WO2025213482A1 - Ai/ml模型或功能的操作处理方法、装置和通信系统 - Google Patents

Ai/ml模型或功能的操作处理方法、装置和通信系统

Info

Publication number
WO2025213482A1
WO2025213482A1 PCT/CN2024/087633 CN2024087633W WO2025213482A1 WO 2025213482 A1 WO2025213482 A1 WO 2025213482A1 CN 2024087633 W CN2024087633 W CN 2024087633W WO 2025213482 A1 WO2025213482 A1 WO 2025213482A1
Authority
WO
WIPO (PCT)
Prior art keywords
csi
layer
model
models
information
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
PCT/CN2024/087633
Other languages
English (en)
French (fr)
Inventor
金立强
王昕�
王国童
张群
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Fujitsu Ltd
Original Assignee
Fujitsu Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Fujitsu Ltd filed Critical Fujitsu Ltd
Priority to PCT/CN2024/087633 priority Critical patent/WO2025213482A1/zh
Publication of WO2025213482A1 publication Critical patent/WO2025213482A1/zh
Pending legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04WWIRELESS COMMUNICATION NETWORKS
    • H04W24/00Supervisory, monitoring or testing arrangements
    • H04W24/04Arrangements for maintaining operational condition

Definitions

  • the embodiments of the present application relate to the field of communication technologies.
  • AI/ML-based CSI feedback compression uses a two-sided model, with the AI/ML model or function located on both the terminal equipment (UE) side and the network (NW) side (i.e., the gNB side).
  • UE terminal equipment
  • NW network
  • AI/ML-based feedback compression compresses CSI in the spatial frequency domain (SF-AI/ML CSI compression).
  • TSF-AI/ML CSI compression requires the use of historical CSI information.
  • the availability and usability of historical CSI information is affected by factors such as rank changes, CSI drop (e.g., CSI drop based on layer priority), and uplink control information (UCI) loss.
  • rank changes e.g., CSI drop based on layer priority
  • UCI uplink control information
  • CSI temporal correlation determines whether historical CSI auxiliary information is helpful for current CSI compression. Too low correlation will lead to invalid input of AI/ML models or functions, which will not only fail to improve CSI compression performance, It may even be treated as noise by AI/ML models or functions, degrading CSI compression performance. Therefore, TSF-AI/ML is more suitable for scenarios with good CSI time-domain correlation, such as those corresponding to layers with low UE mobility, low rank values, and large eigenvalues.
  • life cycle management (LCM) operations of AI/ML models or functions are the same for all layers and/or rank values, that is, there is no distinction between layers and/or rank values.
  • embodiments of the present application provide an operation processing method, device and communication system for an AI/ML model or function.
  • a method for operating and processing an AI/ML model or function is provided, which is applied to a multi-antenna (MIMO) communication system.
  • the method is applied to the terminal side, and the method includes: receiving configuration information from a network device; and sending first information to the network device based on the configuration information, where the first information is information obtained after the terminal device operates the AI/ML model or function according to the precoded layer and/or rank value.
  • MIMO multi-antenna
  • a method for operating and processing an AI/ML model or function is provided, which is applied to a multi-antenna (MIMO) communication system.
  • the method is applied on the network side, and the method includes: sending configuration information to a terminal device; receiving first information, which is information obtained after the terminal device operates the AI/ML model or function according to the precoded layer and/or rank value.
  • MIMO multi-antenna
  • an operation processing device for an AI/ML model or function which is applied to a multi-antenna (MIMO) communication system.
  • the device is applied to a terminal side, and the device includes: a first receiving unit, which receives configuration information from a network device; a first sending unit, which sends first information to the network device based on the configuration information, wherein the first information is information obtained after the terminal device operates the AI/ML model or function according to the precoded layer and/or rank value.
  • MIMO multi-antenna
  • an operation processing device of an AI/ML model or function which is applied to a multi-antenna (MIMO) communication system, and the device is applied to the network side, and the device includes: a second sending unit that sends configuration information to a terminal device; a second receiving unit that receives first information, and the first information The information is obtained after the terminal device operates the AI/ML model or function according to the precoded layer and/or rank value.
  • MIMO multi-antenna
  • a terminal device comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to execute the computer program to implement an operation processing method of the AI/ML model or function on the terminal device side.
  • a network device comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to execute the computer program to implement an operation processing method of the AI/ML model or function on the above-mentioned network device side.
  • a communication system includes: a terminal device, which receives configuration information from a network device; based on the configuration information, sends first information to the network device, the first information being information obtained after the terminal device performs an operation on an AI/ML model or function according to a precoded layer and/or rank value; and/or, a network device, which sends configuration information to the terminal device; receives the first information, the first information being information obtained after the terminal device performs an operation on an AI/ML model or function according to a precoded layer and/or rank value.
  • the AI/ML model or function is operated according to the precoded layer and/or rank value, and the AI/ML model or function can be flexibly utilized according to the layer and/or rank value, giving full play to the advantages of the AI/ML model or function, and ensuring the performance of the AI/ML model or function and the performance of the communication system.
  • these layers can switch to SF-AI/ML or non-AI/ML methods for CSI feedback, which can avoid the performance deterioration of the AI/ML model or function and ensure the performance of the communication system.
  • FIG1 is a schematic diagram of a communication system according to an embodiment of the present application.
  • Figure 2 is a schematic diagram of a use case of the TSF-AI/ML CSI compression feedback enhancer in the spatiotemporal and frequency domains.
  • FIG3 is a schematic diagram of an operation and processing method of an AI/ML model or function according to an embodiment of the present application
  • FIG4 is another schematic diagram of an operation processing method of an AI/ML model or function according to an embodiment of the present application.
  • FIG5 is a schematic diagram of an example of performing layer-by-layer operations on an AI/ML model or function according to an embodiment of the present application
  • FIG6 is a schematic diagram of another example of performing layer-by-layer operations on an AI/ML model or function according to an embodiment of the present application.
  • FIG7 is a schematic diagram of another example of operating an AI/ML model or function by layer according to an embodiment of the present application.
  • FIG8 is a schematic diagram of an example of using the output of the TSF-AI/ML model to acquire or recover historical CSI information according to an embodiment of the present application;
  • FIG9 is a schematic diagram of an operation processing device for an AI/ML model or function according to an embodiment of the present application.
  • FIG10 is another schematic diagram of an operation processing device for an AI/ML model or function according to an embodiment of the present application.
  • FIG. 11 is a schematic block diagram of a system structure of a terminal device according to an embodiment of the present invention.
  • FIG12 is a schematic block diagram of the system structure of the network device according to an embodiment of the present application.
  • the terms “first”, “second”, etc. are used to distinguish different elements from the name, but do not indicate the spatial arrangement or temporal order of these elements, and these elements should not be limited by these terms.
  • the term “and/or” includes any one and all combinations of one or more of the associated listed terms.
  • the terms “comprising”, “including”, “having”, etc. refer to the presence of the stated features, elements, components or components, but do not exclude the presence or addition of one or more other features, elements, components or components.
  • the term “communication network” or “wireless communication network” may refer to a network that complies with any of the following communication standards, such as New Radio (NR), Long Term Evolution (LTE), Enhanced Long Term Evolution (LTE-A, LTE-Advanced), Wideband Code Division Multiple Access (WCDMA), High-Speed Packet Access (HSPA), etc.
  • NR New Radio
  • LTE Long Term Evolution
  • LTE-A Enhanced Long Term Evolution
  • WCDMA Wideband Code Division Multiple Access
  • HSPA High-Speed Packet Access
  • communication between devices in the communication system may be carried out according to communication protocols of any stage, such as but not limited to the following communication protocols: 1G (generation), 2G, 2.5G, 2.75G, 3G, 4G, 4.5G and 5G, New Radio (NR), 6G and future communications, etc., and/or other communication protocols currently known or to be developed in the future.
  • 1G generation
  • 2G 2.5G
  • 2.75G 3G
  • 4G 4G
  • 4.5G and 5G 3G
  • NR New Radio
  • 6G and future communications etc.
  • other communication protocols currently known or to be developed in the future.
  • Network device refers to, for example, a device in a communication system that connects a terminal device to a communication network and provides services for the terminal device.
  • Network devices may include, but are not limited to, the following devices: base station (BS), access point (AP), transmission reception point (TRP), broadcast transmitter, mobile management entity (MME), gateway, server, radio network controller (RNC), base station controller (BSC), etc.
  • the base station may include but is not limited to: Node B (NodeB or NB), evolved Node B (eNodeB or eNB), 5G base station (gNB), 6G base station and future base stations, etc., and may also include remote radio heads (RRH, Remote Radio Head), remote radio units (RRU, Remote Radio Unit), relays (relay) Or low-power nodes (such as femto, pico, etc.).
  • NodeB Node B
  • eNodeB or eNB evolved Node B
  • gNB 5G base station
  • 6G base station and future base stations etc.
  • RRH Remote Radio Head
  • RRU Remote Radio Unit
  • relays relays
  • low-power nodes such as femto, pico, etc.
  • base station can include some or all of their functions, each base station can provide communication coverage for a specific geographical area.
  • the term "cell” can refer to a base station and/or its coverage area, depending on the context in which the term is used.
  • the term "user equipment” (UE) or “terminal equipment” (TE) refers to, for example, a device that accesses a communication network through a network device and receives network services.
  • a user equipment may be fixed or mobile and may also be referred to as a mobile station (MS), a terminal, a user, a subscriber station (SS), an access terminal (AT), a station, a mobile terminal (MT), and so on.
  • terminal devices may include but are not limited to the following devices: cellular phones, personal digital assistants (PDAs), wireless modems, wireless communication devices, handheld devices, machine-type communication devices, laptop computers, cordless phones, smart phones, smart watches, digital cameras, etc.
  • PDAs personal digital assistants
  • wireless modems wireless communication devices
  • handheld devices machine-type communication devices
  • machine-type communication devices laptop computers
  • cordless phones smart phones
  • smart watches digital cameras
  • the user equipment may also be a machine or device for monitoring or measurement, including but not limited to: machine type communication (MTC) terminals, vehicle-mounted communication terminals, device-to-device (D2D) terminals, machine-to-machine (M2M) terminals, terminals supporting sidelink communication, and the like.
  • MTC machine type communication
  • D2D device-to-device
  • M2M machine-to-machine
  • network side or “network device side” refers to one side of the network, which can be a base station or one or more network devices as described above.
  • user side or “terminal side” or “terminal device side” refers to the user or terminal side, which can be a UE or one or more terminal devices as described above.
  • device can refer to either network equipment or terminal equipment.
  • uplink control signal and “uplink control information (UCI)” or “physical uplink control channel (PUCCH)” are interchangeable, and the terms “uplink data signal” and “uplink data information” or “physical uplink shared channel (PUSCH)” are interchangeable.
  • downlink control signal and “downlink control information (DCI)” or “physical downlink control channel (PDCCH)” are interchangeable, and the terms “downlink data signal” and “downlink data information” or “physical downlink shared channel (PDSCH)” are interchangeable.
  • DCI downlink control information
  • PDCCH physical downlink control channel
  • the uplink signal may include an uplink data signal and/or an uplink control signal and/or a PRACH and/or an SRS (sounding reference signal), etc., which may also be referred to as an uplink transmission (UL transmission) or an uplink information or an uplink channel.
  • UL transmission uplink transmission
  • Sending/receiving an uplink transmission on an uplink resource may be understood as sending/receiving the uplink transmission using the uplink resource.
  • the downlink signal may include a downlink data signal and/or a downlink control signal and/or a synchronization signal (SS, such as PSS/SSS) and/or a broadcast channel (PBCH) and/or an SSB (SS/PBCH block, including PSS, SSS and PBCH and its DMRS) and/or a CSI-RS, etc., which may also be referred to as a downlink transmission (DL transmission) or a downlink information or a downlink channel.
  • SS downlink control signal and/or a synchronization signal
  • PBCH broadcast channel
  • SSB SS/PBCH block, including PSS, SSS and PBCH and its DMRS
  • CSI-RS CSI-RS
  • high-layer signaling may be, for example, radio resource control (RRC) signaling;
  • RRC signaling may include, for example, RRC messages (RRC message), for example, broadcast/public RRC messages/signaling (such as master information block (MIB), system information (systeminformation, SI), dedicated RRC messages/signaling; or RRC information elements (RRC information element, RRC IE); or information fields included in RRC messages or RRC information elements (or information fields included in information fields).
  • High-layer signaling may also be, for example, medium access control layer (Medium Access Control, MAC) signaling; or called MAC control element (MAC control element, MAC CE).
  • Medium Access Control Medium Access Control
  • MAC control element MAC control element
  • Configuration/indication refers to direct or indirect configuration/indication by the network device through high-layer signaling and/or physical layer signaling. Configuration/indication can be achieved by introducing high-layer parameters in high-layer signaling, and high-layer parameters refer to information fields (fields) and/or information elements/information units/information elements (IEs) in high-layer signaling.
  • Physical layer signaling refers to, for example, control information (DCI) carried by the physical downlink control channel or control information carried by the sequence, but is not limited thereto.
  • a base station as an example of an access network device.
  • FIG1 is a schematic diagram of a communication system according to an embodiment of the present application, schematically illustrating a situation in which a terminal device and a network device are used as an example.
  • a communication system 100 may include a network device 101, a terminal device 102, and a terminal device 103.
  • FIG1 only illustrates two terminal devices and one network device as an example.
  • the embodiments of the present application are not limited thereto.
  • existing services or future services can be transmitted between the network device 101, the terminal device 102, and the terminal device 103.
  • these services may include, but are not limited to, enhanced mobile broadband (eMBB), massive machine type communication (mMTC), ultra-reliable and low-latency communication (URLLC), and communications related to reduced-capability terminal devices, etc.
  • eMBB enhanced mobile broadband
  • mMTC massive machine type communication
  • URLLC ultra-reliable and low-latency communication
  • communications related to reduced-capability terminal devices etc.
  • terminal devices 102 and 103 can be in the RRC_IDLE state, the RRC_INACTIVE state, or the RRC_CONNECTED state.
  • Terminal devices 102 and 103 can also communicate with network device 101.
  • network device 101 can send a paging message to terminal device 102 or send data to terminal device 102, and terminal device 102 receives data sent by network device 101.
  • different terminal devices can also communicate with each other.
  • terminal device 102 and terminal device 103 can exchange data.
  • terminal device 102 and terminal device 103 are both within the coverage range of network device 101, but the present application is not limited thereto.
  • Terminal device 102 and terminal device 103 may both be outside the coverage range of network device 101, or one of terminal device 102 and terminal device 103 may be within the coverage range of network device 101 and the other may be outside the coverage range of network device 101.
  • one or more AI/ML models may be configured and run in a network device and/or a terminal device.
  • the AI/ML models may be used for various signal processing functions in wireless communications, such as CSI prediction, CSI compression, beamforming, positioning management, and the like; however, the present application is not limited thereto.
  • Figure 2 illustrates a sub-use case for the TSF-AI/ML CSI compression feedback enhancement in the time-space frequency domain.
  • the CSI measurement results undergo matrix decomposition (e.g., SVD (singular value decomposition) or EVD (eigenvalue decomposition)) to generate feature vectors. These feature vectors are then fed into an encoder, which generates CSI feedback information.
  • a decoder decodes the CSI feedback information sent by the UE to generate reconstructed CSI.
  • the UE-side encoder and decoder may employ AI/ML methods (e.g., RNN/LSTM/GRU cascaded Transformer/CNN models). These AI/ML methods may utilize time-domain information (also known as historical CSI information, side information, or similar terms) to assist in current CSI compression and/or decompression, aiming to achieve higher compression rates or higher feedback accuracy.
  • AI/ML methods e.g., RNN/LSTM/GRU cascaded Transformer/CNN models.
  • time-domain information also known as historical CSI information, side information, or similar terms
  • time domain information When using time domain information for CSI compression and/or decompression, is the time domain information available and/or usable? It is susceptible to factors such as rank changes, CSI drop (e.g., CSI drop based on layer priority), uplink control information (UCI) loss, etc. When time domain information cannot be obtained or is unavailable, the performance of the AI/ML model will deteriorate or even fail to work.
  • rank changes e.g., rank changes
  • CSI drop e.g., CSI drop based on layer priority
  • UCI uplink control information
  • CSI time-domain correlation determines whether historical CSI information contributes to current CSI compression and/or decompression. Too low a correlation will result in invalid input to the AI/ML model, which will not only fail to improve CSI compression performance but may even be treated as noise by the AI/ML model, degrading CSI compression performance. Therefore, AI/ML is more suitable for scenarios with good CSI time-domain correlation, which typically corresponds to lower UE mobility or layers with lower rank values and/or larger eigenvalues.
  • CSI compression may also be referred to as "CSI encoding", “CSI generation” or other similar names, and the results of operations such as CSI compression, CSI encoding, or CSI generation may be referred to as CSI feedback information, CSI reporting information or other similar names.
  • CSI decompression may also be referred to as “CSI decoding”, “CSI reconstruction”, “CSI recovery”, “CSI reconstruction” or similar names.
  • the AI/ML model may also be referred to as an AI/ML method, AI/ML unit, AI/ML function, or AI/ML element.
  • the AI/ML model On the UE side, the AI/ML model may also be referred to as an encoder, CSI generation part, or similar term.
  • the AI/ML model On the network device side, the AI/ML model may also be referred to as a decoder, CSI reconstruction part, or similar term.
  • the use cases to which the operation processing method of the AI/ML model or function in the embodiment of the present application is applicable include but are not limited to CSI compression feedback.
  • the embodiment of the present application is also applicable to various other use cases and/or scenarios for applying AI/ML models or functions.
  • An embodiment of the present application provides an operation and processing method for an AI/ML model or function, which is applied to a multi-antenna (MIMO) communication system and is described from the terminal device side and/or the network side.
  • MIMO multi-antenna
  • FIG3 is a schematic diagram of an operation processing method of an AI/ML model or function according to an embodiment of the present application. As shown in FIG3 , from the terminal (UE) side, the method includes:
  • the 302 Send first information to the network device according to the configuration information, where the first information is information obtained after the terminal device operates the AI/ML model or function according to the precoded layer and/or rank value.
  • FIG4 is another schematic diagram of an operation processing method of an AI/ML model or function according to an embodiment of the present application. As shown in FIG4 , from the network (NW) side, the method includes:
  • the operation of the AI/ML model or function is performed according to the precoded layer and/or rank value, and the AI/ML model or function can be flexibly utilized according to the layer and/or rank value, giving full play to the advantages of the AI/ML model or function, and ensuring the performance of the AI/ML model or function and the performance of the communication system.
  • the AI/ML model on the UE side and the AI/ML model on the network side may correspond or match.
  • the present application is not limited thereto, and the AI/ML model may be used only on the UE side or only on the network side.
  • the AI/ML model refers to the model on the UE side and/or the model on the network side.
  • “operating an AI/ML model or function according to a pre-coded layer and/or rank value” may be referred to simply as “operating an AI/ML model or function according to a layer and/or rank value”.
  • the rank value indicates the number of pre-coded layers. For example, a rank value of 1 indicates that the number of pre-coded layers is 1, i.e., layer 1; a rank value of 2 indicates that the number of pre-coded layers is 2, i.e., layer 1 and layer 2; and so on.
  • the configuration information is at least used to send the first information.
  • the configuration information includes but is not limited to resource configuration information and/or reporting configuration information used to send the first information.
  • the configuration information may include one or more predefined rules for operating the AI/ML model or function according to the layer and/or rank value; for example, the rules include but are not limited to: on which layers to perform the operation of one or several AI/ML models or functions, and/or on which layers to perform the operation of one or several AI/ML models or functions when the rank value changes.
  • the rule may not be included in the configuration information, but may be sent through another configuration information or other information.
  • the configuration information may include an indication that the terminal device can perform operations of one or several AI/ML models or functions on a certain layer or layers.
  • the indication may not be included in the configuration information, but may be sent through another configuration information or other information.
  • one or more of the above configuration information may be sent to the terminal device via RRC signaling.
  • the present application is not limited thereto, and one or more of the above configuration information may also be sent via other means.
  • the operation of the AI/ML model or function is also referred to as life cycle management (LCM) operation of the AI/ML model or function.
  • LCM life cycle management
  • the operations of the AI/ML model or function include, but are not limited to, one or more of the following operations:
  • the operation of AI/ML models or functions according to layers and/or rank values is based on predefined rules, which include but are not limited to: on which layers to perform the operation of one or more AI/ML models or functions, and/or on which layers to perform the operation of one or more AI/ML models or functions when the rank value changes.
  • the rule is related to rank indication (RI) changes: when the feedback RI value is different from the last feedback RI value, the AI/ML model or function is operated according to the layer and/or rank value;
  • the terminal device implicitly indicates to the network device whether it has performed the operation of the AI/ML model or function through RI value feedback: when the RI changes, the terminal device takes action; when the RI remains unchanged, the terminal device does not take any action.
  • the operation of AI/ML models or functions according to layers and/or rank values is based on: one or more predefined rules configured on the network side, including but not limited to: on which layers to perform the operation of one or more AI/ML models or functions, and/or on which layers to perform the operation of one or more AI/ML models or functions when the rank value changes;
  • the rules configured on the network side may be included in the configuration information or may not be included in the configuration information.
  • the terminal device selects one of the rules, performs the operation and notifies the network side of the corresponding operation.
  • the notification includes explicit or implicit notification, wherein the implicit notification is, for example, through reporting RI, resource allocation, CSI discard, etc.
  • the notification includes at least one of the following two notifications:
  • Inform network devices about the rules based on which the terminal device performs the operation of the AI/ML model or function.
  • the rule is related to RI changes: when the feedback RI value is different from the last feedback RI value, the terminal device operates the AI/ML model or function according to the layer and/or rank value based on a certain rule.
  • the terminal device implicitly indicates to the network device whether it has performed the operation of the AI/ML model or function through RI value feedback: when the RI changes, the terminal device takes action; when the RI remains unchanged, the terminal device does not take any action.
  • the terminal device uses an indication message to indicate to the network device which rule it is using to operate the AI/ML model or function. If the network device is configured with only one rule, the indication message can be left blank or have a cost of zero.
  • the operation of AI/ML models or functions according to layers and/or rank values is based on: the network indicating that the terminal device can perform the operation of one or more AI/ML models or functions on a certain layer or layers;
  • the indication on the network side may be included in the configuration information or may not be included in the configuration information
  • the configuration information is determined by the network device based on the operation of the terminal device requesting the AI/ML model or function.
  • the operation of AI/ML models or functions according to layers and/or ranks is based on the terminal device determining to perform one or more AI/ML model or function operations on a particular layer or layers and notifying the network device of the operations. For example, the terminal device determines that it needs to perform one or more AI/ML model or function operations on a particular layer or layers and notifies the network device of the specific operations.
  • the method further includes:
  • the terminal device performs an operation of the AI/ML model or function according to the layer and/or rank value to obtain or generate the first information
  • the method further includes:
  • the network device performs an AI/ML model on the received first information according to the layer and/or rank value.
  • the network device may perform an AI/ML model or function operation based on the layer and/or rank value for other information or other situations.
  • the network device may perform an AI/ML model or function operation based on the layer and/or rank value for the first information or for other information or other situations.
  • the terminal device and/or the network device performs AI/ML model or function operations on some or all layers and/or rank values, and/or performs the same or different AI/ML model or function operations on different layers and/or rank values.
  • operations on AI/ML models or functions according to layer and/or rank values are applied to two-sided AI/ML models or functions, or to one-sided AI/ML models or functions.
  • the operation of the AI/ML model or function according to the layer and/or rank value is applied to at least one or more of the following:
  • the layer-specific AI/ML model or function includes: a layer-specific and rank-common AI/ML model or function, and/or a layer-specific and rank-specific AI/ML model or function;
  • the layer-common AI/ML models or functions include: layer-common and rank-common AI/ML models or functions, and/or layer-common and rank-specific AI/ML models or functions.
  • the method further includes:
  • the method further includes:
  • Operation 304 and operation 404 are optional operations.
  • the monitoring of the AI/ML model or function includes: monitoring the performance of the model or function of at least one layer and/or rank; and/or monitoring whether historical CSI information of at least one layer and/or rank is available.
  • Figures 3 and 4 above are merely schematic illustrations of the embodiments of the present application, and the present application is not limited thereto.
  • the execution order of the various operations may be appropriately adjusted, and other operations may be added or some operations may be reduced.
  • Those skilled in the art may make appropriate modifications based on the above content, and are not limited to the descriptions of Figures 3 and 4 above.
  • the following examples illustrate how to operate AI/ML models or functions based on layers and/or rank values.
  • FIG5 is a schematic diagram of an example of performing layer-by-layer operations on an AI/ML model or function according to an embodiment of the present application, wherein the rank value remains unchanged.
  • both Layer 1 and Layer 2 use AI/ML model A for inference.
  • Layer 2's AI/ML model A is switched to AI/ML model B, while Layer 1's AI/ML model remains unchanged.
  • AI/ML model A is TSF-AI/ML and the UE and/or network detects time-domain correlation of CSI in Layer 2 or low TSF-AI/ML feedback accuracy
  • Layer 2 is switched to the less complex SF-AI/ML, or AI/ML model B.
  • FIG6 is a schematic diagram of another example of performing layer-by-layer operations on an AI/ML model or function according to an embodiment of the present application, wherein the rank value changes.
  • the rank value changes from 2 to 1, at which point AI/ML model A only operates in layer 1.
  • the rank value changes from 1 to 2, at which point AI/ML model A in layer 2 is switched to model B.
  • the AI/ML model cannot obtain CSI information in layer 2 at time T2.
  • the second layer switches to the less complex SF-AI/ML, i.e., AI/ML model B.
  • FIG7 is a schematic diagram of another example of performing layer-by-layer operations on an AI/ML model or function according to an embodiment of the present application, wherein the rank value remains unchanged.
  • AI/ML model A2 is TSF-AI/ML
  • its input is 2-layer CSI.
  • the UE and/or network detects the time domain correlation of CSI or the TSF-AI/ML feedback accuracy is too low, and the AI/ML model A2 is switched to B2.
  • the AI/ML model is switched to SF-AI/ML, which is AI/ML model B2.
  • a CSI compression feedback scheme is determined per layer.
  • the same CSI compression feedback scheme can be used on all layers at the same time, or different CSI compression feedback schemes can be used on different layers.
  • the same CSI compression feedback scheme may be used on the same layer, or different CSI compression feedback schemes may be used on the same layer.
  • the CSI compression feedback scheme includes but is not limited to one or more of the following:
  • TSF-AI/ML time-space frequency domain
  • AI/ML-based spatial-frequency domain SF-AI/ML
  • the AI/ML model or function requires CSI information at one or more historical moments.
  • the layer switches to SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods, so that the TSF-AI/ML model or function obtains, restores or resets the historical CSI information; when the historical CSI information is obtained, restored or reset, the TSF-AI/ML CSI compression feedback is activated or used at the layer.
  • the switching is applied to the initial operation or reactivation of the TSF-AI/ML model or function, or to the situation where historical CSI information is unavailable due to rank value change or UCI loss or CSI discard.
  • one or more layers need to switch to SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods.
  • the time window may also be called a warming-up window, an initialization window, or a re-initialization window (warming up/initialization/re-initialization window).
  • the embodiments of the present application do not limit the name of the time window.
  • the time window is defined according to the number of time moments required for the TSF-AI/ML model or function to obtain or recover or reset historical CSI information, and the definition is explicit or implicit.
  • the time window is applied to the initial operation or reactivation of the TSF-AI/ML model or function, or to situations where historical CSI information is unavailable due to rank value changes or UCI loss or CSI discard.
  • the time window is predefined in the standard, or is based on RRC configuration and/or updated and/or indicated through MAC-CE or DCI.
  • the time window is used on the terminal side and/or the network side.
  • the time windows on both sides may be the same or different.
  • the size of the time window is the same or different for different layers and/or rank values.
  • the size of the time window depends on the terminal capabilities of the terminal device.
  • whether to switch to SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods within the time window is configured by predefined configuration, or based on RRC signaling configuration and/or based on MAC CE or DCI update and/or indication.
  • whether to support switching to SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods depends on the terminal capabilities of the terminal device.
  • whether to support simultaneous operation of TSF-AI/ML CSI compression feedback and non-TSF-AI/ML CSI compression feedback depends on the terminal capabilities of the terminal device.
  • whether to switch to SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods within the time window is determined by the terminal device.
  • whether to switch to SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods at one or more layers is reported by the terminal device to the network device.
  • TSF-AI/ML CSI compression feedback is applied only to the first layer, and SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods is applied to other layers, or,
  • SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods is applied to the first layer and other layers.
  • a combination of one or more of the following methods can be used to obtain or restore the historical CSI information:
  • Method 1 The layer switches to SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods, while the TSF-AI/ML encoder output for the layer is still sent to the network side over the air interface for updating the network side's historical CSI information.
  • the network side obtains or recovers the historical CSI information by inputting the fed-back TSF-AI/ML encoder output into the TSF-AI/ML decoder.
  • Method 2 There is a proxy or approximate TSF encoder on the network side.
  • the network side obtains or recovers the historical CSI information at one or more moments through SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods; wherein, the network side reconstructs the eigenvector and/or channel matrix of the layer through SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods, and inputs the reconstructed eigenvector and/or channel matrix into the proxy or approximate TSF encoder, and then inputs the output of the proxy or approximate TSF encoder into the TSF decoder to obtain or recover the historical CSI information on the network side.
  • Method 3 When the historical CSI information on the network side is reconstructed CSI, the layer switches to SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods, and uses the reconstructed CSI based on SF-AI/ML CSI compression feedback or CSI feedback based on non-AI/ML methods to obtain or recover historical CSI information.
  • SF-AI/ML CSI compression feedback or non-AI/ML-based methods are used for CSI feedback.
  • TSF-AI/ML participates in feedback, it does not involve CSI feedback and PDSCH scheduling and is only used for obtaining/recovering historical CSI information.
  • the LCM switches the TSF-AI/ML model to the SF-AI/ML model, or a non-AI/ML method (e.g., falling back to the existing codebook solution).
  • the SF-AI/ML or non-AI/ML methods of these layers are used not only for CSI feedback but also for historical CSI information acquisition/recovery.
  • the method used to obtain or restore historical CSI information is configured by the network side, that is, the network side can configure which method or a combination of multiple methods to use to obtain or restore historical CSI information;
  • the terminal device may report a recommended method (which method or a combination of multiple methods) to the network side, and the network side ultimately determines the method for obtaining or restoring historical CSI information.
  • a recommended method which method or a combination of multiple methods
  • FIG8 is a schematic diagram of an example of using the output of the TSF-AI/ML model to acquire or recover historical CSI information according to an embodiment of the present application.
  • the rank value changes from 1 to 2, so the TSF-AI/ML model If the historical CSI information for Layer 2 cannot be obtained or is unavailable, the LCM switches Layer 2 to the SF-AI/ML model for CSI feedback. Furthermore, at times 2 and 3 (T1 and T2), the TSF-AI/ML model's output for Layer 2 is also fed back to the network for the acquisition or update of historical CSI information. At this time, the time window size is 2 (two moments). At time 4 (T3), the historical CSI information for Layer 2 has been obtained or recovered, so both Layer 1 and Layer 2 use the TSF-AI/ML model for CSI feedback.
  • the AI/ML model or function is operated according to the precoded layer and/or rank value, and the AI/ML model or function can be flexibly utilized according to the layer and/or rank value, giving full play to the advantages of the AI/ML model or function, and ensuring the performance of the AI/ML model or function and the performance of the communication system.
  • these layers can switch to SF-AI/ML or non-AI/ML methods for CSI feedback, which can avoid the performance deterioration of the AI/ML model or function and ensure the performance of the communication system.
  • the present application provides an apparatus for processing the operation of an AI/ML model or function.
  • This apparatus may be, for example, a terminal device, or one or more components or assemblies configured on the terminal device.
  • the apparatus corresponds to the method applied to the terminal device in the embodiment of the first aspect, and the contents identical to those in the embodiment of the first aspect are not repeated here.
  • FIG9 is a schematic diagram of an operation processing device for an AI/ML model or function according to an embodiment of the present application.
  • the operation processing device 900 for an AI/ML model or function includes:
  • a first receiving unit 901 receives configuration information from a network device
  • the first sending unit 902 sends first information to the network device according to the configuration information, where the first information is information obtained after the terminal device operates the AI/ML model or function according to the precoded layer and/or rank value.
  • the operation of the AI/ML model or function includes at least one or more of the following operations:
  • the operation of AI/ML models or functions according to layers and/or rank values is based on predefined rules, which include at least: on which layers the operation of one or more AI/ML models or functions is performed, and/or on which layers the operation of one or more AI/ML models or functions is performed when the rank value changes.
  • the rule is related to rank indication (RI) changes: when the feedback RI value is different from the last feedback RI value, the AI/ML model or function is operated according to the layer and/or rank value;
  • the terminal device implicitly indicates to the network device whether it has performed the operation of the AI/ML model or function through RI value feedback: when the RI changes, the terminal device takes action; when the RI remains unchanged, the terminal device does not take any action.
  • the operation of the AI/ML model or function according to the layer and/or rank value is based on: one or more predefined rules configured on the network side, the rules at least including: on which layers to perform the operation of one or more AI/ML models or functions, and/or on which layers to perform the operation of one or more AI/ML models or functions when the rank value changes; the rules configured on the network side are included or not included in the configuration information,
  • the terminal device selects one of the rules, performs the operation and notifies the network side of the corresponding operation, wherein the notification includes explicit or implicit notification.
  • the notification includes at least one of the following two notifications:
  • Inform network devices about the rules based on which the terminal device performs the operation of the AI/ML model or function.
  • the rule is related to RI changes: when the feedback RI value is different from the last feedback RI value, the terminal device operates the AI/ML model or function according to the layer and/or rank value based on a certain rule.
  • the terminal device implicitly indicates to the network device whether it has performed the operation of the AI/ML model or function through RI value feedback: when the RI changes, the terminal device takes action; when the RI remains unchanged, the terminal device does not take any action.
  • the terminal device displays an indication message to the network device, indicating which rule it is using to operate the AI/ML model or function. If the network device is configured with only one rule, the indication message can be left blank, or The cost is 0.
  • the operation of the AI/ML model or function according to the layer and/or rank value is based on: the network side indicates that the terminal device can perform the operation of one or more AI/ML models or functions on a certain layer or layers; the network side indication is included in the configuration information or not;
  • the configuration information is determined by the network device based on the operation of the terminal device requesting the AI/ML model or function.
  • the operation of AI/ML models or functions according to layers and/or ranks is based on the terminal device determining to perform one or more AI/ML model or function operations on a particular layer or layers and notifying the network device of the operations. For example, the terminal device determines that it needs to perform one or more AI/ML model or function operations on a particular layer or layers and notifies the network device of the specific operations.
  • the apparatus 900 further includes:
  • a processing unit 903 performs operations on an AI/ML model or function according to the layer and/or rank value, wherein the processing unit 903 performs operations on the AI/ML model or function on some or all of the layers and/or rank values, and/or the processing unit 903 performs the same or different AI/ML models or functions on different layers and/or rank values.
  • operations on AI/ML models or functions according to layer and/or rank values are applied to two-sided AI/ML models or functions, or to one-sided AI/ML models or functions.
  • the operation of the AI/ML model or function according to the layer and/or rank value is applied to at least one or more of the following:
  • the layer-specific AI/ML model or function includes: a layer-specific and rank-common AI/ML model or function, and/or a layer-specific and rank-specific AI/ML model or function;
  • the layer-common AI/ML models or functions include: (Layer common and rank common AI/ML) models or functions, and/or, layer common and rank specific (Layer common and rank specific AI/ML) models or functions.
  • the apparatus 900 further includes:
  • a monitoring unit 904 monitors the AI/ML model or function according to the layer and/or rank value,
  • the monitoring of the AI/ML model or function includes: monitoring the performance of the model or function of at least one layer and/or rank; and/or monitoring whether historical CSI information of at least one layer and/or rank is available.
  • the processing unit determines a CSI compression feedback scheme according to layers for CSI compression feedback based on a bilateral AI/ML model.
  • all layers use the same CSI compression feedback scheme, or different layers use different CSI compression feedback schemes.
  • the same CSI compression feedback scheme may be used on the same layer, or different CSI compression feedback schemes may be used on the same layer.
  • the CSI compression feedback scheme includes at least one or more of the following:
  • TSF-AI/ML time-space frequency domain
  • AI/ML-based spatial-frequency domain SF-AI/ML
  • the AI/ML model or function requires CSI information at one or more historical moments.
  • the layer switches to SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods, so that the TSF-AI/ML model or function obtains, restores or resets the historical CSI information; when the historical CSI information is obtained, restored or reset, the TSF-AI/ML CSI compression feedback is activated or used at the layer.
  • one or more layers need to switch to SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods.
  • the time window is predefined in the standard, or is based on RRC configuration and/or updated and/or indicated through MAC-CE or DCI.
  • whether to switch to SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods within the time window is configured by predefined configuration, or based on RRC signaling configuration and/or based on MAC CE or DCI update and/or indication.
  • TSF-AI/ML CSI compression feedback is applied only to the first layer, and SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods is applied to other layers, or, SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods is applied to the first layer and other layers.
  • a combination of one or more of the following methods is used to obtain or restore the historical CSI information:
  • the layer switches to SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods, while the output of the TSF-AI/ML encoder for the layer is still sent to the network side over the air interface for updating historical CSI information on the network side.
  • a proxy or approximate TSF encoder exists on the network side.
  • the network side obtains or recovers historical CSI information at one or more time points through SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods.
  • the layer switches to SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods, and uses the reconstructed CSI based on SF-AI/ML CSI compression feedback or CSI feedback based on non-AI/ML methods to obtain or recover historical CSI information.
  • the operation processing device 900 for AI/ML models or functions may also include other components or modules.
  • the specific content of these components or modules please refer to the relevant art.
  • FIG9 only illustrates the connection relationship or signal direction between various components or modules.
  • various related technologies such as bus connection can be used.
  • the above-mentioned components or modules can be implemented by hardware facilities such as processors, memories, transmitters, and receivers; the implementation of this application is not limited to this.
  • the AI/ML model or function is performed according to the pre-coded layer and/or rank value. Operation, which can flexibly utilize AI/ML models or functions based on layer and/or rank values, give full play to the advantages of AI/ML models or functions, and ensure the performance of AI/ML models or functions and communication system performance;
  • these layers can switch to SF-AI/ML or non-AI/ML methods for CSI feedback, which can avoid the performance deterioration of the AI/ML model or function and ensure the performance of the communication system.
  • the present application provides an operation processing device for an AI/ML model or function.
  • the device may be, for example, a network device or one or more components or assemblies configured on the network device.
  • the device corresponds to the method applied to the network side in the embodiment of the first aspect, and the content that is the same as the embodiment of the first aspect is not repeated here.
  • FIG10 is another schematic diagram of an operation processing device for an AI/ML model or function according to an embodiment of the present application.
  • the operation processing device 1000 for an AI/ML model or function includes:
  • a second sending unit 1001 which sends configuration information to the terminal device
  • the second receiving unit 1002 receives first information, where the first information is information obtained after the terminal device operates the AI/ML model or function according to the precoded layer and/or rank value.
  • the operation of the AI/ML model or function includes at least one or more of the following operations:
  • the operation of AI/ML models or functions according to layers and/or rank values is based on predefined rules, which include at least: on which layers the operation of one or more AI/ML models or functions is performed, and/or on which layers the operation of one or more AI/ML models or functions is performed when the rank value changes.
  • the rule is related to rank indication (RI) changes: when the feedback RI value is different from the previous feedback When the RI values fed back are different, the AI/ML model or function is operated according to the layer and/or rank value;
  • the terminal device implicitly indicates to the network device whether it has performed the operation of the AI/ML model or function through RI value feedback: when the RI changes, the terminal device takes action; when the RI remains unchanged, the terminal device does not take any action.
  • the operation of the AI/ML model or function according to the layer and/or rank value is based on: one or more predefined rules configured on the network side, the rules at least including: on which layers to perform the operation of one or more AI/ML models or functions, and/or on which layers to perform the operation of one or more AI/ML models or functions when the rank value changes; the rules configured on the network side are included or not included in the configuration information,
  • the terminal device selects one of the rules, performs the operation and notifies the network side of the corresponding operation, wherein the notification includes explicit or implicit notification.
  • the notification includes at least one of the following two notifications:
  • Inform network devices about the rules based on which the terminal device performs the operation of the AI/ML model or function.
  • the rule is related to RI changes: when the feedback RI value is different from the last feedback RI value, the terminal device operates the AI/ML model or function according to the layer and/or rank value based on a certain rule.
  • the terminal device implicitly indicates to the network device whether it has performed the operation of the AI/ML model or function through RI value feedback: when the RI changes, the terminal device takes action; when the RI remains unchanged, the terminal device does not take any action.
  • the terminal device uses an indication message to indicate to the network device which rule it is using to operate the AI/ML model or function. If the network device is configured with only one rule, the indication message can be left blank or have a cost of zero.
  • the operation of the AI/ML model or function according to the layer and/or rank value is based on: the network side indicates that the terminal device can perform the operation of one or more AI/ML models or functions on a certain layer or layers; the network side indication is included in the configuration information or not;
  • the configuration information is determined by the network device based on the operation of the terminal device requesting the AI/ML model or function.
  • the operation of the AI/ML model or function according to the layer and/or rank value is based on: the terminal device decides to perform one or more AI/ML model or function operations on a certain layer or certain layers, and notifies the network device of the operation. For example, the terminal device determines that it needs to perform one or more AI/ML model or function operations on a certain layer or certain layers, and notifies the network device of the specific operation. operation.
  • the apparatus 1000 further includes:
  • a processing unit 1003 performs operations on an AI/ML model or function according to the layer and/or rank value, wherein the processing unit 1003 performs operations on the AI/ML model or function on some or all of the layers and/or rank values, and/or the processing unit 1003 performs the same or different AI/ML models or functions on different layers and/or rank values.
  • operations on AI/ML models or functions according to layer and/or rank values are applied to two-sided AI/ML models or functions, or to one-sided AI/ML models or functions.
  • the operation of the AI/ML model or function according to the layer and/or rank value is applied to at least one or more of the following:
  • the layer-specific AI/ML model or function includes: a layer-specific and rank-common AI/ML model or function, and/or a layer-specific and rank-specific AI/ML model or function;
  • the layer-common AI/ML models or functions include: layer-common and rank-common AI/ML models or functions, and/or layer-common and rank-specific AI/ML models or functions.
  • the apparatus 1000 further includes:
  • a monitoring unit 1004 monitors the AI/ML model or function according to the layer and/or rank value for the operation of the AI/ML model or function according to the layer and/or rank value.
  • the monitoring of the AI/ML model or function includes: monitoring the performance of the model or function of at least one layer and/or rank; and/or monitoring whether historical CSI information of at least one layer and/or rank is available.
  • the processing unit determines a CSI compression feedback scheme according to layers for CSI compression feedback based on a bilateral AI/ML model.
  • all layers use the same CSI compression feedback scheme, or different layers use different CSI compression feedback schemes.
  • the same CSI compression feedback scheme may be used on the same layer, or different CSI compression feedback schemes may be used on the same layer.
  • the CSI compression feedback scheme includes at least one or more of the following:
  • TSF-AI/ML time-space frequency domain
  • AI/ML-based spatial-frequency domain SF-AI/ML
  • the AI/ML model or function requires CSI information at one or more historical moments.
  • the layer switches to SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods, so that the TSF-AI/ML model or function obtains, restores or resets the historical CSI information; when the historical CSI information is obtained, restored or reset, the TSF-AI/ML CSI compression feedback is activated or used at the layer.
  • one or more layers need to switch to SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods.
  • the time window is predefined in the standard, or is based on RRC configuration and/or updated and/or indicated through MAC-CE or DCI.
  • whether to switch to SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods within the time window is configured by predefined configuration, or based on RRC signaling configuration and/or based on MAC CE or DCI update and/or indication.
  • TSF-AI/ML CSI compression feedback is applied only to the first layer, and SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods is applied to other layers, or, SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods is applied to the first layer and other layers.
  • a combination of one or more of the following methods is used to obtain or restore the historical CSI information:
  • the layer switches to SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods, and
  • the output of the TSF-AI/ML encoder for the layer is still sent to the network side over the air interface for updating the historical CSI information on the network side.
  • a proxy or approximate TSF encoder exists on the network side.
  • the network side obtains or recovers historical CSI information at one or more time points through SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods.
  • the layer switches to SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods, and uses the reconstructed CSI based on SF-AI/ML CSI compression feedback or CSI feedback based on non-AI/ML methods to obtain or recover historical CSI information.
  • the operation processing device 1000 for AI/ML models or functions may also include other components or modules.
  • the specific content of these components or modules please refer to the relevant art.
  • FIG10 only illustrates the connection relationship or signal direction between various components or modules.
  • various related technologies such as bus connection can be used.
  • the above-mentioned components or modules can be implemented by hardware facilities such as processors, memories, transmitters, and receivers; the implementation of this application is not limited to this.
  • the AI/ML model or function is operated according to the precoded layer and/or rank value, and the AI/ML model or function can be flexibly utilized according to the layer and/or rank value, giving full play to the advantages of the AI/ML model or function, and ensuring the performance of the AI/ML model or function and the performance of the communication system.
  • these layers can switch to SF-AI/ML or non-AI/ML methods for CSI feedback, which can avoid the performance deterioration of the AI/ML model or function and ensure the performance of the communication system.
  • An embodiment of the present application provides a terminal device, which can execute the method applied to the terminal side in the embodiment of the first aspect.
  • the terminal device may include the apparatus described in the embodiment of the second aspect.
  • Figure 11 is a schematic block diagram of the system architecture of a terminal device according to an embodiment of the present invention.
  • terminal device 1100 may include a processor 1110 and a memory 1120; memory 1120 is coupled to processor 1110. It should be noted that this diagram is exemplary; other types of structures may be used to supplement or replace this structure to implement telecommunication or other functions.
  • the processor 1110 is configured to: receive configuration information from a network device; and send first information to the network device based on the configuration information, where the first information is information obtained after the terminal device operates the AI/ML model or function according to the precoded layer and/or rank value.
  • the terminal device 1100 may further include: a communication module 1130, an input unit 1140, a display 1150, and a power supply 1160. It is worth noting that the terminal device 1100 does not necessarily include all the components shown in FIG11 ; in addition, the terminal device 1100 may also include components not shown in FIG11 , and reference may be made to related art for details.
  • the processor 1110 is sometimes also referred to as a controller or an operation control, and may include a microprocessor or other processor device and/or logic device.
  • the processor 1110 receives inputs and controls the operations of various components of the terminal device 1100 .
  • Memory 1120 may be, for example, one or more of a cache, flash memory, a hard drive, removable media, volatile memory, non-volatile memory, or other suitable devices. It may store various data and programs for executing related information. Processor 1110 may execute the programs stored in memory 1120 to implement information storage or processing. The functions of other components are similar to those of existing devices and are not further described here. Each component of terminal device 1100 may be implemented using dedicated hardware, firmware, software, or a combination thereof without departing from the scope of the present invention.
  • the operation of the AI/ML model or function is performed according to the pre-coded layer and/or rank value, and the AI/ML model or function can be flexibly utilized according to the layer and/or rank value, giving full play to the advantages of the AI/ML model or function, and ensuring the performance of the AI/ML model or function and the performance of the communication system;
  • these layers can switch to SF-AI/ML or non-AI/ML methods for CSI feedback, which can avoid the performance deterioration of the AI/ML model or function and ensure the performance of the communication system.
  • An embodiment of the present application provides a network device, which can execute the method applied to the network side in the embodiment of the first aspect.
  • the network device may include the apparatus described in the embodiment of the third aspect.
  • FIG 12 is a schematic block diagram of the system configuration of a network device according to an embodiment of the present application.
  • network device 1200 may include a processor 1210 and a memory 1220; memory 1220 is coupled to processor 1210.
  • Memory 1220 can store various data and also stores an information processing program 1230. This program 1230 is executed under the control of processor 1210 to receive various information sent by terminal devices and to send various information to terminal devices.
  • the processor 1210 can be configured to: send configuration information to the terminal device; receive first information, where the first information is information obtained after the terminal device operates the AI/ML model or function according to the precoded layer and/or rank value.
  • the network device 1200 may further include: a transceiver 1540 and an antenna 1550, etc.; wherein the functions of the above components are similar to those in the prior art and are not described in detail here. It is worth noting that the network device 1200 does not necessarily include all the components shown in FIG12 ; in addition, the network device 1200 may also include components not shown in FIG12 , and reference may be made to the prior art for details.
  • the operation of the AI/ML model or function is performed according to the pre-coded layer and/or rank value, and the AI/ML model or function can be flexibly utilized according to the layer and/or rank value, giving full play to the advantages of the AI/ML model or function, and ensuring the performance of the AI/ML model or function and the performance of the communication system;
  • these layers can switch to SF-AI/ML or non-AI/ML methods for CSI feedback, which can avoid the performance deterioration of the AI/ML model or function and ensure the performance of the communication system.
  • the embodiment of the present application provides a communication system, including the terminal device according to the embodiment of the eighth aspect and/or the network device according to the embodiment of the ninth aspect.
  • a communication system including the terminal device according to the embodiment of the eighth aspect and/or the network device according to the embodiment of the ninth aspect.
  • the structure of the communication system can refer to FIG1.
  • the communication system 100 includes a network device 101 and terminal devices 102, 103.
  • the terminal device 102 and/or the terminal device 103 can be the same as the terminal device described in the embodiment of the fourth aspect, and/or the network device 101 can be the same as the terminal device described in the embodiment of the fifth aspect.
  • the network equipment is the same, so the repeated content will not be repeated.
  • the above devices and methods of the present application can be implemented by hardware or by a combination of hardware and software.
  • the present application relates to such a computer-readable program that, when executed by a logic component, enables the logic component to implement the devices or components described above, or enables the logic component to implement the various methods or steps described above.
  • the present application also relates to a storage medium for storing the above program, such as a hard disk, a magnetic disk, an optical disk, a DVD, a flash memory, etc.
  • the method/device described in conjunction with the embodiments of the present application can be directly embodied as hardware, a software module executed by a processor, or a combination of the two.
  • one or more of the functional block diagrams shown in Figures 9 or 10 and/or one or more combinations of functional block diagrams can correspond to various software modules of a computer program flow or to various hardware modules.
  • These software modules can correspond to the various steps shown in Figures 3 or 4, respectively.
  • These hardware modules can be implemented by solidifying these software modules, for example, using a field programmable gate array (FPGA).
  • FPGA field programmable gate array
  • the software module may be located in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.
  • a storage medium may be coupled to a processor so that the processor can read information from the storage medium and write information to the storage medium; or the storage medium may be an integral part of the processor.
  • the processor and the storage medium may be located in an ASIC.
  • the software module may be stored in the memory of the mobile terminal or in a memory card that can be inserted into the mobile terminal.
  • the software module may be stored in the MEGA-SIM card or the large-capacity flash memory device.
  • One or more of the functional blocks described in FIG9 or FIG10 and/or one or more combinations of functional blocks may be implemented as a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, or any suitable combination thereof for performing the functions described in this application.
  • One or more of the functional blocks described in FIG9 or FIG10 and/or one or more combinations of functional blocks may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in communication with a DSP, or any other such configuration.
  • first information is sent to the network device, where the first information is information obtained after the terminal device operates the AI/ML model or function according to the precoded layer and/or rank value.
  • Receive first information where the first information is information obtained after the terminal device operates an AI/ML model or function according to a precoded layer and/or rank value.
  • the AI/ML model or function requires CSI information at one or more historical moments.
  • the layer switches to SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods to enable the TSF-AI/ML model or function to obtain, restore or reset the historical CSI information; after the historical CSI information is obtained, restored or reset, the TSF-AI/ML CSI compression feedback is activated or used at the layer.
  • the switching is applied to the initial operation or reactivation of the TSF-AI/ML model or function, or to the situation where historical CSI information is unavailable due to rank value change or UCI loss or CSI discard.
  • one or more layers need to switch to SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods.
  • the time window is defined according to the number of time moments required for the TSF-AI/ML model or function to obtain, restore or reset historical CSI information, and the definition is explicit or implicit.
  • the time window applies to the initial operation or reactivation of the TSF-AI/ML model or function, or to situations where historical CSI information is unavailable due to rank value changes or UCI loss or CSI discard.
  • the time window is used on the terminal side and/or the network side.
  • the time windows on both sides are the same or different.
  • the sizes of the time windows are the same or different, and/or,
  • the size of the time window depends on the terminal capability of the terminal device.
  • Whether switching to SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods is supported depends on the terminal capabilities of the terminal device, and/or, whether simultaneous operation of TSF-AI/ML CSI compression feedback and non-TSF-AI/ML CSI compression feedback is supported depends on the terminal capabilities of the terminal device; and/or,
  • Whether to switch to SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods within the time window is determined by the terminal device, and/or whether to switch to SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods at one or more layers is reported by the terminal device to the network device.
  • TSF-AI/ML CSI compression feedback when historical CSI information of one or more layers on the terminal device side and/or the network side cannot be obtained or is unavailable, one or more of the following methods are used to obtain or restore historical CSI information:
  • the layer switches to SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods, while the output of the TSF-AI/ML encoder for the layer is still sent to the network side via the air interface for updating the historical CSI information on the network side.
  • the network side obtains or recovers the historical CSI information by inputting the fed-back output of the TSF-AI/ML encoder into the TSF-AI/ML decoder.
  • a proxy or approximate TSF encoder exists on the network side.
  • the network side obtains or recovers historical CSI information at one or more time points through SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods.
  • the network side reconstructs the eigenvector and/or channel matrix of the layer through SF-AI/ML CSI compression feedback or CSI compression feedback based on non-AI/ML methods, inputs the reconstructed eigenvector and/or channel matrix to the proxy or approximate TSF encoder, and then inputs the output of the proxy or approximate TSF encoder into the TSF decoder to obtain or recover the historical CSI information on the network side.
  • the layer switches to SF-AI/ML CSI compression feedback Or CSI compression feedback based on non-AI/ML methods, reconstructed CSI based on SF-AI/ML CSI compression feedback or CSI feedback based on non-AI/ML methods is used to obtain or recover historical CSI information.
  • the method used to obtain or restore historical CSI information is configured by the network side, or,
  • the terminal device reports the recommended method to the network side, and the network side ultimately determines the method for obtaining or restoring historical CSI information.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

本申请实施例提供一种AI/ML模型或功能的操作处理方法、装置和通信系统。所述方法包括:从网络设备接收配置信息;根据所述配置信息,向所述网络设备发送第一信息,所述第一信息是终端设备按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作后得到的信息。

Description

AI/ML模型或功能的操作处理方法、装置和通信系统 技术领域
本申请实施例涉及通信技术领域。
背景技术
在3GPP Release 18,对人工智能(AI,Artificial Intelligence)和机器学习(ML,Machine Learning)即AI/ML模型或功能应用于空中接口进行了研究,其中包括将AI/ML模型或功能应用于信道状态信息(CSI,Channel State Information)反馈压缩。基于AI/ML的CSI反馈压缩采用了双边(two-sided)模型,即AI/ML模型或功能位于终端设备(UE)侧和网络(NW)侧(即gNB侧)。在Rel-18中,基于AI/ML的反馈压缩在空频域(spatial frequency domain)对CSI进行压缩(SF-AI/ML CSI compression)。
在Release 19阶段,进一步研究了时空频域的基于AI/ML的CSI压缩反馈增强子用例(TSF-AI/ML CSI compression)。该子用例通过AI/ML方法(例如,RNN/GRU/LSTM模型)利用历史CSI信息来帮助当前时刻的CSI压缩,旨在获得更低的压缩率或者更高的反馈精度。
应该注意,上面对技术背景的介绍只是为了方便对本申请的技术方案进行清楚、完整的说明,并方便本领域技术人员的理解而阐述的。不能仅仅因为这些方案在本申请的背景技术部分进行了阐述而认为上述技术方案为本领域技术人员所公知。
发明内容
TSF-AI/ML CSI compression需要利用历史CSI信息,然而历史CSI信息是否可获取或是否可利用受到秩(rank)变化、CSI丢弃(例如基于layer优先级的CSI丢弃)、上行控制信息(UCI,Uplink Control Information)丢失等因素的影响。当历史CSI信息无法获取或不可用时,TSF-AI/ML模型或功能的性能将会恶化,甚至无法工作。
另一方面,CSI时域相关性决定了历史CSI辅助信息是否有助于当前的CSI压缩。相关性太低将会导致AI/ML模型或功能的无效输入,这不仅无法改善CSI压缩性能, 甚至会被AI/ML模型或功能视作噪声,恶化CSI压缩的性能。因此,TSF-AI/ML更适用于CSI时域相关性较好的场景,例如,该场景对应于UE移动速度较低、rank值较低,特征值大的层。
发明人发现,在现有技术中,AI/ML模型或功能的生命周期管理(LCM,life cycle management)操作(例如包括AI/ML模型或功能的选择、激活、去激活、切换、回退、更新等)对于所有的层和/或rank值是相同的,即不区分层和/或rank值。
但是,在某些用例中,例如,对于基于AI/ML的CSI反馈压缩而言,当采用TSF-AI/ML压缩反馈时,如果历史CSI信息无法获取或不可用,可以考虑将某些层切换到SF-AI/ML压缩反馈或者非AI/ML的压缩反馈。而现有的机制无法满足上述需求。
为了解决上述问题中的一个或多个,本申请实施例提供一种AI/ML模型或功能的操作处理方法、装置和通信系统。
根据本申请实施例的一个方面,提供一种AI/ML模型或功能的操作处理方法,应用于多天线(MIMO)通信系统,所述方法应用于终端侧,所述方法包括:从网络设备接收配置信息;根据所述配置信息,向所述网络设备发送第一信息,所述第一信息是终端设备按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作后得到的信息。
根据本申请实施例的另一个方面,提供一种AI/ML模型或功能的操作处理方法,应用于多天线(MIMO)通信系统,所述方法应用于网络侧,所述方法包括:向终端设备发送配置信息;接收第一信息,所述第一信息是终端设备按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作后得到的信息。
根据本申请实施例的另一个方面,提供一种AI/ML模型或功能的操作处理装置,应用于多天线(MIMO)通信系统,所述装置应用于终端侧,所述装置包括:第一接收单元,其从网络设备接收配置信息;第一发送单元,其根据所述配置信息,向所述网络设备发送第一信息,所述第一信息是终端设备按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作后得到的信息。
根据本申请实施例的另一个方面,提供一种AI/ML模型或功能的操作处理装置,应用于多天线(MIMO)通信系统,所述装置应用于网络侧,所述装置包括:第二发送单元,其向终端设备发送配置信息;第二接收单元,其接收第一信息,所述第一信 息是终端设备按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作后得到的信息。
根据本申请实施例的另一方面,提供一种终端设备,包括存储器和处理器,所述存储器存储有计算机程序,所述处理器被配置为执行所述计算机程序而实现上述终端设备侧的AI/ML模型或功能的操作处理方法。
根据本申请实施例的另一方面,提供一种网络设备,包括存储器和处理器,所述存储器存储有计算机程序,所述处理器被配置为执行所述计算机程序而实现上述网络设备侧的AI/ML模型或功能的操作处理方法。
根据本申请实施例的另一方面,提供一种通信系统,其中,所述系统包括:终端设备,其从网络设备接收配置信息;根据所述配置信息,向所述网络设备发送第一信息,所述第一信息是终端设备按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作后得到的信息;和/或,网络设备,其向终端设备发送配置信息;接收第一信息,所述第一信息是终端设备按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作后得到的信息。
本申请实施例的有益效果之一在于:按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作,能够根据层和/或rank值灵活的利用AI/ML模型或功能,充分发挥AI/ML模型或功能的优势,保证AI/ML模型或功能的性能和通信系统性能;
例如,当UE侧和/或网络侧某一层或者某些层的历史时刻CSI信息无法获取/不可用时,基于按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作的机制,这些层能够切换为SF-AI/ML或者非AI/ML方法进行CSI反馈,能够避免AI/ML模型或功能的性能恶化,保证通信系统性能。
另外,除了CSI压缩反馈这一用例,本领域技术人员能够将该方法应用于其他各种应用AI/ML模型或功能的用例和/或场景,并获得相应的有益效果。
参照后文的说明和附图,详细公开了本申请的特定实施方式,指明了本申请的原理可以被采用的方式。应该理解,本申请的实施方式在范围上并不因而受到限制。在所附权利要求的精神和条款的范围内,本申请的实施方式包括许多改变、修改和等同。
针对一种实施方式描述和/或示出的特征可以以相同或类似的方式在一个或更多个其它实施方式中使用,与其它实施方式中的特征相组合,或替代其它实施方式中的特征。
应该强调,术语“包括/包含”在本文使用时指特征、整件、步骤或组件的存在,但并不排除一个或更多个其它特征、整件、步骤或组件的存在或附加。
附图说明
在本申请实施例的一个附图或一种实施方式中描述的元素和特征可以与一个或更多个其它附图或实施方式中示出的元素和特征相结合。此外,在附图中,类似的标号表示几个附图中对应的部件,并可用于指示多于一种实施方式中使用的对应部件。
图1是本申请实施例的通信系统的一示意图;
图2是时空频域的基于AI/ML的CSI压缩(TSF-AI/ML CSI compression)反馈增强子用例的一示意图;
图3是本申请实施例的AI/ML模型或功能的操作处理方法的一示意图;
图4是本申请实施例的AI/ML模型或功能的操作处理方法的另一示意图;
图5是本申请实施例的按层进行AI/ML模型或功能的操作的一个示例的示意图;
图6是本申请实施例的按层进行AI/ML模型或功能的操作的另一个示例的示意图;
图7是本申请实施例的按层进行AI/ML模型或功能的操作的又一个示例的示意图;
图8是本申请实施例的使用TSF-AI/ML模型的输出进行历史CSI信息的获取或恢复的一个示例的示意图;
图9是本申请实施例的AI/ML模型或功能的操作处理装置的一示意图;
图10是本申请实施例的AI/ML模型或功能的操作处理装置的另一示意图;
图11是本发明实施例的终端设备的系统构成的一示意框图;
图12是本申请实施例的网络设备的系统构成的一示意框图。
具体实施方式
参照附图,通过下面的说明书,本申请的前述以及其它特征将变得明显。在说明书和附图中,具体公开了本申请的特定实施方式,其表明了其中可以采用本申请的原则的部分实施方式,应了解的是,本申请不限于所描述的实施方式,相反,本申请包括落入所附权利要求的范围内的全部修改、变型以及等同物。下面结合附图对本申请 的各种实施方式进行说明。这些实施方式只是示例性的,不是对本申请的限制。
在本申请实施例中,术语“第一”、“第二”等用于对不同元素从称谓上进行区分,但并不表示这些元素的空间排列或时间顺序等,这些元素不应被这些术语所限制。术语“和/或”包括相关联列出的术语的一种或多个中的任何一个和所有组合。术语“包含”、“包括”、“具有”等是指所陈述的特征、元素、元件或组件的存在,但并不排除存在或添加一个或多个其他特征、元素、元件或组件。
在本申请实施例中,单数形式“一”、“该”等包括复数形式,应广义地理解为“一种”或“一类”而并不是限定为“一个”的含义;此外术语“所述”应理解为既包括单数形式也包括复数形式,除非上下文另外明确指出。此外术语“根据”应理解为“至少部分根据……”,术语“基于”应理解为“至少部分基于……”,除非上下文另外明确指出。
在本申请实施例中,术语“通信网络”或“无线通信网络”可以指符合如下任意通信标准的网络,例如新无线(New Radio,NR)、长期演进(LTE,Long Term Evolution)、增强的长期演进(LTE-A,LTE-Advanced)、宽带码分多址接入(WCDMA,Wideband Code Division Multiple Access)、高速报文接入(HSPA,High-Speed Packet Access)等等。
并且,通信系统中设备之间的通信可以根据任意阶段的通信协议进行,例如可以包括但不限于如下通信协议:1G(generation)、2G、2.5G、2.75G、3G、4G、4.5G以及5G、新无线(NR,New Radio)、6G以及未来的通信等等,和/或其他目前已知或未来将被开发的通信协议。
在本申请实施例中,术语“网络设备”例如是指通信系统中将终端设备接入通信网络并为该终端设备提供服务的设备。网络设备可以包括但不限于如下设备:基站(BS,Base Station)、接入点(AP、Access Point)、发送接收点(收发节点)(TRP,Transmission Reception Point)、广播发射机、移动管理实体(MME、Mobile Management Entity)、网关、服务器、无线网络控制器(RNC,Radio Network Controller)、基站控制器(BSC,Base Station Controller)等等。
其中,基站可以包括但不限于:节点B(NodeB或NB)、演进节点B(eNodeB或eNB)、5G基站(gNB)、6G基站以及未来的基站,等等,此外还可包括远端无线头(RRH,Remote Radio Head)、远端无线单元(RRU,Remote Radio Unit)、中继(relay) 或者低功率节点(例如femto、pico等等)。并且术语“基站”可以包括它们的一些或所有功能,每个基站可以对特定的地理区域提供通信覆盖。术语“小区”可以指的是基站和/或其覆盖区域,这取决于使用该术语的上下文。
在本申请实施例中,术语“用户设备”(UE,User Equipment)或者“终端设备”(TE,Terminal Equipment)例如是指通过网络设备接入通信网络并接收网络服务的设备。用户设备可以是固定的或移动的,并且也可以称为移动台(MS,Mobile Station)、终端、用户、用户台(SS,Subscriber Station)、接入终端(AT,Access Terminal)、站、移动终端(MT,Mobile Termination),等等。
其中,终端设备可以包括但不限于如下设备:蜂窝电话(Cellular Phone)、个人数字助理(PDA,Personal Digital Assistant)、无线调制解调器、无线通信设备、手持设备、机器型通信设备、膝上型计算机、无绳电话、智能手机、智能手表、数字相机,等等。
再例如,在物联网(IoT,Internet of Things)等场景下,用户设备还可以是进行监控或测量的机器或装置,例如可以包括但不限于:机器类通信(MTC,Machine Type Communication)终端、车载通信终端、设备到设备(D2D,Device to Device)终端、机器到机器(M2M,Machine to Machine)终端、支持边链路(sidelink)通信的终端,等等。
此外,术语“网络侧”或“网络设备侧”是指网络的一侧,可以是某一基站,也可以包括如上的一个或多个网络设备。术语“用户侧”或“终端侧”或“终端设备侧”是指用户或终端的一侧,可以是某一UE,也可以包括如上的一个或多个终端设备。本文在没有特别指出的情况下,“设备”可以指网络设备,也可以指终端设备。
在不引起混淆的情况下,术语“上行控制信号”和“上行控制信息(UCI,Uplink Control Information)”或“物理上行控制信道(PUCCH,Physical Uplink Control Channel)”可以互换,术语“上行数据信号”和“上行数据信息”或“物理上行共享信道(PUSCH,Physical Uplink Shared Channel)”可以互换;
术语“下行控制信号”和“下行控制信息(DCI,Downlink Control Information)”或“物理下行控制信道(PDCCH,Physical Downlink Control Channel)”可以互换,术语“下行数据信号”和“下行数据信息”或“物理下行共享信道(PDSCH,Physical Downlink Shared Channel)”可以互换。
另外,上行信号可以包括上行数据信号和/或上行控制信号和/或PRACH和/或SRS(sounding reference signal,探测参考信号)等,也可以称为上行传输(UL transmission)或上行信息或上行信道。在上行资源上发送/接收上行传输可以理解为使用该上行资源发送/接收该上行传输。下行信号可以包括下行数据信号和/或下行控制信号和/或同步信号(SS,例如PSS/SSS)和/或广播信道(PBCH)和/或SSB(SS/PBCH block,包括PSS,SSS和PBCH及其DMRS)和/或CSI-RS等,也可以称为下行传输(DL transmission)或下行信息或下行信道。在下行资源上发送/接收下行传输可以理解为使用该下行资源发送/接收该下行传输。
在本申请实施例中,高层信令例如可以是无线资源控制(RRC)信令;RRC信令例如包括RRC消息(RRC message),例如包括广播/公共RRC消息/信令(例如主信息块(MIB)、系统信息(systeminformation,SI)、专用RRC消息/信令;或者RRC信息元素(RRC information element,RRC IE);或者RRC消息或RRC信息元素包括的信息域(或信息域包括的信息域)。高层信令例如还可以是媒体接入控制层(Medium Access Control,MAC)信令;或者称为MAC控制元素(MAC control element,MAC CE)。但本申请不限于此。
在本申请实施例中,“至少一个”和“一个或多于一个”可以互换,“多个”和“多于一个”可以互换,“多个”是指至少两个,或者两个或两个以上。
在本申请实施例中,预定义是指协议规定好的或者根据协议规定好的规则确定的,无需另外配置。配置/指示是指网络设备通过高层信令和/或物理层信令直接或间接配置/指示的。可以通过在高层信令中引入高层参数配置/指示,高层参数是指高层信令中的信息域(fields)和/或信息元素/信息单元/信息元(IE)等。物理层信令例如是指物理下行控制信道承载的控制信息(DCI)或序列承载的控制信息,但不限于此。
为了便于描述,下文以基站作为接入网络设备的例子进行描述。
在以下的说明中,在不引起混淆的情况下,“如果…”、“在…情况下”以及“当…时”可以相互替换使用。
以下通过示例对本申请实施例的场景进行说明,但本申请不限于此。
图1是本申请实施例的通信系统的一示意图,示意性说明了以终端设备和网络设备为例的情况,如图1所示,通信系统100可以包括网络设备101、终端设备102以及终端设备103。为简单起见,图1仅以两个终端设备和一个网络设备为例进行说明, 但本申请实施例不限于此。
在本申请实施例中,网络设备101、终端设备102以及终端设备103之间可以进行现有的业务或者未来可实施的业务发送。例如,这些业务可以包括但不限于:增强的移动宽带(eMBB,enhanced Mobile Broadband)、大规模机器类型通信(mMTC,massive Machine Type Communication)、高可靠低时延通信(URLLC,Ultra-Reliable and Low-Latency Communication)和减少能力的终端设备的相关通信,等等。
其中,终端设备102、103可以处于RRC_IDLE状态、或RRC_INACTIVE状态或RRC_CONNECTED状态,终端设备102、103也可以与网络设备101进行通信,例如,以终端设备102为例,终端设备102可以向网络设备101发送数据,或者可以进行数据重传。网络设备101可以向终端设备102发送寻呼消息,也可以向终端设备102发送数据,终端设备102接收网络设备101发送的数据。此外,不同的终端设备之间也可以进行通信,例如,终端设备102和终端设备103之间可进行数据交互。
值得注意的是,图1示出了终端设备102和终端设备103均处于网络设备101的覆盖范围内,但本申请不限于此。终端设备102和终端设备103可以均不在网络设备101的覆盖范围内,或者终端设备102和终端设备103中的一个在网络设备101的覆盖范围之内而另一个在网络设备101的覆盖范围之外。
在本申请实施例中,网络设备和/或终端设备中可以配置并运行一个或多个AI/ML模型。AI/ML模型可以用于无线通信的各种信号处理功能,例如CSI预测、CSI压缩、波束预测、定位管理等等;本申请不限于此。
图2是时空频域的基于AI/ML的CSI压缩(TSF-AI/ML CSI compression)反馈增强子用例的一示意图。如图2所示,在UE侧,CSI测量结果经过矩阵分解(例如,SVD分解(奇异值分解)或者EVD分解(特征值分解))后,得到特征向量,该特征向量被输入到编码器(encoder),由编码器生成CSI反馈信息。在网络侧,通过解码器(decoder)对UE发送的CSI反馈信息进行解码,生成重构的CSI。
在一些实施例中,UE侧的编码器和解码器可以采用AI/ML方法(例如,RNN/LSTM/GRU级联Transformer/CNN等模型)。其中,AI/ML方法可以利用时域信息(又可以被称为历史CSI信息、辅助信息(side information)等类似名称)来辅助当前的CSI压缩和/或解压缩,旨在获得更高的压缩率或者更高的反馈精度。
在利用时域信息进行CSI压缩和/或解压缩时,时域信息是否可获取和/或可利用 容易受到rank变化、CSI丢弃(例如,基于层(layer)优先级的CSI丢弃)、上行控制信息(UCI)丢失等因素的影响。当时域信息无法获取或不可用时,那么AI/ML模型的性能将会恶化,甚至无法工作。
此外,CSI时域相关性决定了历史CSI信息是否有助于当前的CSI压缩和/或解压缩。相关性太低将会导致AI/ML模型的无效输入,这不仅无法改善CSI压缩性能,甚至会被AI/ML模型视为噪声,导致CSI压缩的性能恶化。因此,AI/ML更适用于CSI时域相关性较好的场景,该场景通常对应于UE移动速度较低的情况、或rank值较低和/或特征值较大的层。
在本申请实施例中,CSI压缩也可以称为“CSI编码”、“CSI生成”等类似名称,将CSI压缩、或CSI编码、或CSI生成等操作的结果称为CSI反馈信息、CSI上报信息等类似名称。
在本申请实施例中,CSI解压缩也可以称为“CSI解码”、“CSI重构”、“CSI恢复”、“CSI重建”等类似名称。
在本申请实施例中,AI/ML模型也可以称为AI/ML方法、AI/ML单元(AI/ML unit)、AI/ML功能、或AI/ML元素(AI/ML element)的类似名称。在UE侧,AI/ML模型又可以称为编码器、CSI生成部分(CSI generation part)等类似名称,在网络设备侧,AI/ML模型又可以称为解码器、CSI重建部分(CSI reconstruction part)等类似名称。
需要说明的是,本申请实施例的AI/ML模型或功能的操作处理方法适用的用例包括但不限于CSI压缩反馈,但是,本领域技术人员能够理解,本申请实施例还适用于其他各种应用AI/ML模型或功能的用例和/或场景。
第一方面的实施例
本申请实施例提供一种AI/ML模型或功能的操作处理方法,该方法应用于多天线(MIMO)通信系统,从终端设备侧和/或网络侧进行说明。
图3是本申请实施例的AI/ML模型或功能的操作处理方法的一示意图,如图3所示,从终端(UE)侧来说,该方法包括:
301:从网络设备接收配置信息;以及
302:根据所述配置信息,向所述网络设备发送第一信息,所述第一信息是终端设备按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作后得到的信息。
图4是本申请实施例的AI/ML模型或功能的操作处理方法的另一示意图,如图4所示,从网络(NW)侧来说,该方法包括:
401:向终端设备发送配置信息;
402:接收第一信息,所述第一信息是终端设备按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作后得到的信息。
根据上述实施例,按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作,能够根据层和/或rank值灵活的利用AI/ML模型或功能,充分发挥AI/ML模型或功能的优势,保证AI/ML模型或功能的性能和通信系统性能。
在一些实施例中,UE侧的AI/ML模型与网络侧的AI/ML模型可以是对应地或匹配的。本申请不限于此,也可以仅在UE侧采用AI/ML模型,或者仅在网络侧采用AI/ML模型。在后续的内容中,若无特别说明,AI/ML模型是UE侧的模型和/或网络侧模型。
在一些实施例中,“按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作”可以简称为“按照层和/或rank值进行AI/ML模型或功能的操作”。
在一些实施例中,rank值表示预编码的层的数量。例如,rank值为1表示预编码的层的数量为1,即第1层;rank值为2表示预编码的层的数量为2,即第1层和第2层;以此类推。
在一些实施例中,该配置信息至少用于发送第一信息,例如,该配置信息包括但不限于用于发送第一信息的资源配置信息和/或上报配置信息。
在一些实施例中,该配置信息可以包括一种或者多种预定义的规则,该规则用于按照层和/或rank值进行的AI/ML模型或功能的操作;例如,所述规则包括但不限于:在哪些层上执行某一种或者某几种AI/ML模型或功能的操作,和/或rank值变化时在哪些层上执行某一种或者某几种AI/ML模型或功能的操作。
或者,该规则也可以不包含在该配置信息中,而是通过另一个配置信息或其他信息来发送。
在一些实施例中,该配置信息可以包括终端设备在某一个层上或者某些层上可以执行某一种或者某几种AI/ML模型或功能的操作的指示。
或者,该指示也可以不包含在该配置信息中,而是通过另一个配置信息或其他信息来发送。
在一些实施例中,上述配置信息中的一个或多个可以通过RRC信令发送给终端设备。本申请不限于此,也可以通过其他方式发送上述配置信息中的一个或多个。
在一些实施例中,所述AI/ML模型或功能的操作也称为AI/ML模型或功能的生命管理周期(LCM,life cycle management)操作。
在一些实施例中,所述AI/ML模型或功能的操作包括但不限于以下操作中的一个或者多个:
AI/ML模型或功能的选择;
AI/ML模型或功能的激活;
AI/ML模型或功能的去激活;
AI/ML模型或功能的切换;
AI/ML模型或功能的更新;
AI/ML模型或功能的回退。
在一些实施例中,按照层和/或rank值进行的AI/ML模型或功能的操作基于预定义的规则,所述规则包括但不限于:在哪些层上执行某一种或者某几种AI/ML模型或功能的操作,和/或rank值变化时在哪些层上执行某一种或者某几种AI/ML模型或功能的操作。
在一些实施例中,所述规则与秩指示(RI)变化有关:当反馈RI值与上一个反馈的RI值不同时,按照层和/或rank值进行AI/ML模型或功能的操作;
终端设备通过RI值反馈隐式指示网络设备其是否进行了该AI/ML模型或功能的操作:当RI变化时,终端设备采取操作;当RI不变时,终端设备没有操作。
在一些实施例中,按照层和/或rank值进行的AI/ML模型或功能的操作基于:网络侧配置一种或者多种预定义的规则,所述规则包括但不限于:在哪些层上执行某一种或者某几种AI/ML模型或功能的操作,和/或rank值变化时在哪些层上执行某一种或者某几种AI/ML模型或功能的操作;
另外,如上所述,网络侧配置的所述规则可以包括在所述配置信息中,也可以不包括在所述配置信息中,
所述终端设备选择其中的一种规则,进行所述操作并将相应的操作通知网络侧,例如,所述通知包括显式的或隐式的通知,其中,隐式的通知例如通过上报RI,资源分配、CSI丢弃等。
在一些实施例中,该通知包括以下两种通知中的至少一种:
通知网络设备,终端设备是否进行了AI/ML模型或功能的操作;
通知网络设备,终端设备基于哪一种规则进行AI/ML模型或功能的操作。
在一些实施例中,所述规则与RI变化有关:当反馈RI值与上一个反馈的RI值不同时,终端设备基于某一种规则按照层和/或rank值进行AI/ML模型或功能的操作。
终端设备通过RI值反馈隐式指示网络设备其是否进行了该AI/ML模型或功能的操作:当RI变化时,终端设备采取操作;当RI不变时,终端设备没有操作。
终端设备通过某一指示信息显示指示网络设备,其基于哪一种规则进行AI/ML模型或功能的操作。当网络设备仅配置了一种规则时,那么该指示信息可以缺省,或者开销为0。
在一些实施例中,按照层和/或rank值进行的AI/ML模型或功能的操作基于:网络侧指示终端设备在某一个层上或者某些层上可以执行某一种或者某几种AI/ML模型或功能的操作;
另外,如上所述,网络侧的所述指示可以包括在所述配置信息中,也可以不包括在所述配置信息中;
在一些实施例中,所述配置信息为网络设备基于终端设备请求AI/ML模型或功能的操作确定。
在一些实施例中,按照层和/或rank值进行的AI/ML模型或功能的操作基于:终端设备决定在某一个层上或者某些层上执行某一种或者某几种AI/ML模型或功能的操作,并通知网络设备所述操作。例如,终端设备自身判断需要进行在某一个层上或者某些层上执行某一种或者某几种AI/ML模型或功能的操作,并通知网络设备具体的操作。
在一些实施例中,如图3所示,该方法还包括:
303:按照层和/或秩(rank)值进行AI/ML模型或功能的操作;
例如,终端设备按照层和/或秩(rank)值进行AI/ML模型或功能的操作,得到或生成第一信息;
在一些实施例中,类似的,如图4所示,该方法还包括:
403:按照层和/或秩(rank)值进行AI/ML模型或功能的操作;
例如,网络设备对接收到的第一信息按照层和/或秩(rank)值进行AI/ML模型 或功能的操作;和/或,网络设备对于其他信息或其他情况,按照层和/或秩(rank)值进行AI/ML模型或功能的操作。也就是说,网络设备按照层和/或秩(rank)值进行AI/ML模型或功能的操作可以针对第一信息,也可以针对其他信息或其他情况。
在操作303和/或操作403中,例如,终端设备和/或网络设备对部分或者全部的层和/或rank值进行AI/ML模型或功能的操作,和/或,对不同的层和/或rank值进行相同或者不同的AI/ML模型或功能的操作。
在一些实施例中,按照层和/或rank值进行的AI/ML模型或功能的操作应用于双边(two-sided)AI/ML模型或功能,或者,应用于单边(one-sided)AI/ML模型或功能。
其中,对于按照层和/或rank值进行的AI/ML模型或功能的操作应用于双边AI/ML模型或功能的情况,在终端侧和网络侧的按照层和/或rank值进行的AI/ML模型或功能的操作是对应的。
在一些实施例中,按照层和/或rank值进行的AI/ML模型或功能的操作至少应用于以下的一个或多个:
rank专用的(Rank specific)AI/ML模型或功能;
rank共用的(Rank common)AI/ML模型或功能;
层专用的(Layer specific)AI/ML模型或功能;
层共用的(Layer common)AI/ML模型或功能,
其中,所述层专用的(Layer specific)AI/ML模型或功能包括:层专用且rank共用的(Layer specific and rank common)AI/ML模型或功能,和/或,层专用且层专用的(Layer specific and rank specific AI/ML)模型或功能;
所述层共用的(Layer common)AI/ML模型或功能包括:层共用且rank共用的(Layer common and rank common AI/ML)模型或功能,和/或,层共用且rank专用的(Layer common and rank specific AI/ML)模型或功能。
在一些实施例中,如图3所示,所述方法还包括:
304:对于按照层和/或rank值进行的AI/ML模型或功能的操作,按照层和/或rank值进行所述AI/ML模型或功能的监测。
在一些实施例中,类似的,如图4所示,该方法还包括:
404:对于按照层和/或rank值进行的AI/ML模型或功能的操作,按照层和/或rank 值进行所述AI/ML模型或功能的监测。
操作304和操作404为可选操作。
在一些实施例中,所述AI/ML模型或功能的监测包括:监测至少一个层和/或rank的模型或功能的性能;和/或,监测至少一个层和/或rank的历史CSI信息是否可用。
值得注意的是,以上图3和图4仅对本申请实施例进行了示意性说明,但本申请不限于此。例如可以适当地调整各个操作之间的执行顺序,此外还可以增加其他的一些操作或者减少其中的某些操作。本领域的技术人员可以根据上述内容进行适当地变型,而不仅限于上述图3和图4的记载。
以下通过几个示例对按照层和/或rank值进行AI/ML模型或功能的操作进行说明。
图5是本申请实施例的按层进行AI/ML模型或功能的操作的一个示例的示意图,其中rank值不变。
如图5所示,在T0和T1时刻,第1层和第2层均使用AI/ML模型A进行推理。在T2时刻,第2层的AI/ML模型A被切换为AI/ML模型B,第1层的AI/ML模型保持不变。例如,AI/ML模型A为TSF-AI/ML,UE和/或网络监测到第2层的CSI时域相关性或者TSF-AI/ML反馈精度太低,将第2层切换为复杂度更低的SF-AI/ML,也即AI/ML模型B。
图6是本申请实施例的按层进行AI/ML模型或功能的操作的另一个示例的示意图,其中rank值发生变化。
如图6所示,在T1时刻,rank值由2变为1,此时AI/ML模型A仅工作在第1层。在T2时刻,rank值由1变为2,此时第2层的AI/ML模型A被切换为模型B。例如,对于TSF-AI/ML,由于T1时刻为rank=1,T2时刻AI/ML模型无法获得第2层的CSI信息。考虑到历史CSI信息丢失后的第一个时刻(T2时刻)TSF-AI/ML推理结果可能性能较差,或者相较于SF-AI/ML没有优势,因此第2层切换到复杂度更低的SF-AI/ML,也即AI/ML模型B。
图7是本申请实施例的按层进行AI/ML模型或功能的操作的又一个示例的示意图,其中rank值不变。
如图7所示,模型A2用于Rank=2。在T2时刻,用于Rank=2的AI/ML模型A2被切换为AI/ML模型B2。例如,AI/ML模型A2为TSF-AI/ML,其输入为2层CSI,UE和/或网络监测到CSI时域相关性或者TSF-AI/ML反馈精度太低,将此时的 AI/ML模型切换为SF-AI/ML,也即AI/ML模型B2。
下面以应用于CSI压缩反馈为例对本申请实施例进行说明,但是,本申请实施例不限于此,其还可以适用于其他各种应用AI/ML模型或功能的用例和/或场景。
在一些实施例中,对于基于双边AI/ML模型的CSI压缩反馈,按照层确定CSI压缩反馈方案。
例如,对于基于双边AI/ML模型的CSI压缩反馈,在同一个时刻,所有层上采用相同的CSI压缩反馈方案,或者,不同的层上采用不同的CSI压缩反馈方案;
在不同的时刻,同一个层上采用相同的CSI压缩反馈方案,或者,同一个层上可以采用不同的CSI压缩反馈方案。
在一些实施例中,所述CSI压缩反馈方案包括但不限于以下的一个或多个:
基于AI/ML的时空频域(TSF-AI/ML)CSI压缩反馈;
基于AI/ML的联合CSI预测与CSI压缩反馈;
基于AI/ML的空频域(SF-AI/ML)CSI压缩反馈;
基于非AI/ML方法的CSI压缩反馈。
在一些实施例中,对于TSF-AI/ML CSI压缩反馈,AI/ML模型或功能需要一个或多个历史时刻的CSI信息,
当终端侧和/或网络侧的一个或多个层的历史时刻CSI信息无法获取或不可用时,所述层切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈,以使TSF-AI/ML模型或功能获取或恢复或重置历史CSI信息;当历史CSI信息被获取或恢复或重置后,TSF-AI/ML CSI压缩反馈在所述层被激活或使用。
在一些实施例中,所述切换应用于TSF-AI/ML模型或功能的初始运行或者重新激活,或者,应用于由于rank值变化或UCI丢失或CSI丢弃导致历史CSI信息不可用的情况。
在一些实施例中,在一个时间窗(window)内,一个或多个层需要切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈。
例如,该时间窗也可以称为预热窗或初始化窗或重初始化窗(warming up/initialization/re-initialization window),本申请实施例不对该时间窗的名称进行限制。
在一些实施例中,所述时间窗根据所述TSF-AI/ML模型或功能获取或恢复或重置历史CSI信息所需要的时刻数来定义,所述定义是显式的或者隐式的。
在一些实施例中,所述时间窗应用于TSF-AI/ML模型或功能的初始运行或者重新激活,或者,应用于由于rank值变化或UCI丢失或CSI丢弃导致历史CSI信息不可用的情况。
在一些实施例中,所述时间窗在标准中预定义,或者基于RRC配置和/或通过MAC-CE或DCI更新和/或指示。
在一些实施例中,所述时间窗用于终端侧和/或网络侧,当所述时间窗用于终端侧和网络侧时,两侧的时间窗可以相同,也可以不同。
在一些实施例中,对于不同的层和/或rank值,所述时间窗的大小相同或者不同。
在一些实施例中,所述时间窗的大小取决于终端设备的终端能力。
在一些实施例中,在所述时间窗内是否切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈通过预定义配置,或者基于RRC信令配置和/或基于MAC CE或DCI更新和/或指示。
在一些实施例中,是否支持切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈取决于终端设备的终端能力。
在一些实施例中,是否支持同时运行TSF-AI/ML CSI压缩反馈和非TSF-AI/ML CSI压缩反馈取决于终端设备的终端能力。
在一些实施例中,在所述时间窗内是否切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈由终端设备决定。
在一些实施例中,是否在一个或多个层切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈由终端设备报告给网络设备。
在一些实施例中,TSF-AI/ML CSI压缩反馈仅应用于第一层,SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈应用于其他层,或者,
SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈应用于第一层以及其他层。
下面,对于TSF-AI/ML CSI压缩反馈中,终端设备侧和/或网络侧的某一层或某些层的历史CSI信息无法获取或不可用的情况进行说明。
在一些实施例中,对于TSF-AI/ML CSI压缩反馈,当终端设备侧和/或网络侧的一个或多个层的历史CSI信息无法获取或不可用时,可以采用以下方式中的一种或多种的结合来获取或恢复历史CSI信息:
方式1:所述层切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈,而TSF-AI/ML编码器关于所述层的输出仍然通过空口发送给网络侧,用于网络侧的历史CSI信息的更新;其中,网络侧通过把反馈的TSF-AI/ML编码器的输出输入TSF-AI/ML解码器完成历史CSI信息的获取或恢复;
方式2:网络侧存在代理或近似(proxy/approximate)TSF编码器,在该情况下,网络侧通过SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈来获取或恢复一个或多个时刻的历史CSI信息;其中,网络侧通过SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈重构所述层的特征向量和/或信道矩阵,并将重构的特征向量和/或信道矩阵输入所述代理或近似(proxy/approximate)TSF编码器,再将所述代理或近似(proxy/approximate)TSF编码器的输出输入TSF解码器来获得或恢复网络侧的历史CSI信息。
方式3:当网络侧的历史CSI信息为重构CSI时,所述层切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈,将基于SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI反馈的重构CSI用于历史CSI信息的获取或恢复。
例如,对于上述方式1,SF-AI/ML CSI压缩反馈或基于非AI/ML方法被用于CSI反馈,TSF-AI/ML虽然参与反馈,但不涉及CSI反馈和PDSCH调度,仅用于历史CSI信息的获取/恢复;
例如,对于上述方式2和方式3,对于历史CSI信息无法获取或不可用的层,LCM把TSF-AI/ML模型切换为SF-AI/ML模型,或者非AI/ML方法(例如,回退到现有码本方案)。这些层的SF-AI/ML或非AI/ML方法不仅被用于CSI反馈,还被用于历史CSI信息获取/恢复。
在一些实施例中,获取或恢复历史CSI信息所采用的方式由网络侧配置,也就是说,网络侧可以配置采用哪一种方式或者多种方式结合的方式来获取或恢复历史CSI信息;
或者,也可以由终端设备向网络侧上报推荐使用的方式(使用哪一种方式或者多种方式结合的方式),网络侧最终确定获取或恢复历史CSI信息的方式。
图8是本申请实施例的使用TSF-AI/ML模型的输出进行历史CSI信息的获取或恢复的一个示例的示意图。
如图8所示,在第2时刻(T1时刻),rank值由1变为2,因此TSF-AI/ML模型 无法获取或不可用第2层的历史CSI信息,LCM切换该第2层至SF-AI/ML模型进行CSI反馈。此外,在第2时刻和第3时刻(T1和T2时刻),TSF-AI/ML模型关于第2层的输出也反馈给网络侧用于历史CSI信息的获取或更新,此时时间窗的大小为2(2个时刻)。在第4时刻(T3时刻),第2层的历史CSI信息已获取或恢复,因此第1层和第2层都使用了TSF-AI/ML模型进行CSI反馈。
以上各个实施例仅对本申请实施例进行了示例性说明,但本申请不限于此,还可以在以上各个实施例的基础上进行适当的变型。例如,可以单独使用上述各个实施例,也可以将以上各个实施例中的一种或多种结合起来。
根据上述实施例,按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作,能够根据层和/或rank值灵活的利用AI/ML模型或功能,充分发挥AI/ML模型或功能的优势,保证AI/ML模型或功能的性能和通信系统性能;
例如,当UE侧和/或网络侧某一层或者某些层的历史时刻CSI信息无法获取/不可用时,基于按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作的机制,这些层能够切换为SF-AI/ML或者非AI/ML方法进行CSI反馈,能够避免AI/ML模型或功能的性能恶化,保证通信系统性能。
第二方面的实施例
本申请实施例提供一种AI/ML模型或功能的操作处理装置。该装置例如可以是终端设备,也可以是配置于终端设备的某个或某些部件或者组件,其与第一方面的实施例中应用于终端设备侧的方法相对应,与第一方面的实施例相同的内容不再赘述。
图9是本申请实施例的AI/ML模型或功能的操作处理装置的一示意图。如图9所示,AI/ML模型或功能的操作处理装置900包括:
第一接收单元901,其从网络设备接收配置信息;
第一发送单元902,其根据所述配置信息,向所述网络设备发送第一信息,所述第一信息是终端设备按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作后得到的信息。
在一些实施例中,所述AI/ML模型或功能的操作至少包括以下操作中的一个或者多个:
AI/ML模型或功能的选择;
AI/ML模型或功能的激活;
AI/ML模型或功能的去激活;
AI/ML模型或功能的切换;
AI/ML模型或功能的更新;
AI/ML模型或功能的回退。
在一些实施例中,按照层和/或rank值进行的AI/ML模型或功能的操作基于预定义的规则,所述规则至少包括:在哪些层上执行某一种或者某几种AI/ML模型或功能的操作,和/或rank值变化时在哪些层上执行某一种或者某几种AI/ML模型或功能的操作。
在一些实施例中,所述规则与秩指示(RI)变化有关:当反馈RI值与上一个反馈的RI值不同时,按照层和/或rank值进行AI/ML模型或功能的操作;
终端设备通过RI值反馈隐式指示网络设备其是否进行了该AI/ML模型或功能的操作:当RI变化时,终端设备采取操作;当RI不变时,终端设备没有操作。
在一些实施例中,按照层和/或rank值进行的AI/ML模型或功能的操作基于:网络侧配置一种或者多种预定义的规则,所述规则至少包括:在哪些层上执行某一种或者某几种AI/ML模型或功能的操作,和/或rank值变化时在哪些层上执行某一种或者某几种AI/ML模型或功能的操作;网络侧配置的所述规则包括或不包括在所述配置信息中,
所述终端设备选择其中的一种规则,进行所述操作并将相应的操作通知网络侧,所述通知包括显式的或隐式的通知。
在一些实施例中,该通知包括以下两种通知中的至少一种:
通知网络设备,终端设备是否进行了AI/ML模型或功能的操作;
通知网络设备,终端设备基于哪一种规则进行AI/ML模型或功能的操作。
在一些实施例中,所述规则与RI变化有关:当反馈RI值与上一个反馈的RI值不同时,终端设备基于某一种规则按照层和/或rank值进行AI/ML模型或功能的操作。
终端设备通过RI值反馈隐式指示网络设备其是否进行了该AI/ML模型或功能的操作:当RI变化时,终端设备采取操作;当RI不变时,终端设备没有操作。
终端设备通过某一指示信息显示指示网络设备,其基于哪一种规则进行AI/ML模型或功能的操作。当网络设备仅配置了一种规则时,那么该指示信息可以缺省,或 者开销为0。
在一些实施例中,按照层和/或rank值进行的AI/ML模型或功能的操作基于:网络侧指示终端设备在某一个层上或者某些层上可以执行某一种或者某几种AI/ML模型或功能的操作;网络侧的所述指示包括或不包括在所述配置信息中;
在一些实施例中,所述配置信息为网络设备基于终端设备请求AI/ML模型或功能的操作确定。
在一些实施例中,按照层和/或rank值进行的AI/ML模型或功能的操作基于:终端设备决定在某一个层上或者某些层上执行某一种或者某几种AI/ML模型或功能的操作,并通知网络设备所述操作。例如,终端设备自身判断需要进行在某一个层上或者某些层上执行某一种或者某几种AI/ML模型或功能的操作,并通知网络设备具体的操作。
在一些实施例中,如图9所示,所述装置900还包括:
处理单元903,其按照层和/或秩(rank)值进行AI/ML模型或功能的操作,其中,所述处理单元903对部分或者全部的层和/或rank值进行AI/ML模型或功能的操作,和/或,所述处理单元903对不同的层和/或rank值进行相同或者不同的AI/ML模型或功能的操作。
在一些实施例中,按照层和/或rank值进行的AI/ML模型或功能的操作应用于双边(two-sided)AI/ML模型或功能,或者,应用于单边(one-sided)AI/ML模型或功能。
在一些实施例中,按照层和/或rank值进行的AI/ML模型或功能的操作至少应用于以下的一个或多个:
rank专用的(Rank specific)AI/ML模型或功能;
rank共用的(Rank common)AI/ML模型或功能;
层专用的(Layer specific)AI/ML模型或功能;
层共用的(Layer common)AI/ML模型或功能,
其中,所述层专用的(Layer specific)AI/ML模型或功能包括:层专用且rank共用的(Layer specific and rank common)AI/ML模型或功能,和/或,层专用且层专用的(Layer specific and rank specific AI/ML)模型或功能;
所述层共用的(Layer common)AI/ML模型或功能包括:层共用且rank共用的 (Layer common and rank common AI/ML)模型或功能,和/或,层共用且rank专用的(Layer common and rank specific AI/ML)模型或功能。
在一些实施例中,如图9所示,所述装置900还包括:
监测单元904,其对于按照层和/或rank值进行的AI/ML模型或功能的操作,按照层和/或rank值进行所述AI/ML模型或功能的监测,
所述AI/ML模型或功能的监测包括:监测至少一个层和/或rank的模型或功能的性能;和/或,监测至少一个层和/或rank的历史CSI信息是否可用。
在一些实施例中,所述处理单元对于基于双边AI/ML模型的CSI压缩反馈,按照层确定CSI压缩反馈方案。
在一些实施例中,对于基于双边AI/ML模型的CSI压缩反馈,
在同一个时刻,所有层上采用相同的CSI压缩反馈方案,或者,不同的层上采用不同的CSI压缩反馈方案;
在不同的时刻,同一个层上采用相同的CSI压缩反馈方案,或者,同一个层上可以采用不同的CSI压缩反馈方案。
在一些实施例中,所述CSI压缩反馈方案至少包括以下的一个或多个:
基于AI/ML的时空频域(TSF-AI/ML)CSI压缩反馈;
基于AI/ML的联合CSI预测与CSI压缩反馈;
基于AI/ML的空频域(SF-AI/ML)CSI压缩反馈;
基于非AI/ML方法的CSI压缩反馈。
在一些实施例中,对于TSF-AI/ML CSI压缩反馈,AI/ML模型或功能需要一个或多个历史时刻的CSI信息,
当终端侧和/或网络侧的一个或多个层的历史时刻CSI信息无法获取或不可用时,所述层切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈,以使TSF-AI/ML模型或功能获取或恢复或重置历史CSI信息;当历史CSI信息被获取或恢复或重置后,TSF-AI/ML CSI压缩反馈在所述层被激活或使用。
在一些实施例中,在一个时间窗内,一个或多个层需要切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈。
在一些实施例中,所述时间窗在标准中预定义,或者基于RRC配置和/或通过MAC-CE或DCI更新和/或指示。
在一些实施例中,在所述时间窗内是否切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈通过预定义配置,或者基于RRC信令配置和/或基于MAC CE或DCI更新和/或指示。
在一些实施例中,TSF-AI/ML CSI压缩反馈仅应用于第一层,SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈应用于其他层,或者,SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈应用于第一层以及其他层。
在一些实施例中,对于TSF-AI/ML CSI压缩反馈,当终端设备侧和/或网络侧的一个或多个层的历史CSI信息无法获取或不可用时,采用以下方式中的一种或多种的结合来获取或恢复历史CSI信息:
所述层切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈,而TSF-AI/ML编码器关于所述层的输出仍然通过空口发送给网络侧,用于网络侧的历史CSI信息的更新;
网络侧存在代理或近似(proxy/approximate)TSF编码器,在该情况下,网络侧通过SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈来获取或恢复一个或多个时刻的历史CSI信息;
当网络侧的历史CSI信息为重构CSI时,所述层切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈,将基于SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI反馈的重构CSI用于历史CSI信息的获取或恢复。
以上各个实施例仅对本申请实施例进行了示例性说明,但本申请不限于此,还可以在以上各个实施例的基础上进行适当的变型。例如,可以单独使用上述各个实施例,也可以将以上各个实施例中的一种或多种结合起来。
值得注意的是,以上仅对与本申请相关的各部件或模块进行了说明,但本申请不限于此。AI/ML模型或功能的操作处理装置900还可以包括其他部件或者模块,关于这些部件或者模块的具体内容,可以参考相关技术。
此外,为了简单起见,图9中仅示例性示出了各个部件或模块之间的连接关系或信号走向,但是本领域技术人员应该清楚的是,可以采用总线连接等各种相关技术。上述各个部件或模块可以通过例如处理器、存储器、发射机、接收机等硬件设施来实现;本申请实施并不对此进行限制。
根据上述实施例,按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的 操作,能够根据层和/或rank值灵活的利用AI/ML模型或功能,充分发挥AI/ML模型或功能的优势,保证AI/ML模型或功能的性能和通信系统性能;
例如,当UE侧和/或网络侧某一层或者某些层的历史时刻CSI信息无法获取/不可用时,基于按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作的机制,这些层能够切换为SF-AI/ML或者非AI/ML方法进行CSI反馈,能够避免AI/ML模型或功能的性能恶化,保证通信系统性能。
第三方面的实施例
本申请实施例提供一种AI/ML模型或功能的操作处理装置。该装置例如可以是网络设备,也可以是配置于网络设备的某个或某些部件或者组件,其与第一方面的实施例中应用于网络侧的方法相对应,与第一方面的实施例相同的内容不再赘述。
图10是本申请实施例的AI/ML模型或功能的操作处理装置的另一示意图。如图10所示,AI/ML模型或功能的操作处理装置1000包括:
第二发送单元1001,其向终端设备发送配置信息;
第二接收单元1002,其接收第一信息,所述第一信息是终端设备按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作后得到的信息。
在一些实施例中,所述AI/ML模型或功能的操作至少包括以下操作中的一个或者多个:
AI/ML模型或功能的选择;
AI/ML模型或功能的激活;
AI/ML模型或功能的去激活;
AI/ML模型或功能的切换;
AI/ML模型或功能的更新;
AI/ML模型或功能的回退。
在一些实施例中,按照层和/或rank值进行的AI/ML模型或功能的操作基于预定义的规则,所述规则至少包括:在哪些层上执行某一种或者某几种AI/ML模型或功能的操作,和/或rank值变化时在哪些层上执行某一种或者某几种AI/ML模型或功能的操作。
在一些实施例中,所述规则与秩指示(RI)变化有关:当反馈RI值与上一个反 馈的RI值不同时,按照层和/或rank值进行AI/ML模型或功能的操作;
终端设备通过RI值反馈隐式指示网络设备其是否进行了该AI/ML模型或功能的操作:当RI变化时,终端设备采取操作;当RI不变时,终端设备没有操作。
在一些实施例中,按照层和/或rank值进行的AI/ML模型或功能的操作基于:网络侧配置一种或者多种预定义的规则,所述规则至少包括:在哪些层上执行某一种或者某几种AI/ML模型或功能的操作,和/或rank值变化时在哪些层上执行某一种或者某几种AI/ML模型或功能的操作;网络侧配置的所述规则包括或不包括在所述配置信息中,
所述终端设备选择其中的一种规则,进行所述操作并将相应的操作通知网络侧,所述通知包括显式的或隐式的通知。
在一些实施例中,该通知包括以下两种通知中的至少一种:
通知网络设备,终端设备是否进行了AI/ML模型或功能的操作;
通知网络设备,终端设备基于哪一种规则进行AI/ML模型或功能的操作。
在一些实施例中,所述规则与RI变化有关:当反馈RI值与上一个反馈的RI值不同时,终端设备基于某一种规则按照层和/或rank值进行AI/ML模型或功能的操作。
终端设备通过RI值反馈隐式指示网络设备其是否进行了该AI/ML模型或功能的操作:当RI变化时,终端设备采取操作;当RI不变时,终端设备没有操作。
终端设备通过某一指示信息显示指示网络设备,其基于哪一种规则进行AI/ML模型或功能的操作。当网络设备仅配置了一种规则时,那么该指示信息可以缺省,或者开销为0。
在一些实施例中,按照层和/或rank值进行的AI/ML模型或功能的操作基于:网络侧指示终端设备在某一个层上或者某些层上可以执行某一种或者某几种AI/ML模型或功能的操作;网络侧的所述指示包括或不包括在所述配置信息中;
在一些实施例中,所述配置信息为网络设备基于终端设备请求AI/ML模型或功能的操作确定。
在一些实施例中,按照层和/或rank值进行的AI/ML模型或功能的操作基于:终端设备决定在某一个层上或者某些层上执行某一种或者某几种AI/ML模型或功能的操作,并通知网络设备所述操作。例如,终端设备自身判断需要进行在某一个层上或者某些层上执行某一种或者某几种AI/ML模型或功能的操作,并通知网络设备具体 的操作。
在一些实施例中,如图10所示,所述装置1000还包括:
处理单元1003,其按照层和/或秩(rank)值进行AI/ML模型或功能的操作,其中,所述处理单元1003对部分或者全部的层和/或rank值进行AI/ML模型或功能的操作,和/或,所述处理单元1003对不同的层和/或rank值进行相同或者不同的AI/ML模型或功能的操作。
在一些实施例中,按照层和/或rank值进行的AI/ML模型或功能的操作应用于双边(two-sided)AI/ML模型或功能,或者,应用于单边(one-sided)AI/ML模型或功能。
在一些实施例中,按照层和/或rank值进行的AI/ML模型或功能的操作至少应用于以下的一个或多个:
rank专用的(Rank specific)AI/ML模型或功能;
rank共用的(Rank common)AI/ML模型或功能;
层专用的(Layer specific)AI/ML模型或功能;
层共用的(Layer common)AI/ML模型或功能,
其中,所述层专用的(Layer specific)AI/ML模型或功能包括:层专用且rank共用的(Layer specific and rank common)AI/ML模型或功能,和/或,层专用且层专用的(Layer specific and rank specific AI/ML)模型或功能;
所述层共用的(Layer common)AI/ML模型或功能包括:层共用且rank共用的(Layer common and rank common AI/ML)模型或功能,和/或,层共用且rank专用的(Layer common and rank specific AI/ML)模型或功能。
在一些实施例中,如图10所示,所述装置1000还包括:
监测单元1004,其对于按照层和/或rank值进行的AI/ML模型或功能的操作,按照层和/或rank值进行所述AI/ML模型或功能的监测,
所述AI/ML模型或功能的监测包括:监测至少一个层和/或rank的模型或功能的性能;和/或,监测至少一个层和/或rank的历史CSI信息是否可用。
在一些实施例中,所述处理单元对于基于双边AI/ML模型的CSI压缩反馈,按照层确定CSI压缩反馈方案。
在一些实施例中,对于基于双边AI/ML模型的CSI压缩反馈,
在同一个时刻,所有层上采用相同的CSI压缩反馈方案,或者,不同的层上采用不同的CSI压缩反馈方案;
在不同的时刻,同一个层上采用相同的CSI压缩反馈方案,或者,同一个层上可以采用不同的CSI压缩反馈方案。
在一些实施例中,所述CSI压缩反馈方案至少包括以下的一个或多个:
基于AI/ML的时空频域(TSF-AI/ML)CSI压缩反馈;
基于AI/ML的联合CSI预测与CSI压缩反馈;
基于AI/ML的空频域(SF-AI/ML)CSI压缩反馈;
基于非AI/ML方法的CSI压缩反馈。
在一些实施例中,对于TSF-AI/ML CSI压缩反馈,AI/ML模型或功能需要一个或多个历史时刻的CSI信息,
当终端侧和/或网络侧的一个或多个层的历史时刻CSI信息无法获取或不可用时,所述层切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈,以使TSF-AI/ML模型或功能获取或恢复或重置历史CSI信息;当历史CSI信息被获取或恢复或重置后,TSF-AI/ML CSI压缩反馈在所述层被激活或使用。
在一些实施例中,在一个时间窗内,一个或多个层需要切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈。
在一些实施例中,所述时间窗在标准中预定义,或者基于RRC配置和/或通过MAC-CE或DCI更新和/或指示。
在一些实施例中,在所述时间窗内是否切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈通过预定义配置,或者基于RRC信令配置和/或基于MAC CE或DCI更新和/或指示。
在一些实施例中,TSF-AI/ML CSI压缩反馈仅应用于第一层,SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈应用于其他层,或者,SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈应用于第一层以及其他层。
在一些实施例中,对于TSF-AI/ML CSI压缩反馈,当终端设备侧和/或网络侧的一个或多个层的历史CSI信息无法获取或不可用时,采用以下方式中的一种或多种的结合来获取或恢复历史CSI信息:
所述层切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈,而 TSF-AI/ML编码器关于所述层的输出仍然通过空口发送给网络侧,用于网络侧的历史CSI信息的更新;
网络侧存在代理或近似(proxy/approximate)TSF编码器,在该情况下,网络侧通过SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈来获取或恢复一个或多个时刻的历史CSI信息;
当网络侧的历史CSI信息为重构CSI时,所述层切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈,将基于SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI反馈的重构CSI用于历史CSI信息的获取或恢复。
以上各个实施例仅对本申请实施例进行了示例性说明,但本申请不限于此,还可以在以上各个实施例的基础上进行适当的变型。例如,可以单独使用上述各个实施例,也可以将以上各个实施例中的一种或多种结合起来。
值得注意的是,以上仅对与本申请相关的各部件或模块进行了说明,但本申请不限于此。AI/ML模型或功能的操作处理装置1000还可以包括其他部件或者模块,关于这些部件或者模块的具体内容,可以参考相关技术。
此外,为了简单起见,图10中仅示例性示出了各个部件或模块之间的连接关系或信号走向,但是本领域技术人员应该清楚的是,可以采用总线连接等各种相关技术。上述各个部件或模块可以通过例如处理器、存储器、发射机、接收机等硬件设施来实现;本申请实施并不对此进行限制。
根据上述实施例,按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作,能够根据层和/或rank值灵活的利用AI/ML模型或功能,充分发挥AI/ML模型或功能的优势,保证AI/ML模型或功能的性能和通信系统性能;
例如,当UE侧和/或网络侧某一层或者某些层的历史时刻CSI信息无法获取/不可用时,基于按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作的机制,这些层能够切换为SF-AI/ML或者非AI/ML方法进行CSI反馈,能够避免AI/ML模型或功能的性能恶化,保证通信系统性能。
第四方面的实施例
本申请实施例提供了一种终端设备,该终端设备可以执行第一方面的实施例中应用于终端侧的方法,另外,该终端设备可以包括第二方面的实施例所述的装置。
图11是本发明实施例的终端设备的系统构成的一示意框图。如图11所示,终端设备1100可以包括处理器1110和存储器1120;存储器1120耦合到处理器1110。值得注意的是,该图是示例性的;还可以使用其他类型的结构,来补充或代替该结构,以实现电信功能或其他功能。
在一个实施方式中,处理器1110被配置为:从网络设备接收配置信息;根据所述配置信息,向所述网络设备发送第一信息,所述第一信息是终端设备按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作后得到的信息。
如图11所示,终端设备1100还可以包括:通信模块1130、输入单元1140、显示器1150、电源1160。值得注意的是,终端设备1100也并不是必须要包括图11中所示的所有部件;此外,终端设备1100还可以包括图11中没有示出的部件,可以参考相关技术。
如图11所示,处理器1110有时也称为控制器或操作控件,可以包括微处理器或其他处理器装置和/或逻辑装置,该处理器1110接收输入并控制终端设备1100的各个部件的操作。
其中,存储器1120,例如可以是缓存器、闪存、硬驱、可移动介质、易失性存储器、非易失性存储器或其它合适装置中的一种或更多种。可储存各种数据,此外还可存储执行有关信息的程序。并且处理器1110可执行该存储器1120存储的该程序,以实现信息存储或处理等。其他部件的功能与现有类似,此处不再赘述。终端设备1100的各部件可以通过专用硬件、固件、软件或其结合来实现,而不偏离本发明的范围。
通过本申请实施例,按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作,能够根据层和/或rank值灵活的利用AI/ML模型或功能,充分发挥AI/ML模型或功能的优势,保证AI/ML模型或功能的性能和通信系统性能;
例如,当UE侧和/或网络侧某一层或者某些层的历史时刻CSI信息无法获取/不可用时,基于按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作的机制,这些层能够切换为SF-AI/ML或者非AI/ML方法进行CSI反馈,能够避免AI/ML模型或功能的性能恶化,保证通信系统性能。
第五方面的实施例
本申请实施例提供了一种网络设备,该网络设备可以执行第一方面的实施例中应用于网络侧的方法,另外,该网络设备可以包括第三方面的实施例所述的装置。
图12是本申请实施例的网络设备的系统构成的一示意框图。如图12所示,网络设备1200可以包括:处理器(processor)1210和存储器1220;存储器1220耦合到处理器1210。其中该存储器1220可存储各种数据;此外还存储信息处理的程序1230,并且在处理器1210的控制下执行该程序1230,以接收终端设备发送的各种信息、并且向终端设备发送各种信息。
在一个实施方式中,处理器1210可以被配置为:向终端设备发送配置信息;接收第一信息,所述第一信息是终端设备按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作后得到的信息。
此外,如图12所示,网络设备1200还可以包括:收发机1540和天线1550等;其中,上述部件的功能与现有技术类似,此处不再赘述。值得注意的是,网络设备1200也并不是必须要包括图12中所示的所有部件;此外,网络设备1200还可以包括图12中没有示出的部件,可以参考现有技术。
通过本申请实施例,按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作,能够根据层和/或rank值灵活的利用AI/ML模型或功能,充分发挥AI/ML模型或功能的优势,保证AI/ML模型或功能的性能和通信系统性能;
例如,当UE侧和/或网络侧某一层或者某些层的历史时刻CSI信息无法获取/不可用时,基于按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作的机制,这些层能够切换为SF-AI/ML或者非AI/ML方法进行CSI反馈,能够避免AI/ML模型或功能的性能恶化,保证通信系统性能。
第六方面的实施例
本申请实施例提供了一种通信系统,包括根据第八方面的实施例所述的终端设备和/或根据第九方面的实施例所述的网络设备。具体的内容可以参照第八方面的实施例和第九方面的实施例中的记载。
例如,该通信系统的结构可以参照图1,如图1所示,通信系统100包括网络设备101和终端设备102,103,终端设备102和/或终端设备103可以与第四方面的实施例中记载的终端设备相同,和/或,网络设备101可以与第五方面的实施例中记载的 网络设备相同,重复的内容不再赘述。
本申请以上的装置和方法可以由硬件实现,也可以由硬件结合软件实现。本申请涉及这样的计算机可读程序,当该程序被逻辑部件所执行时,能够使该逻辑部件实现上文所述的装置或构成部件,或使该逻辑部件实现上文所述的各种方法或步骤。本申请还涉及用于存储以上程序的存储介质,如硬盘、磁盘、光盘、DVD、flash存储器等。
结合本申请实施例描述的方法/装置可直接体现为硬件、由处理器执行的软件模块或二者组合。例如,图9或图10中所示的功能框图中的一个或多个和/或功能框图的一个或多个组合,既可以对应于计算机程序流程的各个软件模块,亦可以对应于各个硬件模块。这些软件模块,可以分别对应于图3或图4中所示的各个步骤。这些硬件模块例如可利用现场可编程门阵列(FPGA)将这些软件模块固化而实现。
软件模块可以位于RAM存储器、闪存、ROM存储器、EPROM存储器、EEPROM存储器、寄存器、硬盘、移动磁盘、CD-ROM或者本领域已知的任何其它形式的存储介质。可以将一种存储介质耦接至处理器,从而使处理器能够从该存储介质读取信息,且可向该存储介质写入信息;或者该存储介质可以是处理器的组成部分。处理器和存储介质可以位于ASIC中。该软件模块可以存储在移动终端的存储器中,也可以存储在可插入移动终端的存储卡中。例如,若设备(如移动终端)采用的是较大容量的MEGA-SIM卡或者大容量的闪存装置,则该软件模块可存储在该MEGA-SIM卡或者大容量的闪存装置中。
针对图9或图10中描述的功能方框中的一个或多个和/或功能方框的一个或多个组合,可以实现为用于执行本申请所描述功能的通用处理器、数字信号处理器(DSP)、专用集成电路(ASIC)、现场可编程门阵列(FPGA)或者其它可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件或者其任意适当组合。针对图9或图10描述的功能方框中的一个或多个和/或功能方框的一个或多个组合,还可以实现为计算设备的组合,例如,DSP和微处理器的组合、多个微处理器、与DSP通信结合的一个或多个微处理器或者任何其它这种配置。
以上结合具体的实施方式对本申请进行了描述,但本领域技术人员应该清楚,这些描述都是示例性的,并不是对本申请保护范围的限制。本领域技术人员可以根据本申请的精神和原理对本申请做出各种变型和修改,这些变型和修改也在本申请的范围 内。
根据本申请实施例公开的各种实施方式,还公开了如下附记:
1.一种AI/ML模型或功能的操作处理方法,应用于多天线(MIMO)通信系统,所述方法应用于终端侧,所述方法包括:
从网络设备接收配置信息;
根据所述配置信息,向所述网络设备发送第一信息,所述第一信息是终端设备按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作后得到的信息。
2.一种AI/ML模型或功能的操作处理方法,应用于多天线(MIMO)通信系统,所述方法应用于网络侧,所述方法包括:
向终端设备发送配置信息;
接收第一信息,所述第一信息是终端设备按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作后得到的信息。
3.根据附记1或2所述的方法,其中,
对于TSF-AI/ML CSI压缩反馈,AI/ML模型或功能需要一个或多个历史时刻的CSI信息,
当终端侧和/或网络侧的一个或多个层的历史时刻CSI信息无法获取或不可用时,所述层切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈,以使TSF-AI/ML模型或功能获取或恢复或重置历史CSI信息;当历史CSI信息被获取或恢复或重置后,TSF-AI/ML CSI压缩反馈在所述层被激活或使用,
所述切换应用于TSF-AI/ML模型或功能的初始运行或者重新激活,或者,应用于由于rank值变化或UCI丢失或CSI丢弃导致历史CSI信息不可用的情况。
4.根据附记1-3中的任一项所述的方法,其中,
在一个时间窗内,一个或多个层需要切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈,
所述时间窗根据所述TSF-AI/ML模型或功能获取或恢复或重置历史CSI信息所需要的时刻数来定义,所述定义是显式的或者隐式的。
5.根据附记1-4中的任一项所述的方法,其中,
所述时间窗应用于TSF-AI/ML模型或功能的初始运行或者重新激活,或者,应用于由于rank值变化或UCI丢失或CSI丢弃导致历史CSI信息不可用的情况。
6.根据附记1-5中的任一项所述的方法,其中,
所述时间窗用于终端侧和/或网络侧,
当所述时间窗用于终端侧和网络侧时,两侧的时间窗相同或不同。
7.根据附记1-6中的任一项所述的方法,其中,
对于不同的层和/或rank值,所述时间窗的大小相同或者不同,和/或,
所述时间窗的大小取决于终端设备的终端能力。
8.根据附记1-7中的任一项所述的方法,其中,
是否支持切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈取决于终端设备的终端能力,和/或,是否支持同时运行TSF-AI/ML CSI压缩反馈和非TSF-AI/ML CSI压缩反馈取决于终端设备的终端能力;和/或,
在所述时间窗内是否切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈由终端设备决定,和/或,是否在一个或多个层切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈由终端设备报告给网络设备。
9.根据附记1-8中的任一项所述的方法,其中,
对于TSF-AI/ML CSI压缩反馈,当终端设备侧和/或网络侧的一个或多个层的历史CSI信息无法获取或不可用时,采用以下方式中的一种或多种的结合来获取或恢复历史CSI信息:
所述层切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈,而TSF-AI/ML编码器关于所述层的输出仍然通过空口发送给网络侧,用于网络侧的历史CSI信息的更新,其中,网络侧通过把反馈的TSF-AI/ML编码器的输出输入TSF-AI/ML解码器完成历史CSI信息的获取或恢复;
网络侧存在代理或近似(proxy/approximate)TSF编码器,在该情况下,网络侧通过SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈来获取或恢复一个或多个时刻的历史CSI信息,其中,网络侧通过SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈重构所述层的特征向量和/或信道矩阵,并将重构的特征向量和/或信道矩阵输入所述代理或近似(proxy/approximate)TSF编码器,再将所述代理或近似(proxy/approximate)TSF编码器的输出输入TSF解码器来获得或恢复网络侧的历史CSI信息;
当网络侧的历史CSI信息为重构CSI时,所述层切换为SF-AI/ML CSI压缩反馈 或基于非AI/ML方法的CSI压缩反馈,将基于SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI反馈的重构CSI用于历史CSI信息的获取或恢复。
10.根据附记1-9中的任一项所述的方法,其中,
获取或恢复历史CSI信息所采用的方式由网络侧配置,或者,
终端设备向网络侧上报推荐使用的方式,网络侧最终确定获取或恢复历史CSI信息的方式。

Claims (20)

  1. 一种AI/ML模型或功能的操作处理装置,应用于多天线(MIMO)通信系统,所述装置应用于终端侧,所述装置包括:
    第一接收单元,其从网络设备接收配置信息;
    第一发送单元,其根据所述配置信息,向所述网络设备发送第一信息,所述第一信息是终端设备按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作后得到的信息。
  2. 一种AI/ML模型或功能的操作处理装置,应用于多天线(MIMO)通信系统,所述装置应用于网络侧,所述装置包括:
    第二发送单元,其向终端设备发送配置信息;
    第二接收单元,其接收第一信息,所述第一信息是终端设备按照预编码的层和/或秩(rank)值进行AI/ML模型或功能的操作后得到的信息。
  3. 根据权利要求1或2所述的装置,其中,所述AI/ML模型或功能的操作至少包括以下操作中的一个或者多个:
    AI/ML模型或功能的选择;
    AI/ML模型或功能的激活;
    AI/ML模型或功能的去激活;
    AI/ML模型或功能的切换;
    AI/ML模型或功能的更新;
    AI/ML模型或功能的回退。
  4. 根据权利要求1或2所述的装置,其中,
    按照层和/或rank值进行的AI/ML模型或功能的操作基于预定义的规则,所述规则至少包括:在哪些层上执行某一种或者某几种AI/ML模型或功能的操作,和/或rank值变化时在哪些层上执行某一种或者某几种AI/ML模型或功能的操作。
  5. 根据权利要求1或2所述的装置,其中,
    按照层和/或rank值进行的AI/ML模型或功能的操作基于:网络侧配置一种或者多种预定义的规则,所述规则至少包括:在哪些层上执行某一种或者某几种AI/ML模型或功能的操作,和/或rank值变化时在哪些层上执行某一种或者某几种AI/ML模 型或功能的操作;网络侧配置的所述规则包括或不包括在所述配置信息中,
    所述终端设备选择其中的一种规则,进行所述操作并将相应的操作通知网络侧,所述通知包括显式的或隐式的通知。
  6. 根据权利要求1或2所述的装置,其中,
    按照层和/或rank值进行的AI/ML模型或功能的操作基于:网络侧指示终端设备在某一个层上或者某些层上可以执行某一种或者某几种AI/ML模型或功能的操作;网络侧的所述指示包括或不包括在所述配置信息中。
  7. 根据权利要求6所述的装置,其中,所述配置信息为网络设备基于终端设备请求AI/ML模型或功能的操作确定。
  8. 根据权利要求1或2所述的装置,其中,
    按照层和/或rank值进行的AI/ML模型或功能的操作基于:终端设备决定在某一个层上或者某些层上执行某一种或者某几种AI/ML模型或功能的操作,并通知网络设备所述操作。
  9. 根据权利要求1或2所述的装置,其中,所述装置还包括:
    处理单元,其按照层和/或秩(rank)值进行AI/ML模型或功能的操作,其中,
    所述处理单元对部分或者全部的层和/或rank值进行AI/ML模型或功能的操作,和/或,
    所述处理单元对不同的层和/或rank值进行相同或者不同的AI/ML模型或功能的操作。
  10. 根据权利要求1或2所述的装置,其中,
    按照层和/或rank值进行的AI/ML模型或功能的操作应用于双边(two-sided)AI/ML模型或功能,或者,应用于单边(one-sided)AI/ML模型或功能。
  11. 根据权利要求1或2所述的装置,其中,
    按照层和/或rank值进行的AI/ML模型或功能的操作至少应用于以下的一个或多个:
    rank专用的(Rank specific)AI/ML模型或功能;
    rank共用的(Rank common)AI/ML模型或功能;
    层专用的(Layer specific)AI/ML模型或功能;
    层共用的(Layer common)AI/ML模型或功能,
    其中,所述层专用的(Layer specific)AI/ML模型或功能包括:层专用且rank共用的(Layer specific and rank common)AI/ML模型或功能,和/或,层专用且层专用的(Layer specific and rank specific AI/ML)模型或功能;
    所述层共用的(Layer common)AI/ML模型或功能包括:层共用且rank共用的(Layer common and rank common AI/ML)模型或功能,和/或,层共用且rank专用的(Layer common and rank specific AI/ML)模型或功能。
  12. 根据权利要求1或2所述的装置,其中,所述装置还包括:
    监测单元,其对于按照层和/或rank值进行的AI/ML模型或功能的操作,按照层和/或rank值进行所述AI/ML模型或功能的监测,
    所述AI/ML模型或功能的监测包括:监测至少一个层和/或rank的模型或功能的性能;和/或,监测至少一个层和/或rank的历史CSI信息是否可用。
  13. 根据权利要求7所述的装置,其中,
    所述处理单元对于基于双边AI/ML模型的CSI压缩反馈,按照层确定CSI压缩反馈方案,其中,
    对于基于双边AI/ML模型的CSI压缩反馈,
    在同一个时刻,所有层上采用相同的CSI压缩反馈方案,或者,不同的层上采用不同的CSI压缩反馈方案;
    在不同的时刻,同一个层上采用相同的CSI压缩反馈方案,或者,同一个层上可以采用不同的CSI压缩反馈方案。
  14. 根据权利要求13所述的装置,其中,
    所述CSI压缩反馈方案至少包括以下的一个或多个:
    基于AI/ML的时空频域(TSF-AI/ML)CSI压缩反馈;
    基于AI/ML的联合CSI预测与CSI压缩反馈;
    基于AI/ML的空频域(SF-AI/ML)CSI压缩反馈;
    基于非AI/ML方法的CSI压缩反馈。
  15. 根据权利要求14所述的装置,其中,
    对于TSF-AI/ML CSI压缩反馈,AI/ML模型或功能需要一个或多个历史时刻的CSI信息,
    当终端侧和/或网络侧的一个或多个层的历史时刻CSI信息无法获取或不可用时, 所述层切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈,以使TSF-AI/ML模型或功能获取或恢复或重置历史CSI信息;当历史CSI信息被获取或恢复或重置后,TSF-AI/ML CSI压缩反馈在所述层被激活或使用。
  16. 根据权利要求15所述的装置,其中,
    在一个时间窗内,一个或多个层需要切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈,
    所述时间窗在标准中预定义,或者基于RRC配置和/或通过MAC-CE或DCI更新和/或指示。
  17. 根据权利要求16所述的装置,其中,
    在所述时间窗内是否切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈通过预定义配置,或者基于RRC信令配置和/或基于MAC CE或DCI更新和/或指示。
  18. 根据权利要求14所述的装置,其中,
    TSF-AI/ML CSI压缩反馈仅应用于第一层,SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈应用于其他层,或者,
    SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈应用于第一层以及其他层。
  19. 根据权利要求1或2所述的装置,其中,
    对于TSF-AI/ML CSI压缩反馈,当终端设备侧和/或网络侧的一个或多个层的历史CSI信息无法获取或不可用时,采用以下方式中的一种或多种的结合来获取或恢复历史CSI信息:
    所述层切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈,而TSF-AI/ML编码器关于所述层的输出仍然通过空口发送给网络侧,用于网络侧的历史CSI信息的更新;
    网络侧存在代理或近似(proxy/approximate)TSF编码器,在该情况下,网络侧通过SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈来获取或恢复一个或多个时刻的历史CSI信息;
    当网络侧的历史CSI信息为重构CSI时,所述层切换为SF-AI/ML CSI压缩反馈或基于非AI/ML方法的CSI压缩反馈,将基于SF-AI/ML CSI压缩反馈或基于非 AI/ML方法的CSI反馈的重构CSI用于历史CSI信息的获取或恢复。
  20. 一种通信系统,所述通信系统包括终端设备和网络设备,
    所述终端设备包括权利要求1所述的装置,和/或
    所述网络设备包括权利要求2所述的装置。
PCT/CN2024/087633 2024-04-12 2024-04-12 Ai/ml模型或功能的操作处理方法、装置和通信系统 Pending WO2025213482A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2024/087633 WO2025213482A1 (zh) 2024-04-12 2024-04-12 Ai/ml模型或功能的操作处理方法、装置和通信系统

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2024/087633 WO2025213482A1 (zh) 2024-04-12 2024-04-12 Ai/ml模型或功能的操作处理方法、装置和通信系统

Publications (1)

Publication Number Publication Date
WO2025213482A1 true WO2025213482A1 (zh) 2025-10-16

Family

ID=97349096

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2024/087633 Pending WO2025213482A1 (zh) 2024-04-12 2024-04-12 Ai/ml模型或功能的操作处理方法、装置和通信系统

Country Status (1)

Country Link
WO (1) WO2025213482A1 (zh)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104660315A (zh) * 2013-11-25 2015-05-27 上海益尚信息科技有限公司 新型在协同多点传输系统中控制协作传输下行信号的方法
CN116547935A (zh) * 2021-04-02 2023-08-04 Oppo广东移动通信有限公司 无线通信方法、终端设备和网络设备
WO2023206437A1 (zh) * 2022-04-29 2023-11-02 富士通株式会社 信息传输方法与装置
WO2024031692A1 (zh) * 2022-08-12 2024-02-15 富士通株式会社 Ai/ml模型的监测方法和装置

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104660315A (zh) * 2013-11-25 2015-05-27 上海益尚信息科技有限公司 新型在协同多点传输系统中控制协作传输下行信号的方法
CN116547935A (zh) * 2021-04-02 2023-08-04 Oppo广东移动通信有限公司 无线通信方法、终端设备和网络设备
WO2023206437A1 (zh) * 2022-04-29 2023-11-02 富士通株式会社 信息传输方法与装置
WO2024031692A1 (zh) * 2022-08-12 2024-02-15 富士通株式会社 Ai/ml模型的监测方法和装置

Similar Documents

Publication Publication Date Title
US20240098752A1 (en) Method, device, and system for power saving in wireless networks
EP4503763A1 (en) Dtx transmission method, network node, network device, and storage medium
US20240349392A1 (en) Control signaling monitoring method, apparatus, device and storage medium
US20250016774A1 (en) Indication method and apparatus
US20260107236A1 (en) Communication method and communication apparatus
US20260031879A1 (en) Methods and apparatuses for transmitting and receiving channel state information and communication system
US20250260458A1 (en) Method and apparatus for model management and communication system
WO2025213482A1 (zh) Ai/ml模型或功能的操作处理方法、装置和通信系统
CN115395998A (zh) 一种消息解码方法、发送端和接收端
US20250300708A1 (en) Signal transceiving method and apparatus, and device and storage medium
CN108023668B (zh) 一种无线通信系统中数据传输的方法和装置
US20220408361A1 (en) Energy saving information transmission method, terminal and network device
WO2025208504A1 (zh) 信道状态信息的处理方法、装置和通信系统
JP7827210B2 (ja) 信号送受信方法と装置
WO2025231745A1 (zh) 设定参考信号的装置
US20250038795A1 (en) Methods and apparatuses for transmitting and receiving parameter and communication system
WO2026066228A1 (en) Channel state information (csi) report
WO2026065328A1 (zh) 信道状态信息的增强方法、装置和通信系统
WO2025231764A1 (zh) 发送和接收信息的方法、装置和通信系统
WO2025091339A1 (zh) 功能控制的方法、装置和通信系统
WO2024022157A1 (zh) 定位处理方法、装置及设备
WO2026065113A1 (zh) 数据收发方法以及装置
CN121713392A (zh) 预编码配置方法以及装置
WO2026011314A1 (zh) 波束管理方法以及装置、通信系统
WO2026065513A1 (zh) 信息收发方法及装置、信息发送方法及装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 24934559

Country of ref document: EP

Kind code of ref document: A1