown in FIG. 10.

[0226] Optionally, the processing unit and / or the transceiver unit may be implemented using virtual modules. For example, the processing unit may be implemented using a software functional unit or a virtual device, and the transceiver unit may be implemented using a software functional unit or a virtual device. Alternatively, the processing unit and / or the transceiver unit may be implemented using a physical device (e.g., a circuit system and / or a processor). The following describes the case where the processing unit and the transceiver unit are implemented using a physical device.

[0227] FIG. 12 is a diagram of the structure of a communication device according to the present disclosure. The communication device 1200 may be a UE, a circuit system of a UE, or a circuit system usable in a UE in the embodiment shown in FIG. 5, the embodiment shown in FIG. 8, or the embodiment shown in FIG. 10, and is configured to implement a method corresponding to the UE in the aforementioned method embodiments. Alternatively, the communication device 1200 may be an access network device, a circuit system of an access network device, or a circuit system usable in an access network device in the embodiment shown in FIG. 5, the embodiment shown in FIG. 8, or the embodiment shown in FIG. 10, and is configured to implement a method corresponding to the access network device in the aforementioned method embodiments. For specific functions, please refer to the description of the aforementioned method embodiments. For example, the circuit system is a chip system.

[0228] The communication device 1200 includes one or more processors 1201. The processor 1201 may perform specific control functions. The processor 1201 may be a general-purpose processor or a special-purpose processor. For example, the processor 1201 may include a baseband processor and a central processing unit. The baseband processor may be configured to process communication protocols and communication data. The central processing unit may be configured to control the communication device 1200, execute software programs, and / or process data. The different processors may be independent components or may be arranged in one or more processing circuits, for example, integrated in one or more application-specific integrated circuits.

[0229] Optionally, the communication device 1200 includes one or more memories 1202 for storing instructions 1204. The instructions 1204 may be executed on a processor such that the communication device 1200 executes the methods described in the preceding method embodiments. Optionally, the memory 1202 may further store data. The processor and memory may be located separately or integrated together. The memory may be a non-volatile memory, such as a hard disk drive (HDD) or a solid-state drive (SSD), or a volatile memory, such as a random access memory (RAM). The memory may be, but is not limited to, any other medium accessible by a computer that can hold or store appropriate program code in the form of instructions or data structures. Alternatively, the memory of the present disclosure may be a circuit or any other device capable of performing a storage function and configured to store program instructions and / or data.

[0230] Optionally, the communication device 1200 may store instructions 1203 (which may also be referred to as code or programs). The instructions 1203 may be executed on a processor such that the communication device 1200 executes the methods described in the above embodiments. The processor 1201 may store data.

[0231] For example, the processing unit is implemented using one or more processors 1201, or the processing unit is implemented using one or more processors 1201 and one or more memories 1202, or the processing unit is implemented using one or more processors 1201, one or more memories 1202, and instructions 1203.

[0232] Optionally, the communications device 1200 may further include a transceiver 1205 and an antenna 1206. The transceiver 1205 may be referred to as a transceiver unit, a transceiver, a transceiver circuit, a transceiver device, an input / output interface, etc., and is configured to perform transceiver functions of the communications device 1200 via the antenna 1206. For example, the transceiver unit is implemented using the transceiver 1205, or the transceiver unit is implemented using the transceiver 1205 and the antenna 1206.

[0233] Optionally, communication device 1200 may further include one or more of the following components: a wireless communication module, an audio module, an external memory interface, an internal memory, a universal serial bus (USB) interface, a power management module, an antenna, a speaker, a microphone, an input / output module, a sensor module, a motor, a camera, or a display. It may be understood that in some embodiments, communication device 1200 may include more or fewer components, or integration of some of the components, or separation of some of the components. These components may be implemented by hardware, software, or a combination of software and hardware.

[0234] The processor 1201 and the transceiver 1205 described in this disclosure may be implemented in an integrated circuit (IC), an analog IC, a radio frequency identification (RFID) integrated circuit, a mixed-signal IC, an application specific integrated circuit (ASIC), a printed circuit board (PCB), an electronic device, or the like. The communication apparatus described herein may be an independent device (e.g., an independent integrated circuit or a mobile phone) or may be part of a larger device (e.g., a module that can be incorporated into another device). For details, please refer to the descriptions of the UE and the access network device in the preceding embodiments. The details will not be described again here.

[0235] The present disclosure provides a terminal device, which may be used in the aforementioned embodiments. The terminal device includes corresponding means, units, and / or circuits for implementing the functions of the UE in the embodiment shown in Figure 5, the embodiment shown in Figure 8, or the embodiment shown in Figure 10. For example, the terminal device includes a transceiver module (also referred to as a transceiver unit) configured to support the terminal device in performing transceiver functions, and a processing module (also referred to as a processing unit) configured to support the terminal device in processing signals.

[0236] The present disclosure further provides an access network device, which may be used in the aforementioned embodiments. The access network device includes corresponding means, units, and / or circuits for performing the functions of the access network device in the embodiment shown in Figure 5, the embodiment shown in Figure 8, or the embodiment shown in Figure 10. For example, the access network device includes a transceiver module (also referred to as a transceiver unit) configured to support the access network device in performing transceiver functions, and a processing module (also referred to as a processing unit) configured to support the access network device in processing signals.

[0237] All or part of the technical solutions provided in the present disclosure may be implemented using software, hardware, firmware, or any combination thereof. When software is used to implement the embodiments, all or part of the embodiments may be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the procedures or functions according to the present disclosure are generated, in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, an access network device, a terminal device, an AI node, or another programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, optical fiber, or digital subscriber line (DSL)) or wireless (e.g., infrared, radio, or microwave) methods. The computer-readable storage medium may be any available medium that can be accessed by a computer, or a data storage device, such as a server or data center, that incorporates one or more available media. The available medium may be magnetic media (e.g., floppy disks, hard disk drives, or magnetic tapes), optical media (e.g., digital video discs (DVDs)), semiconductor media, or the like.

[0238] The above description is merely a specific embodiment of the present disclosure and is not intended to limit the scope of protection of the present disclosure. Any variations or replacements that can be easily conceived by those skilled in the art within the technical scope disclosed in the present disclosure shall fall within the scope of protection of the present disclosure. Therefore, the scope of protection of the present disclosure shall be subject to the scope of protection of the claims. [Explanation of symbols]

[0239] 10. Communication Systems 20 Access Network Devices 30 Communication equipment 1200 Communication Equipment 1201 processor 1202 memory 1203 Instructions 1204 Instructions 1205 transceiver 1206 Antenna

Claims

1. A communication method implemented by a first device or a chip of the first device, comprising: obtaining M first downlink channel subdata based on the first downlink channel data, where the M first downlink channel subdata correspond to M data spaces, where M is an integer greater than 1; For an ith first downlink channel subdata among the M first downlink channel subdata, determining first information corresponding to the ith first downlink channel subdata based on a first dictionary corresponding to the ith data space among the M data spaces, wherein a total of M pieces of first information are determined, i is an integer from 1 to M, the ith first downlink channel subdata corresponds to the ith data space, the first dictionary includes a plurality of elements, and the first information corresponding to the ith first downlink channel subdata corresponds to P elements among the plurality of elements, where P is a positive integer; transmitting first indication information, wherein the first indication information indicates the M pieces of first information; A communication method including:

2. The first indication information indicates identifiers of the M pieces of first information, and the step of transmitting the first indication information includes: transmitting the identifiers of the M pieces of first information in a first order, the first order being an arrangement order of the M pieces of data space; 2. The method of claim 1, comprising:

3. the first order is a predefined order, or a second indication is received, the second indication indicating the first order; or the first order is determined, and third instruction information is transmitted, the third instruction information indicating the first order. The method of claim 2.

4. The M first downlink channel subdata are obtained based on the first downlink channel data; the first downlink channel data is a pre-processed result; The first downlink channel data includes F consecutive data streams in the pre-processing result; or the first downlink channel data is compressed information obtained by compressing a pre-processing result; The pre-processing result is obtained by pre-processing second downlink channel data. The method of claim 1.

5. The division method of the M data spaces is predefined, or a fourth instruction is received, the fourth instruction indicating how to divide the M data spaces; or A division method of the M data spaces is determined, and fifth instruction information is transmitted, wherein the fifth instruction information indicates the division method of the M data spaces. The method of claim 1.

6. A communication method implemented by a second device or a chip of said second device, comprising: receiving first indication information from a first device, the first indication information indicating M pieces of first information, where M is an integer greater than 1; reconstructing an i-th second downlink channel subdata for an i-th first information among the M first information based on a first dictionary corresponding to an i-th data space among the M data spaces, wherein a total of M pieces of the second downlink channel subdata are obtained, the i-th first information corresponds to the i-th data space, where i is an integer from 1 to M, the first dictionary includes a plurality of elements, and the first information corresponding to the i-th second downlink channel subdata corresponds to P elements among the plurality of elements; reconstructing downlink channel information based on the M second downlink channel subdata; A communication method including:

7. The step of receiving first indication information includes: receiving the M first information identifiers in a first order, the first order being an arrangement order of the M data spaces; 7. The method of claim 6, comprising:

8. the first order is a predefined order, or a second indication is sent, the second indication indicating the first order; or third indication information is received, the third indication information indicating the first order; The method of claim 7.

9. 7. The method of claim 6, wherein the M data spaces correspond to M dictionaries, each of the data spaces corresponding to one of the dictionaries, or the M data spaces all correspond to the same dictionary.

10. the step of reconstructing downlink channel information based on the M second downlink channel subdata includes: obtaining compression information based on the M second downlink channel subdata; obtaining the downlink channel information based on the compressed information; 7. The method of claim 6, comprising:

11. The division method of the M data spaces is predefined, or A fourth instruction is sent, the fourth instruction indicating how to divide the M data spaces; or fifth instruction information is received, the fifth instruction information indicating how to divide the M data spaces; The method of claim 6.

12. A communications device configured to implement the method according to any one of claims 1 to 5 or configured to implement the method according to any one of claims 6 to 11.

13. A communications device comprising a processor and a memory, the memory coupled to the processor, the processor configured to perform the method of any one of claims 1 to 5 or the method of any one of claims 6 to 11.

14. 12. A computer-readable storage medium configured to store a computer program, the computer program being run on a computer to enable the computer to perform the method of any one of claims 1 to 5, or to enable the computer to perform the method of any one of claims 6 to 11.

15. A communication system comprising an apparatus configured to implement the method according to any one of claims 1 to 5 and / or an apparatus configured to implement the method according to any one of claims 6 to 11.

Images