Channel state information processing
A scalable AI model for CSI processing in wireless networks addresses the limitations of fixed dimensions by employing data processing techniques, improving CSI reconstruction accuracy and reducing overhead.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- ZTE CORP
- Filing Date
- 2025-02-07
- Publication Date
- 2026-06-11
Smart Images

Figure CN2025076126_11062026_PF_FP_ABST
Abstract
Description
CHANNEL STATE INFORMATION PROCESSINGTECHNICAL FIELD
[0001] This disclosure generally relates to handling transmissions in a wireless cellular access network, and is specifically directed to mechanisms for processing Channel State Information (CSI) .BACKGROUND
[0002] Artificial Intelligence / Machine Learning (AI / ML) is a promising enhancement direction for mobile communication system, e.g., 5G (the fifth generation) , 5G-A (5G-Advanced) , and 6G (the sixth generation) . With the introduction of AI / ML technology into the mobile communication system, the performance and operating efficiency of the whole system are expected to be improved, e.g., to improve the accuracy of Channel State Information (CSI) and to further reduce the overhead of CSI feedback via potential AI / ML approaches. For communication system with AI / ML technology, an AI / ML model is adopted.SUMMARY
[0003] This disclosure generally relates to handling transmissions in a wireless cellular access network, and is specifically directed to mechanisms for processing Channel State Information (CSI) .
[0004] In various approaches, a base station transmits Channel State Information-Reference Signal (CSI-RS) to a User Equipment (UE) . The UE measures the CSI-RS and derives the CSI. The CSI can be Rank Indicator (RI) , Precoding Matrix Indicator (PMI) , Channel Quality Indicator (CQI) , Layer Indicator (LI) , Reference Signal Received Power (RSRP) , Reference Signal Received Quality (RSRQ) , Reference signal indicator, the number of non-zero coefficients, etc. In various approaches, the UE may derive the CSI using an AI / ML model. Then, the UE may report the CSI to the base station. The base station may perform scheduling based on the CSI, e.g., determining the coding rate and modulation order, MU pairing, etc.
[0005] During wireless communication, there may be plenty of diverse CSI configurations. However, the input and output dimensions of a typical AI model are often fixed and the supported number of input and / or output dimensions is limited. It is unrealistic to design one AI model corresponding to each CSI configuration. Therefore, a problem to be solved is how to design a scalable AI model to adapt to multiple CSI configurations and implement CSI compression and feedback.
[0006] The present disclosure includes methods and procedures to perform a scalable AI model to adapt to multiple CSI configurations and to report CSI to the base station for accurate reconstruction.
[0007] In some exemplary implementations, a method performed by a wireless terminal device (e.g., UE) includes utilizing a processing module to derive a set of Channel State Information (CSI) , wherein the processing module supports N input and / or output configuration, wherein N is an integer larger than 1. The method also includes transmitting the set of CSI to a wireless access network node (WANN) (e.g., base station) .
[0008] Similarly, a method performed by the WANN (e.g., base station) includes includes receiving, from the wireless terminal device (e.g., UE) , a set of Channel State Information (CSI) , and utilizing a processing module to derive a set of Channel Metrics from the CSI, wherein the processing module supports N input and / or output configurations, wherein N is an integer larger than 1.
[0009] In exemplary implementations of the methods, which may be combined with any of the other exemplary implementations disclosed herein, the processing module comprises an Artificial Intelligence (AI) model.
[0010] In exemplary implementations, which may be combined with any of the other exemplary implementations disclosed herein, the methods include at least one of the following: the wireless terminal device indicating data processing procedure information to the wireless access network node, wherein the wireless access network node can determine a reverse data processing procedure based on the data processing procedure information; or the wireless terminal device receiving an indication of data processing procedure information from the wireless access network node, wherein the wireless terminal device determines a data processing procedure based on the data processing procedure information. In various embodiments, the data processing procedure information is transmitted via high-layer or PHY signaling via an uplink or a downlink channel.
[0011] In exemplary implementations of the methods, which may be combined with any of the other exemplary implementations disclosed herein, the data processing procedure information applies to at least one of: spatial domain processing; frequency domain processing; time domain processing; and / or model output contents.
[0012] In exemplary implementations of the methods, which may be combined with any of the other exemplary implementations disclosed herein, the methods include the WANN (e.g., base station) transmitting, and the wireless terminal device (e.g., UE) receiving, CSI-RS, and the wireless terminal device deriving a first set of channel metrics based on measurement of the CSI-RS, wherein the first set of channel metrics comprises M channel metrics, each of which comprises K components, wherein each component comprises L units, wherein M, K, and L are positive integer numbers; and the wireless terminal device deriving a second set of channel metrics using a data processing procedure for any of the M channel metrics and / or K components and / or L units.
[0013] In exemplary implementations of the methods, which may be combined with any of the other exemplary implementations disclosed herein, the methods include the WANN deriving the set of channel metrics based on the set of CSI by performing the reverse data processing procedure to derive a reconstructed set of channel metrics, wherein the reconstructed set of channel metric comprises M’ channel metrics, each of which comprises K’ components, wherein each component comprises L’ units, wherein M’ , K’ , and L’ are positive integer number, and wherein each L’ is larger than or equal to each L, respectively, and wherein each M’ and K’ are equal to each M and K, respectively.
[0014] In exemplary implementations of the methods, which may be combined with any of the other exemplary implementations disclosed herein, the data processing procedure comprises a padding operation, wherein a largest number of input units supported by the processing module is L_max, wherein L is less than L_max, and wherein the padding operation comprises padding L_max -L units of padded contents in specific components in the second set of channel metrics in order to derive the set of CSI using the processing module.
[0015] In exemplary implementations of the methods, which may be combined with any of the other exemplary implementations disclosed herein, the padded contents comprise at least one of the following: specific constant values of 0 or 1; a constant value derived from a formula or rule; a copy of an existing channel metric for L_max -L instances; a copy of consecutive L_max -L channel metrics in corresponding padding indices when L >= L_max / 2, wherein the consecutive L_max -L channel metrics may be first consecutive L_max –L channel metrics from an index#1, or last consecutive L_max –L channel metrics from index# (L_max –L+1) ; a copy of consecutive L channel metrics with additional padding contents in corresponding padding indices when L <= L_max / 2, wherein the additional padding comprises zero padding an additional copy of the channel metrics for units of L_max-2*L; or new channel metrics derived from the existing channel metrics.
[0016] In exemplary implementations of the methods, which may be combined with any of the other exemplary implementations disclosed herein, a pattern or index of padding comprises at least one of the following: contents are padded at indexes of the first L_max -L units; contents are padded at indexes of the last L_max -L units; contents are padded at X units between two units of channel metrics with a same interval when L_max can be divided by L, wherein X is a positive integer; contents are padded at X units between the two units of channel metrics with a different interval when L_max cannot be divided by L; contents are padded at an index range of Y consecutive units, where Y is a positive integer and Y can be different values for different padding index ranges; or contents are padded at a pattern of indices understood by both the wireless terminal device and the wireless access network node.
[0017] In exemplary implementations of the methods, which may be combined with any of the other exemplary implementations disclosed herein, the data processing procedure comprises a grouping operation, wherein a number of input units supported by the processing module is L_in, wherein L is larger than L_in, and wherein L units of contents in components in the set of channel metrics are divided into groups to derive the set of CSI by the processing module, wherein the grouping operation complies with at least one of the following: consecutive L_in units of channel metrics are divided into a set or group; L_in channel metrics with a same time interval are divided into a set or group; or L_in channel metrics in a set or group consist of consecutive L_in / 2 channel metrics in a first half and the consecutive L_in / 2 channel metrics counted backward in a second half.
[0018] In exemplary implementations of the methods, which may be combined with any of the other exemplary implementations disclosed herein, the methods include the wireless terminal device inputting at least one of the first set of channel metrics and / or the second set of channel metrics into the processing module to derive P sets of CSI, wherein P is a positive integer larger than 1, and the wireless terminal device transmitting, and the WANN receiving, a set of the P sets of CSI.
[0019] In exemplary implementations of the methods, which may be combined with any of the other exemplary implementations disclosed herein, the data processing procedure comprises a masking operation, wherein a largest number of output units supported by the processing module is Q_max, wherein Q is a number of output units for a particular operation of the processing module, and is less than Q_max, and wherein the masking operation comprises masking Q_max -Q units of contents in order to derive the set of CSI using the processing module.
[0020] In exemplary implementations of the methods, which may be combined with any of the other exemplary implementations disclosed herein, a pattern or index of masking comprises at least one of the following: contents are masked at indexes of the first Q_max -Q units; contents are masked at indexes of the last Q_max -Q units; contents are masked at X units between two units of channel metrics with a same interval when Q_max can be divided by Q, wherein X is a positive integer; contents are masked at X units between the two units of channel metrics with a different interval when Q_max cannot be divided by Q; contents are masked at an index range of Y consecutive units, where Y is a positive integer and Y can be different values for different masking index ranges; or contents are masked at a pattern of indices understood by both the wireless terminal device and the wireless access network node.
[0021] In exemplary implementations of the methods, which may be combined with any of the other exemplary implementations disclosed herein, the data processing procedure comprises a truncation operation, wherein a largest number of output elements supported by the processing module is B_max, wherein B is a number of output elements for a particular operation of the processing module, and is less than B_max, and wherein the truncation operation comprises truncating B_max -B elements of contents in order to derive the set of CSI using the processing module.
[0022] In exemplary implementations of the methods, which may be combined with any of the other exemplary implementations disclosed herein, the data processing procedure comprises a switch operation, wherein the processing module supports S branches of output units, wherein each branch comprises O_i units, wherein S and O_i are positive integers, and wherein i = 1, 2, …S, and wherein O_i is a different number for each branch, and wherein the switch operation comprises switching a branch of output units to derive the set of CSI using the processing module.
[0023] In exemplary implementations of the methods, which may be combined with any of the other exemplary implementations disclosed herein, the methods include the wireless terminal device using the data processing procedure information to derive the second set of channel metrics, wherein the data processing procedure information comprises at least one of: information of a pattern or index of padding, comprising at least one of: a starting index, an ending index, a padding number, and / or a type of indication information on padding pattern, including at least one of a binary information, a bitmap, or an index of patterns; or information of padded contents, comprising: a type of indication information on padding contents, including at least one of a binary information, a bitmap, or an index of padded contents.
[0024] In exemplary implementations of the methods, which may be combined with any of the other exemplary implementations disclosed herein, the methods include the wireless terminal device using the data processing procedure information to derive the second set of channel metrics, wherein the data processing procedure information comprises at least one of: information of a grouping pattern comprising at least one of: a group starting index, a group ending index, a number of channel metrics in one group, and / or a type of indication information, including at least one of a binary information, a bitmap, or an index of patterns; or information of grouping methods, comprising: a type of indication information on grouping methods, including at least one of a binary information, a bitmap, or an index of patterns; or information of a padding pattern including at least one of: an index of a padding group, an index for padding channel metric in the group, an index for padding channel metric in total number, a starting index for padding, an ending index for padding, or a padding number.
[0025] In exemplary implementations of the methods, which may be combined with any of the other exemplary implementations disclosed herein, the methods include the wireless terminal device using the data processing procedure information to derive the set of CSI, wherein the data processing procedure information comprises at least one of: information of a pattern or index of masking, comprising at least one of: a starting index of masking, an ending index of masking, a number of masked units, and / or a type of indication information on masking, including at least one of a binary information, a bitmap, or an index of patterns; or information of masking methods, comprising: a type of indication information on masking methods, including at least one of a binary information, a bitmap, or an index of masking contents.
[0026] In exemplary implementations of the methods, which may be combined with any of the other exemplary implementations disclosed herein, the methods include the wireless terminal device indicating, and the WANN receiving an indication of, an encoder model structure, wherein the encoder model structure includes an indication of a hyper-parameter, wherein the indication of the hyper-parameter includes a value of the hyper-parameter or a model ID or a model structure ID.
[0027] In exemplary implementations of the methods, which may be combined with any of the other exemplary implementations disclosed herein, the methods include the WANN transmitting, and the wireless terminal device receiving, at least one of: a complete set of parameters for a structure of the processing module with the data processing procedure information to derive the set of CSI, wherein the data processing procedure comprises at least one of masking, deactivating, freezing, or setting to zero or one values; or a relevant partial set of parameters for the structure of the processing module to derive the set of CSI.
[0028] In exemplary implementations of the methods, which may be combined with any of the other exemplary implementations disclosed herein, the whole set of parameters comprises at least one of: a starting index of activated or deactivated units of parameters, an ending index of activated or deactivated units of parameters, a number of activated or deactivated units of parameters including at least parameters for corresponding operation layers, or a type of indication information including at least one of a binary information, a bitmap, or an index of patterns; or wherein the partial set of parameters comprises at least one of: configuration related information, identification information, additional information, or a type of indication information including at least one of a binary information, a bitmap, or an index of patterns.
[0029] In some other implementations, an apparatus for wireless communication such as a network device is disclosed. The network device may include one or more processors and one or more memories, wherein the one or more processors are configured to read computer code from the one or more memories to implement any one of the methods above. The apparatus for wireless communication may be the wireless access network node (e.g., base station) or the wireless terminal device (e.g., UE) .
[0030] In yet some other implementations, a computer program product is disclosed. The computer program product may include a non-transitory computer-readable medium with computer code stored thereupon, the computer code, when executed by one or more processors, causing the one or more processors to implement any one of the methods above.
[0031] The above embodiments and other aspects and alternatives of their implementations are explained in greater detail in the drawings, the descriptions, and the claims below.BRIEF DESCRIPTION OF THE DRAWINGS
[0032] FIG. 1 shows a wireless access network with an exemplary uplink, downlink, and control channel configuration.
[0033] FIG. 2 shows various example processing components of the wireless terminal device and the wireless access network node of FIG. 1.
[0034] FIGS. 3, 4, 5, 6, and 7 each show a resource diagram that each provides an example of padding in accordance with various embodiments.
[0035] FIGS. 8, 9, 10, 11, 12, and 13 each show a resource diagram that each provides an example of masking in accordance with various embodiments.
[0036] FIG. 14 shows an example bitstream illustrating a truncation operation in accordance with various embodiments.
[0037] FIG. 15 shows an example output diagram illustrating an example of different branches in accordance with various embodiments.
[0038] FIG. 16 shows an example of hyper-parameters illustrated in a typical Transformer block in accordance with various embodiments.DETAILED DESCRIPTION
[0039] The technology and examples of implementations and / or embodiments described in this disclosure can be used to facilitate over-the-air radio resource allocation, configuration, and signaling in wireless access networks as well as operational configuration of a UE and / or a base station within the wireless access networks. The term “exemplary” is used to mean “an example of” and unless otherwise stated, does not imply an ideal or preferred example, implementation, or embodiment. Section headers are used in the present disclosure to facilitate understanding of the disclosed implementations and are not intended to limit the disclosed technology in the sections only to the corresponding section. The disclosed implementations may be further embodied in a variety of different forms and, therefore, the scope of this disclosure or claimed subject matter is intended to be construed as not being limited to any of the embodiments set forth below. The various implementations may be embodied as methods, devices, components, systems, or non-transitory computer readable media. Accordingly, embodiments of this disclosure may, for example, take the form of hardware, software, firmware or any combination thereof.
[0040] This disclosure is directed to handling transmissions in a wireless cellular access network and is specifically directed to mechanisms for processing Channel State Information (CSI) .
[0041] Wireless Network Overview
[0042] A wireless communication network may include a radio access network for providing network access to wireless terminal devices, and a core network for routing data between the access networks or between the wireless network and other types of data networks. In a wireless access network, radio resources are provided for allocation and used for transmitting data and control information. FIG. 1 shows an exemplary wireless access network 100 including a wireless access network node (WANN) or wireless base station 102 (herein referred to as wireless base station, base station, wireless access node, wireless access network node, or WANN) and a wireless terminal device or user equipment (UE) 104 (herein referred to as user equipment, UE, terminal device, or wireless terminal device) that communicates with one another via over-the-air (OTA) radio communication resources 106. The wireless access network 100 may be implemented as, as for example, a 2G, 3G, 4G / LTE, or 5G cellular radio access network. Correspondingly, the base station 102 may be implemented as a 2G base station, a 3G node B, an LTE eNB, or a 5G New Radio (NR) gNB. The user equipment 104 may be implemented as mobile or fixed communication devices installed with mobile identity modules for accessing the base station 102. The user equipment 104 may include but is not limited to mobile phones, laptop computers, tablets, personal digital assistants, wearable devices, distributed remote sensor devices, and desktop computers. Alternatively, the wireless access network 100 may be implemented as other types of radio access networks, such as Wi-Fi, Bluetooth, ZigBee, and WiMax networks.
[0043] FIG. 2 further shows example processing components of the WANN 102 (e.g., base station) and the UE 104 of FIG. 1. The UE 104, for example, may include transceiver circuitry 206 coupled to one or more antennas 208 to effectuate wireless communication with the WANN 102 (or to other UEs) . The transceiver circuitry 206 may also be coupled to a processor 210, which may also be coupled to a memory 212 or other storage devices. The memory 212 may be transitory or non-transitory and may store therein computer instructions or code which, when read and executed by the processor 210, cause the processor 210 to implement various ones of the, functions, methods, and processes of the UE 104 described herein. The memory 212 may also store therein, and the processor 210 may also be configured to execute one or more models (e.g., Artificial Intelligence / Machine Learning (AI / ML) models) to perform one or more functionalities (e.g., AI / ML functionalities) . The memory 212 may also be utilized and allocated for buffering UL and DL transmissions in each band / carrier. The memory 212 may include multiple memory modules assigned to different functions (such as program memory, base band memory, and / or RF memory, to name a few) . Likewise, the WANN 102 may include transceiver circuitry 214 coupled to one or more antennas 216, which may include an antenna tower 218 in various forms, to effectuate wireless communications with the UE 104. The transceiver circuitry 214 may be coupled to one or more processors 220, which may further be coupled to a memory 222 or other storage devices. The memory 222 may be transitory or non-transitory and may store therein instructions or code that, when read and executed by the one or more processors 220, cause the one or more processors 220 to implement various functions, methods, and processes of the WANN 102 described herein.
[0044] Wireless Communication Resource Scheduling / Signaling
[0045] Returning to FIG. 1, the radio communication resources for the over-the-air interface 106 may include a combination of frequency, time, and / or spatial communication resources organized into various resource units or elements in frequency, time, and / or space. The radio communication resources 106 in frequency domain may include portions of licensed radio frequency bands, portions of unlicensed ration frequency bands, or portions of a mix of both licensed and unlicensed radio frequency bands. The radio communication resources 106 available for carrying the wireless communication signals between the base station 102 and user equipment 104 may be further divided into physical downlink channels 110 for transmitting wireless signals from the base station 102 to the user equipment 104 and physical uplink channels 120 for transmitting wireless signals from the user equipment 104 to the base station 102. The physical downlink channels 110 may further include physical downlink control channels (PDCCHs) 112 and physical downlink shared channels (PDSCHs) 114. Likewise, the physical uplink channels 120 may further include physical uplink control channels (PUCCHs) 122 and physical uplink shared channels (PUSCHs) 124. For simplification, other types of downlink and uplink channels are not shown in FIG. 1 but are within the scope of the current disclosure. The control channels PDCCHs 112 and PUCCHs 122 may be used for carrying control information in the form of control messages 116 and 126, herein referred to as Downlink Control Information (DCI) messages or Uplink Control Information (UCI) messages. The shared channels (shared between data and control information) PDSCHs 114 and PUSCHs 124 may be allocated and used for communicating downlink data transmissions 118 and uplink data transmissions 128 between the base station 102 and the user equipment 104.
[0046] The allocation and configuration of the radio communication resources associated with the data channels, such as the PDSCHs and the PUSCHs may be provided by one or more resource scheduling DCIs carried in the PDCCHs. The PDCCHs may be shared by a plurality of UEs in the access network. In various approaches, a particular UE may be configured to perform blind decode procedures on a preconfigured UE-specific Search Space (USS) to detect and identify a payload of a resource scheduling DCI carried in the PDCCH that specifically targets the particular UE. The blind decoding may be performed on preconfigured monitoring occasions of the PDCCH associated with USS. Such monitoring occasions may be referred to as a set of PDCCH candidates. Each PDCCH candidate may be associated with a set of Control Channel Elements (CCEs) . The UE may specifically use its Radio Network Temporary Identifier (RNTI) to decode the PDCCH candidates. The RNTI may be used to demask a PDCCH candidate’s CRC. If no CRC error is detected, the UE determines that PDCCH candidate carries its own control information. The UE may then process the DCI and extract the resource allocation information pertaining to the PDSCH and / or PUSCH for receiving and / or transmitting data.
[0047] Definitions
[0048] In various embodiments, the definition of “time instance” is equivalent to or comprises at least one of the following: slot, sub-slot, symbol, sub-symbol, frame, sub-frame, transmission occasion, occasion, millisecond, microsecond or other typical units for time.
[0049] In various embodiments, a spatial filter can be either a UE-side or a base station-side one, and the spatial filter is also called as spatial-domain filter.
[0050] In various embodiments, a “channel metric” may comprise at least one of the following: raw channel matrix, precoding matrix (i.e., eigenvectors derived from raw channel) , eigenvector (s) , precoder (s) , coefficient matrix (e.g., Rel-16 eType II codebook coefficient matrix W2) , or other channel metrics. A raw channel matrix and precoding matrix can be in any combination of spatial domain, frequency domain, time domain, or projection in any combination of angular domain, delay domain, or doppler domain.
[0051] In various embodiments, a “UL channel” can be PUCCH or PUSCH.
[0052] In various embodiments, a “DL channel” can be PDCCH or PDSCH.
[0053] In various embodiments, “UL RS” can be SRS, PRACH, DMRS (e.g., DMRS for PUSCH or PUCCH) .
[0054] In various embodiments, “DL RS” can be SSB, CSI-RS, DMRS (e.g., DMRS for PDSCH, or PDCCH) .
[0055] In various embodiments, “UL signal” can be UL channel or UL RS (e.g., SRS, PRACH, DMRS, PUSCH or PUCCH) .
[0056] In various embodiments, “DL signal” can be DL channel or DL RS (SSB, CSI-RS, DMRS, PDSCH, or PDCCH) .
[0057] In various embodiments, the power control parameter includes target power (also called as P0) , path loss RS, scaling factor for path loss (also called as alpha) , or closed loop process. Notes that, in this patent, the path-loss can be couple loss.
[0058] In various embodiments, “high layer signaling” can be RRC or MAC-CE, and “PHY layer signaling” can be DCI or UCI.
[0059] In various embodiments, “DCI” may be equivalent to “PDCCH. ”
[0060] In various embodiments, “precoding information” may be equivalent to a PMI, TPMI, precoding, or beam.
[0061] In various embodiments, “TRP” may be equivalent to a RS port, a RS port group, RS resource, or a RS resource set.
[0062] In various embodiments, “port group” may be equivalent to antenna group, or UE port group.
[0063] In various embodiments, “model” may refer to a general term, which is to describe that a UE is capable of doing a processing method, a functionality, a feature, or a feature group. “Model” may refer to functionality, function, functionality module, function module, processing method, information processing method, implementation, feature, feature group, configuration, configuration set, dataset (e.g., for model training) , or data-driven algorithms. Additionally, the instant disclosure refers to a “processing module, ” which may include such models discussed above, including AI and / or ML models.
[0064] Different models may be associated with different configurations (e.g., RRC configuration) . Model activation may refer to activating the corresponding configuration for the UE. Similarly, model deactivation, switching, and fallback may refer to deactivating the corresponding configuration, switching the configuration, and fallback to a configuration without the model, respectively.
[0065] In various embodiments, related issues in the present disclosure may also affect future mobile communication systems (such as 6G mobile communication networks and beyond) and may need to be solved urgently.
[0066] Framework for Processing Channel State Information (CSI)
[0067] Input and / or Output Configurations
[0068] Due to model generalization and scalability issues, in accordance with various embodiments, a model (e.g., a processing module) may support N input and / or output configurations, where N is an integer number larger than one. The number of input configuration and output configuration may be the same, or may not be the same for a particular model. For example, a model may support three kinds of input configurations, but support two output configurations.
[0069] For different model input / output configurations, some data processing procedures (e.g., processing methods) can be performed and / or some rules can be implemented to satisfy the model input / output configurations. Accordingly, some indications and / or mechanisms are needed to align the understanding between the UE 104 and the base station 102 for performing the data processing procedures and performing corresponding reverse data processing procedures, respectively.
[0070] In various embodiments, a “transmitter” indicates data processing procedure information to a “receiver. ” The “receiver” determines the reverse data processing procedure based on the indicated data processing procedure information. In different embodiments, the “transmitter” and “receiver” can be the following:
[0071] Alt. 1: The “transmitter” is the base station 102, and the “receiver” is the UE 104.
[0072] Alt. 2: The “transmitter” is the UE 104, and the “receiver” is the base station 102.
[0073] The indicated data processing procedure information can be transmitted via high-layer signaling or PHY layer signaling, e.g., DCI, UCI, etc. The indicated data processing procedure information can be transmitted via a downlink channel, e.g., DCI carried by PDCCH or PDSCH, data transmission by PDSCH. The indicated data processing procedure information can be transmitted via an uplink channel, e.g., UCI carried by PUCCH or PUSCH, data transmission by PUSCH, etc.
[0074] In various embodiments, the data processing procedure information can apply to spatial domain processing, which may at least comprise the number of transmitting antenna ports, the number of receiving antenna ports, the number of transmitting beams, the number of receiving beams, etc.
[0075] In various embodiments, the data processing procedure information can apply to frequency domain processing, which may at least comprise bandwidths, the number of subcarrier, resource element (RE) , resource element group (REG) , physical resource block (PRB) , resource block group (RBG) , subband (SB) , etc.
[0076] In various embodiments, the data processing procedure information can apply to time domain processing, which may at least comprise frames, half frames, sub-frames, slots, mini-slots, etc.
[0077] The data processing procedure information may also apply to the new domain processing based on the conversion on the above mentioned spatial / frequency / time domain, e.g., angular / delay / doppler domain.
[0078] In various embodiments, the data processing procedure information can apply to model output contents, which may at least comprise AI model output units, quantization bit sequence, etc.
[0079] Some example cases of AI based CSI compression and reconstruction are provided, where one AI model is applied for CSI compression at the UE 104 side (e.g., “encoder” ) , and the other AI model is applied for CSI reconstruction at the base station 102 side (e.g., “decoder” ) . The encoder and decoder may need to be paired for CSI compression and reconstruction. In different examples, the AI model is scalable over the input channel metrics with two different input configurations of numbers of transmitting (Tx) antenna ports, i.e., 32 Tx antenna ports and 16 Tx antenna ports. Then, the encoder can further compress the corresponding channel metric (s) with potential processing to the CSI (s) , and the decoder can reconstruct the channel metric (s) based on the received CSI (s) .
[0080] In various examples, the AI model is scalable over the input channel metrics with two different input configurations of number of bandwidths, i.e., 13 subbands and 15 subbands. Then, the encoder can further compress the corresponding channel metric (s) with potential processing to the CSI (s) , and the decoder can reconstruct the channel metric (s) based on the received CSI (s) .
[0081] In various examples, the AI model is scalable over the input channel metrics with two different input configurations of number of time instances, i.e., 2 slots and 4 slots. Then, the encoder can further compress the corresponding channel metric (s) with potential processing to the CSI (s) , and the decoder can reconstruct the channel metric (s) based on the received CSI (s) .
[0082] In various examples, the AI model is scalable over the output CSIs with three different output configurations of CSI feedback payload sizes, i.e., 100 bits, 200 bits and 300 bits. Then, the decoder can reconstruct the channel metric (s) based on the received CSI (s) corresponding to the CSI payload size (s) .
[0083] In the present disclosure, the embodiments are described principally with respect to CSI-RS. However, the teachings disclosed herein can also be applied to other reference signals as well, including, for example, DMRS, SRS, etc.
[0084] As such, in accordance with various embodiments, a method performed by the wireless terminal device 104 (e.g., UE 104) includes utilizing a processing module to derive a set of Channel State Information (CSI) , wherein the processing module supports N input and / or output configuration, wherein N is an integer larger than 1. The method also includes transmitting the set of CSI to a wireless access network node (WANN) 102 (e.g., base station 102) .
[0085] Similarly, a method performed by the WANN 102 includes receiving, from the wireless terminal device 104, a set of Channel State Information (CSI) , and utilizing a processing module to derive a set of Channel Metrics from the CSI, wherein the processing module supports N input and / or output configurations, wherein N is an integer larger than 1.
[0086] In accordance with various embodiments of the methods, the processing module comprises an Artificial Intelligence (AI) model.
[0087] In accordance with various embodiments, the methods include at least one of the following: the wireless terminal device 104 indicating data processing procedure information to the wireless access network node 102, wherein the wireless access network node 102 can determine a reverse data processing procedure based on the data processing procedure information; or the wireless terminal device 104 receiving an indication of data processing procedure information from the wireless access network node 102, wherein the wireless terminal device 104 determines a data processing procedure based on the data processing procedure information. In various embodiments, the data processing procedure information is transmitted via high-layer or PHY signaling via an uplink or a downlink channel.
[0088] In accordance with various embodiments of the methods, the data processing procedure information applies to at least one of: spatial domain processing; frequency domain processing; time domain processing; and / or model output contents.
[0089] Channel Metric Data Processing Procedures
[0090] In accordance with various embodiments, the base station 102 can transmit Channel State Information-Reference Signal (CSI-RS) to the UE 104. The UE 104 can derive a first set of channel metrics based on the measured CSI-RS. In one embodiment, the first set of channel metrics includes M channel metrics, each of which may comprise K components. Furthermore, each of components may comprise L units, wherein M, K, and L are positive integer numbers. Typically, the first set of channel metrics can be the measured channel metrics or can be the predicted channel metrics (e.g., using an AI model) based on the measured channel metrics. Then, the UE 104 may derive a second set of channel metrics based on the data processing procedures (e.g., processing methods) , e.g., discussed above, wherein the data processing procedures can be performed for any of M channel metrics, K components, and / or L units.
[0091] The second set of channel metrics may be the input into the model (e.g., AI / ML model or “processing module” ) to derive a set of CSI. In various examples, the second set of channel metrics may include one channel metric, so that M equals 1. The UE 104 may transmit the derived set of CSI to the base station 102. Then, the base station 102 may derive the set of channel metrics based on the CSI transmitted from the UE 104. To do this, the base station 102 may perform some reverse data processing procedures or operations to derive the reconstructed set of channel metrics. The reconstructed set of channel metrics may include M’ channel metrics, each of which may comprise K’ components. Furthermore, each of components may comprise L’ units, wherein M’ , K’ , and L’ are positive integer numbers. In one embodiment, L’ are larger than or equal to L, respectively. In some examples, M’ , K’ , and L’ are equal to M, K and L, respectively.
[0092] In various embodiments, the data processing procedure comprises at least a padding operation. In one example, assuming the largest units supported is L_max, if L<L_max, (L_max -L) units of contents are padded in specific components in the set of channel metrics in order to derive a set of CSI by the model.
[0093] The padded contents can comply with the following alternatives:
[0094] Alt 1: Pad the contents with special processing. In one embodiment, specific constant values are padded. For example, all zero values padding, all one values padding, or constant values derived from some specific formulas or rules.
[0095] Alt 2: Pad the contents based on the existing channel metrics, in accordance with one of the following:
[0096] In one embodiment, copy one existing channel metric for (L_max -L) times.
[0097] In one embodiment, copy the consecutive (L_max -L) channel metrics to the corresponding padding indices in case L>= L_max / 2. For example, the first consecutive (L_max –L) channel metrics from index#1, or the last consecutive (L_max –L) channel metrics from index# (L_max –L+1) .
[0098] In one embodiment, copy the consecutive L channel metrics with some additional padding contents to the corresponding padding indices in case L<= L_max / 2. In one embodiment, additional zero padding is performed for the units of L_max-2*L. In one embodiment, additional channel metric copy is performed for the units of L_max-2*L. For example, some channel metrics are duplicated for multiple times.
[0099] In one embodiment, pad the new channel metrics derived from the existing channel metrics. For example, new channel metrics derived by an AI / ML model / filtering methods, e.g., Wiener filtering, Kalman filtering, etc. / interpolation methods, e.g., linear interpolation, non-linear interpolation, etc. / some other operations on the existing channel metrics, e.g., an average operation.
[0100] The above different alternatives can be combined in various embodiments disclosed herein.
[0101] In various embodiments, the padded pattern and / or index (e.g., index of where to pad) can comply with at least one of the following alternatives:
[0102] Alt 1: Pad the contents at the indexes of first (L_max -L) units.
[0103] FIG. 3 shows a resource diagram that provides an example of padding in accordance with various embodiments. With reference to the example of FIG. 3, the AI model may support an input channel metric with L_max =32 ports in spatial domain, for example, the input channel metric is a four-dimensional tensor of {1, 2, 32, 13} , where 1 denotes one time instance, 2 denotes the real part and imaginary part of a complex number, 32 denotes the 32 Tx antenna ports, and 13 denotes 13 subbands. When the input channel metric is with L=16 ports in spatial domain (i.e., L<L_max) , for example, the input channel metric is a four-dimensional tensor of {1, 2, 16, 13} . Therefore, a padding operation is necessary to satisfy the model input requirement {1, 2, 32, 13} . That is some contents are padded at port components to derive the second set of channel metric with 32 ports from the first set of channel metrics, which include only 16 ports. In this example, zero padding is assumed to be performed for the Tx port dimension for the indexes of 1 to 16.
[0104] Alt 2: Pad the contents at the indexes of last (L_max -L) units.
[0105] FIG. 4 shows a resource diagram that provides another example of padding in accordance with various embodiments. With reference to the example in FIG. 4, and based on the assumptions of the example in Alt 1, above, zero padding is assumed to be performed for the Tx port dimension from the indexes of 17 to 32.
[0106] Alt 3: Pad the contents of X units between two units of channel metrics with the same interval, in the case that L_max can be divided by L, where X is a positive integer. Furthermore, X can be different values between two units of channel metric in the case that L_max cannot be divided by L.
[0107] FIG. 5 shows a resource diagram that provides another example of padding in accordance with various embodiments. With reference to the example in FIG. 5, and based on the assumptions of the example in Alt 1, above, zero padding is assumed to be performed for the Tx port dimension with the same interval, where X equals to 1. Alternatively, padding contents can be placed before channel metrics. With reference again to FIG. 5, zero paddings are placed at the odd indices and the channel metrics are placed at even indices. Other variations are possible.
[0108] Alt 4: Pad the contents at some index ranges of Y consecutive units, where Y is a positive integer and Y can be different values for different padding index ranges.
[0109] FIG. 6 shows a resource diagram that provides another example of padding in accordance with various embodiments. With reference to the example in FIG. 6, and based on the assumptions of the example in Alt 1, above, zero padding is assumed to be performed for two index ranges of consecutive Tx port dimension units, where Y equals 8 for both index range of 9 to 16 and index range of 25 to 32. In one embodiment, the padding processing is performed for each antenna polarization. Typically, for example, for dual-polarization, and respectively. Note that and are round up and round down operations, respectively.
[0110] FIG. 7 shows a resource diagram that provides another example of padding in accordance with various embodiments. With reference to the example in FIG. 7, and based again on the assumptions of the example in Alt 1, above, zero padding is assumed to be performed for two index ranges of consecutive Tx port dimension units, where Y equals 8 for both index range of 1 to 8 and index range of 25 to 32. In one embodiment, the padding processing is performed for the first index range of 1 to and the second index range of to L_max, where and respectively. For clarification, in this example, the two index ranges are at head and tail of total L_max units.
[0111] Alt 5: Pad the contents at some specific patterns of indices about which the UE 104 and the base station 102 have a common understanding. For example, the pattern of padded indices can be derived by some specific formulas, tables, or specification. In one embodiment, an index of the patterns can be indicated from the base station 102 to the UE 104, and vice versa. The detailed index indication can be a binary indication, or a bitmap.
[0112] In another embodiment, the data processing procedure may comprise at least a grouping operation.
[0113] Assume the number of input units supported is L_in (e.g., which may be a smallest number of input units supported) . In one embodiment, if L>L_in, then divide L units of contents in the specific components in the set of channel metrics into groups to derive a set of CSI by the model. The UE 104 then can transmit the derived set of CSI to the base station 104. Then, the base station 104 may derive the set of channel metrics based on the CSI transmitted from the UE 104, where some reverse data processing procedures or operations can be performed to derive the reconstructed set of channel metrics.
[0114] In various examples, the group operation can comply with one or more of the following alternatives:
[0115] Alt 1: Consecutive L_in units of channel metrics are divided into a set / group.
[0116] Alt 2: L_in channel metrics with the same time interval are divided into a set / group. For example, specifically, the j-th unit group of channel metric includes the channel metric with the index of (i-1) ×G+j in the total number of channel metrics, where i=1, 2, …, L_in, and j=1, 2, …, G. For example, in case L_in = 8, L=32, and G=4, the unit group#1 comprises the channel metrics with the indices of {1, 5, 9, 13, 17, 21, 25, 29} .
[0117] Alt 3: L_in channel metrics in a set / group consist of the consecutive L_in / 2 channel metrics in the first half and the consecutive L_in / 2 channel metrics counted backward in the second half. For example, in case L_in = 8, L=32, and G=4, the unit group#1 comprises the channel metrics with the indices of {1, 2, 3, 4, 29, 30, 31, 32} .
[0118] In the case that L cannot be divided by L_in, there may be a group comprising less than L_in units of channel metrics. In order to derive a complete set of CSI, the padding operation may be needed for this group, which padding operations are discussed above.
[0119] The method of dividing channel metric into sets / groups can be configured by the base station 102 via high layer configuration, or indicated by the UE 104 to the base station 102. Different models can be configured with different grouping methods, as well.
[0120] As such, in accordance with various embodiments, the methods include the WANN 102 (e.g., base station 102) transmitting, and the wireless terminal device 104 (e.g., UE 104) receiving, CSI-RS, and the wireless terminal device 104 deriving a first set of channel metrics based on measurement of the CSI-RS, wherein the first set of channel metrics comprises M channel metrics, each of which comprises K components, wherein each component comprises L units, wherein M, K, and L are positive integer numbers; and the wireless terminal device 104 deriving a second set of channel metrics using a data processing procedure for any of the M channel metrics and / or K components and / or L units.
[0121] In accordance with various embodiments, the methods include the WANN 102 deriving the set of channel metrics based on the set of CSI by performing the reverse data processing procedure to derive a reconstructed set of channel metrics, wherein the reconstructed set of channel metric comprises M’ channel metrics, each of which comprises K’ components, wherein each component comprises L’ units, wherein M’ , K’ , and L’ are positive integer number, and wherein each L’ is larger than or equal to each L, respectively, and wherein each M’ and K’ are equal to each M and K, respectively.
[0122] In accordance with various embodiments of the methods, the data processing procedure comprises a padding operation, wherein a largest number of input units supported by the processing module is L_max, wherein L is less than L_max, and wherein the padding operation comprises padding L_max -L units of padded contents in specific components in the second set of channel metrics in order to derive the set of CSI using the processing module.
[0123] In accordance with various embodiments of the methods, the padded contents comprise at least one of the following: specific constant values of 0 or 1; a constant value derived from a formula or rule; a copy of an existing channel metric for L_max -L instances; a copy of consecutive L_max -L channel metrics in corresponding padding indices when L >=L_max / 2, wherein the consecutive L_max -L channel metrics may be first consecutive L_max –L channel metrics from an index#1, or last consecutive L_max –L channel metrics from index# (L_max –L+1) ; a copy of consecutive L channel metrics with additional padding contents in corresponding padding indices when L <= L_max / 2, wherein the additional padding comprises zero padding an additional copy of the channel metrics for units of L_max-2*L; or new channel metrics derived from the existing channel metrics.
[0124] In accordance with various embodiments of the methods, a pattern or index of padding comprises at least one of the following: contents are padded at indexes of the first L_max -L units; contents are padded at indexes of the last L_max -L units; contents are padded at X units between two units of channel metrics with a same interval when L_max can be divided by L, wherein X is a positive integer; contents are padded at X units between the two units of channel metrics with a different interval when L_max cannot be divided by L; contents are padded at an index range of Y consecutive units, where Y is a positive integer and Y can be different values for different padding index ranges; or contents are padded at a pattern of indices understood by both the wireless terminal device and the wireless access network node.
[0125] In accordance with various embodiments of the methods, the data processing procedure comprises a grouping operation, wherein a number of input units supported by the processing module is L_in, wherein L is larger than L_in, and wherein L units of contents in components in the set of channel metrics are divided into groups to derive the set of CSI by the processing module, wherein the grouping operation complies with at least one of the following: consecutive L_in units of channel metrics are divided into a set or group; L_in channel metrics with a same time interval are divided into a set or group; or L_in channel metrics in a set or group consist of consecutive L_in / 2 channel metrics in a first half and the consecutive L_in / 2 channel metrics counted backward in a second half.
[0126] Model Output Data Processing Procedures
[0127] Regarding the model scalability over multiple model outputs, in some embodiments, model output data processing procedures or methods may be necessary to derive various sets of CSI by a model. In various embodiments, the set of channel metrics (e.g., the first set, which may not have been processed, or the second set, which has been processed) is the input into the model (e.g., AI / ML model) to derive P sets of CSI, where P is a positive integer larger than one. The UE 104 may transmit a set of CSI (e.g., one set) of the derived P sets of CSI to the base station 102. Then, the base station 102 may derive the set of channel metrics based on the CSI transmitted from the UE 104. Some reverse data processing procedures or operations can be performed to derive the reconstructed set of channel metrics.
[0128] In various embodiments, the data processing procedures (e.g., processing methods) may comprise at least a masking operation.
[0129] Assume the largest model output units supported is Q_max. In one embodiment, if Q<Q_max, (Q_max -Q) units of contents are masked in order to derive a set of CSI by the model.
[0130] The masked pattern and / or indexes can comply with at least one of the following alternatives:
[0131] Alt 1: Mask the contents at the index of first (Q_max -Q) units.
[0132] FIG. 8 shows a resource diagram that provides an example of masking in accordance with various embodiments. With reference to the example in FIG. 8, the AI model supports the model output with Q_max =100 units, and the model output is a vector with the length of 100. When a model output is with Q=60 units (i.e., Q<Q_max) , similarly, the model output is a vector with the length of 60. Therefore, a masking operation is necessary to satisfy the model output requirement of the length of 60 units. That is, the masked 40 units (highlighted in FIG. 8) are not involved to derive the set of CSI for feedback to the base station 102. In this example, the masking operation is assumed to be performed for the model output from the indexes of 1 to 40.
[0133] Alt 2: Mask the contents at the index of last (Q_max -Q) units.
[0134] FIG. 9 shows a resource diagram that provides another example of masking in accordance with various embodiments. With reference to the example in FIG. 9, and based again on the assumptions of the masking example in Alt 1, above, the masking operation is assumed to be performed for the model output units from the index of 61 to 100.
[0135] Alt 3: Mask the contents at X units between two units of channel metrics with the same interval in the case where Q_max can be divided by Q, where X is a positive integer. Furthermore, X can be different values between two units of channel metric in the case where Q_max cannot be divided by Q. The starting index can be an odd index or an even index. In various examples, X equals to 1.
[0136] FIG. 10 shows a resource diagram that provides another example of masking in accordance with various embodiments. In this example, the AI model supports the model output with Q_max =100 units. When the model output is with Q=50 units (i.e., Q<Q_max) , a masking operation may be necessary to satisfy the model output requirement of the length of 50 units, i.e., the masked 50 units (highlighted in FIG. 10) are not involved to derive the set of CSI for feedback to the base station 102. In this example, the masking operation is assumed to be performed for the model output units with the same interval from unit 2, where X equals to 1.
[0137] Alternatively, masked contents can also be started as the odd index. With reference to FIG. 10 again as an example, masking operations could be performed at the odd indices from unit#1 and the even units could be applied for deriving a set of CSI.
[0138] Alt 4: Mask the contents at some index ranges of Y consecutive units, where Y is a positive integer, and Y can be different values for different masking index ranges.
[0139] FIG. 11 shows a resource diagram that provides another example of masking in accordance with various embodiments. With reference to the example in FIG. 9, and based again on the assumptions of the masking example in Alt 3, above, the masking operation is assumed to be performed for two index ranges of consecutive model output units, where Y equals 25 for both index range of 26 to 50 and index range of 25 to 32. In one embodiment, the masking operation is performed for multiple groups of consecutive units with the same interval. Typically, for two groups, and respectively. Note that and are round up and round down operations, respectively.
[0140] FIG. 12 shows a resource diagram that provides another example of masking in accordance with various embodiments. With reference to the example in FIG. 12, the masking operation is assumed to be performed for two index ranges of consecutive model output units, where Y equals to 25 for both index range of 1 to 25 and index range of 76 to 100. In one embodiment, the masking operation is performed for the first index range of 1 to and the second index range of to Q_max, where and respectively. To clarify, the two index ranges in this example are at head and tail of total Q_max units.
[0141] FIG. 13 shows a resource diagram that provides another example of masking in accordance with various embodiments. With reference to the example in FIG. 13, in another embodiment, the masking operation is not performed for the first index range of 1 to and the second index range of to Q_max, where and respectively. For clarification, in this example, the two index ranges of the head and tail of total Q_max units are remained.
[0142] Alt 5: Mask the contents at some specific patterns of indices about which the UE 104 and the base station 102 have a common understanding.
[0143] For example, the pattern of masked indices can be derived by some specific formulas, tables, or specification. In one embodiment, an index of the patterns can be indicated from the base station 102 to the UE 104, and vice versa. The detailed index indication can be a binary indication, bitmap, or an index of masking pattern in the case the base station 102 and the UE 104 have the same understanding on the index.
[0144] In various examples, the model output units can also be the bit information, where the model output contents are quantized into bits. The quantization methods include at least the scalar quantization, which comprises b-bit uniform quantization, b-bit non-uniform quantization, where b is a positive integer larger than 0, and vector quantization.
[0145] In another embodiment, the data processing procedure may comprise at least a truncation operation.
[0146] Assume the largest model output elements supported is B_max. In one embodiment, if B <B_max, (B_max -B) elements of contents are truncated in order to derive a set of CSI. The model output elements may be a binary bitstream. FIG. 14 shows an example bitstream illustrating a truncation operation in accordance with various embodiments. With reference to the example of FIG. 14, the model output can support a bitsream comprising B_max bits, while only B bits are needed to derive one set of CSI.
[0147] The truncation method can refer to the alternatives of the masking methods, e.g., truncate the last (B_max -B) elements, or truncate the first (B_max -B) elements, or truncate (B_max -B) elements via other specific patterns.
[0148] In one embodiment, the data processing procedures may comprise at least a switching operation.
[0149] Assume the model supports S branches of output units, where S, O_i are positive integers, and each branch consists of O_i units, where i = 1, 2, …, S, and O_i is a different number for different branches. In one embodiment, the branch of output units is switched to derive the corresponding set of CSI. FIG. 15 shows an example output diagram illustrating an example of different branches in accordance with various embodiments.
[0150] In various embodiments, configurations on different output branches can be indicated at least by identification information, in accordance with the following alternatives:
[0151] Alt 1: One-level identification information, e.g., model structure branch ID. For example, the model includes 2 output dimensions, which are 100 units and 50 units, respectively. The model structure branch ID#1 is associated with 100 units, and ID#2 is related to 50 units. So, only a one-level ID can indicate the configuration of model output.
[0152] Alt 2: Two-level identification information, e.g., model ID + structure branch ID. For example, there are two models achieving one output dimension, e.g., one model with one output of 50 units (Model ID#1) and the other includes 2 output dimensions, which are 100 units and 50 units, respectively (Model ID#2) . In order to clearly indicate the output configuration of 50 units from the second model, Model ID#2 + structure ID#1 is necessary.
[0153] As such, in accordance with various embodiments, the methods include the wireless terminal device 104 inputting at least one of the first set of channel metrics and / or the second set of channel metrics into the processing module to derive P sets of CSI, wherein P is a positive integer larger than 1, and the wireless terminal device 104 transmitting, and the WANN 102 receiving, a set of the P sets of CSI.
[0154] In accordance with various embodiments of the methods, the data processing procedure comprises a masking operation, wherein a largest number of output units supported by the processing module is Q_max, wherein Q is a number of output units for a particular operation of the processing module, and is less than Q_max, and wherein the masking operation comprises masking Q_max -Q units of contents in order to derive the set of CSI using the processing module.
[0155] In accordance with various embodiments of the methods, a pattern or index of masking comprises at least one of the following: contents are masked at indexes of the first Q_max -Q units; contents are masked at indexes of the last Q_max -Q units; contents are masked at X units between two units of channel metrics with a same interval when Q_max can be divided by Q, wherein X is a positive integer; contents are masked at X units between the two units of channel metrics with a different interval when Q_max cannot be divided by Q; contents are masked at an index range of Y consecutive units, where Y is a positive integer and Y can be different values for different masking index ranges; or contents are masked at a pattern of indices understood by both the wireless terminal device and the wireless access network node.
[0156] In accordance with various embodiments of the methods, the data processing procedure comprises a truncation operation, wherein a largest number of output elements supported by the processing module is B_max, wherein B is a number of output elements for a particular operation of the processing module, and is less than B_max, and wherein the truncation operation comprises truncating B_max -B elements of contents in order to derive the set of CSI using the processing module.
[0157] In accordance with various embodiments of the methods, the data processing procedure comprises a switch operation, wherein the processing module supports S branches of output units, wherein each branch comprises O_i units, wherein S and O_i are positive integers, and wherein i = 1, 2, …S, and wherein O_i is a different number for each branch, and wherein the switch operation comprises switching a branch of output units to derive the set of CSI using the processing module.
[0158] Indication of Data Processing Procedures
[0159] Indication of Channel Metric Padding
[0160] In various embodiments, because the UE 104 may pad (L_max -L) channel metrics and derive a set of CSI reported to the base station 102, the base station 102 may reconstruct the set of channel metrics based on the reported set of CSI. In one embodiment, the base station 102 may indicate data processing procedure information to the UE 104 for deriving the second set of channel metrics. The data processing procedure can be associated with the following contents:
[0161] Information of padding pattern, at least including one or more of information of the starting index, the ending index, the padding number, and / or the type of indication information on the padding pattern, at least including one or more of the following types: a binary information, bitmap, index of patterns in the case the base station 102 and the UE 104 have the same understanding on the index.
[0162] For example, Table 1, below, provides an example of the indication types:
[0163] Table 1
[0164] Information of padding contents and / or methods, at least including one or more of the information of the padding contents, or the type of indication information on padding contents / methods, at least including one or more of the following types: a binary information, bitmap, index of padding contents in the case the base station 102 and the UE 104 have the same understanding on the index.
[0165] With reference again to FIG. 3 as an example, the model supports input channel metrics for 32 Tx ports, and 16 Tx ports are configured. According to the indicated information, the first 16 ports are zero padded for deriving a set of CSI. Therefore, the indication information can be a binary starting index, e.g., 00000, since the base station 102 knows the UE 104 should pad zeros for 16 ports. Or a bitmap is used to indicate the padding index, e.g., 11…. 100…0 (total length of 32 with 16 one values and 16 zero values) , wherein 1 indicates the padding indices.
[0166] In addition, in certain examples, the base station 102 may not know which channel metric in the set is padded. Therefore, in various embodiments, the UE 104 indicates the data processing procedure information to the base station 102 for deriving the reconstructed channel metrics. Symmetrically, the data processing procedures can be associated with the following contents:
[0167] The information of padding pattern, at least including one or more of information of the starting index, the ending index, the padding number, or the type of indication information, at least including one or more of the following types: a binary information, bitmap, index of patterns in the case the base station 102 and the UE 104 have the same understanding on the index
[0168] The information of padding contents / methods, at least including one or more of the information of the padding contents, or the type of indication information on padding contents / methods, at least including one or more of the following types: a binary information, bitmap, index of patterns in the case the base station 102 and the UE 104 have the same understanding on the index.
[0169] Indication of Channel Metric Grouping
[0170] Because the UE 104 may divide L channel metrics into G groups and derive a set of CSI reported to the base station 102, the base station 102 may reconstruct the set of channel metrics based on the reported set of CSI. In one embodiment, the base station 102 indicates the data processing procedure information to the UE 104 for deriving the second set of channel metrics. The data processing procedures can be associated with the following contents:
[0171] Information of grouping pattern, at least including one or more of information of group starting index, the group ending index, the number of channel metrics in one group, the type of indication information, at least including one or more of the following types: a binary information, bitmap, index of patterns in case the base station 102 and the UE 104 have the same understanding on the index.
[0172] Information of grouping methods, at least including one or more of the information of the grouping methods, or the type of indication information on grouping methods, at least including one or more of the following types: a binary information, bitmap, or index of padding contents in case the base station 102 and the UE 104 have the same understanding on the index.
[0173] Information of padding pattern (in case the L cannot be divided by L_in) , at least including one or more of the information of the index of padding group, the index for padding channel metric in the group, the index for padding channel metric in total number, the starting index for padding, the ending index for padding, or the padding number. The type of indication information including at least one or more of the following types: a binary information, bitmap, or index of patterns in case the base station 102 and the UE 104 have the same understanding on the index.
[0174] The information of padding contents / methods, at least including one or more of the information of the padding contents / methods. The type of indication information on the padding contents includes at least one or more of the following types: a binary information, bitmap, index of padding contents in case the base station 102 and the UE 104 have the same understanding on the index.
[0175] In some instances, the base station 102 may not know which channel metric in the set is padded. Therefore, in various embodiments, the UE 104 may indicate the data processing procedure information to the base station 102 for potential reverse data processing procedures to derive the reconstructed channel metrics. Symmetrically, the data processing procedures can be associated with the above mentioned embodiments.
[0176] In various examples, no indication information may be needed to report since the base station 102 and the UE 104 can have the same understanding on the index of padding pattern. For example, the padding operation happens at last (L_max -L) units.
[0177] Indication of Channel Metric Masking
[0178] Because the UE 104 may mask Q units from Q_max units to derive a set of CSI reported to the base station 102, the base station 102 may reconstruct the set of channel metrics based on the reported set of CSI. In one embodiment, the base station 102 may indicate the data processing procedure information to the UE 104 for deriving the set of CSI. The data processing procedures can be associated with the following contents:
[0179] Information of masking pattern, at least including one or more of information of starting index of masking, ending index of masking, the number of masked units, the type of indication information, including at least one or more of the following types: a binary information, bitmap, or index of patterns in case the base station 102 and the UE 104 have the same understanding on the index.
[0180] Information of masking methods, including at least one or more of the information of the masking methods. The type of indication information on masking methods, including at least one or more of the following types: a binary information, bitmap, or index of masking contents in case the base station 102 and the UE 104 have the same understanding on the index.
[0181] Note that this embodiment is also applicable to the truncation operation discussed above.
[0182] Referring again to FIG. 8 as an example, the model supports model output units of 100, and 60 output units are configured. According to the indicated information, the first 40 ports are masked for not deriving a set of CSI. Therefore, the indication information can be a binary starting index and the number of masked units, e.g., 0000000 and 101000, since the base station 102 knows the UE 104 should mask the first 40 units. Or a bitmap is used to indicate the masking index, e.g., 11…. 100…0 (total length of 100 with 40 one values and 60 zero values) , wherein 1 indicates the masking indices.
[0183] As such, in accordance with various embodiments, the methods include the wireless terminal device 104 using the data processing procedure information to derive the second set of channel metrics, wherein the data processing procedure information comprises at least one of:information of a pattern or index of padding, comprising at least one of: a starting index, an ending index, a padding number, and / or a type of indication information on padding pattern, including at least one of a binary information, a bitmap, or an index of patterns; or information of padded contents, comprising: a type of indication information on padding contents, including at least one of a binary information, a bitmap, or an index of padded contents.
[0184] As such, in accordance with various embodiments, the methods include the wireless terminal device 104 using the data processing procedure information to derive the second set of channel metrics, wherein the data processing procedure information comprises at least one of:information of a grouping pattern comprising at least one of: a group starting index, a group ending index, a number of channel metrics in one group, and / or a type of indication information, including at least one of a binary information, a bitmap, or an index of patterns; or information of grouping methods, comprising: a type of indication information on grouping methods, including at least one of a binary information, a bitmap, or an index of patterns; or information of a padding pattern including at least one of: an index of a padding group, an index for padding channel metric in the group, an index for padding channel metric in total number, a starting index for padding, an ending index for padding, or a padding number.
[0185] As such, in accordance with various embodiments, the methods include the wireless terminal device 104 using the data processing procedure information to derive the set of CSI, wherein the data processing procedure information comprises at least one of: information of a pattern or index of masking, comprising at least one of: a starting index of masking, an ending index of masking, a number of masked units, and / or a type of indication information on masking, including at least one of a binary information, a bitmap, or an index of patterns; or information of masking methods, comprising: a type of indication information on masking methods, including at least one of a binary information, a bitmap, or an index of masking contents.
[0186] Model Structure Determination
[0187] In various embodiments, a model includes two parts: model structure and model parameters. Regarding the two-sided model use case, the end-to-end performance can be guaranteed when the model structure is known between the two sides. Typically, the encoder structure at the UE 104 side should be revealed to the base station 102 for the base station 102 to train the decoder deployed at base station 102 side. In one embodiment, the encoder model structure is indicated to the base station 102, including explicit indication of hyper-parameters, such as the values of hyper-parameters or some implicit identification information, such as model ID, model structure ID, etc. In another embodiment, the encoder model structure or hyper-parameters are fixed in the specification or otherwise predetermined. Note that, the specification on model structure can also apply to decoder or encoder plus decoder, other than encoder.
[0188] To describe the model structure, at least the same understanding on the model hyper-parameters is necessary. The following provides some embodiments regarding the hyper-parameters of the model structure.
[0189] In one embodiment, the model structure has a Transformer-related backbone, and at least the following alternatives of hyper-parameters can be included:
[0190] Alternative 1: Typical Transformer backbone:
[0191] -Initial input dimension;
[0192] -Number of Transformer blocks (e.g., 2, 4, or 6; ) ;
[0193] -Model dimension (e.g., 128, 256 or 512) ;
[0194] -Hyper-parameters in the multi-head self-attention modules, at least including number of attention heads (e.g., 4, 8, 16, or 32) and the dimension of each attention head (e.g., model dimension / number of attention heads) , scale factor or bias of QKV matrices, dropout rate (e.g., 0.1 or 0.2) ; and / or
[0195] -Hyper-parameters in the feed-forward modules, at least including number of processing layers (e.g., 1 or 2) , dimension of latent space (e.g., 512, 1024, or 2048) , and the activation function.
[0196] FIG. 16 shows an example of hyper-parameters illustrated in a typical Transformer block in accordance with various embodiments.
[0197] Alternative 2: Swin Transformer backbone:
[0198] -Initial input dimension;
[0199] -Number of Swin-Transformer blocks;
[0200] -Model dimension;
[0201] -Hyper-parameters in the patch partition modules, at least including patch size, and patch resolution;
[0202] -Hyper-parameters in the multi-head self-attention modules, at least including number of attention heads, the dimension of each attention head, window size, scale factor or bias of QKV matrices, dropout rate; and / or
[0203] -Hyper-parameters in the feed-forward modules, at least including number of processing layers, dimension of latent space, and the activation function.
[0204] Alternative 3: Vision transformer backbone:
[0205] -Initial input dimension;
[0206] -Number of Vision-Transformer blocks;
[0207] -Model dimension;
[0208] -Hyper-parameters in the patch partition modules, at least including patch size, patch resolution;
[0209] -Hyper-parameters in the multi-head self-attention modules, at least including number of attention heads, the dimension of each attention head, scale factor or bias of QKV matrices, dropout rate; and / or
[0210] -Hyper-parameters in the feed-forward modules, at least including number of processing layers, dimension of latent space, and the activation function.
[0211] In certain examples, the Embedding and Feed-Forward module may at least include the processing of Multi-Layer Perceptron (MLP) , Full-Connected (FC) layers, Convolutional Neural Network (CNN) , Recurrent Neural Network (RNN) , Long-Short Term Memory (LSTM) , Residual Network (ResNet) , Conv-LSTM, or other processing modules. In addition, the activation function may include at least Sigmoid, Tanh, ReLU, pReLU, LeakyReLU, GeLU, ELU, SELU, softmax, or other activation functions.
[0212] In one embodiment, the model structure has an LSTM-related backbone, and at least the following alternatives of hyper-parameters can be included:
[0213] Alternative 1: Typical LSTM backbone:
[0214] -Input dimension;
[0215] -Hidden dimension; and / or
[0216] -Hyper-parameters for LSTM blocks, at least including Input Gate, Forget Gate, Output Gate, number of LSTM layers, bias, dropout rate, activation function.
[0217] Alternative 2: Bi-directional LSTM backbone:
[0218] -Input dimension;
[0219] -Hidden dimension; and / or
[0220] -Hyper-parameters for LSTM blocks, at least including Input Gate, Forget Gate, Output Gate, number of LSTM layers, bias, dropout rate, activation function.
[0221] Alternative 3: Conv-LSTM backbone:
[0222] -Input dimension;
[0223] -Hidden dimension;
[0224] -Hyper-parameters for CNN blocks, at least including kernel size, strides, paddings, activation function, normalization;
[0225] -Hyper-parameters for Pooling blocks, at least including pooling kernel size, strides; and / or
[0226] -Hyper-parameters for LSTM blocks, at least including Input Gate, Forget Gate, Output Gate, number of LSTM layers, bias, dropout rate, activation function.
[0227] In one embodiment, the model structure has a CNN-related backbone, and at least the following alternatives of hyper-parameters can be included:
[0228] Alternative 1: Typical CNN backbone:
[0229] -Input dimension;
[0230] -Hyper-parameters in the feed-forward modules, at least including number of processing layers, dimension of latent space, and the activation function; and / or
[0231] -Hyper-parameters for CNN blocks, at least including kernel size, strides, paddings, activation function, normalization.
[0232] Alternative 2: ResNet backbone:
[0233] -Input dimension;
[0234] -Number of ResNet blocks;
[0235] -Hyper-parameters in the feed-forward modules, at least including number of processing layers, dimension of latent space, and the activation function; and / or
[0236] -Hyper-parameters for CNN blocks, at least including kernel size, strides, paddings, activation function, normalization.
[0237] In some examples, some high-level hyper-parameters can also be considered, which may apply to any of the above mentioned model structures, such as loss function (e.g., SGCS, NMSE) , optimizer (e.g., Adam, AdamW) , learning rate (e.g., constant learning rate, cosine learning rate) , number of epochs, number of batches, etc.
[0238] As such, in accordance with various embodiments, the methods include the wireless terminal device 104 indicating, and the WANN 102 receiving an indication of, an encoder model structure, wherein the encoder model structure includes an indication of a hyper-parameter, wherein the indication of the hyper-parameter includes a value of the hyper-parameter or a model ID or a model structure ID.
[0239] Indication of Model Parameters
[0240] In various embodiments, based on the model structure mentioned above, model parameters may also be needed for model inference. In one embodiment, model parameters are transferred from the base station 102 to the UE 104. In various examples, the encoder parameters are indicated to the UE 104 from the base station 102. For example, the encoder and decoder are trained as a pair at the base station 102 with the known structure of encoder, and then the encoder parameters are transferred to the UE 104 from the base station 102. In another embodiment, the model parameters are specified in a standardization or are otherwise predefined. Note that the specification on model parameters can also apply to decoder or encoder plus decoder, other than encoder.
[0241] In addition, according to the scalable model structure mentioned above, there are at least the following alternatives regarding the model parameters transfer:
[0242] Alternative 1: Transfer a whole set of parameters for the model structure with potential data processing procedure information to derive a set of CSI. The potential data processing procedures comprise at least masking / deactivating, freezing, or some special processing (e.g., set to zero / one values) . For example, the whole model parameters may be transferred to the UE 104, which are compatible with the maximum model output configuration (e.g., 100 output units) . When model output units of 50 are needed, the parameters of some units are needed to be masked / deactivated (i.e., not applied for inference) to derive a set of CSI.
[0243] Alternative 2: Transfer a relevant partial set of parameters for the model structure to derive a set of CSI. For example, when model output supports multiple branches, e.g., 100 units, 50 units, and 20 units, while the current configuration needs output 50 units, thus, only the parameters of 50 units are transferred to be activated (i.e., applied for inference, and other irrelevant parameters may not be transferred) to derive a set of CSI.
[0244] Regarding the data processing procedure information for the potential data processing procedures to achieve scalability, at least the following alternatives regarding the indication for processing information on model parameters can be considered:
[0245] Alternative 1: Transfer a whole set of parameters for the model structure with potential processing to derive a set of CSI. The indication information about model parameters, may include at least one or more of information of the starting index of activated / deactivated units of parameters, the ending index of activated / deactivated units of parameters, the number of activated / deactivated units of parameters, which include at least parameters for the corresponding operation layer (s) , and the type of indication information, at least including one or more of the following types: a binary information, bitmap, or index of patterns in case the base station 102 and the UE 104 have the same understanding on the index.
[0246] Alternative 2: Transfer a partial set of parameters for the model structure with potential processing to derive a set of CSI. The indication information about model parameters, may including at least one or more of information of the configuration related information, identification information (e.g., model parameter ID) , or additional information (e.g., channel scenario, some implementation related information, e.g., TxRU, antenna tilt angle, etc. ) . The type of indication information, may including at least one or more of the following types: a binary information, bitmap, or index of patterns in case the base station 102 and the UE 104 have the same understanding on the index.
[0247] As such, in accordance with various embodiments, the methods include the WANN 102 transmitting, and the wireless terminal device 104 receiving, at least one of: a complete set of parameters for a structure of the processing module with the data processing procedure information to derive the set of CSI, wherein the data processing procedure comprises at least one of masking, deactivating, freezing, or setting to zero or one values; or a relevant partial set of parameters for the structure of the processing module to derive the set of CSI.
[0248] In accordance with various embodiments of the methods, the whole set of parameters comprises at least one of: a starting index of activated or deactivated units of parameters, an ending index of activated or deactivated units of parameters, a number of activated or deactivated units of parameters including at least parameters for corresponding operation layers, or a type of indication information including at least one of a binary information, a bitmap, or an index of patterns; or wherein the partial set of parameters comprises at least one of: configuration related information, identification information, additional information, or a type of indication information including at least one of a binary information, a bitmap, or an index of patterns.
[0249] The description and accompanying drawings above provide specific example embodiments and implementations. The described subject matter may, however, be embodied in a variety of different forms and, therefore, covered or claimed subject matter is intended to be construed as not being limited to any example embodiments set forth herein. A reasonably broad scope for claimed or covered subject matter is intended. Among other things, for example, subject matter may be embodied as methods, devices, components, systems, or non-transitory computer-readable media for storing computer codes. Accordingly, embodiments may, for example, take the form of hardware, software, firmware, storage media or any combination thereof. For example, the method embodiments described above may be implemented by components, devices, or systems including memory and processors by executing computer codes stored in the memory.
[0250] Throughout the specification and claims, terms may have nuanced meanings suggested or implied in context beyond an explicitly stated meaning. Likewise, the phrase “in one embodiment / implementation / example / approach” as used herein does not necessarily refer to the same embodiment and the phrase “in another embodiment / implementation / example / approach” as used herein does not necessarily refer to a different embodiment. It is intended, for example, that claimed subject matter includes combinations of example embodiments in whole or in part.
[0251] In general, terminology may be understood at least in part from usage in context. For example, terms, such as “and” , “or” , or “and / or, ” as used herein may include a variety of meanings that may depend at least in part on the context in which such terms are used. Typically, “or” if used to associate a list, such as A, B or C, is intended to mean A, B, and C, here used in the inclusive sense, as well as A, B or C, here used in the exclusive sense. In addition, the term “one or more” as used herein, depending at least in part upon context, may be used to describe any feature, structure, or characteristic in a singular sense or may be used to describe combinations of features, structures or characteristics in a plural sense. Similarly, terms, such as “a, ” “an, ” or “the, ” may be understood to convey a singular usage or to convey a plural usage, depending at least in part upon context. In addition, the term “based on” may be understood as not necessarily intended to convey an exclusive set of factors and may, instead, allow for existence of additional factors not necessarily expressly described, again, depending at least in part on context.
[0252] Reference throughout this specification to features, advantages, or similar language does not imply that all of the features and advantages that may be realized with the present solution should be or are included in any single implementation thereof. Rather, language referring to the features and advantages is understood to mean that a specific feature, advantage, or characteristic described in connection with an embodiment is included in at least one embodiment of the present solution. Thus, discussions of the features and advantages, and similar language, throughout the specification may, but do not necessarily, refer to the same embodiment.
[0253] Furthermore, the described features, advantages and characteristics of the present solution may be combined in any suitable manner in one or more embodiments. One of ordinary skill in the relevant art will recognize, in light of the description herein, that the present solution can be practiced without one or more of the specific features or advantages of a particular embodiment. In other instances, additional features and advantages may be recognized in certain embodiments that may not be present in all embodiments of the present solution.
Claims
1.A method performed by a wireless terminal device comprising:utilizing a processing module to derive a set of Channel State Information (CSI) , wherein the processing module supports N input and / or output configuration, wherein N is an integer larger than 1; andtransmitting the set of CSI to a wireless access network node.2.The method according to claim 1,wherein the processing module comprises an Artificial Intelligence (AI) model.3.The method according to one of claims 1-2, comprising:at least one of:the wireless terminal device indicating data processing procedure information to the wireless access network node, wherein the wireless access network node can determine a reverse data processing procedure based on the data processing procedure information; orthe wireless terminal device receiving an indication of data processing procedure information from the wireless access network node, wherein the wireless terminal device determines a data processing procedure based on the data processing procedure information,wherein the data processing procedure information is transmitted via high-layer or PHY signaling via an uplink or a downlink channel.4.The method according to claim 3,wherein the data processing procedure information applies to at least one of:spatial domain processing;frequency domain processing;time domain processing; and / ormodel output contents.5.The method according to one of claims 1-4, comprising:receiving, from the wireless access network node, CSI-RS;deriving a first set of channel metrics based on measurement of the CSI-RS, wherein the first set of channel metrics comprises M channel metrics, each of which comprises K components, wherein each component comprises L units, wherein M, K, and L are positive integer numbers; andderiving a second set of channel metrics using a data processing procedure for any of the M channel metrics and / or K components and / or L units.6.The method according to claim 5,wherein the wireless access network node can derive the set of channel metrics based on the set of CSI by performing the reverse data processing procedure to derive a reconstructed set of channel metrics,wherein the reconstructed set of channel metric comprises M’ channel metrics, each of which comprises K’ components, wherein each component comprises L’ units, wherein M’ , K’ , and L’ are positive integer number, and wherein each L’ is larger than or equal to each L, respectively, and wherein each M’ and K’ are equal to each M and K, respectively.7.The method according to one of claims 5-6,wherein the data processing procedure comprises a padding operation,wherein a largest number of input units supported by the processing module is L_max, wherein L is less than L_max, andwherein the padding operation comprises padding L_max -L units of padded contents in specific components in the second set of channel metrics in order to derive the set of CSI using the processing module.8.The method according to claim 7,wherein the padded contents comprise at least one of the following:specific constant values of 0 or 1;a constant value derived from a formula or rule;a copy of an existing channel metric for L_max -L instances;a copy of consecutive L_max -L channel metrics in corresponding padding indices when L >= L_max / 2, wherein the consecutive L_max -L channel metrics may be first consecutive L_max –L channel metrics from an index#1, or last consecutive L_max –L channel metrics from index# (L_max –L+1) ;a copy of consecutive L channel metrics with additional padding contents in corresponding padding indices when L <= L_max / 2, wherein the additional padding comprises zero padding an additional copy of the channel metrics for units of L_max-2*L; ornew channel metrics derived from the existing channel metrics.9.The method according to one of claims 7-8,wherein a pattern or index of padding comprises at least one of the following:contents are padded at indexes of the first L_max -L units;contents are padded at indexes of the last L_max -L units;contents are padded at X units between two units of channel metrics with a same interval when L_max can be divided by L, wherein X is a positive integer;contents are padded at X units between the two units of channel metrics with a different interval when L_max cannot be divided by L;contents are padded at an index range of Y consecutive units, where Y is a positive integer and Y can be different values for different padding index ranges; orcontents are padded at a pattern of indices understood by both the wireless terminal device and the wireless access network node.10.The method according to one of claims 1-9,wherein the data processing procedure comprises a grouping operation,wherein a number of input units supported by the processing module is L_in, wherein L is larger than L_in, andwherein L units of contents in components in the set of channel metrics are divided intogroups to derive the set of CSI by the processing module,wherein the grouping operation complies with at least one of the following:consecutive L_in units of channel metrics are divided into a set or group;L_in channel metrics with a same time interval are divided into a set or group;orL_in channel metrics in a set or group consist of consecutive L_in / 2 channel metrics in a first half and the consecutive L_in / 2 channel metrics counted backward in a second half.11.The method according to one of claims 1-10, comprising:inputting at least one of the first set of channel metrics and / or the second set of channel metrics into the processing module to derive P sets of CSI, wherein P is a positive integer larger than 1; andtransmitting, to the wireless access network node, a set of the P sets of CSI.12.The method according to one of claims 1-11,wherein the data processing procedure comprises a masking operation,wherein a largest number of output units supported by the processing module is Q_max, wherein Q is a number of output units for a particular operation of the processing module, and is less than Q_max, andwherein the masking operation comprises masking Q_max -Q units of contents in order to derive the set of CSI using the processing module.13.The method according to one of claims 12,wherein a pattern or index of masking comprises at least one of the following:contents are masked at indexes of the first Q_max -Q units;contents are masked at indexes of the last Q_max -Q units;contents are masked at X units between two units of channel metrics with a same interval when Q_max can be divided by Q, wherein X is a positive integer;contents are masked at X units between the two units of channel metrics with a different interval when Q_max cannot be divided by Q;contents are masked at an index range of Y consecutive units, where Y is a positive integer and Y can be different values for different masking index ranges; orcontents are masked at a pattern of indices understood by both the wireless terminal device and the wireless access network node.14.The method according to one of claims 1-13,wherein the data processing procedure comprises a truncation operation,wherein a largest number of output elements supported by the processing module is B_max, wherein B is a number of output elements for a particular operation of the processing module, and is less than B_max, andwherein the truncation operation comprises truncating B_max -B elements of contents in order to derive the set of CSI using the processing module.15.The method according to one of claims 1-14,wherein the data processing procedure comprises a switch operation,wherein the processing module supports S branches of output units, wherein each branch comprises O_i units, wherein S and O_i are positive integers, and wherein i = 1, 2, …S, and wherein O_i is a different number for each branch, andwherein the switch operation comprises switching a branch of output units to derive the set of CSI using the processing module.16.The method according to one of claims 1-15, comprising:using the data processing procedure information by the wireless terminal device to derive the second set of channel metrics,wherein the data processing procedure information comprises at least one of:information of a pattern or index of padding, comprising at least one of:a starting index,an ending index,a padding number, and / ora type of indication information on padding pattern, including at least one of a binary information, a bitmap, or an index of patterns;orinformation of padded contents, comprising:a type of indication information on padding contents, including at least one of a binary information, a bitmap, or an index of padded contents.17.The method according to one of claims 1-16, comprising:using the data processing procedure information by the wireless terminal device to derive the second set of channel metrics,wherein the data processing procedure information comprises at least one of:information of a grouping pattern comprising at least one of:a group starting index,a group ending index,a number of channel metrics in one group, and / ora type of indication information, including at least one of a binary information, a bitmap, or an index of patterns;orinformation of grouping methods, comprising:a type of indication information on grouping methods, including at least one of a binary information, a bitmap, or an index of patterns;orinformation of a padding pattern including at least one of:an index of a padding group,an index for padding channel metric in the group,an index for padding channel metric in total number,a starting index for padding,an ending index for padding, ora padding number.18.The method according to one of claims 1-17, comprising:using the data processing procedure information by the wireless terminal device to derive the set of CSI,wherein the data processing procedure information comprises at least one of:information of a pattern or index of masking, comprising at least one of:a starting index of masking,an ending index of masking,a number of masked units, and / ora type of indication information on masking, including at least one of a binary information, a bitmap, or an index of patterns;orinformation of masking methods, comprising:a type of indication information on masking methods, including at least one of a binary information, a bitmap, or an index of masking contents.19.The method according to one of claims 1-18, comprising:indicating, to the wireless access network node, an encoder model structure, wherein the encoder model structure includes an indication of a hyper-parameter, wherein the indication of the hyper-parameter includes a value of the hyper-parameter or a model ID or a model structure ID.20.The method according to one of claims 1-19, comprising at least one of:receiving, from the wireless access network node, a complete set of parameters for a structure of the processing module with the data processing procedure information to derive the set of CSI, wherein the data processing procedure comprises at least one of masking, deactivating, freezing, or setting to zero or one values; orreceiving, from the wireless access network node, a relevant partial set of parameters for the structure of the processing module to derive the set of CSI.21.The method according to claim 20,wherein the whole set of parameters comprises at least one of:a starting index of activated or deactivated units of parameters,an ending index of activated or deactivated units of parameters,a number of activated or deactivated units of parameters including at least parameters for corresponding operation layers, ora type of indication information including at least one of a binary information, a bitmap, or an index of patterns;orwherein the partial set of parameters comprises at least one of:configuration related information,identification information,additional information, ora type of indication information including at least one of a binary information, a bitmap, or an index of patterns.22.A method performed by a wireless access network node comprising:receiving a set of Channel State Information (CSI) from a wireless terminal device; andutilizing a processing module to derive a set of Channel Metrics from the CSI, wherein the processing module supports N input and / or output configurations, wherein N is an integer larger than 1.23.The method according to claim 22,wherein the processing module comprises an Artificial Intelligence (AI) model.24.The method according to one of claims 22-23, comprising:at least one of:the wireless terminal device indicating data processing procedure information to the wireless access network node, wherein the wireless access network node determines a reverse data processing procedure based on the data processing procedure information; orthe wireless terminal device receiving an indication of data processing procedure information from the wireless access network node, wherein the wireless terminal device can determine a data processing procedure based on the data processing procedure information,wherein the data processing procedure information is transmitted via high-layer or PHY signaling via an uplink or a downlink channel.25.The method according to claim 24,wherein the data processing procedure information applies to at least one of:spatial domain processing;frequency domain processing;time domain processing; and / ormodel output contents.26.The method according to one of claims 22-25, comprising:transmitting, to the wireless terminal device, CSI-RS,wherein the wireless terminal device derives a first set of channel metrics based on measurement of the CSI-RS, wherein the first set of channel metrics comprises M channel metrics, each of which comprises K components, wherein each component comprises L units, wherein M, K, and L are positive integer numbers, andwherein the wireless terminal device derives a second set of channel metrics using a data processing procedure for any of the M channel metrics and / or K components and / or L units.27.The method according to claim 26, comprising:deriving the set of channel metrics based on the set of CSI by performing the reverse data processing procedure to derive a reconstructed set of channel metrics,wherein the reconstructed set of channel metric comprises M’ channel metrics, each of which comprises K’ components, wherein each component comprises L’ units, wherein M’ , K’ , and L’ are positive integer number, and wherein each L’ is larger than or equal to each L, respectively, and wherein each M’ and K’ are equal to each M and K, respectively.28.The method according to one of claims 26-27,wherein the data processing procedure comprises a padding operation,wherein a largest number of input units supported by the processing module is L_max, wherein L is less than L_max, andwherein the padding operation comprises the wireless terminal device padding L_max -L units of padded contents in specific components in the second set of channel metrics in order to derive the set of CSI using the processing module.29.The method according to claim 28,wherein the padded contents comprise at least one of the following:specific constant values of 0 or 1;a constant value derived from a formula or rule;a copy of an existing channel metric for L_max -L instances;a copy of consecutive L_max -L channel metrics in corresponding padding indices when L >= L_max / 2, wherein the consecutive L_max -L channel metrics may be first consecutive L_max –L channel metrics from an index#1, or last consecutive L_max –L channel metrics from index# (L_max –L+1) ;a copy of consecutive L channel metrics with additional padding contents in corresponding padding indices when L <= L_max / 2, wherein the additional padding comprises zero padding an additional copy of the channel metrics for units of L_max-2*L; ornew channel metrics derived from the existing channel metrics.30.The method according to one of claims 28-29,wherein a pattern or index of padding comprises at least one of the following:contents are padded at indexes of the first L_max -L units;contents are padded at indexes of the last L_max -L units;contents are padded at X units between two units of channel metrics with a same interval when L_max can be divided by L, wherein X is a positive integer;contents are padded at X units between the two units of channel metrics with a different interval when L_max cannot be divided by L;contents are padded at an index range of Y consecutive units, where Y is a positive integer and Y can be different values for different padding index ranges; orcontents are padded at a pattern of indices understood by both the wireless terminal device and the wireless access network node.31.The method according to one of claims 22-30,wherein the data processing procedure comprises a grouping operation,wherein a number of input units supported by the processing module is L_in, wherein L is larger than L_in, andwherein L units of contents in components in the set of channel metrics are divided intogroups to derive the set of CSI by the processing module at the wireless terminal device,wherein the grouping operation complies with at least one of the following:consecutive L_in units of channel metrics are divided into a set or group;L_in channel metrics with a same time interval are divided into a set or group;orL_in channel metrics in a set or group consist of consecutive L_in / 2 channel metrics in a first half and the consecutive L_in / 2 channel metrics counted backward in a second half.32.The method according to one of claims 22-31,wherein the wireless terminal device inputs at least one of the first set of channel metrics and / or the second set of channel metrics into the processing module to derive P sets of CSI, wherein P is a positive integer larger than 1, andthe method comprising receiving, from the wireless terminal device, a set of the P sets of CSI.33.The method according to one of claims 22-32,wherein the data processing procedure comprises a masking operation,wherein a largest number of output units supported by the processing module is Q_max, wherein Q is a number of output units for a particular operation of the processing module, and is less than Q_max, andwherein the masking operation comprises the wireless terminal device masking Q_max -Q units of contents in order to derive the set of CSI using the processing module.34.The method according to one of claims 33,wherein a pattern or index of masking comprises at least one of the following:contents are masked at indexes of the first Q_max -Q units;contents are masked at indexes of the last Q_max -Q units;contents are masked at X units between two units of channel metrics with a same interval when Q_max can be divided by Q, wherein X is a positive integer;contents are masked at X units between the two units of channel metrics with a different interval when Q_max cannot be divided by Q;contents are masked at an index range of Y consecutive units, where Y is a positive integer and Y can be different values for different masking index ranges; orcontents are masked at a pattern of indices understood by both the wireless terminal device and the wireless access network node.35.The method according to one of claims 22-34,wherein the data processing procedure comprises a truncation operation,wherein a largest number of output elements supported by the processing module is B_max, wherein B is a number of output elements for a particular operation of the processing module, and is less than B_max, andwherein the truncation operation comprises the wireless terminal device truncating B_max -B elements of contents in order to derive the set of CSI using the processing module.36.The method according to one of claims 22-35,wherein the data processing procedure comprises a switch operation,wherein the processing module supports S branches of output units, wherein each branch comprises O_i units, wherein S and O_i are positive integers, and wherein i = 1, 2, …S, and wherein O_i is a different number for each branch, andwherein the switch operation comprises switching a branch of output units to derive the set of CSI using the processing module.37.The method according to one of claims 22-36,wherein the data processing procedure information is used by the wireless terminal device to derive the second set of channel metrics,wherein the data processing procedure information comprises at least one of:information of a pattern or index of padding, comprising at least one of:a starting index,an ending index,a padding number, and / ora type of indication information on padding pattern, including at least one of a binary information, a bitmap, or an index of patterns;orinformation of padded contents, comprising:a type of indication information on padding contents, including at least one of a binary information, a bitmap, or an index of padded contents.38.The method according to one of claims 22-37,wherein the data processing procedure information is used by the wireless terminal device to derive the second set of channel metrics,wherein the data processing procedure information comprises at least one of:information of a grouping pattern comprising at least one of:a group starting index,a group ending index,a number of channel metrics in one group, and / ora type of indication information, including at least one of a binary information, a bitmap, or an index of patterns;orinformation of grouping methods, comprising:a type of indication information on grouping methods, including at least one of a binary information, a bitmap, or an index of patterns;orinformation of a padding pattern including at least one of:an index of a padding group,an index for padding channel metric in the group,an index for padding channel metric in total number,a starting index for padding,an ending index for padding, ora padding number.39.The method according to one of claims 22-38,wherein the data processing procedure information is used by the wireless terminal device to derive the set of CSI,wherein the data processing procedure information comprises at least one of:information of a pattern or index of masking, comprising at least one of:a starting index of masking,an ending index of masking,a number of masked units, and / ora type of indication information on masking, including at least one of a binary information, a bitmap, or an index of patterns;orinformation of masking methods, comprising:a type of indication information on masking methods, including at least one of a binary information, a bitmap, or an index of masking contents.40.The method according to one of claims 22-39, comprising:receiving, from the wireless terminal device, an indication of an encoder model structure, wherein the encoder model structure includes an indication of a hyper-parameter, wherein the indication of the hyper-parameter includes a value of the hyper-parameter or a model ID or a model structure ID.41.The method according to one of claims 22-40, comprising at least one of:transmitting, to the wireless terminal device, a complete set of parameters for a structure of the processing module with the data processing procedure information to derive the set of CSI, wherein the data processing procedure comprises at least one of masking, deactivating, freezing, or setting to zero or one values; ortransmitting, to the wireless terminal device, a relevant partial set of parameters for the structure of the processing module to derive the set of CSI.42.The method according to claim 41,wherein the whole set of parameters comprises at least one of:a starting index of activated or deactivated units of parameters,an ending index of activated or deactivated units of parameters,a number of activated or deactivated units of parameters including at leastparameters for corresponding operation layers, ora type of indication information including at least one of a binary information, a bitmap, or an index of patterns;orwherein the partial set of parameters comprises at least one of:configuration related information,identification information,additional information, ora type of indication information including at least one of a binary information, a bitmap, or an index of patterns.43.An apparatus for wireless communication comprising a processor that is configured to carry out the method of any of claims 1 to 42.44.A non-transitory computer readable medium having code stored thereon, the code when executed by a processor, causing the processor to implement the method recited in any of claims 1 to 42.