Channel state information prediction method and wireless communication device

Abstract

Provided in the embodiments of the present application are a channel state information (CSI) prediction method and a wireless communication device. The CSI prediction method is executed at a user equipment, and comprises: receiving measurement configuration information sent by a base station, wherein the measurement configuration information comprises at least one of the following pieces of information: measurement reference signal configuration information and CSI reporting configuration information; receiving a measurement reference signal sent by the base station, acquiring predicted CSI on the basis of the measurement reference signal, and performing compressed reporting on the basis of the predicted CSI; and reporting a monitoring result to the base station, wherein the monitoring result is configured to indicate a performance monitoring result of at least one piece of predicted CSI relative to actual measured CSI.

Description

Channel State Information Prediction Method and Wireless Communication Equipment Technical Field

[0001] This application relates to the field of mobile communication technology, specifically to a channel state information prediction method and a wireless communication device. Background Technology

[0002] In wireless communication systems, the measurement of Channel State Information (CSI) is crucial for optimizing Multiple-Input Multiple-Output (MIMO) systems. Time Division Duplex (TDD) systems can derive downlink CSI using the reciprocity of uplink and downlink channels, while Frequency Division Duplex (FDD) systems and some TDD scenarios rely on CSI feedback from user equipment to ensure the accuracy of precoding. As wireless networks evolve towards highly mobile scenarios, traditional CSI estimation methods face challenges due to rapid channel changes. Therefore, there is an urgent need to improve CSI prediction methods to enhance the accuracy of channel state information and system transmission efficiency. Summary of the Invention

[0003] This application provides a channel state information prediction method and a wireless communication device.

[0004] This application provides a Channel State Information (CSI) prediction method, executed in a user equipment, comprising: receiving measurement configuration information sent by a base station, the measurement configuration information including at least one of the following: measurement reference signal configuration information and CSI reporting configuration information; receiving a measurement reference signal sent by the base station; obtaining a predicted CSI based on the measurement reference signal; compressing and reporting the predicted CSI; and reporting a monitoring result to the base station, the monitoring result being used to indicate the performance monitoring result of at least one predicted CSI relative to the actual measured CSI.

[0005] By introducing the above technical solutions and CSI prediction and performance monitoring mechanisms, the accuracy and timeliness of CSI reporting are improved, feedback overhead is reduced, and channel adaptability in high-mobility scenarios is enhanced, thereby optimizing the transmission efficiency of wireless communication systems.

[0006] This application provides a CSI prediction method, executed at a base station, comprising: sending measurement configuration information to a user equipment, the measurement configuration information including at least one of the following: measurement reference signal configuration information and CSI reporting configuration information; sending a measurement reference signal to the user equipment, the measurement reference signal being used to obtain the predicted CSI; and receiving monitoring results reported by the user equipment, the monitoring results being used to indicate the performance monitoring results of at least one predicted CSI relative to the actual measured CSI.

[0007] By introducing the above technical solutions and CSI prediction and performance monitoring mechanisms, the accuracy and timeliness of CSI reporting are improved, feedback overhead is reduced, and channel adaptability in high-mobility scenarios is enhanced, thereby optimizing the transmission efficiency of wireless communication systems.

[0008] This application provides a wireless communication device, including a processor and a memory. The memory is used to store computer programs, and the processor is used to call and run the computer programs stored in the memory to perform the methods described above.

[0009] This application provides a user equipment including a processor and a memory. The memory stores computer programs, and the processor calls and runs the computer programs stored in the memory to perform the methods described above.

[0010] This application provides a base station including a processor and a memory. The memory stores computer programs, and the processor calls and runs the computer programs stored in the memory to perform the methods described above.

[0011] This application provides a network element including a processor and a memory. The memory stores computer programs, and the processor calls and runs the computer programs stored in the memory to perform the methods described above.

[0012] This application provides a chip for implementing the above-described method.

[0013] Specifically, the chip includes a processor for calling and running a computer program from a memory, causing a device on which the chip is installed to perform the methods described above.

[0014] This application provides a computer-readable storage medium for storing a computer program that causes a computer to perform the above-described method.

[0015] This application provides a computer program product including computer program instructions that cause a computer to execute the above-described method.

[0016] This application provides a computer program that, when run on a computer, causes the computer to perform the above-described method. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0018] Figure 1 is a schematic diagram of a general artificial intelligence (AI) / machine learning (ML) architecture in a wireless communication system;

[0019] Figure 2 is a schematic diagram of a wireless communication system architecture provided in an embodiment of this application;

[0020] Figure 3A is a flowchart illustrating a Channel State Information (CSI) prediction method provided in an embodiment of this application.

[0021] Figure 3B is a flowchart illustrating a CSI prediction method provided in an embodiment of this application;

[0022] Figure 3C is a flowchart illustrating a CSI prediction method provided in an embodiment of this application;

[0023] Figure 3D is a flowchart illustrating a CSI prediction method provided in an embodiment of this application;

[0024] Figure 4 is a schematic diagram showing the difference in the distribution of CSI-RS resources within the measurement window and prediction window in an embodiment of this application;

[0025] Figure 5A is a schematic diagram of the counting method for the monitoring results of predicted CSI within the monitoring window provided in the embodiment of this application;

[0026] Figure 5B is a schematic diagram of the counting method for the monitoring results of predicted CSI within the monitoring window provided in the embodiment of this application;

[0027] Figure 5C is a schematic diagram of the counting method for the monitoring results of predicted CSI within the monitoring window provided in the embodiment of this application;

[0028] Figure 6 is a schematic structural diagram of a wireless communication device provided in an embodiment of this application;

[0029] Figure 7 is a schematic structural diagram of a chip according to an embodiment of this application;

[0030] Figure 8 is a schematic block diagram of a wireless communication system provided in an embodiment of this application. Detailed Implementation

[0031] The technical solutions of the embodiments of this application will now be described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0032] In Multiple-Input Multiple-Output (MIMO) systems, whether Time Division Duplex (TDD) or Frequency Division Duplex (FDD), Channel State Information (CSI) measurement is crucial. For TDD systems, the network can obtain the uplink channel through uplink measurement signals and deduce the downlink channel based on the reciprocity of the uplink and downlink channels, thereby calculating the precoding information for downlink data transmission. However, for FDD systems, due to the lack of reciprocity between uplink and downlink channels, or for TDD systems due to limitations in uplink measurement, the network needs to rely on user equipment (UE) feedback of CSI to determine the downlink channel precoding information. The 3rd Generation Partnership Project (3GPP) studied CSI reporting in the first standard of New Radio (NR) and has continuously evolved and optimized it in subsequent versions. In 3GPP Release 18 (Rel-18), considering the rapid changes in channel state information (CSI) of terminals in highly mobile scenarios, traditional CSI estimation methods based on historical data may be difficult to adapt to such rapid changes. Therefore, predictive models can dynamically adjust the precoding matrix and other parameters using the latest channel information, thereby avoiding suboptimal transmission caused by outdated CSI.

[0033] With the rapid development of Artificial Intelligence (AI) technology, 3GPP began researching AI-based CSI prediction technology in Rel-18. However, due to time constraints, Rel-18 was unable to complete the research. Therefore, in Rel-19, 3GPP continued to advance the research on AI-based CSI prediction technology, and finally decided at RAN#105 meeting to upgrade it from a Study Item (SI) to a Work Item (WI).

[0034] Based on the results of the Rel-18 discussion, a general AI / Machine Learning (ML) architecture for wireless communication systems is shown in Figure 1. As described in Figure 1, the management function module includes monitoring capabilities for models / functionalities. In Rel-18, the main functions considered for implementation via AI / ML include CSI prediction and compression, beam management, and localization. Currently, regarding the application of AI / ML in these functions, the standard has reached the following consensus on monitoring AI / ML models / functionalities: the purpose of researching AI / ML model monitoring should include at least the following aspects: model activation, deactivation, selection, switching, rollback, and updating (including retraining).

[0035] Furthermore, based on the Rel-19 research, the following conclusions have been reached. For example, regarding CSI prediction models on the User Equipment (UE) side, the traditional Type II-Doppler-r18 feedback mechanism can serve as a starting point for discussion. And regarding model monitoring, three candidate schemes—Type 1, Type 2, and Type 3—have been proposed.

[0036] Type 1: The UE calculates performance standards and reports performance monitoring results, providing the network side with a decision on whether to revert to the traditional CSI prediction algorithm.

[0037] Type 2: The UE reports the predicted CSI and / or the corresponding ground truth to the network side, which calculates the performance standard and decides whether to fall back to the traditional CSI prediction algorithm.

[0038] Type 3: The UE calculates the performance standard and reports it directly to the network side. Based on the data reported by the UE, the network side decides whether to fall back to the traditional CSI prediction algorithm.

[0039] Rel 16 eTypeII codebook: The existing Rel 16 eTypeII codebook uses a three-level codebook architecture. Where W1∈C P×2L Describes the spatial basis matrix. These are the projection coefficients obtained by projecting the precoding matrix onto the basis matrices in the spatial and frequency domains. It is the frequency domain basis matrix, P is the number of antenna ports, L represents the number of spatial basis vectors selected for a single polarization direction, and M is the frequency domain basis matrix. vNv represents the number of frequency domain basis vectors corresponding to layer v, and N3 represents the number of precoding matrix indicator (PMI) subbands. In the above codebook architecture, the dimensional information of different matrices is partly or entirely indicated by the base station to the user equipment, for example, L, M... v The base station indicates the number of user equipment spatial frequency domain bases and the control factor for the number of non-zero coefficients reported by the user equipment in the form of codebook parameter combinations.

[0040] in, R represents the number of PMI subbands included in the Channel Quality Indicator (CQI) subband; β is used to control the number of reported maximum non-zero coefficients. For example, the number of maximum non-zero coefficients reported by the user equipment at Layer 1 can be expressed as... Furthermore, the positions of the non-zero coefficients reported in W2 are indicated by a bitmap, which has a length of 2LM. v Furthermore, for the Rel-16 eTypeII codebook, the spatial-frequency domain basis matrices are selected based on sets of orthogonal Discrete Fourier Transform (DFT) vectors. For example, the spatial basis matrix W1 is a block diagonal matrix. The latitude of w is w∈C P / 2×L The L column vectors of matrix w are selected from the set of orthogonal DFT vectors with dimension P / 2; and W f M of the matrix v The column vectors are selected from the set of orthogonal DFT vectors with dimension N3.

[0041] The eTypeIIPMI prediction codebook uses a three-level codebook architecture, represented as follows: Where W1∈C P×2L Describes the spatial basis matrix. These are the projection coefficients obtained by projecting the precoding matrix onto the basis matrices in the spatial and frequency domains. It is the frequency domain basis matrix, P is the number of antenna ports, L represents the number of spatial basis vectors selected for a single polarization direction, and M is the frequency domain basis matrix. v Nv represents the number of frequency domain basis vectors corresponding to the v-th layer, and N3 represents the number of PMI subbands. In the above codebook architecture, the dimensional information of different matrices is partially or completely indicated to the user equipment by the base station, for example, L, M v The base station uses codebook parameter combinations to indicate the number of spatial and frequency domain bases of the user equipment, as well as the control factor for the number of non-zero coefficients reported by the user equipment.

[0042] in, R represents the number of PMI subbands included in the CQI subband; β is used to control the number of maximum non-zero coefficients reported. For example, the number of maximum non-zero coefficients reported by the user equipment at layer 1 can be expressed as... Furthermore, in matrix W2, the positions of the reported non-zero coefficients are indicated by a bitmap with a length of 2LM. v Furthermore, for the Rel-16 eTypeII codebook, the spatial frequency domain basis matrix is ​​selected based on a set of orthogonal DFT vectors. For example, the spatial basis matrix W1 is a block diagonal matrix, represented as: The latitude of w is w∈C P / 2×L The L column vectors of matrix w are selected from the set of orthogonal DFT vectors with dimension P / 2; and W f M of the matrix v The column vectors are selected from the set of orthogonal DFT vectors with dimension N3.

[0043] For the eTypeIIPMI prediction codebook, a parameter N4 is introduced, where N4 represents the dimension of the time-domain channel, N4∈{1,2,4,8}. The dimension after time-domain compression is Q.

[0044] Rel-18 MIMO CSI prediction is designed to support various Channel State Information Reference Signal (CSI-RS) configurations, including periodic, aperiodic, and semi-persistent CSI-RS. For periodic and semi-persistent configurations, each CSI-RS resource set contains only one resource. However, in the aperiodic configuration, an offset parameter m is introduced to represent the temporal interval between two CSI-RS resources. This parameter m can be set to 1 or 2, and each CSI-RS resource set contains multiple resources.

[0045] When using Rel-18 MIMO CSI prediction, the reported PMI is associated with N4 consecutive time slot intervals. The value of N4 can be 1, 2, 4, or 8. For aperiodic CSI-RS, the number of time slots (d) contained in each time slot interval is d = 1 or m. For periodic or semi-persistent CSI-RS, the number of time slots in each time slot interval corresponds to the period of the CSI-RS resource. The number of P / SP-CSI-RS instances that the UE needs to measure in each report is implementation-specific. Among the N4 time slot intervals, the earliest time slot interval starts from time slot l = n + δ, where n represents the uplink time slot for the CSI report, and δ belongs to the set δ ∈ {-n}. CSI_ref ,0,1,2}.

[0046] When N4 > 1, Rel-18 MIMO CSI prediction allows for the configuration of multiple CQIs. Here, X represents the number of CQIs, which can be 1 or 2. If X = 2, the two CQIs are calculated independently and do not affect each other.

[0047] Rel-18 MIMO CSI prediction employs a new eType II codebook for PMI feedback. This new codebook is a time-domain extension of the Rel-16 eType II codebook. It uses Q time-domain basis vectors, with Q fixed at 2.

[0048] CSI reporting priority calculation formula: Pri iCSI (y,k,c,s)=2·N cells ·M s ·y+N cells ·M s ·k+M s ·c+s.

[0049] The smaller the calculated parameter, the higher the priority. Other parameters are as follows:

[0050] 1) A-CSI: y=0; SP-CSI on PUSCH: y=1; SP-CSI on PUCCH: y=2; P-CSI: y=3.

[0051] 2) Carrying L1-RSRP or L1-SINR: k = 0; other cases: k = 1.

[0052] 3) c: the index of the cell; N_{cells}: the number of cells, configured through higher-level parameters maxNrofServingCells.

[0053] 4)s:reportConfigID;M_{s}:maxNrofCSI-ReportConfigurations.

[0054] The above research provides technical direction for monitoring AI / ML models for CSI prediction and promotes the standardization process.

[0055] Based on the above background, some embodiments of this application further consider a series of related issues such as model data collection and function-based model monitoring, building upon the research presented above. Some embodiments of this application mainly target AI-based CSI prediction technology in MIMO scenarios, involving data collection, model monitoring, lifecycle management (LCM), and related content.

[0056] In some embodiments of this application, the predicted CSI mentioned can refer to the predicted real channel or the predicted precoding matrix information. The real channel and precoding matrix information can be predicted at the resource block (RB) granularity or at the sub-band granularity.

[0057] Furthermore, in some embodiments of this application, the monitoring results primarily consider the performance monitoring result of at least one predicted CSI relative to the actual measured CSI. The performance monitoring result can be evaluated, for example, by the Normalized Mean Square Error (NMSE), Squared Generalized Cosine Similarity (SGCS), or Generalized Cosine Similarity (GCS) between the actual measured CSI and the predicted CSI. However, for the reporting scheme of the monitored content, other performance metrics are also applicable.

[0058] Furthermore, all steps in the air interface interaction flowcharts involved in some embodiments of this application are not mandatory, nor do they represent all processes of interaction between the network and user equipment. Rather, they are possible air interface interaction processes proposed based on some embodiments of this application. In addition, the order of different signaling is not fixed, that is, there are no strict timing constraints between different signaling.

[0059] In some embodiments of this application, the length and position of the measurement window can be represented by the following two parameters: the number of measurement channels in the time domain, and the time domain interval between adjacent measurement channels. For example, the time domain interval between adjacent measurement channels can be at least one of 1 to 50 time slots.

[0060] In some embodiments of this application, the length and position of the prediction window can be represented by the following three parameters: the number of predicted CSIs contained within the prediction window, the time interval between adjacent predicted CSIs, and the time interval between the last measured CSI in the measurement window and the first predicted CSI in the prediction window. These time intervals can take any value from 1 to 50 time slots.

[0061] The prediction window can be defined using the traditional Rel-18 Doppler codebook, containing CSIs for N4 consecutive time slot intervals, each lasting d time slots. The earliest CSI in the N4 consecutive time slot intervals is located in time slot l = n + δ, where n is the uplink time slot reported by the CSI, and the time slot offset δ ranges from...

[0062] In summary, some embodiments of this application are applicable to the above-mentioned scenarios and to NR and future communication systems, such as 6G and 7G; at the same time, they can also be applied to WiFi or other similar wireless communication systems.

[0063] Some embodiments of this application include several aspects. Dedicated Scheduling Request (SR) or Uplink Control Information (UCI) message types are introduced to trigger data collection by the AI ​​model for CSI prediction at different Lifecycle Management (LCM) stages, thereby improving the effectiveness of the AI ​​prediction model. Furthermore, some embodiments of this application define event types that trigger the reporting of performance monitoring results within a monitoring window. These include changes in monitoring results relative to a specific threshold, and situations where the difference between multiple adjacent monitoring results meets certain constraints, ensuring that the monitoring information accurately reflects the effectiveness of CSI prediction.

[0064] Furthermore, for different event types, user equipment may need to report the event's status, event type, and corresponding performance monitoring results within the monitoring window, so that the network side can better adjust the CSI prediction mechanism. Meanwhile, some embodiments of this application propose a network-triggered AI prediction CSI rollback mechanism. That is, when the network side triggers a traditional CSI prediction reporting message, it can explicitly or implicitly instruct the user equipment to perform a rollback operation for the AI ​​prediction CSI, thereby ensuring the system's flexibility and adaptability.

[0065] Furthermore, some embodiments of this application specify the reporting priority of the AI ​​prediction model under different LCM states, and define the processing time and CSI processing unit occupancy information when AI CSI prediction and traditional CSI reporting are jointly processed, ensuring the reasonable allocation of computing resources and improving the overall system performance. This solution not only improves the accuracy and timeliness of CSI prediction, but also optimizes network scheduling strategies, making it suitable for 5G NR, future 6G, 7G, WiFi, and other wireless communication systems.

[0066] Some embodiments of this application enable user equipment to support CSI prediction using AI-based user equipment-side models, thereby improving the accuracy of CSI prediction and increasing system capacity. Meanwhile, some embodiments of this application achieve the following technical effects:

[0067] Reduce the configuration overhead of channel measurement resources. Some embodiments of this application optimize the configuration of CSI-RS measurement resources so that not every predicted CSI within the prediction window requires corresponding measurement resources, thereby reducing unnecessary resource allocation and lowering system overhead.

[0068] Furthermore, the feedback overhead on the user equipment is reduced. Some embodiments of this application optimize the reporting content during performance monitoring. For example, only the performance index between a predicted CSI and the actual measured CSI within the monitoring window is reported, rather than the complete CSI data. Additionally, some embodiments of this application introduce an event-triggered reporting mechanism, performing CSI feedback only when specific conditions are met, thereby reducing the feedback overhead on the UE side.

[0069] Furthermore, the overhead of control signaling is reduced. Some embodiments of this application reduce the signaling burden on the network side by explicitly or implicitly instructing the function fallback of CSI prediction. For example, monitoring reporting can be activated by Medium Access Control-Control Element (MAC CE) or Downlink Control Information (DCI) messages, carrying corresponding indication information; or when a message triggering traditional Rel-18 Doppler codebook reporting is received, it can be considered that a fallback from AI-based CSI prediction to traditional Rel-18 Doppler codebook is required, thereby reducing additional signaling overhead.

[0070] Furthermore, the robustness of the system is enhanced. Some embodiments of this application monitor the AI-based CSI prediction results to ensure the accuracy of the predicted CSI. If the monitoring results are unsatisfactory, the system can quickly revert to the traditional Rel-18 Doppler codebook, thereby ensuring the reliability of CSI acquisition while maintaining system capacity. In addition, some embodiments of this application also set certain priorities for the content reported by user devices, ensuring that high-priority content (such as monitoring results and event-triggered reports) is reported first when resources are limited, while the reporting priority of predicted CSI data is relatively low. This priority management mechanism helps to further improve the robustness of the system and ensure the timeliness and reliability of data in constrained environments.

[0071] The technical solutions of this application can be applied to various wireless communication systems, such as: Long Term Evolution (LTE) systems, LTE Frequency Division Duplex (FDD) systems, LTE Time Division Duplex (TDD) systems, 5G communication systems, or future wireless communication systems.

[0072] For example, the wireless communication system 100 used in this application embodiment is shown in FIG2. The wireless communication system 100 may include a network-side device 110, which may be a device communicating with a user equipment (UE) 120. The network-side device 110 can provide communication coverage for a specific geographical area and can communicate with user equipment located within that coverage area. Optionally, the network-side device 110 may be a base station or a Location Management Function (LMF) for providing location services. Optionally, the base station may be an evolved Node B (eNB or eNodeB) in an LTE system, or the base station may be a mobile switching center, relay station, access point, vehicle-mounted equipment, wearable device, hub, switch, bridge, router, network equipment in a 5G network, or a base station in a future communication system, etc.

[0073] The wireless communication system 100 also includes at least one user equipment 120 located within the coverage area of ​​the network-side device 110. As used herein, "user equipment" includes, but is not limited to, devices configured to receive / transmit communication signals via wired connections, such as via Public Switched Telephone Networks (PSTN), Digital Subscriber Line (DSL), digital cable, direct cable connection; and / or another data connection / network; and / or via a wireless interface, such as for cellular networks, Wireless Local Area Networks (WLAN), digital television networks such as DVB-H networks, satellite networks, AM-FM broadcast transmitters; and / or other user equipment. User equipment configured to communicate via a wireless interface may be referred to as "wireless user equipment 120," "wireless user equipment 120," or "mobile user equipment 120." Examples of mobile user equipment 120 include, but are not limited to, satellite or cellular phones; personal communications system (PCS) user equipment 120 that can combine cellular radiotelephony with data processing, fax, and data communication capabilities; PDAs that may include radiotelephones, pagers, Internet / intranet access, web browsers, notebooks, calendars, and / or Global Positioning System (GPS) receivers; and conventional laptop and / or handheld receivers or other electronic devices that include radiotelephone transceivers. User equipment can refer to access user equipment 120, user units, user stations, mobile stations, mobile stations, remote stations, remote user equipment, mobile devices, wireless communication equipment, or user agents. The access user equipment 120 can be a cellular phone, cordless phone, Session Initiation Protocol (SIP) phone, Wireless Local Loop (WLL) station, Personal Digital Assistant (PDA), handheld device with wireless communication capabilities, computing device or other processing device connected to a wireless modem, vehicle-mounted device, wearable device, user equipment in a 5G network, or user equipment in a future PLMN, etc.

[0074] This application provides a CSI prediction method executed in a user equipment 120, comprising: receiving measurement configuration information sent by a base station, the measurement configuration information including at least one of the following: measurement reference signal configuration information and CSI reporting configuration information; receiving a measurement reference signal sent by the base station; obtaining a predicted CSI based on the measurement reference signal; compressing and reporting the predicted CSI; and reporting a monitoring result to the base station, the monitoring result being used to indicate the performance monitoring result of at least one predicted CSI relative to the actual measured CSI. Through the above technical solution, a CSI prediction and performance monitoring mechanism is introduced to improve the accuracy and timeliness of CSI reporting, reduce feedback overhead, and enhance channel adaptability in high-mobility scenarios, thereby optimizing the transmission efficiency of the wireless communication system. The order, combination, specific content, and execution of the above operations do not constitute a limitation on this application embodiment and can be flexibly adjusted according to actual needs to meet different scenarios and technical requirements.

[0075] This application provides a CSI prediction method executed at a base station, comprising: sending measurement configuration information to a user equipment 120, the measurement configuration information including at least one of the following: measurement reference signal configuration information and CSI reporting configuration information; sending a measurement reference signal to the user equipment 120, the measurement reference signal being used to obtain the predicted CSI; and receiving monitoring results reported by the user equipment 120, the monitoring results being used to indicate the performance monitoring results of at least one predicted CSI relative to the actual measured CSI. Through the above technical solution, a CSI prediction and performance monitoring mechanism is introduced, improving the accuracy and timeliness of CSI reporting, reducing feedback overhead, and enhancing channel adaptability in high-mobility scenarios, thereby optimizing the transmission efficiency of the wireless communication system. The order, combination, specific content, and execution of the above operations do not constitute a limitation on this application embodiment and can be flexibly adjusted according to actual needs to meet different scenarios and technical requirements.

[0076] Optionally, "Network 130 side" can refer to a base station. "Network 130 side" can refer to the current serving cell, a candidate cell, a primary serving cell, or a secondary cell.

[0077] Optionally, user equipment 120 can perform device-to-device (D2D) communication with each other.

[0078] The wireless communication system 100 also includes a network 130. The network 130 may be an IP mobile communication network operated by a mobile communication operator. For example, the network 130 may be the core network used by the mobile communication operator that operates and manages the wireless communication system 100, or it may be the core network used by a virtual mobile communication operator such as a Mobile Virtual Network Operator (MVNO).

[0079] Network 130 can be connected to network-side device 110 as a relay device for transmitting user data. User equipment 120 sends and receives user data via network 130. It should be noted that user data communication is not limited to IP communication, but can also be non-IP communication.

[0080] Figure 2 exemplarily illustrates a network-side device 110, two user devices 120, and a network 130. Optionally, the wireless communication system 100 may include multiple network devices, and each network device may include other numbers of user devices within its coverage area. This application embodiment does not limit this.

[0081] Optionally, the wireless communication system 100 may also include other network entities such as a network controller, a mobility management entity, and network elements; this application embodiment does not limit this. For example, network 130 may include other network entities such as a network controller, a mobility management entity, and network elements; this application embodiment does not limit this.

[0082] It should be understood that devices with wireless communication functions in the network / system of this application embodiment can be referred to as wireless communication devices. Taking the wireless communication system 100 shown in FIG2 as an example, the wireless communication device may include a network-side device 110, a user equipment 120, and a network 130 with communication functions. The network-side device 110 and the user equipment 120 can be the specific devices described above, which will not be repeated here. The wireless communication device may also include other devices (network 130) in the wireless communication system 100. For example, the network 130 may include other network entities such as a network controller and a mobility management entity. This is not limited in this application embodiment.

[0083] It should be understood that the terms "system" and "network" are often used interchangeably in this document. The term "and / or" in this document merely describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects are in an "or" relationship. In some embodiments, the term "configuration" can refer to "pre-configuration" and "network configuration." The terms "definition" or "pre-defined" in the embodiments of this application can be implemented by pre-storing corresponding codes, tables, or other information indicative indices in the device (e.g., including UE and network devices). This application does not limit specific implementations. For example, "definition" or "pre-defined" can refer to those defined in a protocol. It should also be understood that "protocol" in this invention can refer to standard protocols in the field of communications, such as Long Term Evolution (LTE) protocols, new radio protocols, 5G NR, future 6G, 7G, WiFi, and other related protocols used in future communication systems. This application does not limit this.

[0084] In some embodiments, Figure 3A is a flowchart illustrating a CSI prediction method provided in this application. The CSI prediction method, executed on a user equipment, includes: Operation 301A: receiving measurement configuration information sent by a base station, the measurement configuration information including at least one of the following: measurement reference signal configuration information or CSI reporting configuration information; Operation 302A: receiving a measurement reference signal sent by the base station, obtaining a predicted CSI based on the measurement reference signal, and compressing and reporting the predicted CSI; and Operation 303A: reporting monitoring results to the base station, the monitoring results indicating the performance monitoring results of at least one predicted CSI relative to the actual measured CSI. Through the above technical solution, a CSI prediction and performance monitoring mechanism is introduced, improving the accuracy and timeliness of CSI reporting, reducing feedback overhead, and enhancing channel adaptability in high-mobility scenarios, thereby optimizing the transmission efficiency of the wireless communication system. The order, combination, specific content, and execution of operations 301A, 302A, and 303A do not constitute a limitation on the embodiments of this application and can be flexibly adjusted according to actual needs to meet different scenarios and technical requirements.

[0085] In some embodiments, Figure 3B is a flowchart illustrating a CSI prediction method provided in this application embodiment. The CSI prediction method, executed at a base station, includes: Operation 301B: sending measurement configuration information to a user equipment, the measurement configuration information including at least one of the following: measurement reference signal configuration information or CSI reporting configuration information; Operation 302B: sending a measurement reference signal to the user equipment, the measurement reference signal being used to obtain the predicted CSI; and Operation 303B: receiving monitoring results reported by the user equipment, the monitoring results being used to indicate the performance monitoring results of at least one predicted CSI relative to the actual measured CSI. Through the above technical solution, a CSI prediction and performance monitoring mechanism is introduced, improving the accuracy and timeliness of CSI reporting, reducing feedback overhead, and enhancing channel adaptability in high-mobility scenarios, thereby optimizing the transmission efficiency of the wireless communication system. The order, combination, specific content, and execution of operations 301B, 302B, and 303B do not constitute a limitation on the embodiments of this application and can be flexibly adjusted according to actual needs to meet different scenarios and technical requirements.

[0086] The operations in Figures 3A and 3B will be further explained below with reference to Figures 3C and 3D. It should be understood that the steps shown in Figures 3C and 3D are only examples, and their execution order can be deleted, added, or adjusted according to actual needs.

[0087] In some embodiments, Figure 3C is a flowchart illustrating a CSI prediction method provided by an embodiment of this application. In some embodiments, as shown in Figure 3C, AI-based CSI prediction collects data based on network-side instructions. The reporting process of AI-based CSI prediction, building upon the process of PMI prediction and CSI reporting based on the eType-II codebook in 3GPP Rel-18, further considers a series of related issues such as model data collection and function-based model monitoring. The main difference lies in whether AI is used to implement CSI prediction. However, AI-based CSI prediction also involves a series of operations such as data collection, inference, model monitoring, and LCM management.

[0088] Some embodiments of this application primarily focus on the signaling and process design related to data collection based on network-side instructions, as well as air interface signaling interaction and LCM management involved in the model monitoring process. The signaling interaction process between the base station and the UE is at least as shown in Figure 3C, and mainly includes at least one of the following steps. To make the technical solutions of the embodiments of this application more adaptable and flexible, the order, number, and whether all steps are performed are not limited to the embodiments of this application. Specifically, the AI-based CSI prediction process may include, but is not limited to, at least one of the following steps:

[0089] Step 1: Predefine the relevant rules to support AI-based CSI prediction functionality.

[0090] In some embodiments, the base station or UE determines a configuration that supports AI-based CSI prediction functionality. This configuration includes at least one of the following: a definition of performance monitoring events; reporting priority information for content reported by the AI ​​model at different LCM stages; and indication information of the CSI processing time corresponding to CSI reporting at different stages of the AI ​​model, as well as CSI processing unit occupancy information. In some embodiments, the definition of performance monitoring events includes at least one of the following: the performance monitoring result of the first predicted CSI within a monitoring window is lower than a threshold; the mean, median, and / or variance of predicted CSIs within the monitoring window are lower than the threshold; or the first predicted CSI is lower than the threshold in multiple consecutive monitoring windows. In some embodiments, the reporting priority information of the AI ​​model's reported content at different LCM stages includes at least one of the following: the reporting priority of the performance monitoring results is between the reporting of Layer 1 Reference Signal Received Power (L1-RSRP) or Layer 1 Signal-to-Interference-plus-Noise Ratio (L1-SINR) and the reporting of traditional CSI prediction or AI-based CSI prediction; the reporting priority of the performance monitoring results corresponding to CSI reporting under different time-domain behaviors is higher than that of the corresponding CSI reporting.

[0091] For example, predefined rules supporting AI-based CSI prediction functionality include at least one of the following: Definition of performance monitoring events: such as the first predicted CSI within a monitoring window falling below a threshold, the mean / median / variance falling below a threshold, or the first predicted CSI within multiple consecutive monitoring windows falling below a threshold. Priority of LCM stage reporting: The reporting priority of monitoring results is higher than or between L1-RSRP, L1-SINR reporting, and traditional / AI predicted CSI reporting; for CSI reporting under different time-domain behaviors, the monitoring results have a higher priority than the CSI itself.

[0092] Specifically, in step 1, it is first necessary to predefine the relevant rules to support the AI-based CSI prediction function, so as to ensure that the AI ​​model can effectively predict CSI and report it reasonably in different scenarios.

[0093] The definition of a performance monitoring event mainly includes at least one of the following situations: if the first predicted CSI is lower than the set threshold within a monitoring window, a monitoring event is triggered; if the mean, median, or variance of all predicted CSIs within a monitoring window is lower than the set threshold, a monitoring report is made; if the first predicted CSI is consistently lower than the threshold within multiple consecutive monitoring windows, the corresponding monitoring mechanism is triggered to ensure the stability and reliability of CSI prediction.

[0094] Meanwhile, the reporting content in the LCM phase needs to have its priority clearly defined to ensure reasonable resource allocation on the network side. Specifically, the reporting priority of monitoring results should be higher than or at least between the reporting of measurement results such as L1-RSRP and L1-SINR and the reporting of traditional / AI-predicted CSI, ensuring that key performance indicators are fed back first. In addition, under different time-domain behaviors (such as short-term or long-term CSI prediction), the monitoring results of the corresponding CSI reports can have a higher priority than the CSI report itself, so as to adjust the AI ​​model or prediction mechanism more promptly, thereby improving the overall performance of the communication system.

[0095] Step 2: User devices report the ability to support AI-based CSI prediction.

[0096] In some embodiments, the user equipment reports to the base station its ability to support AI-based CSI prediction, wherein the AI-based CSI prediction capability includes information on the number of CSIs predicted within a supported prediction window. For example, the user equipment may report information such as the supported prediction window size and the number of predicted CSIs so that the network side can perform resource allocation and scheduling.

[0097] Specifically, in step 2, the user equipment (UE) needs to report its AI-based CSI prediction capability to the base station so that the network side can allocate and schedule resources appropriately based on the device's capabilities. In some embodiments, the UE reports its supported AI-based CSI prediction capability to the base station, including information such as the size of the prediction window and the number of CSIs that can be predicted within that window. This capability reporting mechanism ensures that the AI ​​prediction model can run efficiently on different devices, while avoiding prediction errors caused by resource waste or insufficient computing power.

[0098] Step 3: The user equipment receives the measurement configuration information.

[0099] In some embodiments, the user equipment receives measurement configuration information sent by the base station, the measurement configuration information including at least one of the following: measurement reference signal configuration information or CSI reporting configuration information.

[0100] Step 4: The user equipment receives the measurement reference signal and performs CSI prediction.

[0101] In some embodiments, the user equipment receives a measurement reference signal transmitted by the base station, obtains a predicted CSI based on the measurement reference signal, and compresses and reports the predicted CSI. For example, the user equipment inputs the received measurement reference signal into an AI model for inference to obtain predicted CSI data. The predicted CSI can be compressed before reporting to reduce uplink overhead.

[0102] Specifically, in step 4, the user equipment uses the measurement reference signal input to the AI ​​model for inference calculations to obtain the predicted CSI data. Furthermore, to reduce uplink transmission overhead and improve network transmission efficiency, the user equipment can compress the predicted CSI before reporting it. This compression not only reduces the size of the reported data but also optimizes the use of wireless resources, avoiding unnecessary bandwidth consumption, thereby improving the overall performance and response speed of the communication system.

[0103] Step 5: User devices report predicted CSI.

[0104] For example, the predicted CSI data is compressed before being reported to the network side.

[0105] Step 6: The user equipment receives the monitoring instruction information.

[0106] In some embodiments, the user equipment receives monitoring indication information sent by the base station, the monitoring indication information being used to instruct the user equipment to initiate performance monitoring. For example, the monitoring indication information is used to notify the user equipment to perform monitoring of a model or function to evaluate the accuracy of the AI ​​model.

[0107] In step 6, the user equipment needs to receive monitoring instruction information to initiate performance monitoring of the AI ​​model or function. In some embodiments, the base station sends monitoring instruction information to the user equipment to notify the equipment to perform monitoring and evaluation of the AI ​​model, ensuring the accuracy and reliability of CSI predictions.

[0108] Specifically, upon receiving the monitoring instruction, the user equipment will evaluate the performance of the AI ​​model's predictions according to preset monitoring rules. This includes comparing the error between the predicted CSI and the actual measured CSI, or monitoring the adaptability of the prediction model under different network environments. Through this monitoring mechanism, the network side can monitor the AI ​​model's operational status in real time and adjust model parameters or execute rollback strategies when necessary to improve overall communication quality and system stability.

[0109] Step 7: The user equipment receives the monitoring and measurement signals.

[0110] In some embodiments, the user equipment receives monitoring and measurement signals sent by the base station, and these signals are used for CSI measurement. For example, after receiving the monitoring and measurement signals, the user equipment performs a real CSI measurement and uses it as the ground truth for AI model monitoring. The monitoring and measurement signals can be categorized into performance monitoring signals, functional monitoring signals, and model monitoring signals according to different monitoring requirements.

[0111] Performance monitoring measurement signals are primarily used to evaluate the overall accuracy and stability of CSI predictions, ensuring the adaptability of the AI ​​model in different network environments. Functional monitoring measurement signals are used to monitor whether the CSI prediction function is working properly, such as checking whether the prediction results meet expectations and whether there are any abnormal deviations. Model monitoring measurement signals focus on evaluating the operation of the AI ​​model itself, such as the model's convergence, the effect of parameter optimization, and whether updates or rollbacks are needed. By receiving these different types of monitoring measurement signals, user equipment can accurately measure and provide feedback on the performance, functionality, and model operation of CSI predictions, thereby providing more comprehensive data support for LCM on the network side.

[0112] In step 7, the user equipment needs to receive monitoring and measurement signals to perform actual CSI measurements and for monitoring and evaluation of the AI ​​model. In some embodiments, the base station sends monitoring and measurement signals to the user equipment, which instruct the user equipment to perform CSI measurements to obtain actual channel state information.

[0113] Specifically, after receiving the monitoring and measurement signal, the user equipment performs a real CSI measurement and uses the measured CSI as the ground truth value to compare and evaluate the prediction results of the AI ​​model. This mechanism can effectively verify the prediction accuracy of the AI ​​model and provide data support for LCM on the network side. If a large error is found between the AI-predicted CSI and the real CSI, the network side can adjust the parameters of the AI ​​model, optimize the CSI prediction strategy, or even revert to the traditional CSI reporting method to ensure the stability and reliability of the communication system.

[0114] Step 8: User devices report monitoring results.

[0115] In some embodiments, the user equipment reports monitoring results to the base station, the monitoring results indicating at least one performance monitoring result of the predicted CSI relative to the actual measured CSI. The actual measured CSI may be obtained based on a measurement reference signal or based on a monitoring measurement signal. Therefore, the monitoring measurement signal portion is represented by dashed lines, meaning that the situation where this signal is specifically used for monitoring may or may not exist.

[0116] In some embodiments, the performance monitoring results include NMSE results, SGCS results, and / or GCS results between the at least one predicted CSI and the actual measured CSI. In some embodiments, the reporting of the monitoring results is either non-event-triggered monitoring result reporting or event-triggered monitoring result reporting. In some embodiments, the non-event-triggered monitoring result reporting includes at least one of the following: the quantized value of the performance monitoring result within the monitoring window; a bitmap corresponding to the state of the performance monitoring result relative to a threshold value within the monitoring window; a predicted CSI index; and combination number indication information. In some embodiments, the event-triggered monitoring result reporting includes at least one of the following: the state information of the event occurrence within the monitoring window; the event index within the monitoring window and / or the quantized value of the performance monitoring result within the monitoring window; the monitoring result; and the predicted CSI. In some embodiments, the non-event-triggered monitoring results and / or the event-triggered monitoring results and the predicted CSI are reported in a single reporting instance.

[0117] Specifically, in step 8, during the process of user equipment reporting monitoring results, the monitoring results are mainly used by the network side to make decisions on the LCM of the AI ​​model or function. The reporting method for monitoring results can be periodic, semi-persistent, or aperiodic, and can also be triggered based on specific events. Specifically, the content of the monitoring report can include the following forms:

[0118] Non-event-triggered monitoring result reporting is suitable for continuously monitoring the prediction accuracy of AI models. For example, user equipment can report the quantized values ​​of NMSE, SGCS, or GCS between at least one predicted CSI and the actual measured CSI within a monitoring window. In addition, it can also report the quantized statistical results of NMSE, SGCS, or GCS between all predicted CSIs and the actual measured CSIs within the entire monitoring window, or provide bitmaps, predicted CSI indices, or combination number indications corresponding to the NMSE, SGCS, or GCS states between predicted CSIs and the actual measured CSIs based on a preset threshold, so that the network side can perform analysis and decision-making.

[0119] Furthermore, event-triggered monitoring result reporting is primarily used to identify specific anomalies or important situations. For example, user equipment can report status information on whether a specific event has occurred within the monitoring window, along with the corresponding event index. Additionally, the device can provide quantified values ​​of NMSE, SGCS, or GCS between the predicted CSI and the actual measured CSI within the monitoring window to help the network side determine if the AI ​​model's predictions have deviated. Moreover, to provide more complete monitoring data, user equipment can attach predicted CSI data along with the monitoring results, enabling the network side to perform more in-depth analysis.

[0120] Furthermore, monitoring results, whether event-triggered or not, can be reported within the same reporting instance as the predicted CSI, improving data integrity and validity. This reporting method helps the network side more comprehensively evaluate the performance of the AI ​​model and optimize CSI prediction strategies based on monitoring data, thereby improving the overall stability and efficiency of the communication system.

[0121] Step 9: The user equipment receives the LCM indication information.

[0122] In some embodiments, the user equipment receives LCM indication information sent by the base station, the LCM indication information being used to instruct the user equipment to perform a corresponding LCM operation. In some embodiments, the LCM indication information includes function rollback operation indication information. In some embodiments, the function rollback operation indication information instructs the cessation of AI-based CSI prediction reporting and / or the activation of conventional Doppler codebook CSI reporting, the function rollback operation indication information indicating the function rollback operation via MAC CE or DCI.

[0123] In some embodiments, the LCM indication information can be model or function LCM indication information. Model LCM indication information is mainly used to manage the running status of AI models, such as triggering model updates, optimizing model parameters, or reverting to traditional CSI reporting methods when prediction results are poor. Function LCM indication information focuses on the overall management of CSI prediction functions, including enabling functions, reverting to traditional CSI prediction methods, adjusting reporting strategies, or optimizing computing resource allocation. By receiving LCM indication information, user equipment can dynamically adjust CSI prediction models or functions according to instructions from the network side, ensuring that the communication system maintains efficient and stable operation in different network environments.

[0124] Specifically, in step 9, during the process of the user equipment receiving LCM indication information, the LCM indication information is mainly used to instruct the user equipment to perform the corresponding LCM operation. The corresponding LCM operation is, for example, a function rollback operation. Specifically, the signaling bearer method for the rollback operation may include at least one of the following:

[0125] In some embodiments, a 1-bit rollback indication can be added to the MAC CE or DCI instruction through the traditional Rel-18 Doppler codebook CSI reporting activation mechanism to explicitly notify the user equipment to perform a rollback operation.

[0126] In some embodiments, management can be carried out using an instruction plus predefined approach. That is, when a user equipment receives an instruction to activate traditional Rel-18 Doppler codebook CSI reporting, and the instruction also includes an operation to deactivate AI-based predictive CSI reporting, the user equipment understands that it is falling back from AI-based CSI prediction to traditional Rel-18 Doppler codebook CSI reporting, thereby ensuring that the communication system can operate stably when the prediction effect decreases or network requirements change.

[0127] Step 10: The user equipment receives downlink data information.

[0128] In some embodiments, the user equipment receives downlink data information sent by the base station.

[0129] For example, based on the CSI reported by the user equipment, the network side performs data precoding and sends downlink data to the user equipment to optimize communication quality.

[0130] The order, combination, specific content, and execution of the above steps do not limit the embodiments of this application and can be flexibly adjusted according to actual needs to meet different scenarios and technical requirements. This process ensures that AI-based CSI prediction can operate efficiently and has a complete monitoring, reporting, and adjustment activation mechanism to adapt to dynamic network environments.

[0131] In some embodiments, Figure 3D is a flowchart illustrating a CSI prediction method provided by an embodiment of this application. In some embodiments, as shown in Figure 3D, AI-based CSI prediction collects data based on user equipment requests. The reporting process of AI-based CSI prediction, building upon the process of PMI prediction and CSI reporting based on the eType-II codebook in 3GPP Rel-18, further considers a series of related issues such as model data collection and function-based model monitoring. The main difference lies in whether AI is used to implement CSI prediction. However, AI-based CSI prediction also involves a series of operations such as data collection, inference, model monitoring, and LCM management.

[0132] Some embodiments of this application primarily focus on the signaling and process design related to data collection based on user equipment requests, as well as air interface signaling interaction and LCM management involved in model monitoring. The signaling interaction process between the base station and the UE is at least as shown in Figure 3D, and mainly includes at least one of the following steps. To make the technical solutions of the embodiments of this application more adaptable and flexible, the order, number, and whether all steps are performed are not limited to the embodiments of this application. Specifically, the AI-based CSI prediction process may include, but is not limited to, at least one of the following steps:

[0133] Step 1: Predefine the relevant rules to support AI-based CSI prediction functionality.

[0134] In some embodiments, the base station or UE determines a configuration that supports AI-based CSI prediction functionality. This configuration includes at least one of the following: a definition of performance monitoring events; reporting priority information for content reported by the AI ​​model at different LCM stages; and indication information of the CSI processing time corresponding to CSI reporting at different stages of the AI ​​model, as well as CSI processing unit occupancy information. In some embodiments, the definition of performance monitoring events includes at least one of the following: the performance monitoring result of the first predicted CSI within a monitoring window is lower than a threshold; the mean, median, and / or variance of predicted CSIs within the monitoring window are lower than the threshold; or the first predicted CSI is lower than the threshold in multiple consecutive monitoring windows. In some embodiments, the reporting priority information of the AI ​​model's reported content at different LCM stages includes at least one of the following: the reporting priority of the performance monitoring results is between the reporting of Layer 1 Reference Signal Received Power (L1-RSRP) or Layer 1 Signal-to-Interference-plus-Noise Ratio (L1-SINR) and the reporting of traditional CSI prediction or AI-based CSI prediction; the reporting priority of the performance monitoring results corresponding to CSI reporting under different time-domain behaviors is higher than that of the corresponding CSI reporting.

[0135] For example, predefined rules supporting AI-based CSI prediction functionality include at least one of the following: Definition of performance monitoring events: such as the first predicted CSI within a monitoring window falling below a threshold, the mean / median / variance falling below a threshold, or the first predicted CSI within multiple consecutive monitoring windows falling below a threshold. Priority of LCM stage reporting: The reporting priority of monitoring results is higher than or between L1-RSRP, L1-SINR reporting, and traditional / AI predicted CSI reporting; for CSI reporting under different time-domain behaviors, the monitoring results have a higher priority than the CSI itself.

[0136] Specifically, in step 1, it is first necessary to predefine the relevant rules to support the AI-based CSI prediction function, so as to ensure that the AI ​​model can effectively predict CSI and report it reasonably in different scenarios.

[0137] The definition of a performance monitoring event mainly includes at least one of the following situations: if the first predicted CSI is lower than the set threshold within a monitoring window, a monitoring event is triggered; if the mean, median, or variance of all predicted CSIs within a monitoring window is lower than the set threshold, monitoring and reporting are required; if the first predicted CSI is consistently lower than the threshold within multiple consecutive monitoring windows, the corresponding monitoring mechanism is triggered to ensure the stability and reliability of CSI predictions.

[0138] Meanwhile, the reporting content in the LCM phase needs to have its priority clearly defined to ensure reasonable resource allocation on the network side. Specifically, the reporting priority of monitoring results should be higher than or at least between the reporting of measurement results such as L1-RSRP and L1-SINR and the reporting of traditional / AI-predicted CSI, ensuring that key performance indicators are fed back first. In addition, under different time-domain behaviors (such as short-term or long-term CSI prediction), the monitoring results of the corresponding CSI reports can have a higher priority than the CSI report itself, so as to adjust the AI ​​model or prediction mechanism more promptly, thereby improving the overall performance of the communication system.

[0139] Step 2: User devices report the ability to support AI-based CSI prediction.

[0140] In some embodiments, the user equipment reports to the base station its ability to support AI-based CSI prediction, wherein the AI-based CSI prediction capability includes information on the number of CSIs predicted within a supported prediction window. For example, the user equipment may report information such as the supported prediction window size and the number of predicted CSIs so that the network side can perform resource allocation and scheduling.

[0141] Specifically, in step 2, the user equipment (UE) needs to report its AI-based CSI prediction capability to the base station so that the network side can allocate and schedule resources appropriately based on the device's capabilities. In some embodiments, the UE reports its supported AI-based CSI prediction capability to the base station, including information such as the size of the prediction window and the number of CSIs that can be predicted within that window. This capability reporting mechanism ensures that the AI ​​prediction model can run efficiently on different devices, while avoiding prediction errors caused by resource waste or insufficient computing power.

[0142] Step 3: The user equipment receives the measurement configuration information.

[0143] In some embodiments, the user equipment receives measurement configuration information sent by the base station, the measurement configuration information including at least one of the following: measurement reference signal configuration information or CSI reporting configuration information.

[0144] Step 4: The user equipment receives the measurement reference signal and performs CSI prediction.

[0145] In some embodiments, the user equipment receives a measurement reference signal transmitted by the base station, obtains a predicted CSI based on the measurement reference signal, and compresses and reports the predicted CSI. For example, the user equipment inputs the received measurement reference signal into an AI model for inference to obtain predicted CSI data. The predicted CSI can be compressed before reporting to reduce uplink overhead.

[0146] Specifically, in step 4, the user equipment uses the measurement reference signal input to the AI ​​model for inference calculations to obtain the predicted CSI data. Furthermore, to reduce uplink transmission overhead and improve network transmission efficiency, the user equipment can compress the predicted CSI before reporting it. This compression not only reduces the size of the reported data but also optimizes the use of wireless resources, avoiding unnecessary bandwidth consumption, thereby improving the overall performance and response speed of the communication system.

[0147] Step 5: User devices report predicted CSI.

[0148] For example, the predicted CSI data is compressed before being reported to the network side.

[0149] Step 6: The user equipment sends a data collection request message.

[0150] In some embodiments, the user equipment sends a data collection request message to the base station. The data collection request message includes at least one of the following: a data collection request message for model training, a data collection request message for model inference, and a data collection request message for model monitoring. In some embodiments, the data collection request message carries at least one bit of indication information via at least one dedicated scheduling request (SR) or uplink control information (UCI).

[0151] For example, in step 6, during the process of the user equipment sending a data collection request message, the data collection request message is mainly used for data collection, including data collection required for model training, model inference, and model monitoring. The user equipment can send the corresponding request to the network side through different air interface signaling methods, specifically including at least one of the following methods:

[0152] In some embodiments, a user equipment can individually request data collection for model training, inference, or monitoring from the network side via multiple dedicated SRs or dedicated UCIs carrying 1 bit of indication information. This approach allows user equipment to request different types of data more flexibly.

[0153] In some embodiments, the user equipment can also send a merged data collection request to the network side by carrying 2 bits of information in a dedicated SR or dedicated UCI. For example, 00 indicates model training, 01 indicates model inference, and 10 indicates model monitoring. This reduces signaling overhead and improves the utilization efficiency of uplink resources. This data collection request mechanism ensures that the AI ​​model can obtain sufficient data support during CSI prediction to optimize the accuracy of model training, inference, and monitoring, thereby improving the overall performance of the communication system.

[0154] Step 7: The user equipment receives the monitoring and measurement signals.

[0155] In some embodiments, the user equipment receives monitoring and measurement signals sent by the base station, and these signals are used for CSI measurement. For example, after receiving the monitoring and measurement signals, the user equipment performs a real CSI measurement and uses it as the ground truth for AI model monitoring. The monitoring and measurement signals can be categorized into performance monitoring signals, functional monitoring signals, and model monitoring signals according to different monitoring requirements.

[0156] Performance monitoring measurement signals are primarily used to evaluate the overall accuracy and stability of CSI predictions, ensuring the adaptability of the AI ​​model in different network environments. Functional monitoring measurement signals are used to monitor whether the CSI prediction function is working properly, such as checking whether the prediction results meet expectations and whether there are any abnormal deviations. Model monitoring measurement signals focus on evaluating the operation of the AI ​​model itself, such as the model's convergence, the effect of parameter optimization, and whether updates or rollbacks are needed. By receiving these different types of monitoring measurement signals, user equipment can accurately measure and provide feedback on the performance, functionality, and model operation of CSI predictions, thereby providing more comprehensive data support for LCM on the network side.

[0157] In step 7, the user equipment needs to receive monitoring and measurement signals to perform actual CSI measurements and for monitoring and evaluation of the AI ​​model. In some embodiments, the base station sends monitoring and measurement signals to the user equipment, which instruct the user equipment to perform CSI measurements to obtain actual channel state information.

[0158] Specifically, after receiving the monitoring and measurement signal, the user equipment performs a real CSI measurement and uses the measured CSI as the ground truth value to compare and evaluate the prediction results of the AI ​​model. This mechanism can effectively verify the prediction accuracy of the AI ​​model and provide data support for LCM on the network side. If a large error is found between the AI-predicted CSI and the real CSI, the network side can adjust the parameters of the AI ​​model, optimize the CSI prediction strategy, or even revert to the traditional CSI reporting method to ensure the stability and reliability of the communication system.

[0159] Step 8: User devices report monitoring results.

[0160] In some embodiments, the user equipment reports monitoring results to the base station, the monitoring results indicating at least one performance monitoring result of the predicted CSI relative to the actual measured CSI. The actual measured CSI may be obtained based on a measurement reference signal or based on a monitoring measurement signal. Therefore, the monitoring measurement signal portion is represented by dashed lines, meaning that the situation where this signal is specifically used for monitoring may or may not exist.

[0161] In some embodiments, the performance monitoring results include the NMSE result and / or SGCS, or GCS result between the at least one predicted CSI and the actual measured CSI. In some embodiments, the reporting of the monitoring results is either non-event-triggered monitoring result reporting or event-triggered monitoring result reporting. In some embodiments, the non-event-triggered monitoring result reporting includes at least one of the following: the quantized value of the performance monitoring result within the monitoring window; a bitmap corresponding to the state of the performance monitoring result relative to a threshold value within the monitoring window; the predicted CSI index; and combination number indication information. In some embodiments, the event-triggered monitoring result reporting includes at least one of the following: the state information of the event occurrence within the monitoring window; the event index within the monitoring window and / or the quantized value of the performance monitoring result within the monitoring window; the monitoring result; and the predicted CSI. In some embodiments, the non-event-triggered monitoring results and / or the event-triggered monitoring results and the predicted CSI are reported in a single reporting instance.

[0162] Specifically, in step 8, during the process of user equipment reporting monitoring results, the monitoring results are mainly used by the network side to make decisions on the LCM of the AI ​​model or function. The reporting method for monitoring results can be periodic, semi-persistent, or aperiodic, and can also be triggered based on specific events. Specifically, the content of the monitoring report can include the following forms:

[0163] Non-event-triggered monitoring result reporting is suitable for continuously monitoring the prediction accuracy of AI models. For example, user equipment can report the quantized values ​​of NMSE, SGCS, or GCS between at least one predicted CSI and the actual measured CSI within the monitoring window. In addition, it can also report the quantized statistical results of NMSE, SGCS, or GCS between all predicted CSIs and the actual measured CSIs within the entire monitoring window, or provide bitmaps, predicted CSI indices, or combination number indications corresponding to the NMSE, SGCS, or GCS states between predicted CSIs and the actual measured CSIs based on a preset threshold or a threshold configured on the network side, so that the network side can perform analysis and decision-making.

[0164] Furthermore, event-triggered monitoring result reporting is primarily used to identify specific anomalies or important situations. For example, user equipment can report status information on whether a specific event has occurred within the monitoring window, along with the corresponding event index. Additionally, the device can provide quantified values ​​of NMSE, SGCS, or GCS between the predicted CSI and the actual measured CSI within the monitoring window to help the network side determine if the AI ​​model's predictions have deviated. Moreover, to provide more complete monitoring data, user equipment can attach predicted CSI data along with the monitoring results, enabling the network side to perform more in-depth analysis.

[0165] Furthermore, monitoring results, whether event-triggered or not, can be reported within the same reporting instance as the predicted CSI, improving data integrity and validity. This reporting method helps the network side more comprehensively evaluate the performance of the AI ​​model and optimize CSI prediction strategies based on monitoring data, thereby improving the overall stability and efficiency of the communication system.

[0166] Step 9: The user equipment receives the LCM indication information.

[0167] In some embodiments, the user equipment receives LCM indication information sent by the base station, the LCM indication information being used to instruct the user equipment to perform a corresponding LCM operation. In some embodiments, the LCM indication information includes function rollback operation indication information. In some embodiments, the function rollback operation indication information instructs the cessation of AI-based CSI prediction reporting and / or the activation of conventional Doppler codebook CSI reporting, the function rollback operation indication information indicating the function rollback operation via MAC CE or DCI.

[0168] In some embodiments, the LCM indication information can be model or function LCM indication information. Model LCM indication information is mainly used to manage the running status of AI models, such as triggering model updates, optimizing model parameters, or reverting to traditional CSI reporting methods when prediction results are poor. Function LCM indication information focuses on the overall management of CSI prediction functions, including enabling functions, reverting to traditional CSI prediction methods, adjusting reporting strategies, or optimizing computing resource allocation. By receiving LCM indication information, user equipment can dynamically adjust CSI prediction models or functions according to instructions from the network side, ensuring that the communication system maintains efficient and stable operation in different network environments.

[0169] Specifically, in step 9, during the process of the user equipment receiving LCM indication information, the LCM indication information is mainly used to instruct the user equipment to perform the corresponding LCM operation. The corresponding LCM operation is, for example, a function rollback operation. Specifically, the signaling bearer method for the rollback operation may include at least one of the following:

[0170] In some embodiments, a 1-bit rollback indication can be added to the MAC CE or DCI instruction through the traditional Rel-18 Doppler codebook CSI reporting activation mechanism to explicitly notify the user equipment to perform a rollback operation.

[0171] In some embodiments, management can be carried out using an instruction plus predefined approach. That is, when a user equipment receives an instruction to activate traditional Rel-18 Doppler codebook CSI reporting, and the instruction also includes an operation to deactivate AI-based predictive CSI reporting, the user equipment understands that it is falling back from AI-based CSI prediction to traditional Rel-18 Doppler codebook CSI reporting, thereby ensuring that the communication system can operate stably when the prediction effect decreases or network requirements change.

[0172] Step 10: The user equipment receives downlink data information.

[0173] In some embodiments, the user equipment receives downlink data information sent by the base station.

[0174] For example, based on the CSI reported by the user equipment, the network side performs data precoding and sends downlink data to the user equipment to optimize communication quality.

[0175] The order, combination, specific content, and execution of the above steps do not limit the embodiments of this application and can be flexibly adjusted according to actual needs to meet different scenarios and technical requirements. This process ensures that AI-based CSI prediction can operate efficiently and has a complete monitoring, reporting, and adjustment activation mechanism to adapt to dynamic network environments.

[0176] To facilitate understanding of the technical solutions of the embodiments of this application, the technical solutions related to the embodiments of this application are described below. The technical solutions of the first embodiment, second embodiment, third embodiment, fourth embodiment, fifth embodiment, sixth embodiment, seventh embodiment, eighth embodiment, ninth embodiment, tenth embodiment, eleventh embodiment, twelfth embodiment, thirteenth embodiment, fourteenth embodiment, and fifteenth embodiment are listed below for description, but this application is not limited thereto.

[0177] In some embodiments of this application, the solutions of the first embodiment can be implemented in conjunction with the solutions of the second, third, fourth, fifth, sixth, seventh, eighth, ninth, tenth, eleventh, twelfth, thirteenth, and fourteenth embodiments, and / or the fifteenth embodiment, or can be implemented independently of the solutions of the second, third, fourth, fifth, sixth, seventh, eighth, ninth, tenth, eleventh, twelfth, thirteenth, fourteenth, and fifteenth embodiments. In some embodiments of this application, solutions of multiple embodiments can be implemented in combination or independently.

[0178] First embodiment: Resource configuration during data collection.

[0179] During model training, the temporal distribution of measurement windows typically employs a denser configuration of measurement resources, while the temporal distribution of CSI for prediction windows is relatively sparse. Therefore, using traditional periodic or semi-persistent CSI-RS resource configuration methods may lead to a waste of measurement resources. To address this issue, some embodiments of this application propose an optimization scheme specifically for the rational configuration of performance monitoring resources for measurement and prediction windows during model training, thereby improving resource utilization efficiency and ensuring the effectiveness of model training. Furthermore, for non-periodic measurements and reporting, since the measurement and prediction windows are located at different positions in the temporal domain, continuing to use the traditional Rel-18 Doppler codebook-based non-periodic CSI-RS measurement resource configuration clearly cannot meet the monitoring needs of the model or function. Therefore, some embodiments of this application further consider how to optimize non-periodic CSI reporting through resource configuration, enabling it to better assist in model or function monitoring, improve the accuracy of CSI prediction, and enhance the overall system performance. For specific solutions, please refer to the first embodiment for more detailed design ideas and implementation methods.

[0180] In AI-based CSI prediction scenarios, AI models can be deployed on the user equipment side. Training the AI ​​model on the user equipment side relies on measurement signals transmitted from the network side, such as the CSI-RS signal used for CSI measurements. When training the AI / ML model for CSI prediction, the training data needs to include not only channel measurements within the observation window but also actual measurements within the prediction window, in order to design and optimize the loss function during model training.

[0181] However, for every CSI prediction, it's not necessary to collect all the actual measurements within the measurement window. Using traditional CSI-RS resource configuration to support AI model training for CSI prediction can lead to unnecessary consumption of measurement resources, resulting in resource waste. Therefore, a more efficient resource management strategy is needed to optimize CSI-RS configuration, ensuring that wireless resource utilization is improved while meeting the needs of AI model training.

[0182] To address the aforementioned issues, this embodiment provides at least one solution to ensure more reasonable resource allocation during AI / ML model training for CSI prediction, thereby improving model training efficiency while reducing wireless resource overhead.

[0183] Option 1: In some examples, the measurement resource set of the measurement configuration information includes at least one Channel State Information Reference Signal (CSI-RS) resource for CSI measurement, wherein at least the first resource is used for CSI measurement corresponding to the measurement window during CSI prediction model training, and at least the second resource is used for CSI measurement corresponding to the prediction window during CSI prediction model training.

[0184] For example, during the training of a CSI prediction model, a set of measurement resources includes at least one CSI-RS resource for CSI measurements. A portion of the CSI-RS resources are used for CSI measurements within a measurement window, while another portion is used for CSI measurements within a prediction window, providing complete training data to support AI model optimization.

[0185] For example, a measurement resource set may contain two sets of CSI measurement resources, referred to as the first set and the second set. The first set of CSI measurement resources is used for CSI measurements within the measurement window, providing historical CSI data for the AI ​​model to learn channel change trends. This set may include 4, 8, 12, 20, or 24 channel measurement resources. The second set of CSI measurement resources is used for CSI measurements within the prediction window, providing ground truth CSI values ​​for verifying and optimizing the prediction accuracy of the AI ​​model. This set may include 1, 2, 4, or 8 channel measurement resources. In this way, the measurement resource set can allocate CSI-RS resources more efficiently during AI model training, avoiding resource waste caused by traditional configuration methods, while ensuring that the AI ​​model can obtain sufficient training data, improving the accuracy and stability of CSI predictions.

[0186] Option 2: In some examples, the channel measurement resources in the first set of measurement resources in the measurement configuration information are used for CSI measurements corresponding to the measurement window during the CSI prediction model training process, and the channel measurement resources in the second set of measurement resources in the measurement configuration information are used for CSI measurements corresponding to the prediction window during the CSI prediction model training process.

[0187] For example, during the training of a CSI prediction model, the network side can configure at least one set of measurement resources, ensuring that each set contains at least one channel measurement resource. Specifically, the channel measurement resource in one set is used for the CSI measurement corresponding to the measurement window during the CSI prediction model training process, while the channel measurement resource in the other set is used for the CSI measurement corresponding to the prediction window during the CSI prediction model training process.

[0188] For example, the network side can configure two CSI-RS resource sets for user equipment. Each measurement resource set contains at least one channel measurement resource, such as 4, 8, 12, 20, or 24 channel measurement resources. The channel measurement resources in the first set are primarily used for CSI measurements corresponding to the measurement window during the CSI prediction model training process, providing historical CSI data to support the AI ​​model in learning channel change trends. The channel measurement resources in the second set are primarily used for CSI measurements corresponding to the prediction window during the CSI prediction model training process, providing true CSI values ​​to optimize the prediction accuracy of the AI ​​model. This measurement resource set can contain 1, 2, 4, or 8 channel measurement resources. This configuration of measurement resources can rationally allocate CSI-RS resources, ensuring the AI ​​model can acquire sufficient training data while avoiding unnecessary resource waste, thus improving the accuracy of CSI prediction and overall system performance.

[0189] Option 3: In some examples, the measurement resources in the first resource set subset of the measurement configuration information are used for CSI measurements corresponding to the measurement window during the CSI prediction model training process, and the measurement resources in the second resource set subset of the measurement configuration information are used for CSI measurements corresponding to the prediction window during the CSI prediction model training process.

[0190] For example, during the training of a CSI prediction model, the network side can configure at least one set of measurement resources, ensuring that each set contains at least one channel measurement resource. The measurement resources in at least one set are used for CSI measurements corresponding to the measurement window during the CSI prediction model training process; this portion can be referred to as the first subset of resource sets. The measurement resources in the remaining at least one set are used for CSI measurements corresponding to the prediction window during the CSI prediction model training process; this portion can be referred to as the second subset of resource sets.

[0191] For example, the network side can configure multiple CSI-RS resource sets for user equipment, with each measurement resource set containing at least one channel measurement resource. The measurement resources in the first 4, 8, 12, 20, or 24 measurement resource sets are primarily used for CSI measurements corresponding to the measurement window during the CSI prediction model training process, providing historical channel information to the AI ​​model to support model learning and optimization. The measurement resources in the remaining 1, 2, 4, or 8 measurement resource sets are mainly used for CSI measurements corresponding to the prediction window during the CSI prediction model training process, providing real CSI values ​​to verify and optimize the AI ​​model's predictive capabilities. In this way, the configuration of measurement resources can more rationally allocate CSI-RS resources, ensuring that the AI ​​model receives sufficient data support during training, while reducing unnecessary resource waste and improving the accuracy of CSI predictions and system performance.

[0192] For the aforementioned schemes, within the measurement window, the interval between two adjacent CSI-RS measurement resources or two adjacent measurement opportunities can be set to m measurement units, where m can be at least one value from 1 to 10, or not exceeding a certain threshold value x (x can be at least one value from 2 to 10). The unit of measurement unit can be at least one time slot or at least one symbol. Meanwhile, within the prediction window, the interval between two adjacent CSI-RS measurement resources should not be less than the time interval between two adjacent predicted CSIs within the prediction window. For example, if the time interval between two adjacent predicted CSIs within the prediction window is n, then the time interval between two adjacent CSI-RS measurement resources within the prediction window should not be less than n, where n can be any value from 1 to 50 time slots or milliseconds. Furthermore, the number of CSI-RS measurement resources within the prediction window cannot exceed the number of predicted CSIs included within the prediction window.

[0193] Meanwhile, the configuration of CSI-RS measurement resources within the measurement window and / or CSI-RS measurement resources within the prediction window must meet at least one of the following constraints: the same time domain type (periodic, semi-continuous, or aperiodic), the same period, the same number of antenna ports, the same PCoffset (power control offset), the same PCOffsetSS (power control offset subset), the same CDM type (code division multiplexing type), the same RS density (reference signal density), and the same QCL configuration (quasi-co-channel configuration).

[0194] Furthermore, the periodicity of measurement resources can also be distinguished. For example, CSI-RS resources within the measurement window may be periodic, while CSI-RS resources within the prediction window may be periodic, semi-persistent, or aperiodic; for example, CSI-RS resources within the measurement window may be semi-persistent, while CSI-RS resources within the prediction window may be semi-persistent or aperiodic.

[0195] If CSI-IM (CSI-Interference Measurement) is used for interference measurement, only one resource is configured in the corresponding CSI-IM resource set.

[0196] To enhance model generalization capabilities, the temporal distribution of measurement resources within the measurement window may be denser during model training, meaning the interval between two adjacent measurement opportunities is smaller. For example, during model training, the interval between two measurement opportunities or measurement resources may be m time slots, where m takes at least one value from 1 to 10. Specifically, in terms of resource configuration, this can be represented by the interval between two adjacent CSI-RS resources within the measurement window being smaller than the interval between two adjacent CSI-RS resources within the prediction window. For periodic or semi-persistent CSI-RS resources, this can be represented by the interval between the offset values ​​of two adjacent resources within the measurement window being smaller than the interval between the offset values ​​of two adjacent resources within the prediction window. A specific example configuration can be seen in Figure 4 to more intuitively illustrate the difference in the distribution of CSI-RS resources within the measurement window and the prediction window.

[0197] The first embodiment addresses the resource allocation problem for data collection during the training of AI-based CSI prediction models, proposing at least one optimization scheme to improve training efficiency, optimize resource utilization, and reduce unnecessary measurement overhead. Scheme 1: Within a measurement resource set, CSI-RS resources are divided into measurement window CSI measurement resources (for providing historical CSI data) and prediction window CSI measurement resources (for providing true CSI values ​​as ground truth). Scheme 2: Two independent CSI-RS resource sets are configured on the network side, respectively for CSI measurements in the measurement window and prediction window, improving the flexibility and effectiveness of resource management. Scheme 3: Multiple measurement resource sets are divided into a first subset (measurement window CSI measurements) and a second subset (prediction window CSI measurements), ensuring the AI ​​model receives sufficient data support while optimizing resource allocation and reducing unnecessary measurement overhead. Furthermore, this embodiment further explores the temporal interval of CSI-RS measurement resources, the configuration constraints of measurement windows and prediction windows, and optimization strategies to enhance the generalization ability of AI models, such as adjusting the density of measurement resources to ensure the diversity and effectiveness of training data. This embodiment optimizes CSI-RS resource configuration, making the allocation of measurement resources during AI model training more reasonable, effectively improving the accuracy of CSI prediction, while reducing the consumption of wireless resources and improving the overall system performance.

[0198] Second embodiment: Differentiation of content reported by user devices.

[0199] Regarding the reporting method of performance monitoring results, according to the standard discussion, periodic, semi-persistent, or non-periodic reporting mechanisms can be adopted. However, the specific content of the performance monitoring results still needs further definition to ensure that the network side can accurately understand and process this data. At the same time, the relationship and reporting format between the reporting of performance monitoring results and the reporting of model inference output also need to be further clarified. For example, how the network side distinguishes the reporting content under different LCM states, and how to ensure the validity and consistency of various types of reported information, are issues that the standard needs to discuss in depth. Therefore, some embodiments of this application provide corresponding solutions; for specific solutions, please refer to the second embodiment for a more complete design concept and implementation method.

[0200] Based on current standard discussions, when user equipment-side AI models are used for CSI prediction, the traditional codebook type "typeII-Doppler-r18" feedback mechanism can be used as a starting point. However, from the network side's perspective, based solely on the reported content, it's impossible to distinguish whether the CSI feedback from the user equipment is obtained based on traditional prediction methods (e.g., using the "typeII-Doppler-r18" codebook) or on AI model prediction. This ambiguity may affect the network side's parsing and further optimization of CSI data.

[0201] Furthermore, for different CSI prediction AI models, some of the AI ​​model's output results may not need to be reported to the network side during model training. Therefore, user devices need to align the purpose of the model's output with the network side to ensure that the use of training data meets the network side's optimization requirements. Additionally, during model monitoring, the CSI predicted by the AI ​​model may need to be reported jointly with performance metrics from performance monitoring so that the network side can accurately assess the AI ​​model's prediction accuracy and operational status.

[0202] Meanwhile, considering that model training, inference, and monitoring may use the same CSI-RS resource configuration, the network side needs to be able to distinguish the specific content that needs to be reported at different model stages, so as to reasonably parse and utilize CSI data in different application scenarios. To address this requirement, this embodiment provides at least one solution to ensure that user devices and the network side can effectively collaborate at different CSI prediction model stages, optimize the CSI reporting mechanism, and improve the applicability and accuracy of AI-predicted CSI.

[0203] Option 1: In some examples, the user equipment determines whether the predicted CSI is used for AI-based CSI prediction by using the CSI reporting configuration field.

[0204] For example, this embodiment reuses the existing standard CSI-ReportConfig field and adds an indicator field under it to identify whether the current CSI reporting configuration is suitable for AI-based CSI prediction. Specifically, a 1-bit indicator is added under the CSI-ReportConfig field. When this bit is set to 1, it indicates that the current configuration is used for AI-based CSI prediction; if this field is not configured or is defaulted, it means that the CSI-ReportConfig field is suitable for traditional CSI reporting.

[0205] In addition, to distinguish the case where CSI predictions are used for model training (i.e., the predicted CSI data does not need to be reported to the network side), the reportQuantity field under the CSI-ReportConfig field should be configured as "None" (no reporting), and / or the resource configuration for periodic or semi-continuous reporting should be defaulted in the reportConfigType field under the CSI-ReportConfig field, to indicate that the CSI prediction data is only used for local training of the AI ​​model and will not be reported to CSI.

[0206] On the other hand, in order to distinguish that the reported CSI is based on AI model inference, the reportQuantity field under the CSI-ReportConfig field is configured to be the same as the "typeII-Doppler-r18" codebook, namely cri-RI-PMI-CQI, so as to ensure that the AI-predicted CSI can be reported according to the existing standards, while facilitating the network side to identify and parse the source of the CSI reported data.

[0207] Option 2: In some examples, the user equipment determines the predicted CSI based on AI-based CSI prediction through the codebook configuration field under the CSI reporting configuration field.

[0208] For example, this embodiment reuses the existing standard CSI-ReportConfig field and associates it with a specific codebook type through CodebookConfig to distinguish the source of CSI reported data. When the codebook type of CodebookConfig is configured as "AI-predict-CSI-r19", it indicates that the currently reported CSI information is based on AI model prediction calculations, thus enabling the network side to clearly distinguish between AI-predicted CSI and traditional CSI reporting methods.

[0209] In addition, to distinguish the case where CSI predictions are used for model training, i.e., the predicted CSI does not need to be reported, the reportQuantity field under the CSI-ReportConfig field should be configured as none, and / or the resource configuration for periodic or semi-continuous reporting in the reportConfigType field under the CSI-ReportConfig field should be defaulted, so as to indicate that the CSI prediction is only used for model training and will not be uploaded to the network side.

[0210] On the other hand, to distinguish that the reported CSI is based on AI model inference, the reportQuantity field under the CSI-ReportConfig field should be configured with the same value as the "typeII-Doppler-r18" codebook, i.e., cri-RI-PMI-CQI. This configuration method ensures that the reporting of AI-predicted CSI is compatible with existing standards, enabling the network side to parse and utilize the AI-predicted CSI data according to the traditional CSI reporting mechanism.

[0211] Option 3: In some examples, the user equipment determines the CSI prediction used for model training as AI-based CSI prediction and / or determines the reported CSI as obtained from AI model inference through the reporting quantity field under the CSI reporting configuration field, and / or determines the default reporting resource for periodic or semi-continuous reporting through the reporting configuration type field under the CSI reporting configuration field.

[0212] For example, this embodiment reuses the existing standard CSI-ReportConfig field and uses specific field configurations to distinguish the purpose of CSI predictions, so as to ensure that the network side can correctly parse the CSI prediction data of the AI ​​model.

[0213] To distinguish the use of CSI predictions for model training (i.e., when predicted CSI data does not need to be reported), the `reportQuantity` field under `CSI-ReportConfig` should be configured as `AI-CSI-for-training`. This configuration indicates that during model training, the user device does not need to report the prediction results to the network side, or that the user device reports the prediction results while simultaneously performing the necessary steps for model training. Furthermore, the periodic or semi-continuous reporting configuration in the `reportConfigType` field under `CSI-ReportConfig` can be left blank to further clarify that the CSI prediction data is only used for local model training and will not participate in the CSI reporting process. If periodic or semi-continuous reporting is configured in the `reportConfigType` field under `CSI-ReportConfig`, then the predicted CSI results must also be reported.

[0214] To distinguish the reported CSIs from those derived from AI model inference, the `reportQuantity` field under `CSI-ReportConfig` should be configured as `AI-cri-RI-PMI-CQI`, differentiating it from the traditional `cri-RI-PMI-CQI` configuration under the `typeII-Doppler-r18` codebook. This configuration ensures compatibility between AI-predicted CSI reporting and traditional CSI reporting mechanisms, while enabling the network side to clearly identify the source of CSI data, thereby optimizing network scheduling and resource allocation.

[0215] Option 4: In some examples, the user equipment determines, through the information element IE, that the predicted CSI is based on AI-based CSI prediction and / or the reporting of CSI information based on traditional CSI prediction.

[0216] For example, this embodiment defines a dedicated Information Element (IE) for configuring the reporting of AI-based CSI predictions and / or compressed CSI information. This field is named AI-CSI-ReportConfig. The introduction of this field aims to optimize the management of CSI prediction data, enabling the network side to more accurately parse the source and purpose of AI-predicted CSI.

[0217] To distinguish whether CSI predictions are for model training (i.e., predicted CSI does not need to be reported), the `reportQuantity` field under the `CSI-ReportConfig` field should be configured to `none`, `AI-CSI-for-training`, or `MAC CE (Medium Access Control Control Element)` with no corresponding `reportConfigure`. This clearly indicates that the CSI data is only used for AI model training and will not be reported. Furthermore, the `reportConfigType` field under the `CSI-ReportConfig` field can be configured to the default configuration for periodic or semi-continuous reporting to ensure that predicted CSI during model training does not consume uplink resources.

[0218] To distinguish that the reported CSI is derived from AI model inference, the reportQuantity field under the CSI-ReportConfig field should be configured with the same value as the "typeII-Doppler-r18" codebook configuration, i.e., cri-RI-PMI-CQI. This configuration scheme ensures that AI-predicted CSI reporting conforms to existing standards, while enabling the network side to correctly identify the source of CSI data, thereby optimizing resource scheduling and improving the adaptability and efficiency of the communication system.

[0219] Option 5: In some examples, the user equipment determines various reporting quantities through the CSI reporting configuration field or the reporting quantity field under the AI-based CSI reporting configuration field.

[0220] For example, in AI-based CSI prediction, the content and format of reports required by the AI ​​model on the user equipment side may differ at different stages. According to existing standards, the CSI reporting volume of user equipment is distinguished by the `reportQuantity` field under the `CSI-ReportConfig` field, and this field is semi-statically configured. If the AI ​​CSI reporting function is configured in the traditional way, multiple sets of `CSI-ReportConfig` need to be configured for different reporting requirements, which will significantly increase the overhead of RRC resources.

[0221] To address this issue, this solution proposes configuring multiple reporting quantities simultaneously in the `reportQuantity` field of either the traditional `CSI-ReportConfig` field or the newly added `AI-CSI-ReportConfig` field specifically for AI reporting. MAC CE or DCI is then used to further indicate which type of CSI information is being reported. If CSI reporting is periodic, a specific reporting quantity under the `reportQuantity` field will be reported by default, for example, the first configured reporting quantity will be reported by default.

[0222] Specifically, the reportQuantity field can be configured with at least one of the following reporting quantities:

[0223] “None”: No reporting required.

[0224] "AI-CSI-for-training" indicates that the current model is in the training phase, and the predicted CSI does not need to be reported or the predicted CSI needs to be reported.

[0225] “cri-RI-PMI-CQI”: CSI reporting volume consistent with the traditional “typeII-Doppler-r18” codebook configuration.

[0226] "Performance-metric": Performance monitoring results reported by user devices.

[0227] “PM-cri-RI-PMI-CQI”: Combines the CSI reporting volume with the performance monitoring results, monitors the accuracy of historical CSI predictions while predicting CSI, thereby ensuring the accuracy of CSI predictions and allowing for timely adjustments to the CSI prediction strategy.

[0228] To dynamically indicate which specific reporting volume to report, control can be achieved through MAC CE or DCI messages.

[0229] If the reportQuantity field is configured as "None", the user equipment will not report CSI.

[0230] If the reportQuantity field is configured as "AI-CSI-for-training", it indicates that the current stage is the AI ​​model training phase, and the predicted CSI does not need to be reported or the predicted CSI needs to be reported.

[0231] The specific indication of a MAC CE or DCI message depends on the number of reports configured in the reportQuantity field. For example, if multiple different reporting types are configured, the specific reporting type can be indicated by a 2-bit MAC CE or DCI field.

[0232] The MAC CE or DCI indication can be an activation message used to activate CSI reporting. This involves adding a certain number of bits to the traditional MAC CE or DCI activation message to indicate the corresponding reporting quantity. The specific number of bits to add depends on the number of reporting quantities configured in the reportQuantity field.

[0233] In addition, this instruction can also be a new type of MAC CE or DCI message. No specific restrictions are imposed here, and all of them fall within the protection scope of this scheme.

[0234] Meanwhile, under the CSI-ReportConfig field or the AI-CSI-ReportConfig field specifically for AI reporting, different reporting resources can be configured for different reporting volumes under the periodic reporting field or the semi-persistent OnPUCCH field. Different reporting volumes may be configured with the same reporting resources. This mechanism ensures that AI-based CSI prediction reporting is more flexible and efficient, while reducing RRC resource overhead and improving the adaptability of CSI reporting.

[0235] This solution proposes multiple reporting configuration mechanisms to support the reporting needs of AI-based CSI prediction at different stages, including training, inference, and monitoring. These reporting options include, but are not limited to: "None": No reporting. "AI-CSI-for-training": Indicates that the system is currently in the AI ​​model training phase, and CSI prediction either does not require reporting or requires reporting. "cri-RI-PMI-CQI": The traditional CSI reporting method, identical to the "typeII-Doppler-r18" codebook. "Performance-metric": UE reports performance monitoring results to enhance the flexibility of CSI feedback. "PM-cri-RI-PMI-CQI": A reporting mode combining performance monitoring results and traditional CSI feedback, improving network-side scheduling optimization capabilities. It monitors the accuracy of historical CSI predictions while predicting CSI, ensuring the accuracy of CSI predictions and allowing for timely adjustments to the CSI prediction strategy. By dynamically indicating which type of CSI data to report via MAC CE or DCI messages, multiple CSI-ReportConfig sets are avoided for different CSI reporting needs, thereby reducing RRC resource overhead and improving the adaptability and reporting efficiency of CSI prediction. This mechanism optimizes RRC resource utilization by dynamically adjusting the CSI reporting content, enabling AI-predicted CSI to flexibly adapt to network requirements at different stages, thus improving the efficiency and intelligence of the communication system.

[0236] For schemes one to five above, in order to further distinguish between the CSI reporting of model monitoring output results and the joint reporting of model monitoring output results and predicted CSI information, this embodiment optimizes the configuration method of the reporting volume so that the network side can accurately parse and manage CSI data at different stages.

[0237] To distinguish whether the reported CSI is the output of model monitoring, the reportQuantity field can be configured as "Performance-metric". When the UE reports the performance monitoring results, the CSI data reported by the user equipment only includes the output of model monitoring, which is used by the network side to evaluate the prediction accuracy and performance of the AI ​​model.

[0238] If the CSI information predicted by the model and the output results of the model monitoring need to be reported together, the reportQuantity field is configured as "PM-cri-RI-PMI-CQI". This means that the CSI data includes both the output results of the model monitoring and the prediction results of the model inference. This ensures that the network side can comprehensively evaluate the prediction effect of the AI ​​model and make corresponding optimizations. While predicting CSI, the accuracy of historical CSI predictions is monitored to ensure the accuracy of CSI predictions and adjust the CSI prediction strategy in a timely manner.

[0239] Furthermore, during model training, the CSI-ReportConfig field, or the reportQuantity field under the AI-CSI-ReportConfig field specifically for AI reporting, can be configured as "cri-RI-PMI-CQI". However, whether the user equipment actually performs CSI reporting depends on whether the MAC CE or DCI message carries relevant indication information. For example, in the MAC CE or DCI message, a 1-bit indication can be used to activate or deactivate CSI reporting to determine whether the corresponding content is currently being uploaded to the network side.

[0240] For periodic CSI reporting, the decision to report can be made in the RRC configuration using appropriate indication information. For example, a 1-bit indication can be used to control whether the CSI data corresponding to the reportQuantity field is uploaded to the network side.

[0241] It is important to note that the CSI information reported by model inference and the performance metrics reported by model monitoring output can correspond to different CSI-ReportConfig IDs. That is, the network side can use different IDs to identify the category of CSI reporting, ensuring that CSI prediction data and model monitoring data are not confused, thereby improving the accuracy and adaptability of data management.

[0242] The second embodiment proposes a user device-reported content differentiation mechanism to optimize the parsing and management of AI-based CSI prediction data. This ensures the network side can accurately identify the source and purpose of CSI data and rationally parse and utilize it at different stages of training, inference, and monitoring. A new indicator field is added to the CSI-ReportConfig field to distinguish between traditional and AI-based CSI reports. The source of CSI data is differentiated by associating codebook types (e.g., "AI-predict-CSI-r19") through Codebook Configuration. Separate management of AI-predicted CSI and traditional CSI is achieved through the Information Element (IE) AI-CSI-ReportConfig. Multiple reporting volume mechanisms (e.g., "None", "AI-CSI-for-training", "Performance-metric", "PM-cri-RI-PMI-CQI") are used for differentiation. Specific CSI reporting content is dynamically indicated through MAC CE or DCI, improving the flexibility of CSI reporting. Model monitoring reporting can be configured as follows: The `reportQuantity` field is set to "Performance-metric," reporting only the output of model monitoring for network evaluation of AI model performance. Model inference and monitoring reporting can be combined: The `reportQuantity` field is set to "PM-cri-RI-PMI-CQI," including both predicted CSI from AI inference and monitoring output. This allows the network to comprehensively evaluate model performance, monitoring the accuracy of historical CSI predictions while simultaneously predicting CSI, ensuring accurate predictions and enabling timely adjustments to the CSI prediction strategy. During training, the `reportQuantity` field can be configured to "cri-RI-PMI-CQI," but whether it is reported is indicated by a 1-bit indicator in MAC CE or DCI. Periodic CSI reporting is controlled by a 1-bit indicator in the RRC configuration, controlling whether CSI data is reported to the network. Different CSI-ReportConfig IDs can be used for reporting CSI information from model inference and performance metrics from model monitoring, ensuring clear data classification, avoiding confusion, and improving the accuracy and adaptability of data management. The second embodiment achieves efficient management of AI-predicted CSI through a flexible CSI reporting differentiation mechanism, optimizes RRC resource utilization, ensures that CSI data can accurately adapt to network requirements, and improves the intelligence and scheduling efficiency of the communication system.

[0243] Third embodiment: Resource allocation during performance monitoring.

[0244] During model training, the temporal distribution of measurement windows typically employs a denser configuration of measurement resources, while the temporal distribution of CSI in prediction windows is relatively sparse. Therefore, using traditional periodic or semi-persistent CSI-RS resource configuration methods may lead to a waste of measurement resources. To address this issue, some embodiments of this application propose an optimization scheme specifically for the rational configuration of performance monitoring resources for measurement and prediction windows during model training, thereby improving resource utilization efficiency and ensuring the effectiveness of model training. Furthermore, for non-periodic measurements and reporting, since the measurement and prediction windows are located at different positions in the temporal domain, continuing to use the traditional Rel-18 Doppler codebook-based non-periodic CSI-RS measurement resource configuration clearly cannot meet the monitoring needs of the model or function. Therefore, some embodiments of this application further consider how to optimize non-periodic CSI reporting through resource configuration, enabling it to better assist in model or function monitoring, improve the accuracy of CSI prediction, and enhance the overall system performance. For a more detailed design approach and implementation method, please refer to the third embodiment.

[0245] For AI-based CSI prediction, the AI ​​model can reside on the user device side. For AI models on the user device side, the measurement resources used during model or function monitoring can be configured differently. Specifically, the measurement resources may be the same as those used during model inference, or they may be separately configured specifically for model monitoring to optimize the accuracy of CSI prediction and the efficiency of system resource utilization.

[0246] Especially for aperiodic CSI-RS measurement resources, the reporting of predicted CSI is also aperiodic, so the measurement window and the prediction window will not overlap. In this case, the CSI-RS resources for the measurement window and the CSI-RS resources for the prediction window are different CSI-RS resources. That is, the CSI-RS resources used for the measurement window focus on collecting historical CSI information, while the CSI-RS resources used for the prediction window are used to generate predicted CSI information for the AI ​​model to infer and report.

[0247] To address the aforementioned needs, this embodiment provides at least one resource configuration scheme to ensure that the CSI-RS resource configuration can reasonably divide the measurement window and prediction window during AI model training, inference, and monitoring, thereby improving the accuracy of CSI prediction, optimizing network-side resource management, and reducing unnecessary resource waste.

[0248] Option 1: In some examples, the measurement resource set of the measurement configuration information includes at least one non-periodic CSI-RS measurement resource, and a portion of the measurement resources in the measurement resource set are used for CSI measurements under the measurement window during the CSI prediction process.

[0249] For example, the network side configures a measurement resource set for the user equipment. This measurement resource set includes at least one aperiodic CSI-RS measurement resource, and the measurement resources can be divided according to different stages of CSI prediction. Specifically, a portion of the measurement resources in the measurement resource set is used for CSI measurements within the measurement window of CSI prediction; this portion of measurement resources can be defined as measurement resource group one. Furthermore, at least one CSI-RS measurement resource in the measurement resource set is used for CSI measurements within the CSI prediction window; this portion of measurement resources can be defined as measurement resource group two.

[0250] In terms of resource allocation, measurement resource group one contains M aperiodic CSI-RS measurement resources for acquiring CSI data within the measurement window, while measurement resource group two contains N aperiodic CSI-RS measurement resources for performing CSI measurements within the prediction window. Furthermore, the number of CSIs predicted within the prediction window is N⁴, where N can be greater than, equal to, or less than N⁴, depending on the system's CSI measurement and prediction requirements.

[0251] Meanwhile, the time-domain locations of the N aperiodic CSI-RS measurement resources can overlap, partially overlap, or not overlap at all with the time-domain location of the predicted CSI. This configuration ensures the flexibility of measurement resources, enabling the AI ​​model to rationally adjust the resource distribution of the measurement and prediction windows according to different CSI measurement and prediction needs, thereby improving the accuracy of CSI prediction, optimizing network-side resource management, and reducing unnecessary resource waste.

[0252] Option 2: In some examples, the measurement resource set of the measurement configuration information includes at least one non-periodic CSI-RS measurement resource. The first measurement resource set includes M non-periodic CSI-RS measurement resources for CSI measurement under the measurement window during the CSI prediction process; the second measurement resource set includes N non-periodic CSI-RS measurement resources for CSI measurement under the prediction window during the CSI prediction process, where M and N are greater than or equal to 1.

[0253] For example, the network side configures two measurement resource sets for the user equipment, and each measurement resource set contains at least one aperiodic CSI-RS measurement resource. One measurement resource set contains M aperiodic CSI-RS measurement resources, used for CSI measurements within the measurement window of CSI prediction; the other measurement resource set contains N aperiodic CSI-RS measurement resources, used for CSI measurements within the CSI prediction window.

[0254] In addition, the number of CSIs predicted within the prediction window is N4, where the value of N can be greater than, equal to, or less than N4, and the specific value is determined by the CSI measurement requirements of the system. At the same time, the time-domain positions of the N aperiodic CSI-RS measurement resources can completely coincide, partially coincide, or not coincide at all with the time-domain positions of the predicted CSIs.

[0255] For example, when N = N4, the time-domain position of each aperiodic CSI-RS measurement resource can completely coincide with the time-domain position of the predicted CSI, that is, located in the same slot, so as to ensure the close correspondence between the measurement resources and the predicted resources and improve the accuracy and utilization efficiency of the CSI data.

[0256] When N < N4, the time-domain position of each aperiodic CSI-RS measurement resource can still coincide with the time-domain position of the predicted CSI. For example, it can be selected to coincide with the time-domain position of the first predicted CSI. This method helps to report the performance monitoring results faster, enables the network side to judge the prediction results of the AI model more quickly, optimizes the network resource scheduling, and improves the adaptability and accuracy of the CSI prediction.

[0257] Solution 3: In some examples, the measurement resource set of the measurement configuration information includes at least one aperiodic CSI-RS measurement resource. The first measurement resource set includes M aperiodic CSI-RS measurement resources for CSI measurement under the measurement window in the CSI prediction process; the second measurement resource set includes N aperiodic CSI-RS measurement resources for CSI measurement under the prediction window in the CSI prediction process, and M and N are greater than or equal to 1.

[0258] Illustrate with an example. The network side configures at least two measurement resource sets for the user equipment, and each measurement resource set includes at least one aperiodic CSI-RS measurement resource. Among them, one measurement resource set includes M aperiodic CSI-RS measurement resources for CSI measurement within the measurement window of the CSI prediction; the remaining N measurement resource sets are used for CSI measurement within the prediction window of the CSI prediction.

[0259] In addition, the number of CSIs predicted within the prediction window is N4, where the value of N can be greater than, equal to, or less than N4, and the specific value depends on the CSI measurement requirements and the system resource allocation strategy. At the same time, the time-domain positions of the N aperiodic CSI-RS measurement resources can completely coincide, partially coincide, or not coincide at all with the time-domain positions of the predicted CSIs.

[0260] For example, when N = N4, the time-domain position where each aperiodic CSI-RS measurement resource is located can completely coincide with the time-domain position where the predicted CSI is located, that is, in the same slot, so as to ensure the correspondence between the measurement resource and the predicted resource, and improve the measurement accuracy and utilization efficiency of CSI data.

[0261] When N < N4, the time-domain position where each aperiodic CSI-RS measurement resource is located can still coincide with the time-domain position where the predicted CSI is located. For example, it can be selected to coincide with the time-domain position where the first predicted CSI is located. This method helps to report the performance monitoring results faster, enabling the network side to evaluate and optimize the prediction results of the AI model more quickly, thereby improving the network resource scheduling efficiency and enhancing the adaptability and accuracy of CSI prediction.

[0262] In the above scheme, the value of N4 can be any one of {1, 2, 4, 8, 12}, and the value of M can be any one of {4, 8, 12}, and the specific values are determined by the CSI measurement requirements of the system. In addition, when the aperiodic CSI-RS measurement resource coincides, partially coincides or does not completely coincide with the time-domain position where the predicted CSI is located, there is a corresponding correlation relationship between the predicted CSI and the actually measured CSI. The specific correlation relationship can refer to the eighth embodiment, but in addition to the correlation relationship described in the eighth embodiment, other correlation methods are not excluded, and the specific correlation method can be adjusted according to the system optimization requirements.

[0263] In addition, in the above scheme, multiple CSI-RS measurement resources under the same resource set or multiple resource sets need to meet at least one of the following constraints: multiple CSI-RS resources have the same number of antenna ports to ensure the consistency of CSI measurement data; the same PCoffset (power control offset) to ensure the relative timing consistency of CSI measurement; the same PCOffsetSS (power control subset offset) to ensure the stability of measurement configuration; the same CDM (code division multiplexing) type to optimize the multiplexing efficiency of CSI resources; the same RS density (reference signal density) to maintain the CSI measurement accuracy; the same QCL (quasi-co-located channel) configuration to ensure the channel consistency during CSI measurement. This configuration method can improve the accuracy of CSI measurement and prediction, enable the network side to more effectively analyze and utilize CSI data, and then optimize the system performance.

[0264] The third embodiment proposes a resource allocation scheme for performance monitoring, aiming to optimize the measurement resource management of the AI model during CSI prediction on the user equipment side, ensure a reasonable division of the CSI measurement window and the prediction window, and improve the accuracy of CSI prediction and the utilization efficiency of system resources. The measurement resources can be the same as those in the inference stage or can be separately configured for performance monitoring. The CSI-RS resources of the measurement window and the CSI-RS resources of the prediction window are independently configured to prevent overlap and improve the integrity of data collection.

[0265] Provide at least one different resource allocation scheme: Scheme 1: Use a measurement resource set, divided into Measurement Resource Group 1 (M CSI-RS resources) for the measurement window and Measurement Resource Group 2 (N CSI-RS resources) for the prediction window. Scheme 2: Use two independent measurement resource sets, respectively for the measurement window and the prediction window, to improve the flexibility of resource management. Scheme 3: Use multiple measurement resource sets to ensure the accuracy of CSI measurement and prediction in different stages and adapt to more complex network requirements. The number of predicted CSIs N4 in the prediction window can take values {1, 2, 4, 8, 12}, and the number of resources M in the measurement window can take values {4, 8, 12}. When N = N4, the predicted CSI completely coincides with the aperiodic CSI-RS resources, improving the timing synchronization. When N < N4, the CSI-RS resources can partially coincide or the first CSI-RS resource coincides with the first predicted CSI, accelerating the feedback of performance monitoring results and improving the prediction evaluation efficiency of the AI model on the network side.

[0266] The CSI-RS resources under the same measurement resource set or multiple resource sets need to meet at least one of the following constraints: the same number of antenna ports, the same PCoffset and PCOffsetSS (power control offset parameters), the same CDM type (code division multiplexing type), the same RS density (reference signal density), the same QCL configuration (quasi-co-located channel configuration). These constraints ensure the consistency of CSI measurement data, improve the measurement accuracy, and optimize the network resource scheduling. The third embodiment improves the CSI prediction accuracy of the AI model, reduces resource waste, and optimizes the parsing and management of CSI data on the network side by reasonably configuring CSI measurement and prediction resources, enhancing the intelligence and adaptability of the communication system.

[0267] Fourth embodiment: The reporting content (Type 3) during performance monitoring.

[0268] Regarding the reporting method of performance monitoring results, according to the standard discussion, periodic, semi-persistent, or non-periodic reporting mechanisms can be adopted. However, the specific content of the performance monitoring results still needs further definition to ensure that the network side can accurately understand and process this data. At the same time, the relationship and reporting format between the reporting of performance monitoring results and the reporting of model inference output also need to be further clarified. For example, how the network side distinguishes the reporting content under different LCM states, and how to ensure the validity and consistency of various types of reported information, are issues that the standard needs to discuss in depth. Therefore, some embodiments of this application provide corresponding solutions; for specific solutions, please refer to the fourth embodiment for a more complete design concept and implementation method.

[0269] In some instances, performance monitoring results may be presented in at least one of the following ways. Performance monitoring results may include, for example, NMSE, SGCS, or GCS results between predicted CSI and associated measured CSI.

[0270] Method 1: In some examples, the performance monitoring results are quantized, with the number of quantized bits being Y, where Y is greater than 1.

[0271] For example, the quantization of NMSE, SGCS, or GCS results can be encoded using the number of quantization bits Y to reduce uplink transmission overhead while maintaining effective feedback on prediction errors. For instance, if 3 bits (Y=3) are used for quantization, the error ranges of NMSE, SGCS, or GCS can be divided into different levels, each corresponding to a predefined bit value. Specifically, Table 1 shows a mapping relationship for 3-bit quantization, where different error ranges correspond to different bit codes. For example, smaller errors (lower NMSE, higher SGCS, or GCS) can be assigned lower bit values, indicating more accurate AI model predictions; larger errors (higher NMSE, lower SGCS, or GCS) can be assigned higher bit values, indicating decreased prediction accuracy of the AI ​​model, requiring network-side attention and optimization.

[0272] This quantification method allows user equipment to reduce uplink signaling burden, improve the reporting efficiency of CSI prediction errors, and ensure that the network side can accurately analyze the prediction performance of the AI ​​model, thereby further optimizing radio resource management and CSI prediction mechanisms.

[0273] Table 1: The performance monitoring results are quantized, and the number of bits quantized is Y, where Y is greater than 1.

[0274] Method 2: In some examples, the performance monitoring results are quantized at equal intervals, with the number of quantization bits being Y, and the value of Y depending on the size of the equal interval division.

[0275] For example, for quantization of NMSE, SGCS, or GCS results, an equal-interval quantization method can be used, where the number of quantization bits Y depends on the size of the equal interval division. Equal-interval quantization can effectively reduce the complexity of data reporting while maintaining accurate feedback on CSI prediction errors.

[0276] For example, when Y = 2 bits, the values ​​of NMSE, SGCS, or GCS can be divided into equally spaced intervals, each interval being 1 / 3 the size of the range, and each interval corresponding to a bit encoding value. Thus, when user equipment reports NMSE, SGCS, or GCS, it only needs to convert the measured value into the corresponding bit value for transmission based on the interval to which it belongs.

[0277] Specifically, Table 2 illustrates a mapping relationship for 2-bit equal-interval quantization, where: lower errors (high GCS, high SGCS, or low NMSE) correspond to smaller bit values, indicating more accurate AI model predictions; higher errors (low GCS, low SGCS, or high NMSE) correspond to larger bit values, indicating decreased AI model prediction accuracy, requiring network-side attention and optimization. This equal-interval quantization method allows user equipment to reduce uplink signaling overhead, improve the efficiency of accuracy information reporting, and ensure that the network side can quickly evaluate AI model prediction performance based on the quantization results, thereby optimizing CSI prediction management and scheduling strategies.

[0278] Table 2: Mapping relationship of 2-bit equal-interval quantization, with the size of the equal-interval division being 1 / 3.

[0279] Method 3: In some examples, the performance monitoring results are quantized at unequal intervals, with the number of quantization bits being Y. The value of Y depends on the quantization boundary values. The boundary values ​​can be predefined.

[0280] For example, for the quantization of NMSE, SGCS, or GCS results, an unequal-interval quantization method can be used, where the number of quantization bits Y depends on the quantization boundary values ​​in a predefined table. These boundary values ​​can be predefined or preset based on system requirements or empirical data.

[0281] For example, when Y = 2 bits, the values ​​of NMSE, SGCS, or GCS can be divided into different intervals by using two predefined boundary values ​​A = 1 / 2 and B = 3 / 4, and a corresponding bit encoding value can be assigned to each interval.

[0282] Under this unequal interval quantization method, smaller errors (higher GCS, higher SGCS, or lower NMSE) can be assigned to the interval with the smallest error range and a smaller bit value, indicating that the AI ​​model's prediction is more accurate; larger errors (lower GCS, lower SGCS, or higher NMSE) can be assigned to the interval with the larger error range and a larger bit value, indicating that the AI ​​model's prediction accuracy is lower and needs adjustment or optimization. Specifically, Table 3 shows a mapping relationship for 2-bit unequal interval quantization, where: when NMSE, SGCS, or GCS is less than A (1 / 2), it corresponds to the minimum bit value; when NMSE, SGCS, or GCS is between A (1 / 2) and B (3 / 4), it corresponds to the middle bit value; when NMSE, SGCS, or GCS is greater than B (3 / 4), it corresponds to the maximum bit value.

[0283] Compared to equal-interval quantization, unequal-interval quantization can more accurately capture the impact of different error ranges, especially providing higher resolution in low-error regions to optimize the accuracy feedback of AI model predictions. This approach can reduce unnecessary data reporting, improve uplink signaling utilization, and ensure that the network side can more accurately analyze the prediction performance of the AI ​​model, thereby optimizing the CSI prediction mechanism and radio resource management strategies.

[0284] Table 3: The performance monitoring results are quantized at unequal intervals, with the number of quantization bits being Y. The value of Y depends on the boundary value of the quantization.

[0285] Method 4: In some examples, when the performance monitoring result is less than a threshold value, a first quantization interval is used; when the performance monitoring result is greater than the threshold value, a second quantization interval is used, where the first quantization interval is greater than the second quantization interval. In some examples, the performance monitoring result is quantized at unequal intervals, with the number of quantization bits being Y, and the quantization interval is larger when the result is less than a certain threshold value, and smaller when the result is greater than a certain threshold value.

[0286] For example, quantization of NMSE, SGCS, or GCS results can be performed using unequal interval quantization, where the number of quantization bits Y is affected by a quantization threshold. Specifically, when NMSE, SGCS, or GCS is less than a certain threshold, a larger quantization interval is used, while when NMSE, SGCS, or GCS is greater than the threshold, a smaller quantization interval is used. This improves accuracy when prediction errors are large, while reducing redundant reporting information when errors are low.

[0287] For example, when Y = 3 bits, the values ​​of NMSE, SGCS, or GCS can be divided into multiple unequal intervals. When the error is small (i.e., low NMSE, high SGCS, or high GCS), a larger quantization interval is used to reduce uplink signaling overhead while maintaining reasonable feedback on AI prediction accuracy. When the error is large (i.e., high NMSE, low SGCS, or low GCS), a smaller quantization interval is used to provide more refined accuracy information, enabling the network side to accurately adjust the model or optimize resource scheduling strategies.

[0288] Specifically, Table 4 illustrates a mapping relationship for 3-bit unequal-interval quantization. When the error is low, NMSE, SGCS, or GCS are divided into fewer intervals, each with a larger span. As the error exceeds a certain threshold, the quantization intervals become finer, providing higher resolution to more accurately reflect the prediction error of the AI ​​model. In this way, the network can more accurately obtain the precision information of the AI ​​model while avoiding redundant data transmission when the error is small, improving the system's reporting efficiency and resource utilization. Compared to traditional equal-interval quantization or simple unequal-interval quantization methods, this scheme can dynamically adjust the quantization interval, reducing unnecessary data transmission at low errors and improving measurement accuracy at high errors, thereby optimizing the CSI prediction performance of the AI ​​model and enhancing the network's intelligent scheduling capabilities.

[0289] Table 4: Mapping relationship of 3-bit unequal interval quantization.

[0290] In the above schemes, the calculation methods for NMSE, SGCS, or GCS results can have various dimensions, including but not limited to: averaging across all sub-bands, i.e., the overall error mean calculated across all subcarriers; averaging across all RB levels, i.e., the average error calculated at the resource block level; averaging across all layers, i.e., calculating the overall error for all MIMO antenna layers; calculating for each layer individually, i.e., calculating the error separately for different MIMO antenna layers; averaging across all CSIs within the prediction window, i.e., averaging all predicted CSI data within the entire prediction window; and calculating for a specific CSI moment within the prediction window, i.e., calculating the error value at a specified moment.

[0291] Furthermore, the aforementioned CSI information can be a predicted or measured channel matrix, or a precoding matrix calculated based on the predicted or measured channel matrix. If the calculation of NMSE, SGCS, or GCS is based on the predicted or measured channel matrix, then the layer-level reporting amount is not involved in the CSI report.

[0292] Assuming that the evaluation method for monitoring performance uses the calculation results of NMSE, SGCS, or GCS between the predicted CSI and the associated measured CSI, and that the accuracy information is reported using X-bit quantization, then the specific reporting content for monitoring performance indicators should at least include the following information, and at least one of the following schemes can be adopted:

[0293] Option 1: In some examples, the performance monitoring results include, within the monitoring window, the performance monitoring results between at least one predicted CSI and the associated actual measured CSI, the performance monitoring results including information on all layers and / or all subbands corresponding to the CSI.

[0294] For example, within each monitoring window, at least the NMSE, SGCS, or GCS between a CSI and its associated measured CSI should be calculated. This result should cover all MIMO layers and / or all subbands corresponding to the CSI. For instance, the average of the NMSE, SGCS, or GCS of the channel state information for all layers and / or all subbands can be used to provide a comprehensive assessment of the AI ​​model's prediction error.

[0295] For example, assuming there are M predicted CSIs within the monitoring window, the reported NMSE, SGCS, or GCS results will require M×X bits, where M is the number of CSIs to be reported within the monitoring window, and satisfies M≤N4 (N4 is the number of CSIs within the prediction window); X is the number of quantization bits for a single NMSE, SGCS, or GCS value.

[0296] Furthermore, to reduce uplink transmission overhead, an optimized reporting strategy can be adopted, such as reporting only the NMSE, SGCS, or GCS results between the CSI at the last moment of the prediction window and its associated measured CSI. This approach reduces the number of bits required while still providing relatively accurate accuracy information, allowing the network side to evaluate and optimize the prediction accuracy of the AI ​​model.

[0297] By using the above methods, we can reduce uplink signaling overhead, improve the adaptability of CSI prediction, and optimize radio resource management while ensuring the integrity of CSI prediction accuracy information, thereby improving the prediction efficiency of AI models and the resource scheduling capabilities of the network side.

[0298] Option 2: In some examples, the performance monitoring results include, within the monitoring window, the performance monitoring results between the precoding vector of at least one layer contained in the predicted CSI and the corresponding layer of the actual measured CSI, the performance monitoring results containing information on all subbands of the corresponding layer in the CSI.

[0299] For example, within each monitoring window, at least the NMSE, SGCS, or GCS results between the layers contained in a CSI and the precoding vectors of the corresponding layers in the associated measured CSI need to be calculated. These results should cover all subbands of the corresponding layer in the CSI. For instance, the average of the NMSE, SGCS, or GCS of the channel state information for all subbands corresponding to each layer can be used to comprehensively evaluate the error of the AI ​​model predictions.

[0300] For example, assuming there are N4 predicted CSIs within the monitoring window, and each CSI corresponds to L layers, the reported NMSE, SGCS, or GCS results require M×L×X bits, where M is the number of monitored CSIs that need to be reported within the monitoring window, and satisfies M≤N4 (N4 is the number of CSIs within the prediction window); L is the number of layers corresponding to each CSI; and X is the number of quantization bits for a single NMSE, SGCS, or GCS value.

[0301] Furthermore, to reduce uplink transmission overhead, an optimized reporting strategy can be adopted. For example, only the NMSE, SGCS, or GCS results of the precoding vectors corresponding to each layer between the CSI at the last moment of the prediction window and its associated measured CSI can be reported. This approach can reduce uplink signaling overhead while ensuring the validity of accuracy information, enabling the network side to accurately assess the prediction accuracy of the AI ​​model and optimize the CSI prediction mechanism and resource management strategy accordingly.

[0302] By using the above methods, we can optimize the reporting of CSI prediction accuracy information, reduce network overhead, improve the accuracy of CSI prediction, optimize wireless resource scheduling, and thus enhance the prediction performance of AI models and the intelligent decision-making capabilities of the network side.

[0303] Option 3: Within each monitoring window, calculate the NMSE, SGCS, or GCS results between all predicted CSIs and their associated measured CSIs. This result should cover all layers and / or all subbands corresponding to all predicted CSIs. For example, the average NMSE, SGCS, or GCS of the channel state information for all layers and / or subbands across all CSIs can be calculated to provide a comprehensive assessment of prediction error.

[0304] For example, assuming there are N4 predicted CSIs within the monitoring window, the reported NMSE, SGCS, or GCS results require X bits, where X bits represent the overall NMSE, SGCS, or GCS values ​​of the channel state information for all layers and / or subbands of the N4 predicted CSIs.

[0305] This approach reduces the amount of data reported, enabling the network to quickly obtain the overall CSI prediction error across the entire monitoring window, which can then be used as a basis for AI model optimization and resource scheduling. Simultaneously, because the accuracy information of all CSIs is merged into a single overall value for reporting, this approach reduces uplink signaling overhead while improving the reporting efficiency of CSI prediction information, optimizing radio resource management, and enhancing the predictive adaptability of the AI ​​model.

[0306] Option 4: Within each monitoring window, calculate the NMSE, SGCS, or GCS results between the precoding vectors of all layers included in the predicted CSI and the corresponding layers of the measured CSI. This result should cover all subbands corresponding to each layer in the CSI. For example, the average NMSE, SGCS, or GCS of the channel state information for all CSIs and all subbands corresponding to each layer can be calculated to provide a more comprehensive assessment of prediction error.

[0307] For example, assuming there are N4 predicted CSIs within the monitoring window, and each CSI corresponds to L layers, the reported NMSE, SGCS, or GCS results require L×X bits, where X bits represent the NMSE, SGCS, or GCS values ​​of the channel state information of all subbands in each layer within the N4 predicted CSIs.

[0308] Furthermore, the aforementioned subbands can be selected according to different application scenarios. For example, the CQI subband is used for scheduling optimization based on CSI prediction; the PMI subband is used for optimizing MIMO transmission strategies; and the RB-level subband is used for finer-grained CSI error assessment. In terms of time window configuration, the monitoring window can correspond to the CSI prediction window to ensure alignment between the monitoring period and the prediction period, thereby improving the accuracy and comparability of precision information.

[0309] Furthermore, the above-mentioned content is reported within a single reporting instance, and this reported content can consist of multiple parts to ensure data integrity and resolvability. The length of the monitoring window can be configured by the network side to match different prediction requirements. For example, when inference and model / functional monitoring use the same CSI-RS measurement resources (such as using the same periodic or semi-persistent CSI-RS resources), the length of the monitoring window can be consistent with the prediction window, i.e., M = N4. The values ​​of M and N4 can be any of {1, 2, 4, 8}, and the specific configuration can be optimized by the network side according to system requirements to improve the accuracy of model monitoring and reduce uplink signaling overhead.

[0310] Considering that when N4 values ​​are large, reporting all monitoring results in a single reporting instance may increase the uplink transmission burden. Therefore, the terminal side can adopt methods such as batch reporting, compressed reporting, and event-triggered reporting to reduce unnecessary data overhead, improve reporting efficiency, and optimize the use of system resources.

[0311] The above solution ensures that the accuracy of CSI predictions can be accurately and efficiently fed back to the network side, while taking into account resource optimization, uplink transmission efficiency and prediction accuracy, providing reliable data support for the optimization of AI models.

[0312] Option 5: The network side can determine the maximum number of K reported by user equipment through configuration or predefinition. max Each performance monitoring result, and K max A performance monitoring result can be K max The performance monitoring results of the predicted CSI can also be K. max All performance monitoring results predicting CSI within a performance monitoring window. The user equipment reports K performance monitoring results based on the monitoring data, where K ≤ K. max It may also report the index information corresponding to the performance monitoring results, as well as the number of K values. For example, when the monitoring window contains N4 predicted CSIs, the user equipment reports the performance monitoring results of K predicted CSIs, along with the index information of those K predicted CSIs within the monitoring window, and may also include the number of K values. Under this scheme, K... max The value of K can be configured on the network side or determined by the user equipment. The network side configures it through RRC, MAC CE, or DCI, where K... max ≤N4. For example, when N4=8, K max =8 or 4. Performance monitoring results can be the quantized result of the NMSE or SGCS value between the predicted CSI and the associated actual measured CSI, or the quantized result of other performance indicators; the specific value method is not limited. Furthermore, if N4 ≤ 2 or 4, all performance monitoring results of predicted CSI must be reported.

[0313] Option 6: The network side can determine the K performance monitoring results to be reported by the user equipment through configuration or predefined methods. These K performance monitoring results can be K predicted CSIs, or they can be all predicted CSIs within the K performance monitoring windows. The user equipment reports these K performance monitoring results based on the monitoring data, and may also report the corresponding index information. For example, if the monitoring window contains N4 predicted CSIs, the user equipment reports the performance monitoring results for K predicted CSIs, and also reports the index information of these K predicted CSIs within the monitoring window, where K ≤ N4. Under this scheme, the value of K can be configured on the network side via RRC, MAC CE, or DCI. For example, when N4 = 8, K = 8 or 4. Performance monitoring results can be the quantized result of the NMSE or SGCS values ​​between the predicted CSI and the associated actual measured CSI, or the quantized result of other performance indicators; the specific value method is not limited. Furthermore, if N4 ≤ 2 or 4, all performance monitoring results for predicted CSI must be reported.

[0314] In addition, the above content can be reported in conjunction with the predicted CSI or reported independently. The specific triggering and reporting methods can adopt at least one of the following schemes to optimize reporting efficiency and adapt to different network requirements.

[0315] Option 1: Periodic or semi-continuous prediction of CSI reporting, and joint reporting of monitoring performance results.

[0316] In some examples, for periodic or semi-continuous predicted CSI reporting, the performance monitoring results are reported in a periodic or semi-continuous manner, and the predicted CSI is reported jointly with the monitored performance results.

[0317] In this scheme, both the reporting of performance monitoring results and the reporting of predicted CSI are carried out in a periodic or semi-continuous manner, and the two are reported jointly.

[0318] Both reporting can be completed in the same reporting instance, that is: joint reporting on the same Physical Uplink Control Channel (PUCCH), joint reporting on the same Physical Uplink Shared Channel (PUSCH), or configuration under the same CSI-ReportConfig field.

[0319] In the CSI-ReportConfig field, the reported quantity can be configured as "P-CSI-And-MR", indicating that predicted CSI and performance monitoring results are reported jointly. (Note: "P-CSI-And-MR" is just an example name; other names can be used.)

[0320] Since there are no predictive CSI and actual measured CSI available for comparison within the initial measurement window, no performance monitoring results are reported during this period. To distinguish the reported content, a 1-bit instruction can be added to the beginning of the reported content to indicate whether the current joint report includes performance monitoring results, ensuring the accuracy of data parsing.

[0321] Option 2: Periodic or semi-continuous prediction of CSI reporting, with independent reporting of monitoring performance results.

[0322] In some instances, for periodic or semi-persistent predicted CSI reporting, the reporting of performance monitoring results is periodic, semi-persistent, or aperiodic, and the predicted CSI is reported independently of the monitored performance results.

[0323] In this scheme, the reporting of predicted CSI is still periodic or semi-continuous, while the reporting of performance monitoring results can be periodic, semi-continuous, or aperiodic, and the two are reported independently.

[0324] The reporting of predicted CSI and the monitoring performance results each correspond to a reporting instance, meaning that the two can be reported on different PUCCHs, or the predicted CSI can be reported on the PUCCH while the monitoring results are reported on the PUSCH, ensuring that the reporting of monitoring results will not affect the regular reporting of predicted CSI.

[0325] In the CSI-ReportConfig field, the reporting quantity can be configured as follows: "P-CSI" indicates reporting predicted CSI; "Performance-metric" indicates reporting performance monitoring results.

[0326] The advantage of this scheme is that it allows for flexible configuration of the reporting time of CSI prediction and monitoring results, thereby optimizing the allocation of uplink channel resources and reducing the impact of invalid data transmission on reporting.

[0327] Option 3: Non-periodic prediction of CSI and non-periodic monitoring of performance results.

[0328] In some examples, for non-periodic predicted CSI reporting, the reporting of performance monitoring results is non-periodic, and the reporting of predicted CSI and the reporting of monitoring results are triggered by the same trigger instance.

[0329] In this scheme, both the predicted CSI and the monitoring performance results are reported aperiodically. The reporting of both predicted CSI and monitoring performance results can be triggered by the same trigger instance; that is, when an event (such as the CSI prediction error exceeding a threshold) occurs, the network side requires the UE to report both the predicted CSI and the monitoring results simultaneously. However, the reporting of predicted CSI and monitoring performance results can be carried on different PUSCHes to ensure the independence of data parsing and avoid channel conflicts or data loss caused by joint reporting.

[0330] The above-mentioned solutions flexibly adapt to different CSI prediction and monitoring scenarios, providing the following options: Joint reporting (Solution 1): reducing control signaling overhead and improving data reporting efficiency; Independent reporting (Solution 2): ensuring the independence of predicted CSI and monitoring results, providing greater flexibility and improving uplink channel utilization; Aperiodic triggering (Solution 3): suitable for dynamically changing CSI prediction needs, improving the adaptability and flexibility of monitoring. These solutions can be configured appropriately according to network requirements, optimizing the utilization of wireless resources while ensuring the accuracy of AI model predictions, and improving the applicability and reliability of CSI feedback.

[0331] Fifth Example: Reporting content during performance monitoring (Type 1).

[0332] Regarding the reporting method of performance monitoring results, according to the standard discussion, periodic, semi-persistent, or non-periodic reporting mechanisms can be adopted. However, the specific content of the performance monitoring results still needs further definition to ensure that the network side can accurately understand and process this data. At the same time, the relationship and reporting format between the reporting of performance monitoring results and the reporting of model inference output also need to be further clarified. For example, how the network side distinguishes the reporting content under different LCM states, and how to ensure the validity and consistency of various types of reported information, are issues that the standard needs to discuss in depth. Therefore, some embodiments of this application provide corresponding solutions; for specific solutions, please refer to the fifth embodiment for a more complete design concept and implementation method.

[0333] When using user equipment-side models for performance monitoring of AI-based CSI prediction, it is necessary to evaluate the similarity between the predicted CSI and the associated measured CSI based on certain performance metrics, such as NMSE, SGCS, or GCS. However, directly reporting NMSE, SGCS, or GCS values ​​can lead to significant reporting overhead and may offer limited support for network-side decision-making, as the network typically does not require overly granular similarity comparisons during scheduling and resource allocation.

[0334] Based on the conclusions of Rel-19, the network side may configure corresponding threshold standards to assist user equipment in performing performance monitoring locally, thereby reducing the amount of data reported while ensuring that the prediction accuracy of the AI ​​model meets network requirements. Therefore, this embodiment proposes reporting performance monitoring results based on certain threshold standards. The specific reporting scheme can adopt at least one of the following methods:

[0335] This embodiment proposes at least one of the following performance monitoring result reporting schemes to optimize the reporting efficiency of CSI predictive performance monitoring.

[0336] Option 1: In some examples, within the monitoring window, the performance monitoring result between at least one predicted CSI and the associated actual measured CSI is compared with a threshold value to determine whether the performance monitoring result exceeds the threshold value. In some examples, the comparison result can be represented using 1 bit: bit = 1 indicates that the performance monitoring result is greater than the threshold value; bit = 0 indicates that the performance monitoring result is less than or equal to the threshold value.

[0337] For example, in AI-based CSI prediction, when user equipment performs performance monitoring, it can evaluate the similarity between the predicted CSI and the associated measured CSI, for example, by measuring NMSE, SGCS, or GCS. However, to reduce reporting overhead and avoid the network receiving too much irrelevant information, the monitoring results can be classified based on threshold values, and only binary threshold comparison results can be reported, i.e., whether the NMSE, SGCS, or GCS between the predicted CSI and the measured CSI exceeds a preset threshold value.

[0338] For example, within each monitoring window, the NMSE, SGCS, or GCS values ​​between the predicted result of at least one CSI and its associated measured CSI are calculated and compared with a threshold value. The specific comparison result can be represented using 1 bit: bit = 1 indicates that the NMSE, SGCS, or GCS value is greater than the threshold value; bit = 0 indicates that the NMSE, SGCS, or GCS value is less than or equal to the threshold value. This result includes information from all layers and / or all subbands corresponding to the CSI, and can be calculated using, for example, the average of the NMSE, SGCS, or GCS values ​​from the channel state information of all layers and / or subbands.

[0339] Assuming there are N4 predicted CSIs within the monitoring window, the reporting overhead of this scheme requires M bits, where M ≤ N4 (i.e., the number of CSIs that need to be reported within the monitoring window). To reduce the reporting overhead, one can choose to only report the comparison results of NMSE, SGCS, or GCS between the CSI at the last moment of the prediction window and its associated measured CSI.

[0340] Option 2: Based on the performance monitoring results at the CSI level, determine whether the performance monitoring results exceed the threshold value.

[0341] For example, within each monitoring window, the NMSE, SGCS, or GCS values ​​between the predicted CSI and the precoding vector of the corresponding layer of at least one CSI are calculated and compared with a threshold value. The specific comparison result can also be represented using 1 bit: bit = 1 indicates that the NMSE, SGCS, or GCS value is greater than the threshold value; bit = 0 indicates that the NMSE, SGCS, or GCS value is less than or equal to the threshold value.

[0342] The NMSE, SGCS, or GCS calculation results of this scheme include all layers of each CSI, and the average value is taken over all subbands. Assuming there are N4 predicted CSIs within the monitoring window, and each CSI has L layers, the reporting overhead of this scheme requires M×L bits, where M≤N4 (i.e., the number of CSIs that need to be reported within the monitoring window). To reduce the amount of reported data, only the comparison results of the NMSE, SGCS, or GCS of the precoding vectors of each layer of the CSI at the last moment of the prediction window and its associated measured CSI can be reported.

[0343] Option 3: Based on the performance monitoring results of all predicted CSIs, determine whether the performance monitoring results exceed the threshold value.

[0344] For example, within each monitoring window, the NMSE, SGCS, or GCS values ​​for all predicted CSIs are calculated and compared to a threshold value. The comparison result is represented using 1 bit: bit = 1 indicates that the NMSE, SGCS, or GCS value is greater than the threshold value; bit = 0 indicates that the NMSE, SGCS, or GCS value is less than or equal to the threshold value. The NMSE, SGCS, or GCS calculation results of this scheme include information from all layers and / or all subbands of all CSIs. For example, the average NMSE, SGCS, or GCS of all layers and / or subbands of all CSIs can be used for calculation. Assuming there are N4 predicted CSIs within a monitoring window, the reporting overhead of this scheme requires N4 bits to indicate whether the NMSE, SGCS, or GCS of each predicted CSI exceeds the threshold value. If only the NMSE, SGCS, or GCS values ​​of predicted CSIs need to be reported as greater than or equal to a threshold, compression can be achieved through index indication. For example: CSI indexing method: Indicates the location of CSIs greater than or equal to the threshold using an index. For instance, when N4 = 4, the CSI index requires 2 bits. Combination number indication method: Only reports the number of CSIs that meet the conditions, without listing the specific index, to further reduce the amount of data reported, lower indication overhead, and improve the flexibility of the indication.

[0345] Option 4: Based on the performance monitoring results of all predicted CSIs at the CSI level, determine whether the performance monitoring results exceed the threshold value.

[0346] For example, within each monitoring window, the NMSE, SGCS, or GCS values ​​between the precoding vectors of all predicted CSI layers are calculated and compared with a threshold value. The comparison result is also represented using 1 bit: bit = 1 indicates that the NMSE, SGCS, or GCS value is greater than the threshold value; bit = 0 indicates that the NMSE, SGCS, or GCS value is less than or equal to the threshold value.

[0347] The NMSE, SGCS, or GCS calculation results of this scheme include all layers of all predicted CSIs, and the average value is taken over all sub-bands. Assuming there are N4 predicted CSIs within the monitoring window, and each CSI has L layers, the reporting overhead of this scheme requires N4×L bits to indicate whether the NMSE, SGCS, or GCS of each layer of each CSI exceeds the threshold value. If it is only necessary to report which predicted CSIs have NMSE, SGCS, or GCS values ​​greater than or equal to the threshold value, the indication overhead can be reduced and the indication flexibility improved by using an index indication method. As described in Scheme 3, the CSI index method or the combination number indication method can be used for optimization to further reduce the indication overhead and improve the indication flexibility.

[0348] Option 5: The network side determines the maximum K reported by the user equipment through configuration or predefinition.max Performance monitoring results. K max A performance monitoring result can be K max The performance monitoring results of the predicted CSI can also be K. max All performance monitoring results predicting CSI are collected within a performance monitoring window. The user equipment reports K performance monitoring results based on these results, where K ≤ K. max Furthermore, it may simultaneously report the index information corresponding to the performance monitoring results and the number of representations of K. For example, if the monitoring window contains N4 predicted CSIs, the user equipment reports the performance monitoring results of K predicted CSIs, and simultaneously reports the index information of the K predicted CSIs in the monitoring window. In addition, it may also include information representing the number of representations of K. Under this scheme, K... max The value of K can be configured by the network side or determined by the user equipment. The network side can configure it through RRC, MAC CE, or DCI, where K... max ≤N4. For example, when N4=8, K max =8 or 4. Performance monitoring results can be the comparison of the NMSE or SGCS values ​​between the predicted CSI and its associated actual measured CSI with the threshold value, or the comparison of the values ​​of other performance indicators with the threshold value; no specific restrictions are imposed here. In addition, if N4 ≤ 2 or 4, all performance monitoring results of predicted CSI must be reported.

[0349] Option 6: The network side determines the K performance monitoring results to be reported by the user equipment through configuration or predefined methods. These K results can be the performance monitoring results of K predicted CSIs, or the performance monitoring results of all predicted CSIs within the K performance monitoring windows. The user equipment reports these K performance monitoring results based on the monitoring situation, and may also report the corresponding index information for each performance monitoring result. For example, if the monitoring window contains N4 predicted CSIs, the user equipment reports the performance monitoring results of K predicted CSIs, and simultaneously reports the index information of these K predicted CSIs within the monitoring window, where K ≤ N4. In this option, the value of K can be configured by the network side through RRC, MAC CE, or DCI. For example, if N4 ≤ 2 or 4, K can be 8 or 4.

[0350] In the above schemes, the CSI included in the monitoring window can be complete channel information or precoding matrix information obtained based on channel information. The definition of the monitoring window can be found in the seventh embodiment. The threshold values ​​in the above schemes can be configured by the network side or predefined. For example, the network side can configure them through RRC, MAC CE, or DCI, and the above schemes may correspond to different threshold values.

[0351] The fifth embodiment uses threshold-based comparison results for reporting to reduce data volume while ensuring the AI ​​model's prediction accuracy meets network requirements. The fifth embodiment provides several different performance monitoring and reporting schemes to optimize the reporting efficiency of CSI prediction performance monitoring. Scheme 1: Calculate the NMSE, SGCS, or GCS value of at least one predicted CSI within the monitoring window, compare it with the threshold value, and represent it with 1 bit (1 = exceeding the threshold, 0 = not exceeding the threshold). This scheme is applicable to the average results of all layers and / or sub-bands and can report only the predicted CSI at the last moment to reduce overhead. Scheme 2: Calculate independently for each layer of at least one CSI within the monitoring window, compare NMSE, SGCS, or GCS, and represent the result with 1 bit (1 = exceeding the threshold, 0 = not exceeding the threshold). This scheme is more granular, applicable to layer-level monitoring, and can report only the predicted CSI at the last moment to reduce data volume. Option 3: Calculate all predicted CSIs within the monitoring window, compare them with the threshold value, and represent the result with 1 bit (1 = exceeding the threshold, 0 = not exceeding the threshold). Applicable to the average result of all layers and / or sub-bands, the reporting overhead is N4 bits. The CSI indexing method or combination number indication method can be used to further reduce the indication overhead and improve the flexibility of the indication. Option 4: Calculate each layer of all predicted CSIs within the monitoring window and independently compare the NMSE, SGCS, or GCS values, using 1 bit (1 = exceeding the threshold, 0 = not exceeding the threshold). Applicable to detailed layer-level monitoring, the reporting overhead is N4×L bits. The amount of reported data can be optimized through the CSI indexing method or combination number indication method to improve the flexibility of the indication. Options 5 and 6 both involve user equipment reporting performance monitoring results based on network-side configuration or predefined methods. The difference is that Option 5 sets a maximum reporting quantity K. max (This can be determined by the network side, user equipment, or a predefined method), while Scheme 6 directly reports the K value through network side configuration or a predefined method, and both support the additional reporting of index information.

[0352] The fifth embodiment uses threshold-based CSI performance monitoring results reporting, which greatly reduces data overhead and ensures that the network side can effectively evaluate the accuracy of AI-predicted CSI, thereby achieving more optimized scheduling and resource management.

[0353] Sixth embodiment: Triggering of data collection.

[0354] Data collection primarily involves model training, model inference, and monitoring of the model or its functions. Since CSI predictions utilize models from the User Equipment (UE) side, the data collection triggering mechanism is crucial for ensuring the model's effectiveness and accuracy. Based on current standard discussions, data collection is primarily triggered in two ways: one is by network-side indication, and the other is based on an active request from the UE. For the UE-based request scenario, it is necessary to further clarify the content of the request information sent by the UE and the main carrier method of the request to ensure that the network side can correctly parse and respond. Therefore, some embodiments of this application provide corresponding solutions; for specific solutions, please refer to Embodiment Six for detailed design ideas and implementation methods.

[0355] In some instances, during model training, model inference, and / or model monitoring, the user equipment sends the data collection request message to the base station to trigger data collection, and / or receives reference signals from the base station for training, inference, and / or monitoring.

[0356] Based on current standard discussions, there are two main ways to trigger data collection for AI-based CSI prediction: one is based on network-side instructions, and the other is based on user device-side requests. Furthermore, data collection involves multiple aspects such as model training, inference, and performance monitoring. This embodiment primarily considers the triggering methods for data collection during model training and performance monitoring, and focuses on designing a solution for data collection based on user device-side requests.

[0357] Specifically, this embodiment proposes at least one of the following feasible solutions:

[0358] Option 1: During model training, data collection is triggered by the user equipment sending a 1-bit request message. The user equipment uses this request message to notify the network side to activate the distribution of the corresponding reference resources for training. This 1-bit indication message is carried in a dedicated UCI message, and the specific delivery method can be based on periodic PUCCH resources. The corresponding PUCCH format can be selected from at least one of Format 0, Format 1, Format 2, Format 3, and Format 4.

[0359] To support the aforementioned data collection triggering methods, configuration can be performed using Radio Resource Control (RRC). The RRC configuration introduces the parameter 'PUCCHResourceConfig-AITraining' to define the PUCCH resources related to training data collection. This parameter can be associated with or not associated with a SchedulingRequestId to accommodate different resource scheduling requirements. Furthermore, the RRC parameter configuration must include at least periodicityAndOffset and / or PUCCH-ResourceID to precisely define the PUCCH resource allocation and scheduling strategy.

[0360] Option 2: During model monitoring, data collection is triggered by the user equipment (UE) sending a 1-bit request message. The UE notifies the network side of this request message to activate the corresponding monitoring result reporting. This 1-bit indication message is carried in a dedicated UCI message, and the specific carrying method can be based on periodic PUCCH resources. The corresponding PUCCH format can be selected from at least one of Format 0, Format 1, Format 2, Format 3, and Format 4. To support the above data collection triggering method, it can be configured via RRC. The parameter 'PUCCHResourceConfig-AIMonitoring' is introduced into the RRC configuration to define the PUCCH resources related to monitoring data collection. This parameter can be associated with or not associated with SchedulingRequestId to adapt to different resource scheduling requirements. Furthermore, the RRC parameter configuration includes at least periodicityAndOffset and / or PUCCH-ResourceID to precisely define the allocation and scheduling strategy of PUCCH resources.

[0361] Option 3: During model training, data collection is triggered by the user equipment sending a 1-bit request message. The user equipment uses this request message to notify the network side to activate the allocation of corresponding reference resources for training. Unlike Option 1, this 1-bit indication message is carried in a dedicated SR message. The specific delivery method is still based on periodic PUCCH resources, and the corresponding PUCCH format can be selected from at least one of Format 0, Format 1, Format 2, Format 3, and Format 4. To support this option, it can be configured via RRC, introducing the parameter 'DataCollectionRequest-AITraining' to define the PUCCH resources related to training data collection. This parameter must be associated with the SchedulingRequestId to ensure coordination between request scheduling and resource allocation. Furthermore, the RRC parameter configuration must include at least periodicityAndOffset and / or PUCCH-ResourceID for precise configuration of PUCCH resources.

[0362] Option 4: During model monitoring, data collection is triggered by the user equipment (UE) sending a 1-bit request message. The UE notifies the network side of this request message to activate the corresponding monitoring result reporting. This 1-bit indication message is carried in a dedicated UCI message, specifically based on periodic PUCCH resources. The corresponding PUCCH format can be selected from at least one of Format 0, Format 1, Format 2, Format 3, and Format 4. To support this option, configuration can be done via RRC. The RRC configuration introduces the parameter 'DataCollectionRequest-AIMonitoring' to define the PUCCH resources related to monitoring data collection. This parameter can be associated with or not associated with the SchedulingRequestId to adapt to different scheduling requirements. Furthermore, the RRC parameter configuration must include at least periodicityAndOffset and / or PUCCH-ResourceID to precisely define the allocation and scheduling strategy of PUCCH resources.

[0363] Option 5: During model inference, data collection is triggered by the user equipment sending a 1-bit request message, which prompts the network side to activate and distribute the corresponding reference resources for inference. This 1-bit indication message is carried in a dedicated UCI message, specifically using periodic PUCCH resources. The PUCCH format can be at least one of Format 0, Format 1, Format 2, Format 3, and Format 4, and the specific configuration can be defined through RRC. A new parameter, "PUCCHResourceConfig-AIInference," is introduced in the RRC configuration for configuring periodic PUCCH resources. This parameter can be associated with SchedulingRequestId or not. Furthermore, the RRC parameter configuration should at least include periodicityAndOffset and / or PUCCH-ResourceID to ensure proper management and scheduling of PUCCH resources.

[0364] Solution Six: A method is proposed to trigger data collection based on a 1-bit request message sent by the user equipment. Specifically, the user equipment carries this 1-bit indication information through a dedicated SR message, which can be transmitted through periodic PUCCH resources. The PUCCH format can include at least one of Format 0, Format 1, Format 2, Format 3, and Format 4, and the specific format can be configured via RRC. To support this mechanism, the parameter 'DataCollectionRequest-AIInference' is introduced into the RRC configuration. This parameter can be associated with the ScheduledRequestId, and the RRC parameter configuration must include at least periodicityAndOffset and / or PUCCH-ResourceID, thereby ensuring that the network side can activate the corresponding reference resource distribution based on the user equipment's request for data collection for model training.

[0365] Solution 7: A method is proposed to trigger data collection during model training, inference, and monitoring by sending a 2-bit request message from the user equipment. This 2-bit indication message is carried through a dedicated UCI message, specifically transmitted via periodic PUCCH resources. The PUCCH format can include at least one of Format 0, Format 1, Format 2, Format 3, and Format 4, and can be configured via RRC. To implement this mechanism, the parameter 'PUCCHResourceConfig-AI' is introduced into the RRC configuration. This parameter can be associated with or not associated with the ScheduledRequestId, and the RRC parameter configuration must include at least periodicityAndOffset and / or PUCCH-ResourceID. Furthermore, the encoding method of the 2-bit indication message can be further defined, such as "00" representing a data collection request during training, "01" representing a data collection request during inference, and "10" representing a data collection request during monitoring. This ensures that the network side can accurately activate the corresponding reference resource distribution to meet the data requirements of different application scenarios.

[0366] Option 8: This option also triggers data collection based on a 2-bit request message sent by the user equipment, but the difference lies in that this 2-bit indication message is carried through a dedicated SR message. Specifically, this information can be transmitted through periodic PUCCH resources and supports at least one of Format 0, Format 1, Format 2, Format 3, and Format 4, with the specific format configurable via RRC. To support this mechanism, the parameter 'DataCollectionRequest-AI' is introduced into the RRC configuration. This parameter can be associated with the ScheduleRequestId, and the RRC parameter configuration must include at least periodicityAndOffset and / or PUCCH-ResourceID to ensure that the network side can activate the corresponding reference resource distribution based on the user equipment's request, thereby meeting the data collection needs of different stages such as model training, inference, and monitoring.

[0367] Option 9: Data collection for model training, inference, and / or performance monitoring can be triggered by the user equipment sending a request message of at least 1 bit. This request message is carried via a MAC CE and includes at least one of the following: First, the MAC CE message must carry at least 1 bit of a request message to trigger data collection. Second, the purpose of this request message can clearly indicate whether it is a data collection request for model training, or for model inference and / or performance monitoring, or indicate through relevant fields that the MAC CE message is for data collection. The specific data collection type can be distinguished by multiple bits of the request message. For example, in Option 7 or Option 8, 2 bits are used to distinguish the purpose of the data collection request.

[0368] In addition, MAC CE messages can also carry information related to the amount of data required for data collection, such as the amount of data needed for model training. This can be reflected in the duration of data collection, the amount of data required (e.g., how many time slots of channel matrix information or precoding matrix information are needed), or the number of channel measurements required.

[0369] In all the above schemes, data collection during model training, inference, and monitoring can be achieved through user device requests. For example, when the value of 1 bit is 1, the user device requests data collection from the network side for model training, inference, or monitoring; when the value of the bit is 0, the network side does not respond to the information.

[0370] Furthermore, when the network receives another 1-bit message with a value of 1, the user device will stop requesting data collection from the network for model training, inference, or monitoring. In other words, the network will respond to the user device's data collection requests when it receives an odd number of 1-bit request messages with a value of 1; and will stop responding to the user device's data collection requests when it receives an even number of 1-bit request messages with a value of 1.

[0371] Similarly, for multi-bit request messages, when the network side receives a certain value of the multi-bit request information an odd number of times, it will respond to the user equipment's data collection request; while when the network side receives the same value of the multi-bit request information an even number of times, it will stop responding to the user equipment's data collection request.

[0372] Furthermore, in the above scheme, some data collection request messages do not include information related to the required data volume. In this regard, information related to the required data volume can be carried in the user equipment's capability reporting information, specifically reflecting the duration of data collection, the amount of data required for data collection (e.g., how many time slots of channel matrix information or precoding matrix information are needed), or the number of channel measurements required for data collection.

[0373] Alternatively, data volume-related information can also be carried in the user equipment's auxiliary information, which takes the same form as described above. This includes the duration of data collection, the amount of data required, and the number of channel measurements, to ensure that the network side can reasonably configure and manage the data collection process.

[0374] The sixth embodiment explores the triggering mechanism for data collection, focusing primarily on the triggering method of user device-side requests. It designs several different schemes to support data collection needs at different stages, such as model training, inference, and monitoring. Schemes one through five mainly use 1-bit request information, while schemes six through eight introduce more flexible 2-bit request information to differentiate between different application scenarios. These schemes are all carried through dedicated UCI or SR messages, combined with periodic PUCCH resource transmission, and rely on RRC parameter configuration, such as PUCCHResourceConfig-AI or DataCollectionRequest-AI, to achieve precise resource scheduling and management. Overall, these schemes provide a flexible and efficient mechanism, making data collection for AI models more accurate and efficient in wireless network environments. Through fine-grained request information and flexible resource scheduling, these schemes improve data collection efficiency, ensure data quality during model training, inference, and monitoring, while reducing network resource consumption and achieving intelligent data acquisition and optimization.

[0375] Seventh Implementation Example: Temporal Behavior and Triggering Methods During Model or Function Monitoring.

[0376] Regarding the reporting method of performance monitoring results, according to the standard discussion, periodic, semi-persistent, or non-periodic reporting mechanisms can be adopted. However, the specific content of the performance monitoring results still needs further definition to ensure that the network side can accurately understand and process this data. At the same time, the relationship and reporting format between the reporting of performance monitoring results and the reporting of model inference output also need to be further clarified. For example, how the network side distinguishes the reporting content under different LCM states, and how to ensure the validity and consistency of various types of reported information, are issues that the standard needs to discuss in depth. Therefore, some embodiments of this application provide corresponding solutions; for specific solutions, please refer to Embodiment Seven for a more complete design concept and implementation method.

[0377] Based on the discussion process of the Rel-19 standard, there are two main ways to trigger data collection for CSI prediction: one is through network-side indication, and the other is through user equipment requests. However, the specific air interface signaling design involved in data collection during training, inference, and monitoring still needs further discussion. Therefore, this proposal suggests that performance monitoring results can be reported jointly with predicted CSI, or independently. To address this, this embodiment mainly considers the air interface signaling interactions that may be involved in data collection during model or function monitoring. This embodiment further proposes some reporting behaviors and triggering methods related to model / function monitoring, specifically including at least one of the following methods.

[0378] Option 1: For the collection of model monitoring data, a network-side indication method is used for triggering. Specifically, this indication method can be implicit or explicit, depending on the resource configuration and reporting method during performance monitoring. For example, assuming that the measurement resource configuration during model inference is periodic, and the reporting of inference results is also periodic, the measurement resources used to obtain the actual CSI measurement during performance monitoring can reuse the CSI-RS measurement resource configuration during model inference. Based on the progress of the Rel-19 standardization discussion, for CSI prediction, the reporting method of model or function monitoring results can be periodic, semi-persistent, non-periodic, or event-triggered. However, the specific reporting behavior and triggering method of monitoring results still need further research.

[0379] For periodic or semi-continuous CSI prediction reporting, the performance monitoring results are also reported periodically or semi-continuously, and both are reported jointly, meaning the monitoring results and predicted CSI are transmitted within the same reporting instance. This joint reporting can be carried out via PUCCH or PUSCH and configured under the CSI-ReportConfig field, or the AI-CSI-ReportConfig field can be used to define the output of AI-related models through user device reporting configuration. If the model-predicted CSI and the model monitoring output are reported jointly, the reportQuantity field under the CSI-ReportConfig field can be configured as "PM-cri-RI-PMI-CQI", indicating that the reported CSI includes both the model monitoring output and the model inference result. This monitors the accuracy of historical predicted CSI while predicting CSI, ensuring the accuracy of predicted CSI and allowing for timely adjustments to the CSI prediction strategy. Furthermore, the number of monitored CSIs included in the monitoring window should be less than or equal to the number of CSIs in the prediction window to ensure reasonable data matching.

[0380] Option 2: For periodic or semi-continuous predictive CSI reporting, the performance monitoring results are also reported periodically or semi-continuously, and both are reported independently, meaning that predictive CSI and performance monitoring results each correspond to a separate reporting instance. Specifically, these two can be transmitted on the same or two PUCCHs, or predictive CSI can be reported via PUCCH, while monitoring results can be reported via PUSCH. Furthermore, each reporting instance corresponds to a CSI-ReportConfig, or an AI-CSI-ReportConfig used to configure the user device's reporting configuration for AI-related model outputs.

[0381] Furthermore, the reporting cycle for monitoring performance results is shorter than the reporting cycle for predicted CSI, especially when N4 is large, such as when N4 is 8. This design allows the network side to obtain monitoring performance results earlier and make LCM decisions based on them. Specifically, this can be achieved by shortening the monitoring window length to be shorter than the predicted CSI window length. For example, with the same CSI-RS resource period, if the number of CSI measurements included in the monitoring window is M, then M must satisfy M≤N4.

[0382] In addition, if both the predicted CSI reporting and the monitoring performance results reporting adopt a semi-persistent mode, and both are reported via PUCCH, the same MAC CE activation message can be shared to optimize signaling overhead and improve transmission efficiency.

[0383] For Scheme 1 and Scheme 2, if the user equipment reporting configuration used to configure the AI-related model output adopts the dedicated AI-CSI-ReportConfig, and assuming that the reporting of predicted CSI is semi-persistent, and the reporting of monitoring results is also semi-persistent and reported through PUCCH, then under the existing MAC CE activation message mechanism, the user equipment will not be able to distinguish whether the current MAC CE activation message is used to activate the traditional CSI reporting configuration (CSI-ReportConfig) or to activate the dedicated AI-CSI-ReportConfig.

[0384] To address this issue, a 1-bit indication can be added to the existing MAC CE activation message to clearly distinguish whether the current MAC CE activation message is for activating traditional CSI reporting or for activating AI-related dedicated AI-CSI-ReportConfig. For example, specific MAC CE configurations can be found in Table 5, where the newly added 1-bit indication field can be used to identify the applicable scope of the MAC CE, thereby ensuring that user equipment can correctly parse and execute the corresponding CSI reporting configuration.

[0385] Table 5: MAC CE Configuration.

[0386] The AI ​​field indicates whether the MAC CE message is applied to semi-persistent CSI reporting user equipment based on AI-based PUCCH activation / deactivation. If this field is AI (configured as 1), it means the MAC CE message is used to activate the AI-specific AI-CSI-ReportConfig. If the AI ​​field is configured as 1, then the S... i The field is used to indicate the activation / deactivation status of semi-persistent CSI reporting under AI-CSI-ReportConfigToAddModList. S0 is used to indicate the PUCCH resource in the specified BWP used for semi-persistent CSI reporting, and has the lowest AI-CSI-ReportConfigID in the list of those with type set to semiPersistentOnPUCCH.

[0387] Similarly, if the user device reporting configuration used to configure the AI-related model output is a dedicated AI-CSI-ReportConfig, and the predicted CSI reporting is semi-persistent, while the monitoring result reporting is also semi-persistent and reported on the PUSCH, or if one of them is semi-persistently reported on the PUSCH, then 1 bit can be added to the DCI trigger message to indicate that the DCI trigger message activates CSI reporting based on the AI-dedicated AI-CSI-ReportConfig.

[0388] In the above scheme, the reporting period for monitoring results can be less than or equal to, or greater than or equal to, the reporting period for predicted CSI, and the monitoring results reported by user equipment are related to the length of the monitoring window. The length of the monitoring window specifies the number of predicted CSIs corresponding to a single monitoring result report, and the specific monitoring window length can be defined in at least one of the following ways.

[0389] Method 1: The network side instructs the user equipment that the number of predicted CSIs included in the monitoring window corresponding to the performance monitoring result report is M. The value of M can be less than or equal to N4, where N4 represents the number of predicted CSIs or precoding matrices included in the prediction window. For example, the value of M can be at least one of {1, 2, 4, 6, 8, 12, 16, 24, 32}. Furthermore, when the prediction window contains a large number of predicted CSIs, to ensure timely and accurate prediction results, the value of M can be configured to be no greater than N4; conversely, to reduce the amount of monitoring results reported, the value of M can be configured to be no less than N4. For example, when N4 is small (e.g., N4 = 1), the value of M can be at least one of {1, 2, 4, 8, 12}; while when N4 is large (e.g., N4 = 8), the value of M can be at least one of {1, 2, 4, 8}.

[0390] Method 2: A predefined approach is used, specifying that the length of the monitoring window is the same as the length of the prediction window. That is, the number of CSIs monitored in the monitoring window equals the number of CSIs predicted in the prediction window, i.e., M = N⁴. Furthermore, when the prediction window contains a large number of predicted CSIs, delayed reporting of monitoring results may lead to a decrease in system performance. Therefore, a predefined approach can be used to specify that when the prediction window is long, the length of the monitoring window should not exceed the length of the prediction window. For example, when N⁴ = 4 or N⁴ = 8, the value of M can be at least one of {1, 2, 4} to ensure the timeliness of monitoring result reporting.

[0391] Method 3: The network side instructs the user equipment on the ratio of the monitoring window length to the predicted window length. For example, the ratio factor is f, and f can take at least one value from {1 / 4, 1 / 2, 1, 2, 3, 4}. This method allows for flexible adaptation to different application scenarios by adjusting the ratio factor, improving the system's adaptability.

[0392] In the methods described above, the values ​​of M or f can be configured via RRC, or multiple candidate values ​​can be configured via RRC and indicated via MAC CE or DCI. M represents the number of predicted CSIs included in the monitoring window, or the predicted CSIs corresponding to the reported monitoring results are associated with M consecutive time slot intervals, each consecutive time slot interval lasting d time slots. If the monitoring results are obtained based on periodic or semi-persistent CSI-RS measurement resources, then d corresponds to the measurement period of the CSI-RS resources; if the monitoring results are obtained based on aperiodic CSI-RS measurement resources, then the value of d ranges from {1, m}, where m represents the time slot interval corresponding to any two adjacent resources in the aperiodic CSI-RS resources, and the value of m can be 1 or 2.

[0393] It is important to note that in the above methods, if the reporting period of the monitoring results is the same as the reporting period of the predicted CSI, then the corresponding offset of their periods is also the same. In this case, the length of the monitoring window is determined based on the reference resource reported by the predicted CSI, that is, the reference resource reported by the monitoring results is the same as the reference resource reported by the predicted CSI.

[0394] For periodic or semi-persistent predicted CSI reporting, performance monitoring results can be reported periodically, semi-persistently, or aperiodically. In this case, the reporting of predicted CSI and performance monitoring results is done independently, each corresponding to a separate reporting instance. This means they can be reported on one or two PUCCHs respectively, or the predicted CSI can be reported on the PUCCH while the monitoring results are reported on the PUSCH. Each corresponds to a separate CSI-ReportConfig field. Under this configuration, the reporting quantity in CSI-ReportConfig can be configured as P-CSI and Performance-metric, representing the reported predicted CSI and performance monitoring results, respectively. Furthermore, depending on the UE's capability requirements, if the UE is configured with a CSI-ReportConfig that includes the higher-layer parameter N4 and whose reportQuantity is set to "cri-RI-PMI-CQI", it is assumed that the UE supports UE-side CSI prediction. The reported PMI represents the prediction precoding matrix associated with N4 consecutive time slot intervals, each with a duration of d time slots, where the value of N4 is configured by the higher-layer parameter N4 and can be {1,2,4,8}.

[0395] Option 3: For non-periodic predicted CSI reporting, performance monitoring result reporting is also non-periodic. In this case, predicted CSI reporting and monitoring result reporting can be triggered by a single trigger instance, but the two reports can be carried out on different PUSCHs to ensure reporting flexibility and optimized resource utilization.

[0396] Option 4: For semi-persistent predictive CSI reporting, performance monitoring results can be reported semi-persistently or aperiodically. In this case, the reporting of predicted CSI and performance monitoring results is done independently, each corresponding to a separate reporting instance. This means they can be reported on one or two PUCCHs respectively, or the predicted CSI can be reported on the PUCCH, while the monitoring results can be reported on the PUSCH. Each corresponds to a CSI-ReportConfig field. In CSI-ReportConfig, the reporting quantity can be configured as P-CSI or "cri-RI-PMI-CQI" and Performance-metric, representing the reported predicted CSI and performance monitoring results, respectively. This approach helps improve data accuracy and system flexibility, allowing CSI prediction and performance monitoring results to be configured and reported independently, better adapting to different network requirements and UE capabilities.

[0397] The seventh embodiment focuses on the temporal behavior and triggering methods during model or functional monitoring. It primarily discusses the air interface signaling interaction methods for data collection and proposes multiple reporting methods, including joint reporting and independent reporting, as well as the bearer methods (PUCCH or PUSCH) for different reporting instances. Solution 1 uses a network-side indication method to trigger the collection of model monitoring data and supports periodic, semi-persistent, non-periodic, or event-triggered monitoring result reporting, ensuring flexibility in data acquisition. Solution 2, in periodic or semi-persistent predictive CSI reporting scenarios, adopts an independent reporting mode, enabling the network side to obtain monitoring performance data earlier for optimized decision-making. Solution 3 is suitable for non-periodic predictive CSI reporting and allows control of data transmission on different PUSCHs through trigger instances, improving resource utilization. Solution 4 focuses on semi-persistent predictive CSI reporting, employing an independent reporting method to improve the independence of CSI prediction and monitoring performance data, adapting to various network configurations. Regarding CSI reporting for AI-related models, the seventh embodiment also proposes adding 1 bit of indication information to the MAC CE or DCI trigger message to ensure correct parsing and execution of different reporting configurations. The seventh embodiment optimizes the acquisition of CSI prediction and performance monitoring data through flexible reporting methods, improving system adaptability and transmission efficiency while reducing signaling overhead, enabling CSI prediction and monitoring results to more reasonably match network requirements.

[0398] Eighth Implementation Example: Definition of the correlation between predicted CSI and measured actual CSI.

[0399] For monitoring models or functions, key performance indicators include NMSE and SGNC. However, when calculating these performance indicators, the measured actual CSI and the predicted CSI may not correspond perfectly in the time domain, posing a challenge to accurately evaluating the performance of the predictive model. Therefore, it is necessary to establish a certain correlation so that the measured CSI and the predicted CSI can be reasonably compared, thereby ensuring the accuracy and interpretability of the monitoring results. Especially when the two time domains do not completely overlap, how to reasonably define their correlation becomes an important issue that needs to be discussed in the standard. To address this issue, some embodiments of this application provide corresponding solutions. For specific solutions, please refer to Embodiment 8 for more detailed design ideas and implementation methods.

[0400] Considering that the time-domain location corresponding to the predicted CSI within the prediction window may not be completely consistent with the time-domain location of the measured actual CSI within the prediction window, it is necessary to establish a correlation between the two to more accurately determine the degree of approximation between the predicted CSI and the actual measured CSI. To this end, this embodiment proposes at least one of the following schemes to optimize the accuracy of predicted CSI and the rationality of performance monitoring.

[0401] Option 1: In some cases, if the predicted CSI and the actual measured CSI are located in the same time slot, then the predicted CSI is associated with the actual measured CSI. For example, if the predicted CSI and the actual measured CSI are located in the same time slot, they are directly associated, meaning that the predicted CSI is considered to correspond to the actual measured CSI within that time slot, and performance monitoring is evaluated based on this.

[0402] Option 2: In some cases, if the predicted CSI and the actual measured CSI are located in different time slots, the predicted CSI is associated with the actual measured CSI that has the smallest absolute time-domain deviation from the predicted CSI. For example, if the predicted CSI and the actual measured CSI are located in different time slots, the predicted CSI should be associated with the actual measured CSI that has the smallest absolute time-domain deviation from it. If there are two actual measured CSIs whose time-domain positions deviate from the predicted CSI by the same amount, the predicted CSI can be associated with the actual measured CSI located earlier or later in the time domain, or no constraint may be imposed, depending on the implementation of the user equipment. Alternatively, it is also possible to simultaneously associate two actual measured CSIs and evaluate the performance monitoring results of the predicted CSI by calculating the average of their NMSE, SGCS, or GCS.

[0403] Option 3: In some cases, if the predicted CSI and the actual measured CSI are located in different time slots, the predicted CSI is associated with the actual measured CSI that is later in the time domain and has the smallest time domain deviation from the predicted CSI. For example, if the predicted CSI and the actual measured CSI are located in different time slots, the predicted CSI is associated with the actual measured CSI that is later in the time domain and has the smallest time domain deviation from it. This approach helps to utilize updated data for comparison, improving the effectiveness and real-time performance of the predicted CSI.

[0404] Option 4: In some cases, if the predicted CSI and the actual measured CSI are located in different time slots, the predicted CSI is associated with the actual measured CSI that is earlier in the time domain and has the smallest time domain deviation from the predicted CSI. For example, if the predicted CSI and the actual measured CSI are located in different time slots, the predicted CSI is associated with the actual measured CSI that is earlier in the time domain and has the smallest time domain deviation from it. This method can complete the verification of the predicted CSI more quickly and reduce the impact of delay on monitoring accuracy.

[0405] Furthermore, the time-domain interval between the predicted CSI and the actual measured CSI cannot exceed a certain threshold T. The value of T can be predefined or configured by the network side. For example, the value of T can be at least one value in the range of 1 to 40 time slots or milliseconds (ms), or the value of T cannot exceed the time interval between two consecutive predicted CSIs, or half of the time interval. If this threshold is exceeded, the value of the performance monitoring NMSE, SGCS, or GCS is considered to be set to 0 or the lowest quantization value under the quantization method in the fourth embodiment, or, according to the fifth embodiment, it is considered to be below a certain threshold T, and therefore not included in the valid monitoring data.

[0406] When the temporal location deviation between the predicted CSI and the actual measured CSI is the same, it can be associated with any CSI, or a real measured CSI with high coherence to the predicted CSI can be selected. Furthermore, the system should provide explicit instructions to confirm which specific CSI is associated, ensuring the accuracy and consistency of the data analysis.

[0407] The eighth embodiment defines the correlation between predicted CSI and actual measured CSI to optimize the accuracy of predicted CSI and the rationality of performance monitoring. To address potential inconsistencies in the temporal positions of predicted CSI and actual measured CSI within the prediction window, several correlation schemes are proposed: Scheme 1 directly correlates predicted CSI and actual measured CSI within the same time slot; Scheme 2 correlates the predicted CSI to the actual measured CSI with the smallest absolute value of temporal deviation, or correlates two CSIs simultaneously and calculates their average error; Scheme 3 selects the actual measured CSI with the later temporal position and the smallest deviation to improve the real-time performance of data matching; Scheme 4 selects the actual measured CSI with the earlier temporal position to reduce the impact of delay on monitoring accuracy. Furthermore, a maximum temporal interval T between predicted CSI and actual measured CSI is specified. The value of T can be predefined or configured by the network side. If the threshold is exceeded, the performance monitoring data will be considered invalid or at the lowest quantization level. If multiple CSIs have the same temporal position deviation, they can be correlated to the CSI with higher coherence, and clear indication information will be provided to ensure the accuracy and consistency of data analysis. This solution improves the matching accuracy of predicted CSI through flexible association strategies, ensures the reliability of performance monitoring, optimizes the efficiency of data analysis, and enhances the network side's ability to judge the accuracy of CSI predictions.

[0408] Ninth Implementation Example: Definition of Trigger Events for Reporting Performance Monitoring Results.

[0409] Performance monitoring results can also be reported using an event-based approach. However, the specific event types need further clarification to ensure the rationality and applicability of the reporting mechanism. Different types of events may trigger the reporting of performance monitoring results in different network environments and application scenarios. Therefore, the definition and classification of these events need to be further discussed and clarified in the standard. To this end, some embodiments of this application provide corresponding solutions, which analyze and optimize the possible event types in detail. For specific solutions, please refer to Embodiment 9 for a more complete design concept and implementation method.

[0410] In some examples, the performance monitoring results are triggered in at least one of the following ways: within a single or multiple consecutive monitoring windows, the predicted CSI and the associated actual measured CSI satisfy certain constraints; or between a single or multiple consecutive predicted CSI and the associated actual measured CSI, certain constraints are satisfied.

[0411] For example, in AI-based CSI prediction, the reporting of performance monitoring results can be triggered by events. However, the specific event definitions still need to be clarified. This embodiment mainly discusses the definition of triggering events for performance monitoring result reporting and proposes the following possible methods.

[0412] Triggering events based on statistical characteristics include: Event 1, if the variance of the performance monitoring results of predicted CSI within the monitoring window is not greater than a certain threshold value T; Event 2, if at least one performance monitoring result of predicted CSI within the monitoring window is not greater than a certain threshold value T; Event 3, if the performance monitoring results of P consecutive predicted CSI within the monitoring window are not greater than a certain threshold value T, and the value of N satisfies N≤N4; Event 4, if the average value of the performance monitoring results of predicted CSI within the monitoring window is not greater than a certain threshold value T; Event 5, if the performance monitoring results of M predicted CSI within the monitoring window are not greater than a certain threshold value T, and the value of N satisfies N≤N4.

[0413] Triggering events based on CSI performance change trends include: Event 6, if the absolute value of the difference between the performance monitoring results of two consecutive predicted CSIs within the monitoring window is greater than or equal to a certain threshold value T; Event 7, if the absolute value of the difference between the performance monitoring results of two or more consecutive predicted CSIs within the monitoring window is greater than or equal to a certain threshold value K, for example, the first and last predicted CSIs; Event 8, if the performance monitoring results of all predicted CSIs within the monitoring window are not greater than a certain threshold value T; Event 9, if the median of the predicted CSI performance monitoring results within the monitoring window is not greater than a certain threshold value T.

[0414] In addition, the triggering events based on specific CSI monitoring values ​​include: Event 10, if the performance monitoring result of the first predicted CSI in the monitoring window is not greater than a certain threshold value T; Event 11, if the performance monitoring result of the last predicted CSI in the monitoring window is not greater than a certain threshold value T.

[0415] For CSI performance changes across monitoring windows, possible triggering events include: Event 12, within P consecutive monitoring windows, the performance monitoring results of M consecutive or non-consecutive predicted CSIs at the same location are not greater than a certain threshold value T; Event 13, within P consecutive monitoring windows, at least one of the absolute values ​​of the mean of the differences between the performance monitoring results of P predicted CSIs at the same location, or the mean of the absolute values, is greater than a certain threshold value T, where any monitoring window can be used as a reference when calculating the difference, such as the first monitoring window; Event 14, within P consecutive monitoring windows, at least one of the mean values ​​of the performance monitoring results of P predicted CSIs at the same location is not greater than a certain threshold value T, such as the monitoring result of the first predicted CSI.

[0416] The triggering events for the CSI performance decline trend include: Event 15, within P consecutive monitoring windows, the performance monitoring result of the last predicted CSI in the previous monitoring window is better than the performance monitoring result of the first predicted CSI in the next monitoring window; Event 16, within P consecutive monitoring windows, the performance monitoring result of the predicted CSI continues to decline, where the performance monitoring result of the CSI can be the average of the monitoring results of all predicted CSIs within the monitoring window, or the monitoring result of at least one predicted CSI at the same position within the monitoring window; Event 17, within P consecutive monitoring windows, the performance monitoring result of the predicted CSI continues to decline to below a certain threshold, where the performance monitoring result of the CSI can be the average of the monitoring results of all predicted CSIs within the monitoring window, or the monitoring result of at least one predicted CSI at the same position within the monitoring window.

[0417] In addition, events based on continuity and threshold triggering include: Event 18, if the performance monitoring results of the predicted CSI within P consecutive monitoring windows are not greater than a certain threshold value T; Event 19, if the absolute value of the difference between the performance monitoring results of the predicted CSI within two consecutive monitoring windows is greater than or equal to a certain threshold value T; Event 20, if the difference between the performance monitoring results of the predicted CSI before and after two consecutive monitoring windows is greater than or equal to a certain threshold value T.

[0418] For a sustained decline in CSI, possible triggering events include: Event 21 describes a situation where the performance monitoring results of predicted CSI continuously decline over P or P consecutive monitoring windows, meaning each monitoring result is lower than the previous one, indicating a continuous deterioration in CSI prediction performance. Event 22 further specifies the endpoint condition for the decline based on Event 21, namely, that the performance monitoring results of predicted CSI continuously decline over P or P consecutive monitoring windows, and the last monitoring result is less than or equal to a certain threshold value T. This means that CSI prediction not only continues to decline throughout the monitoring process but also eventually reaches a specific critical level. Event 23 is also based on the downward trend of Event 21 but adds a requirement for the magnitude of the decline. Specifically, the performance monitoring results of predicted CSI continuously decline over P or P consecutive monitoring windows, and the overall magnitude of the decline is greater than or equal to a certain threshold value T. This not only indicates that CSI prediction performance is continuously deteriorating but also requires that the degree of deterioration reaches or exceeds a certain specific value.

[0419] Event 24 focuses on whether the performance monitoring results of the predicted CSI fall below a certain threshold T multiple times within P consecutive monitoring windows. Specifically, this event requires that at least M monitoring results be less than or equal to the threshold T within the monitoring period, without emphasizing whether these points below the threshold are consecutive.

[0420] Event 25 further emphasizes continuity, requiring that within P consecutive monitoring windows, the performance monitoring results for predicted CSI must include M consecutive monitoring results less than or equal to a certain threshold T. This means that CSI prediction performance not only drops to the critical level multiple times, but these drops must also be continuous, further strengthening the monitoring requirements for severe declines in CSI prediction performance.

[0421] Event 26 describes a scenario where, within P or more consecutive monitoring windows, the percentage X of the predicted CSI performance monitoring results is less than or equal to a certain threshold T. The value of X can be predefined or configured by the network side, for example, X = 0.5.

[0422] Event 27 describes that within P or more consecutive monitoring windows, the mean of the predicted CSI performance monitoring results is less than or equal to a certain threshold value T.

[0423] Event 28 describes a situation where, within P or more consecutive monitoring windows, the median of the predicted CSI performance monitoring results is less than or equal to a certain threshold value T.

[0424] Event 29 describes the variance of the predicted CSI performance monitoring results being less than or equal to a certain threshold value T within P or more consecutive monitoring windows.

[0425] In the above scheme, the performance monitoring results of the predicted CSI can be specifically represented as the NMSE, SGCS, or GCS values ​​between the predicted CSI and its associated measured CSI. Furthermore, within the monitoring window, the performance monitoring results of the predicted CSI can be calculated in two ways: one is based on the performance monitoring results corresponding to each predicted CSI within the monitoring window, and the other is based on the overall performance monitoring results corresponding to all predicted CSIs within the monitoring window, i.e., the summation or comprehensive calculation result of the NMSE, SGCS, or GCS values ​​corresponding to all predicted CSIs within the monitoring window.

[0426] In the above scheme, the performance monitoring results of predicted CSI within multiple consecutive monitoring windows can be the average of all predicted CSI monitoring results within the monitoring window; or the monitoring results of at least one predicted CSI at the same location within the monitoring window; or the monitoring results of all predicted CSI within the monitoring window; or the performance monitoring results of predicted CSI at a certain time domain location.

[0427] Furthermore, the values ​​of parameters P, M, and / or T in the above scheme can be configured by the network side, for example, through RRC, MAC CE, or DCI, or they can be configured using standard predefined methods. For example, the threshold value T can be in the range of T∈(0,1], the number of monitoring windows P can be in the range of P∈[1,1000], and the value of M can satisfy M≤P.

[0428] Furthermore, the definition of the monitoring window in the above scheme can be found in the seventh embodiment. In addition, this scheme is also applicable to the cases where the number of CSIs in the prediction window is N4=1 and N4>1, and in the above multiple events, at least two events can be combined to form new composite events to adapt to more complex performance monitoring needs.

[0429] Furthermore, in the above scheme, there are several different ways to select the starting point for counting the monitoring results of predicted CSIs within multiple consecutive monitoring windows. For example, the starting point for counting can be the predicted CSI associated with the latest measured CSI as the last CSI, or the monitoring window containing the last predicted CSI in the monitoring window associated with the latest measured CSI as the last monitoring window. Alternatively, the starting point can be the new measured CSI associated with the first predicted CSI after the latest measured CSI associated with the last predicted CSI in multiple consecutive predicted CSIs, or the first new monitoring window after the last monitoring window in multiple consecutive monitoring windows. Another approach is to use the first predicted CSI in multiple consecutive predicted CSIs where the performance monitoring result is less than the threshold value T as the starting point, or the first monitoring window in multiple consecutive monitoring windows where the performance monitoring result is less than the threshold value T as the starting point. If no performance monitoring result is less than the threshold value T in multiple consecutive predicted CSIs or monitoring windows, the counting can be restarted starting from the next predicted CSI or monitoring window. Specific examples of the above counting methods are shown in Figures 5A, 5B, and 5C.

[0430] Referring to Figure 5C, the aforementioned events can be triggered either for a single monitoring window or for each monitoring window individually. The monitoring window can be configured on the network side. Specifically, the monitoring window can be a fixed window pre-defined by the network side, within which monitoring is performed; alternatively, periodic monitoring can be used, where the monitoring window is always present and monitoring occurs within a fixed period. However, in the case of periodic monitoring, whether or not monitoring results need to be reported still depends on the event triggering conditions; that is, reporting is only performed when specific event conditions are met.

[0431] Furthermore, network-side decisions can be based on a comprehensive assessment of monitoring results from multiple events. For example, the network can evaluate performance results during periodic monitoring. If an event is detected during monitoring and its duration reaches a certain threshold, a report is triggered. Subsequently, the network can make further decisions based on the results of multiple event reports. Alternatively, user equipment can be considered for counting, whereby user equipment autonomously monitors CSI performance across multiple monitoring windows and decides whether to report based on the accumulated event triggers, thereby improving the flexibility of monitoring and decision-making.

[0432] In summary, the ninth embodiment primarily defines the reporting trigger events for performance monitoring results, covering various methods such as statistical characteristics, CSI performance change trends, specific CSI monitoring values, cross-monitoring window changes, CSI performance decline trends, continuity, and threshold triggering. The triggering conditions for these events can be calculated based on the NMSE, SGCS, or GCS values ​​between the predicted and measured CSI, and the monitoring window configuration can be preset by the network side or use a periodic monitoring method. The counting methods are also diverse, including triggering strategies based on the latest measured CSI, predicted CSI, or threshold values. Furthermore, network-side decisions can be based on the combined results of multiple events, while user equipment can also autonomously count and decide whether to report, enhancing flexibility and adaptability. The ninth embodiment provides a flexible and configurable performance monitoring mechanism that, through various trigger events and monitoring strategies, achieves accurate monitoring and efficient reporting of predicted CSI performance, contributing to the optimization of resource management and service quality in wireless communication networks.

[0433] Tenth Implementation Example: Reporting content and reporting method based on event reporting.

[0434] The content reported by user equipment may differ depending on the event. Furthermore, the method of carrying the reported content needs further clarification in event-triggered reporting scenarios. These factors directly affect the efficiency of network-side acquisition and processing of CSI information. Therefore, the standard needs to further discuss how to define the reported content of user equipment under different events, as well as the corresponding reporting methods, to ensure the effectiveness and consistency of information transmission. To this end, some embodiments of this application provide corresponding solutions, standardizing and optimizing the reported content of user equipment and its carrying methods for different events. For specific solutions, please refer to the tenth embodiment for more detailed design ideas and implementation methods.

[0435] For AI-based CSI prediction, performance monitoring result reporting can be triggered in an event-driven manner. Within this framework, various events may trigger user equipment (UE) to report performance monitoring results. However, the specific content that UEs need to report for different event types requires further discussion. Therefore, this embodiment mainly explores the main reporting content based on event reporting and proposes at least one of the following solutions.

[0436] Option 1: Report the corresponding event type based on the performance monitoring results.

[0437] In some cases, the decision is made based on the type of event reported by the performance monitoring results to whether to revert from AI-based predictive CSI reporting to traditional Doppler codebook CSI reporting.

[0438] For example, in this scheme, after a user equipment detects a specific event, it reports the corresponding event index. The network side can then decide whether to revert to the traditional CSI prediction scheme based on the reported event index. For cases with only one event type, a 1-bit information can be used as an indication, and the network side can then determine whether to revert to the traditional CSI prediction scheme.

[0439] If multiple event types exist, the user equipment needs to report the specific event index. There are several reporting methods available, such as using event index IDs, bitmaps, or combination numbers. Specifically, for example, if there are S>1 events, each event can be reported via... The indication can be provided using a single bit, or by using S bits to indicate each event separately, with each bit corresponding to one event. Alternatively, a combination of bits can be used for joint indication. For example, when there are S events, the user equipment may satisfy only one event, or multiple events, or even all events. Therefore, the specific event types satisfied can be jointly indicated using a combination of bits to more effectively express the event state. For specific indication methods, please refer to the implementation schemes in Table 6.

[0440] Table 6: Event indexes are combined using a combination of numbers to indicate events.

[0441] As shown in Table 6 above, when the maximum number of configured events is 1, it is necessary to... The required bit indication overhead is k bits, where k ≤ S when the maximum number of configured events is k. Or through The number of events is indicated by 1 bit. Assuming there are k events, the bit overhead is used to indicate which specific events correspond to which number of events.

[0442] Option 2: Directly report the status of the event.

[0443] In this scheme, regardless of the specific event that occurs, the user equipment only needs to report the event's status using 1 bit of information. When the bit is 1, it indicates that an event has occurred; if the event has not occurred, no report is made. This method simplifies the reporting content, reduces communication overhead, and is suitable for scenarios where distinguishing between event types is not critical.

[0444] Option 3: Periodic or semi-continuous event status reporting.

[0445] In this scheme, user equipment (UE) periodically or semi-persistently reports the status of events using a single bit. If an event condition is met, the reported bit is 1; otherwise, it is 0. The reporting period can be configured by the network side or predefined, for example, made the same as the reporting period for predicted CSI. Furthermore, this scheme supports joint reporting with predicted CSI, meaning that predicted CSI and event occurrence information are transmitted simultaneously within the same reporting instance. If a semi-persistent reporting method is used, an event can trigger the reporting instance, thereby simultaneously reporting relevant information for both predicted CSI and the monitored event.

[0446] Option 4: Periodic or semi-continuous event information reporting.

[0447] In this scheme, user equipment reports event information in a periodic or semi-persistent manner, with specific reporting content referring to Scheme 1. The reporting period can be configured by the network side or predefined, for example, set to be the same as the reporting period of predicted CSI, and supporting joint reporting with predicted CSI. If a semi-persistent reporting method is used, a single reporting instance can trigger simultaneous reporting of predicted CSI and monitored events, thereby improving reporting efficiency.

[0448] Option 5: User devices report specific event indexes and monitoring results simultaneously.

[0449] In some examples, the user device reports an event index, reports the status of the event, and / or reports the corresponding performance monitoring results.

[0450] In this scheme, the user equipment not only reports the specific event index, but also the corresponding monitoring results. The specific event index can be found in Embodiment 9, and the reporting method can be found in Scheme 1, i.e., through... Each bit indicates a specific event that has occurred. Furthermore, the reporting of monitoring results can refer to the fourth and fifth embodiments, which may include specific output content based on the performance monitoring results. For example, in Scheme 1, which supports S events, the user equipment can indicate the events that have occurred using an appropriate number of bits, and simultaneously provide performance monitoring data related to the events, so that the network side can perform further analysis and decision-making.

[0451] Option 6: User devices report the status of the event and simultaneously report the monitoring results.

[0452] This scheme is similar to Scheme 5, but adopts the event reporting method of Scheme 2. That is, the user equipment only indicates whether an event has occurred using 1 bit of information, without distinguishing the specific event type. In addition, the user equipment also needs to report relevant monitoring results simultaneously; the content of the specific monitoring results can be found in Embodiments 4 and 5. For example, if Scheme 2 uses 1 bit of information to indicate that an event has occurred, the user equipment also needs to report possible detailed information based on performance monitoring results to provide a more comprehensive basis for network optimization.

[0453] Option 7: The information reported by user devices includes at least one of the following: the type of event. Assume events can be categorized in different ways. For example, based on whether the performance monitoring results of the triggering event correspond to multiple predicted CSIs, events can be classified into multiple categories, such as short-term events and long-term events; or based on the severity of the event, events can be classified into multiple categories, such as Type 1 events and Type 2 events. Type 1 events are more severe and may directly affect the normal operation of the model or function, while Type 2 events are relatively less severe, and may only show that at least one predicted CSI performance monitoring result did not meet expectations, such as failing to reach a certain threshold. The reporting of event types can be achieved through specific indication information, such as the index information of the event type. For example, if there are two types of events, the event category can be indicated by 1 bit.

[0454] Event Index: Event index information can be indicated by event ID, bitmap, or combination number. For details, please refer to the indication method in Scheme 1.

[0455] Furthermore, the index of an event can also be the index of a specific event under a certain category of events. Event occurrence status: The status information of an event can be represented by 1 bit, where a bit of 1 indicates that the event has occurred, and a bit of 0 indicates that the event has not occurred.

[0456] Performance monitoring information for triggering events: This performance monitoring information may include at least one of the following: performance monitoring results under a single monitoring window, which may refer to the fourth embodiment and / or the fifth embodiment; performance monitoring results under multiple monitoring windows; performance monitoring results for at least one predicted CSI; performance monitoring results for predicted CSIs that directly trigger events; performance monitoring results for predicted CSIs that do not directly trigger events, such as performance monitoring results for other predicted CSIs located in the same performance monitoring window as the predicted CSIs that directly trigger events; performance monitoring results for multiple consecutive predicted CSIs; and performance monitoring results corresponding to the scheme, as described in the fourth embodiment and / or the fifth embodiment.

[0457] The index information of the predicted CSI corresponding to the performance monitoring results can include at least one of the following: index information of at least one predicted CSI under a single monitoring window; index information of at least one predicted CSI at the same position under multiple monitoring windows. For example, in P monitoring windows, each monitoring window corresponds to N4 predicted CSIs, then the index information of at least one predicted CSI among the N4 predicted CSIs in the P monitoring windows is reported. For example, it can be achieved through... Bits are used for indication; the index information of at least one predicted CSI among multiple consecutive predicted CSIs, for example, one predicted CSI among K consecutive predicted CSIs, can be obtained through... The information is indicated by bits; if multiple predicted CSIs are involved, it can be indicated by a bitmap, the index of the predicted CSIs, or the number of combinations.

[0458] Index information of monitoring windows: Assuming that the user equipment monitors multiple monitoring windows at the same time, if the monitoring results of at least one window need to be reported, the user equipment needs to report the index information of the corresponding monitoring window.

[0459] In the aforementioned schemes, whether to report the type and / or index of the event can be configured by the network side, for example, via RRC. Furthermore, whether to report the performance monitoring results of one or more predicted CSIs that directly trigger the event can also be configured by the network side.

[0460] Furthermore, in the above scheme, the definition of the monitoring window can be referred to the seventh embodiment, and the definition of the starting position of the monitoring window can be referred to the ninth embodiment.

[0461] In summary, the above-mentioned solutions provide reporting methods with different granularities and complexities. They can adopt the simplest 1-bit event occurrence status indication or a more detailed reporting method that combines event types and monitoring data to meet the needs of different network scenarios.

[0462] In summary, the tenth embodiment explores the reporting content and methods based on event reporting, proposing various reporting schemes to meet different communication needs. Schemes one through four primarily focus on reporting event types or event occurrence states, using one or more bits to indicate whether an event has been triggered, combined with periodic or semi-persistent reporting mechanisms. Schemes five and six further extend this approach, requiring user equipment to provide corresponding monitoring results while reporting event indexes, thereby enhancing the network's decision-making capabilities. These schemes can be used independently or optimized for different communication scenarios to ensure that AI-predicted CSI monitoring and reporting are both efficient and accurate, thus optimizing wireless network resource management and improving service quality. Scheme seven involves the type, index, and status of events reported by user equipment, as well as performance monitoring information for triggered events, and provides multiple indexing methods (ID, bitmap, combination number) to indicate specific events and predicted CSI monitoring results. Whether to report event types, indexes, and predicted CSI monitoring results for triggered events can be configured by the network side via RRC. The monitoring window and its starting position can be referenced in embodiments seven and nine, respectively. The tenth embodiment provides a flexible event-triggered reporting mechanism. By combining different reporting content and methods, it can achieve simplified reporting with minimal communication overhead, and can also provide detailed monitoring data when necessary, optimize the performance monitoring of AI prediction CSI, reduce terminal reporting overhead, and improve the intelligent control capability of wireless networks.

[0463] Eleventh embodiment: The method of carrying LCM decision instruction information.

[0464] Based on the performance monitoring results reported by user equipment, the network side needs to decide whether to fall back to the traditional CSI prediction and compressed reporting based on the Rel-18 Doppler codebook. This decision is crucial for ensuring the accuracy of CSI reporting and optimizing network performance. However, how to effectively instruct user equipment to fall back to the traditional CSI prediction method still needs further discussion in the standard to ensure the rationality and feasibility of the fallback mechanism. In this context, some embodiments of this application provide a corresponding instruction scheme that can clearly convey the fallback instruction to the user equipment, ensuring that the system can switch to the optimal CSI reporting strategy under appropriate conditions. For a detailed understanding of the design and implementation of the fallback mechanism, please refer to the eleventh embodiment.

[0465] Based on the current progress of standard discussions, the network will implement a rollback operation based on the performance monitoring results reporting decision function, i.e., a rollback from AI-based CSI prediction to the traditional Rel-18 Doppler codebook. However, the specific instructions on the network side and the main signaling bearer method are still uncertain. Therefore, this embodiment provides at least one of the following solutions:

[0466] Option 1: For AI-based CSI prediction, if the predicted CSI reporting is periodic or semi-persistent, assuming the network side determines based on performance monitoring results that a fallback to the Rel-18 Doppler codebook is necessary, and a fallback to AI-based CSI compression is also possible, but the CSI prediction uses a traditional method, the network side can implicitly instruct the user equipment to fall back to the traditional Rel-18 Doppler codebook. The periodic or semi-persistent CSI measurements can remain unchanged, i.e., the periodic or semi-persistent CSI-RS measurement resources remain unchanged. The network side activates semi-persistent or aperiodic Rel-18 Doppler codebook-based CSI reporting via MAC CE or DCI, and the CSI measurement resources corresponding to the CSI-ReportConfig associated with the MAC CE or DCI activation message are the same as the CSI-RS measurement resources corresponding to the AI-based CSI prediction reporting. The MAC CE or DCI activation message can reuse traditional signaling content or adopt the MAC CE activation message type in the seventh embodiment.

[0467] Furthermore, for periodic AI-based CSI prediction reporting, the network side can also revert to traditional Rel-18 Doppler codebook-based reporting via RRC reconfiguration.

[0468] In addition, the above scheme requires the network side to configure at least two related configurations for predictive CSI reporting for the user equipment during RRC configuration: one for AI-based predictive CSI reporting and one for traditional Rel-18 Doppler codebook CSI reporting. The reporting configuration for AI-based predictive CSI reporting refers to the reporting configuration method in the second embodiment. If there are also reports of monitoring results for AI functions or model outputs, these reports should also be deactivated.

[0469] Similarly, to switch from the traditional Rel-18 Doppler codebook to AI-based CSI prediction reporting, simply activate AI-based CSI prediction reporting via the MAC CE or DCI command. In this case, traditional Rel-18 Doppler codebook reporting will be deactivated or stopped.

[0470] Option 2: For AI-based CSI prediction, if the predicted CSI reporting is semi-persistent, assuming the network side determines based on performance monitoring results that it needs to fall back to the Rel-18 Doppler codebook, it may also fall back to AI-based CSI compression. However, if the CSI prediction adopts the traditional method, the network side can activate semi-persistent or aperiodic Rel-18 Doppler codebook-based CSI reporting through MAC CE or DCI. Furthermore, the CSI measurement resources corresponding to the CSI-ReportConfig associated with the MAC CE or DCI activation message are the same as the CSI-RS measurement resources corresponding to the AI-based CSI prediction reporting.

[0471] If the configuration for AI-based CSI prediction reporting also reuses the traditional CSI-ReportConfig, then the MAC CE or DCI activation message for activating Rel-18 Doppler codebook-based CSI reporting can simultaneously deactivate AI-based CSI prediction reporting. At the same time, other operations related to AI-based CSI prediction can be stopped using a predefined method, such as reporting performance monitoring results for AI CSI prediction, or training, inference, and monitoring of other AI-based CSI models. Specific MAC CE or DCI messages can take the following form, for example, the S... i The field corresponds to a CSI-ReportConfig index position of 0 for AI-based CSI prediction reporting, while the corresponding CSI-ReportConfig index position based on the Rel-18 Doppler codebook is configured to 1. For example, a new CSI request field can be added to the DCI message to indicate deactivation of AI-based CSI prediction reporting, while the original CSI request field indicates activation of Rel-18 Doppler codebook-based CSI reporting. Alternatively, a 1-bit indication of whether to perform a rollback operation can be used in the MAC CE or DCI message. It should be noted that the MAC CE or DCI messages in the above schemes may have different forms. This invention patent does not impose specific constraints, as long as the rollback operation is indicated by a MAC CE or DCI message used to activate traditional Rel-18 Doppler codebook-based CSI reporting, it is within the protection scope of this scheme.

[0472] If the AI-based CSI prediction reporting configuration is based on the AI-specific AI-CSI-ReportConfig, then the MAC CE or DCI activation message for activating Rel-18 Doppler codebook-based CSI reporting can simultaneously deactivate AI-based CSI prediction reporting. At the same time, other operations related to AI-based CSI prediction will be stopped based on predefined methods, such as reporting performance monitoring results for AI CSI prediction, or training, inference, and monitoring of other AI-based CSI models. Specific MAC CE or DCI messages can take the following form, for example, the S... i The field corresponds to the CSI-ReportConfig index position of the AI's CSI prediction report, which is configured as 0, while the CSI-ReportConfig index position based on the Rel-18 Doppler codebook is configured as 1. It's important to note that the AI-specific AI-CSI-ReportConfig index can be merged with the traditional CSI-ReportConfig index. For example, assume the traditional CSI-ReportConfig ID number is counted first, and then the AI-specific AI-CSI-ReportConfig index is counted based on the traditional CSI-ReportConfig ID number. Specifically, if the traditional CSI-ReportConfig ID index is 1 and 2, then the AI-specific AI-CSI-ReportConfig index is 3. Alternatively, the order can be reversed, or another S can be added to the MAC CE message. j The field is used to specifically indicate the activation or deactivation message for AI-based dedicated AI-CSI-ReportConfig reports; for example, a new CSI request field can be added to the DCI message to indicate the deactivation of the dedicated AI-CSI-ReportConfig corresponding to AI-based CSI prediction reporting, while the original CSI request field is used to indicate the activation of CSI reporting based on the Rel-18 Doppler codebook. Alternatively, a 1-bit indication can be used in the MAC CE or DCI message to indicate whether a rollback operation is performed. It should be noted that the MAC CE or DCI messages in the above schemes may have different forms, and this invention patent does not impose specific restrictions. As long as the rollback operation is indicated by the MAC CE or DCI message used to activate traditional CSI reporting based on the Rel-18 Doppler codebook, it is within the protection scope of this scheme.

[0473] The above scheme requires the network side to configure at least two related configurations for predictive CSI reporting for the user equipment during RRC configuration: one for AI-based predictive CSI reporting and one for traditional Rel-18 Doppler codebook CSI reporting. The reporting configuration for AI-based predictive CSI reporting refers to the reporting configuration method in the second embodiment. If there are also reports of monitoring results for AI functions or model outputs, these reports should also be deactivated.

[0474] Similarly, if switching from the traditional Rel-18 Doppler codebook to the AI-based CSI prediction method, it is only necessary to send the corresponding MAC CE or DCI activation command to activate the AI-based CSI prediction reporting. At this time, the field in the MAC CE or DCI activation command that indicates the rollback operation can be configured to 0 or 1.

[0475] Option 3: In some cases, for AI-based CSI prediction, if the reported predicted CSI is periodic, non-periodic, or semi-persistent, based on performance monitoring results, it is determined whether to revert from AI-based predicted CSI reporting to traditional Doppler codebook CSI reporting. The revert operation of the AI-based CSI prediction is performed by receiving MAC CE or DCI sent by the base station.

[0476] For AI-based CSI prediction, if the predicted CSI reporting is periodic, aperiodic, or semi-persistent, assuming the network side determines based on performance monitoring results that a fallback to the Rel-18 Doppler codebook is necessary, and a fallback to AI-based CSI compression is also possible, but the CSI prediction uses a traditional method, the network side can instruct the AI-based CSI prediction fallback operation via dedicated MAC CE or DCI commands. Considering that the main difference between AI-based CSI prediction reporting and Rel-18 Doppler codebook-based reporting lies in the different methods of calculating the predicted CSI, the CSI-ReportConfig associated with MAC CE or DCI remains unchanged; it only indicates which method is used to calculate the predicted CSI. Therefore, the MAC CE or DCI message may contain at least one of the following indicative information: Serving Cell ID, BWP ID, and index information related to the CSI-ReportConfig corresponding to the current AI-based CSI prediction reporting. This can be CSI-ReportConfig ID information, or other information that can explicitly or implicitly represent the CSI-ReportConfig ID corresponding to the AI-based CSI prediction reporting, such as S... iThis field corresponds to the CSI-ReportConfig used for AI-based CSI prediction reporting, and includes indication information to tell the user equipment which method to use to calculate the predicted CSI. For example, a 1-bit indication specifies whether to use AI-based CSI prediction or the traditional Rel-18 Doppler codebook-based calculation method; a bit of 1 / 0 indicates AI-based CSI prediction, and a bit of 0 / 1 indicates Rel-18 Doppler codebook-based calculation. Similarly, the MAC CE indication message can also be used to switch from the traditional Rel-18 Doppler codebook-based calculation method to AI-based CSI prediction.

[0477] If the indication is given via DCI, in addition to at least one indication information included in the traditional DCI message used to activate / deactivate semi-persistent or aperiodic DCI messages, an additional 1 bit is added to indicate whether AI-based CSI prediction or traditional Rel-18 Doppler codebook-based calculation is used. A bit of 1 / 0 indicates AI-based CSI prediction, and a bit of 0 / 1 indicates Rel-18 Doppler codebook-based calculation.

[0478] In addition, the MAC CE or DCI message may also indicate the number of CSIs predicted in the prediction window. The specific indication method can be based on the RRC configuration or the number of CSIs in the predefined prediction window, indicating the index information of selecting one of the CSIs. For example, if the number of CSIs in the RRC or the predefined prediction window is {1,2,4,8}, then the specific number of predicted CSIs can be indicated in the MAC CE or DCI message using 2 bits.

[0479] Option 4: In some cases,

[0480] For AI-based CSI prediction, assuming the network side instructs the user equipment's AI model to predict a maximum of N4 CSIs, and the network side observes from performance monitoring results that the first X predicted CSIs within the prediction window have good performance, while the subsequent N4-X predicted CSIs have poor performance, the network side can instruct the user equipment to reduce the number of output predicted CSIs. For example, when N4=8 or N4=4, if the network side finds from the performance monitoring results reported by the user equipment that the first X=4 / 2 / 1 or the first X=2 / 1 CSIs have good prediction results, while the subsequent predicted CSIs have poor results, the network side can instruct the user equipment's AI model to output X CSIs, or a maximum of X CSIs. The specific signaling instruction can be RRC, MAC CE, or DCI.

[0481] Furthermore, for the model's inference output, the network side can be configured with a maximum number N of predicted CSIs. max Or the maximum possible number N of predicted CSIs to be reported max However, the actual number of predicted CSIs reported by the user equipment depends on the user equipment's choice or on instructions from the network side. If based on the user equipment's choice, the user equipment includes the actual number of predicted CSIs reported in the compressed CSI information it submits, so that the network side can understand the specific reporting situation. Conversely, if based on instructions from the network side, the user equipment does not need to include the number of predicted CSIs reported in the compressed CSI information it submits, but instead reports the number directly as instructed by the network side.

[0482] Option 5: In some cases, for AI-based CSI prediction, if the reported predicted CSI is non-periodic, based on performance monitoring results, it is determined to fall back from AI-based predicted CSI reporting to traditional Doppler codebook CSI reporting. This is done by activating the codebook type via a DCI activation message, which is configured to match the CSI reporting configuration corresponding to the traditional Doppler codebook CSI reporting.

[0483] For AI-based CSI prediction, if the number of predicted CSIs generated by the AI ​​model inference is N4, and the first X predicted CSIs within the prediction window have first-class performance, while the next N4-X predicted CSIs have second-class performance, then the number of predicted CSIs output should be reduced.

[0484] For AI-based CSI prediction, if the CSI prediction reporting is non-periodic, assuming that the network side determines that it needs to fall back to the Rel-18 Doppler codebook based on the performance monitoring results, the network side can directly activate the CSI-ReportConfig configuration corresponding to the CSI reporting of a codebook type configured as the traditional Rel-18 Doppler codebook through the DCI activation message.

[0485] This scheme also requires the network side to configure at least two related configurations for predictive CSI reporting for the user equipment when configuring RRC. One is for AI-based predictive CSI reporting, and the other is for CSI reporting based on the traditional Rel-18 Doppler codebook. The reporting configuration for AI-based predictive CSI reporting refers to the reporting configuration method in the second embodiment.

[0486] The aforementioned solutions provide fallback mechanisms for different scenarios, ensuring that AI-based CSI prediction can flexibly switch to traditional Rel-18 Doppler codebook CSI reporting when performance is substandard. They also allow for a smooth switch between AI CSI prediction and Doppler codebook CSI to meet the dynamic needs of 5G networks. Through different signaling methods such as MAC CE, DCI, and RRC, the network side can effectively control the CSI reporting mode, optimizing the adaptability and controllability of AI CSI prediction.

[0487] The eleventh embodiment provides five fallback mechanisms based on performance monitoring results to ensure a flexible switch to traditional Rel-18 Doppler codebook CSI reporting when AI-based CSI prediction fails to meet performance standards. These schemes cover periodic, semi-persistent, and aperiodic CSI reporting modes and achieve a smooth switch between AI CSI prediction and Doppler codebook CSI through different signaling methods such as MAC CE, DCI, and RRC. By adaptively adjusting the CSI reporting method, the network side can dynamically optimize the stability of CSI prediction while ensuring the flexibility and reliability of the 5G network in different application scenarios. The eleventh embodiment, through its flexible fallback mechanism, achieves seamless switching between AI CSI prediction and Doppler codebook CSI, improving the adaptability and reliability of the 5G network, helping to optimize communication performance, reduce uncertainty caused by model inference, and reduce control signaling overhead.

[0488] Twelfth Implementation Example: Reporting Priority Issues.

[0489] Regarding AI-based CSI prediction, standard discussions primarily employ user equipment-side models. Within this framework, the types of data that user equipment needs to report include: predicted CSI, performance monitoring results, and joint reporting of predicted CSI and performance monitoring results. However, the priority of different data types in the reporting process needs further clarification. Specifically, predicted CSI provides information on future channel states, which is crucial for network optimization, while performance monitoring results are used to evaluate the accuracy of the prediction model and the network's operational status. Furthermore, joint reporting can simultaneously provide prediction information and monitoring data, enhancing the system's decision-making capabilities. Therefore, determining the reporting priority of these data in different application scenarios is critical and requires further in-depth analysis and optimization based on actual needs. Refer to Embodiment Twelve for details on clarifying the priority order of different types of reported content in practical applications.

[0490] In some examples, for event-triggered reporting, its reporting priority is the highest. For AI-based CSI prediction, the content reported by the user equipment includes the predicted CSI based on model inference and the monitoring performance results during the performance monitoring process. Based on the above, this embodiment mainly considers the priority issues of these two aspects of reported content, and specifically, at least one of the following solutions can be referred to:

[0491] Solution 1: Still use the existing CSI reporting priority formula: Pri iCSI (y,k,c,s) = 2·N cells ·M s ·y + N cells ·M s ·k + M s ·c + s.

[0492] Then, let the value of k corresponding to the reporting of the performance monitoring result be 1, and in other cases k = 2, so that the reporting priority of the monitoring result can be obtained between L1-RSRP or L1-SINR and traditional or AI-based CSI prediction.

[0493] Or let the value of k corresponding to the reporting of the performance monitoring result be 2, and in other cases k = 1, so that the reporting priority of the monitoring result is lower than that of L1-RSRP or L1-SINR and traditional or AI-based CSI prediction. [[ID=二十二]]

[0494] For this solution, the reporting of the performance monitoring result is equivalent to the reporting of CSI. For example, the aperiodic monitoring result reporting can be analogous to the aperiodic CSI reporting, and the semi-persistent or periodic monitoring result reporting can be analogous to the semi-persistent or periodic CSI reporting.

[0495] Solution 2: Still use the existing CSI reporting priority formula: Pri iCSI (y,k,c,s) = 2·N cells ·M s ·y + N | cells ·M s ·k + M s ·c + s.

[0496] The reporting of the performance monitoring result is associated with the time-domain behavior of different CSI reports. For A-CSI reporting, the value of y corresponding to the reporting of its performance monitoring result is 0 < y < 1; for SP-CSI on PUSCH reporting, the value of y corresponding to the reporting of its performance monitoring result is 1 < y < 2; for SP-CSI on PUCCH reporting, the value of y corresponding to the reporting of its performance monitoring result is 2 < y < 3. For the CSI reported periodically on PUCCH, the value of y corresponding to the reporting of its performance monitoring result is 3 < y < 4.

[0497] Alternatively, the reporting of the performance monitoring results is associated with the time-domain behavior of different CSI reports. For A-CSI reporting, the value of y corresponding to the reporting of its performance monitoring results is -1 < y < 0; for SP-CSI on PUSCH reporting, the value of y corresponding to the reporting of its performance monitoring results is 0 < y < 1; for SP-CSI on PUCCH reporting, the value of y corresponding to the reporting of its performance monitoring results is 1 < y < 2. For the CSI reported periodically on PUCCH, the value of y corresponding to the reporting of its performance monitoring results is 2 < y < 3.

[0498] Alternatively, the reporting of the performance monitoring results can be equivalent to a CSI report, but there is a certain priority relationship with the traditional CSI report. For example, the corresponding priority can be determined according to the time-domain behavior of the reporting of the monitoring results: if the reporting of the monitoring results is aperiodic, the value of y corresponding to the reporting is 0 < y < 1. If the reporting of the monitoring results is semi-persistent and on PUSCH reporting, the value of y corresponding to the reporting is 1 < y < 2. If the reporting of the monitoring results is semi-persistent and on PUCCH reporting, the value of y corresponding to the reporting is 2 < y < 3. If the reporting of the monitoring results is periodic and on PUCCH reporting, the value of y corresponding to the reporting is 3 < y < 4.

[0499] Alternatively, the reporting of the monitoring results can be equivalent to a CSI report, but there is a certain priority relationship with the traditional CSI report. For example, the corresponding priority can be determined according to the time-domain behavior of the reporting of the monitoring results: if the reporting of the monitoring results is aperiodic, the value of y corresponding to the reporting is -1 < y < 0. If the reporting of the monitoring results is semi-persistent and on PUSCH reporting, the value of y corresponding to the reporting is 0 < y < 1. If the reporting of the monitoring results is semi-persistent and on PUCCH reporting, the value of y corresponding to the reporting is 1 < y < 2. If the reporting of the monitoring results is periodic and on PUCCH reporting, the value of y corresponding to the reporting is 2 < y < 3.

[0500] Solution 3: Still use the existing CSI reporting priority formula: Pri iCSI (y,k,c,s) = 2·N cells ·M s ·y + N cells ·M s ·k + M s ·c + s.

[0501] Differentiate the reporting of performance monitoring results from the reporting of CSI and the reporting of L1-RSRP or L1-SINR, and let the value of k corresponding to the reporting of performance monitoring results be 1 < k < 2, indicating that the priority of reporting performance monitoring results is higher than that of CSI reporting but lower than that of reporting L1-RSRP or L1-SINR.

[0502] Or differentiate the reporting of performance monitoring results from the reporting of CSI and the reporting of L1-RSRP or L1-SINR, and let the value of k corresponding to the reporting of performance monitoring results be 0 < k < 1, indicating that the priority of reporting performance monitoring results is higher than that of CSI reporting and reporting L1-RSRP or L1-SINR.

[0503] For this solution, the reporting of performance monitoring results is equivalent to the reporting of CSI. For example, the reporting of aperiodic monitoring results can be analogized to the reporting of aperiodic CSI, and the reporting of semi-persistent or periodic monitoring results can be analogized to the reporting of semi-persistent or periodic CSI. However, at the same time, compared with the traditional CSI reporting, there are certain differences in its priority, which can be specifically distinguished by the value of k.

[0504] Solution 4: The reporting of performance monitoring results is equivalent to a type of CSI reporting, but there is a certain priority relationship with the traditional CSI reporting, and its priority can be expressed by the following formula: Pri iCSI (y, k, c, s) = 2·N cells ·M s ·y + N cells ·M s ·k + M s ·c + s.

[0505] Where x is a number greater than or equal to 0 and less than or equal to 1, for example, x = 1 / 2; or x is a number less than or equal to 0 and greater than or equal to -1, for example, x = -1 / 2.

[0506] Furthermore, the above formula may also be subtracting x from the original value of y, where x is a number greater than or equal to 0 and less than or equal to 1, for example, x = 1 / 2; or x is a number less than or equal to 0 and greater than or equal to -1, for example, x = -1 / 2.

[0507] If the reporting of the monitoring results is aperiodic, the value of y corresponding to its reporting is 0; for the reporting of the monitoring results that are semi-persistent and reported on PUSCH, the value of y corresponding to its reporting is 1; for the reporting of the monitoring results that are semi-persistent and reported on PUCCH, the value of y corresponding to its reporting is 2; for the reporting of the monitoring results that are periodic and reported on PUCCH, the value of y corresponding to its reporting is 3.

[0508] Alternatively, associate the reporting of the performance monitoring results with the time-domain behavior of different CSI reports, where x is a number greater than or equal to 0 and less than or equal to 1, such as x = 1 / 2; or x is a number less than or equal to 0 and greater than or equal to -1, such as x = -1 / 2. That is, for the time-domain behavior of different CSI reports, the reporting of the corresponding monitoring results needs to follow the above formula.

[0509] The above scheme means that whether it is the CSI report obtained by the traditional method or the CSI report obtained based on AI, their reporting priorities are the same.

[0510] The reporting of the performance monitoring results is associated with the time-domain behavior of different CSI reports. For A-CSI reports, the value range of y corresponding to the reporting of the performance monitoring results is 0 < y < 1; for SP-CSI on PUSCH reports, the value range of y corresponding to the reporting of the performance monitoring results is 1 < y < 2; for SP-CSI on PUCCH reports, the value range of y corresponding to the reporting of the performance monitoring results is 2 < y < 3.

[0511] Or associate the reporting of the performance monitoring results with the time-domain behavior of different CSI reports. For A-CSI reports, the value range of y corresponding to the reporting of the performance monitoring results is -1 < y < 0; for SP-CSI on PUSCH reports, the value range of y corresponding to the reporting of the performance monitoring results is 0 < y < 1; for SP-CSI on PUCCH reports, the value range of y corresponding to the reporting of the performance monitoring results is 1 < y < 2.

[0512] The above scheme means that whether it is the CSI report obtained by the traditional method or the CSI obtained based on AI, their reporting priorities are the same.

[0513] Solution Five: Distinguish the reporting of the performance monitoring results from the reporting of CSI and the reporting of carrying L1-RSRP or L1-SINR. The specific formula can be in the following form: Pri iCSI (y,k,c,s) = 2·N cells ·M s ·y + N cells ·M s ·(k + x) + M s ·c + s. Where x is a number greater than or equal to 0 and less than or equal to 1, such as x = 1 / 2; or x is a number less than or equal to 0 and greater than or equal to -1, such as x = -1 / 2.

[0514] Furthermore, the above formula may also subtract x from the original value of k, where x is a number greater than or equal to 0 and less than or equal to 1, such as x = 1 / 2; or x is a number less than or equal to 0 and greater than or equal to -1, such as x = -1 / 2.

[0515] For this scheme, the reporting of performance monitoring results is equivalent to the reporting of CSI (Continuous System Indication). For example, reporting non-periodic monitoring results is analogous to non-periodic CSI reporting. Reporting semi-continuous or periodic monitoring results is analogous to semi-continuous or periodic CSI reporting. However, compared with traditional CSI reporting, there are certain differences in priority, which are specifically distinguished by the formula mentioned above.

[0516] Option Six: Differentiate between performance monitoring result reporting, CSI reporting, and reporting of L1-RSRP or L1-SINR. The specific formula can be in the following form: Pri iCSI (y,k,c,s)=2·N cells ·M s ·(y+z)+N cells ·M s ·(k+x)+M s ·c+s.

[0517] Where x is a number greater than or equal to 0 and less than or equal to 1, for example x = 1 / 2; or x is a number less than or equal to 0 and greater than or equal to -1, for example x = -1 / 2; z is a number greater than or equal to 0 and less than or equal to 1, for example z = 1 / 2; or x is a number less than or equal to 0 and greater than or equal to -1, for example z = -1 / 2.

[0518] Secondly, the above formula may also be based on the original value of k minus x, where x is a number greater than or equal to 0 and less than or equal to 1, for example x = 1 / 2; or x is a number less than or equal to 0 and greater than or equal to -1, for example x = -1 / 2; and based on y minus z, where z is a number greater than or equal to 0 and less than or equal to 1, for example z = 1 / 2; or x is a number less than or equal to 0 and greater than or equal to -1, for example z = -1 / 2.

[0519] Furthermore, if the monitoring results are reported non-periodicly, the corresponding y value is 0; if the monitoring results are reported semi-continuously and reported on PUSCH, the corresponding y value is 1; if the monitoring results are reported semi-continuously and reported on PUCCH, the corresponding y value is 2; and if the monitoring results are reported periodically and reported on PUCCH, the corresponding y value is 3.

[0520] Option 7: For scenarios involving the joint reporting of performance monitoring results and predicted CSI, both can be reported through a single reporting instance. User equipment can determine whether to perform joint reporting based on the reporting volume configured on the network side. For example, as mentioned in the second embodiment, when reportQuantity is configured as "PM-cri-RI-PMI-CQI", both are jointly reported.

[0521] However, in joint reporting, the performance monitoring results are coded independently as a separate part, analogous to part 1 in traditional CSI reporting, while predictive CSI can be divided into part 1 and part 2. The joint reporting of both can be carried in a single PUCCH or PUSCH. When reporting resources are limited, the performance monitoring results have higher priority than part 1 and part 2 in predictive CSI, and are encoded first during the encoding process. Furthermore, the monitoring results are either reported entirely or discarded entirely.

[0522] Alternatively, performance monitoring results can be encoded as a separate part, analogous to part 1 in traditional CSI reporting, and given the same priority as part 1 in predicted CSI reporting. When resources are limited, they can be discarded or reported together with part 1 of predicted CSI.

[0523] Alternatively, the performance monitoring results can be encoded as a separate part, analogous to part 1 in traditional CSI reporting, but with a lower reporting priority than part 1 in predictive CSI reporting, and the monitoring results can either be reported as a whole or discarded entirely.

[0524] The reporting of performance monitoring results in the above scheme can be event-triggered, or it can be periodic, semi-persistent, or non-periodic. For specific reporting content, please refer to the fourth, fifth, and / or tenth embodiments.

[0525] Furthermore, the reporting priority of performance monitoring results also applies to scenarios where CSI predictions and performance monitoring results are reported jointly.

[0526] In the twelfth embodiment, event-triggered reporting has the highest priority, particularly reporting of AI-based CSI predictions and performance monitoring results. Schemes one through four all utilize existing CSI reporting priority formulas, adjusting them for different variable values ​​to ensure reasonable priority allocation for performance monitoring result reporting under various circumstances. These schemes adjust the priority of performance monitoring results to maintain an appropriate order of priority relative to L1-RSRP, L1-SINR, and traditional or AI-predicted CSI reporting. Furthermore, these methods ensure consistent reporting priorities for traditional CSI and AI-predicted CSI, avoiding unfair resource allocation due to different data sources, thereby optimizing signaling management and resource scheduling in the wireless communication system. The twelfth embodiment ensures a reasonable priority allocation between performance monitoring results and CSI reporting, giving traditional CSI and AI-predicted CSI reporting consistent priority, thus optimizing signaling management and improving network resource utilization efficiency.

[0527] Thirteenth embodiment: Bearer method for event-triggered reporting.

[0528] In some examples, the monitoring results are event-triggered, and the uplink resources used for reporting these event-triggered monitoring results are indicated or requested through a Schedule Request (SR) or Uplink Control Information (UCI). For example, for AI-based CSI prediction, performance monitoring result reporting can be event-driven. When certain events are met, the user equipment reports relevant information to the network side or requests resources from the network side to report event-related information. Specific implementations can adopt at least one of the following schemes:

[0529] Option 1: When at least one event occurs, the user equipment requests PUCCH and / or PUSCH resources from the network side via a new SR message or a new UCI message for reporting performance monitoring results based on the event. The new SR or UCI message is carried on periodic PUCCH resources and can be configured via dedicated RRC signaling, using at least one of PUCCH format 0, 1, 2, 3, or 4. The new SR or UCI message requests the reporting of performance monitoring results from the network side using at least 1 bit. For example, it requests the report using 1 bit; if at least one event occurs, the corresponding bit is 1; otherwise, it is 0. If it is a new SR message, the parameter `reportResourceRequest-UEIAIEvent` is added to the RRC message, and this parameter is associated with a dedicated scheduling request ID, `SchedulingRequestId`. If a new UCI message is added, the parameter firstPUCCHResourceConfig-UEIAIEvent will be added to the RRC message, and this parameter must contain at least the following configuration information: periodicityAndOffset of the PUCCH resource, and PUCCH-ResourceID.

[0530] Option 2: When at least one event occurs, the user equipment (UE) informs the network side via a new SR message or a new UCI message to occupy PUCCH and / or PUSCH resources for reporting performance monitoring results based on the event. The new SR or UCI message is carried on periodic PUCCH resources and can be configured via dedicated RRC signaling, using at least one of PUCCH format 0, 1, 2, 3, or 4. The new SR or UCI message requests the reporting of performance monitoring results from the network side using at least one bit. For example, it requests the report using one bit; if at least one event occurs, the corresponding bit is 1; otherwise, it is 0. If it is a new SR message, the parameter `reportResourceIndicate-UEIAIEvent` is added to the RRC message, and this parameter is associated with a dedicated scheduling request ID (`SchedulingRequestId`). If a new UCI message is added, the parameter firstPUCCHResourceConfig-UEIAIEvent will be added to the RRC message, and this parameter must contain at least the following configuration information: periodicityAndOffset of the PUCCH resource, and PUCCH-ResourceID.

[0531] For the PUCCH or PUSCH resources used by user equipment to report performance monitoring results, Type 1 CG-PUSCH resources can be used, and the corresponding period can be the same as the period of adding new SR messages or adding new UCI messages.

[0532] Furthermore, the performance monitoring results triggered by the event can be jointly reported with the predicted CSI. For example, it can be stipulated that when at least one event occurs, the user equipment notifies the network side of the event occurrence through newly added SR information or newly added UCI information, and jointly reports based on the event's performance monitoring results and predicted CSI. The reported resources can reuse the resources corresponding to the predicted CSI report. In addition, the performance monitoring results are independently encoded as a whole, similar to part 1 information in traditional CSI reporting. The specific reporting content can be referred to in the tenth embodiment, and the corresponding reporting priority can be referred to in the twelfth embodiment.

[0533] Option 3: When at least one event occurs, the user equipment (UE) informs the network side of the event occurrence or its type by adding a new SR (Search Request) message or a new UCI (Unique Event Message). The new SR or UCI message is carried on periodic PUCCH resources and can be configured via dedicated RRC (Register Responsible Code) signaling, using at least one of PUCCH formats 0, 1, 2, 3, or 4. The new SR or UCI message uses at least one bit to inform the network side of the event's status or type. For example, it uses one bit to inform the network side of the event's status; if at least one event occurred, the corresponding bit is 1; otherwise, it is 0. If it is a new SR message, the parameter `eventStateIndicate-UEIAIEvent` is added to the RRC message, and this parameter is associated with a dedicated SchedulingRequestId. If a new UCI message is added, the parameter `firstPUCCHResourceConfig-UEIAIEvent` is added to the RRC message. This parameter must contain at least the following configuration information: the periodicity and offset information of the PUCCH resource, `periodicityAndOffset`, and `PUCCH-ResourceID`. For example, if multiple events exist, and at least one event occurs, the user equipment informs the network side of the type of event. The user equipment can indicate the specific event type using an event index, bitmap, or combination of events. For specific indication methods, refer to Example 10. If the indication is provided through a new SR message, the parameter `eventIdIndicate-UEIAIEvent` is added to the RRC message, and this parameter is associated with a dedicated scheduling request ID, `SchedulingRequestId`. If a new UCI message is added, the parameter `firstPUCCHResourceConfig-UEIAIEvent` is added to the RRC message. This parameter must contain at least the following configuration information: the periodicity and offset information of the PUCCH resource, `periodicityAndOffset`, and `PUCCH-ResourceID`.

[0534] The thirteenth embodiment describes the event-triggered reporting method, mainly targeting event-driven reporting of performance monitoring results based on AI-based CSI prediction. When certain events are met, the user equipment can report relevant information or request reporting resources to the network side. Option 1: The user equipment adds an SR or UCI message to request PUCCH / PUSCH resources for performance monitoring result reporting. This message is carried on periodic PUCCH resources and configured via dedicated RRC signaling, using PUCCH format 0, 1, 2, 3, or 4. The request information occupies at least 1 bit; if an event occurs, the corresponding bit is 1, otherwise it is 0. Option 2: The user equipment adds an SR or UCI message to directly indicate the occupancy of PUCCH / PUSCH resources for event-based performance monitoring result reporting. Similar to Option 1, this message can be configured via RRC signaling and carried using PUCCH format 0, 1, 2, 3, or 4. Option 3: The user equipment adds SR or UCI messages solely to inform of event occurrence or event type, using 1 bit to represent event status (occurred / not occurred) or an event index, bitmap, or combination number to represent the specific event type. Furthermore, event-triggered performance monitoring results can be jointly reported with predicted CSI. The user equipment notifies the event of occurrence in the added SR / UCI message and reuses the same reporting resources for the performance monitoring results and predicted CSI. Simultaneously, the performance monitoring results are encoded independently as a whole, similar to part 1 information in traditional CSI reporting. Specific reporting content can be found in Embodiment 10, and reporting priorities can be found in Embodiment 12. This option optimizes the reporting of performance monitoring results for AI-based CSI prediction through an event-driven approach, ensuring timely event reporting with limited resources while flexibly adapting to different types of CSI prediction needs, thus improving the overall efficiency of the communication system.

[0535] Fourteenth Implementation: CPU Usage Rules Based on AI-Based CSI Prediction.

[0536] In some examples, the model training, inference, and / or performance monitoring in the monitoring results meet certain CSI processing timeline requirements. For instance, AI-based CSI prediction CSI reporting can use the same reporting method as traditional Rel-18 Doppler codebooks; however, due to the difference in processing complexity between AI-based prediction CSI computation and traditional prediction CSI, the number of CSI processing units corresponding to AI-based prediction CSI reporting needs to be redefined. Similarly, for model monitoring reporting, CSI processing units also need to be defined. This embodiment addresses this through O... CPU To represent the number of CPUs used, at least one of the following solutions is provided:

[0537] Option 1: AI-based CSI Prediction Reporting For aperiodic CSI measurements, performance monitoring requires monitoring the performance of the CSIs output by the prediction window. Assuming the number of CSIs output by the prediction window is N⁴, the performance of these N⁴ predicted CSIs needs to be monitored. This performance monitoring process not only requires measuring the actual CSIs associated with the predicted CSIs but also comparing the similarity between the predicted CSIs and the associated actual measured CSIs. For example, similarity can be measured using NMSE, SGCS, or GCS. Therefore, when using aperiodic channel measurement resources, the number of CPUs required for reporting performance monitoring results can be counted using at least one of the following methods:

[0538] Method 1: Considering that the number of CPUs used for reporting performance monitoring results is related to the number of predicted CSIs that need to be monitored, or the number of CSIs that need to be measured in real channels, or the number of non-periodic CSI-RS resources configured for channel measurements, we can assume that the number of CPUs used is 0. CPU =N4, or the number of CPUs used is 0. CPU =K, where K represents the number of CSI-RS resources used for actual channel measurements.

[0539] Method 2: Considering that the calculation of NMSE, SGCS, or CGS does not require feature decomposition of the channel, to simplify CPU counting, we can assume that the number of CPUs used is independent of the number of aperiodic CSI-RS resources used for channel measurement, and uniformly assume that one CPU is used, i.e., O CPU =1.

[0540] Method 3: CPU counting is differentiated based on the value of N4 or the number of aperiodic CSI-RS resources. When the number of N4 resources is 1 or the number of aperiodic CSI-RS resources is 1, the number of CPUs used is 1; when N4 > 1 or the number of aperiodic CSI-RS resources is greater than 1, the number of CPUs used is 0. CPU =X·K, where X can take at least one value from {1 / 4, 1 / 3, 1 / 2, 1}, and K is the number of CSI-RS resources used for real channel measurements, where K can take at least one value from {2, 4, 8, 12}.

[0541] Method 4: Considering that the calculation of NMSE, SGCS, or CGS does not require feature decomposition of the channel, to simplify CPU counting, we can assume that the number of CPUs used is related to the number of aperiodic CSI-RS resources used for channel measurement or to the number N4 of predicted CSIs that need to be monitored. When the value of K and / or N4 is no greater than 2 or 4, the number of CPUs used is 1, or the number of CPUs used is 0. CPU =X·K or O CPU = X·N4, where X = at least one of {1 / 4, 1 / 2, 1}; when the value of K and / or N4 is greater than 2 or 4, the number of CPUs used is 2, or the number of CPUs used is 0. CPU =Y·K or O CPU =Y·N4, where Y = at least one of {1 / 2, 1 / 4, 1 / 3, 1}.

[0542] The specific number of CPUs used by this scheme can also be a combination of the methods mentioned above.

[0543] Option 2: AI-based CSI prediction may not require reporting the predicted CSI during the model training phase. In this case, the reporting volume configuration, according to Example 2, may be "none" or "AI-CSI-for-training". Although reporting the predicted CSI may not be required, CSI prediction is still necessary. Therefore, this option provides the CPU usage of AI-based CSI prediction during the model training phase. Here, it is assumed that the channel measurement resources used in the model training phase are periodic CSI-RS resources or semi-persistent CSI-RS resources. The CPU usage during this model training phase can be determined using at least one of the following methods:

[0544] Method 1: If single-period or semi-continuous CSI-RS measurement resources are configured, although it is not necessary to report the predicted CSI, calculate the precoding matrix based on the predicted CSI, or compress the precoding matrix, the computational complexity of AI-based CSI prediction may be higher than that of traditional CSI prediction algorithms. Therefore, during the model training phase, CPU usage can reuse traditional CPU usage rules. For example, when N4=1, the number of CPUs used is O. CPU =4, when N4>1, the number of CPUs used is O. CPU =Y2·N4≥4, where Y2∈{2 / 3,1,2,3} at least one, and the specific value depends on the reporting capability of the user device.

[0545] Method 2: If single-period or semi-continuous CSI-RS measurement resources are configured, considering that it is not necessary to report the predicted CSIs, calculate the precoding matrix based on the predicted CSIs, or compress the precoding matrix, the CPU usage during the model training phase is related to the number of predicted CSIs N4. For example, when N4 = 1, the CPU usage is O. CPU =X, where X can take at least one value from {1, 2, 3, 4}. When N4 > 1, the number of CPUs used is O. CPU =Y·N4≥X, where Y∈ at least one of {1 / 3,1 / 2,2 / 3,1,2,3}, and the specific values ​​of X and / or Y depend on the reporting capability of the user device.

[0546] Method 3: If K periods or semi-persistent CSI-RS measurement resources are configured, considering that the predicted CSIs do not need to be reported, and the precoding matrix does not need to be calculated based on the predicted CSIs, and the precoding matrix does not need to be compressed, the CPU usage during the model training phase is related to the number of predicted CSIs N4 and the number of CSI-RS resources K used for channel measurements. For example, when N4 = 1, the CPU usage is O. CPU =K, when N4>1, the number of CPUs used is O. CPU =Y·K, where Y∈ at least one of {1 / 4,1 / 3,1 / 2,2 / 3,1,2,3}, and the specific value depends on the reporting capability of the user device.

[0547] Method 4: If K periods or semi-persistent CSI-RS measurement resources are configured, considering that the predicted CSIs do not need to be reported, and that the precoding matrix does not need to be calculated based on the predicted CSIs, and that the precoding matrix does not need to be compressed, but considering that the computational complexity of AI-based CSI prediction may be higher than that of traditional CSI prediction algorithms, the CPU usage during the model training phase is related to the number of predicted CSIs N4 and the number of CSI-RS resources K used for channel measurement. For example, if the CPU usage is O CPU =X·K, where X takes at least one value from {1 / 4,1 / 3,1 / 2,2 / 3,3 / 4,1,2,3,4,8}. This occupancy rule applies to cases where N4 takes different values, and the specific value depends on the reporting capability of the user equipment.

[0548] Option 3: For AI-based CSI prediction, if the model inference results and performance monitoring results are reported independently, the CPU usage for reporting performance monitoring results needs to be defined. When performing performance monitoring, the performance of the CSIs output by the prediction window needs to be monitored. Assuming the number of CSIs output by the prediction window is N4, the performance of these N4 predicted CSIs needs to be monitored. The performance monitoring process not only requires measuring the real CSIs associated with the predicted CSIs but also comparing the similarity between the predicted CSIs and the associated real measured CSIs, for example, by using NMSE, SGCS, or GCS to measure the similarity. If the channel measurement resources during performance monitoring are periodic or semi-persistent CSI-RS measurement resources, the CPU usage during performance monitoring can be handled using at least one of the following methods:

[0549] Method 1: Considering that the number of CPUs used for reporting performance monitoring results is related to the number of CSI-RS resources used for channel measurement, the number of CPUs used for reporting performance monitoring results can be 0. CPU = 1, 2 or 4.

[0550] Method 2: Considering that the number of CPUs used for reporting performance monitoring results is related to the number of predicted CSIs that need to be monitored, when N4 = 1, O CPU = 1, 2, or 4; when N4 > 1, O CPU =Y2·N4≥1 or O CPU =Y2·N4≥2 or O CPU =Y2·N4≥4, Y2∈{1 / 2,1 / 4,2 / 3,1,2,3,4}, and the specific value of Y2 depends on the reporting capability of the user equipment.

[0551] Method 3: Considering that the number of CPUs used for reporting performance monitoring results is related to the number of predicted CSIs that need to be monitored, O CPU =Y2·N4, Y2∈{1 / 2,1 / 4,2 / 3,1,2,3,4}, and the specific value of Y2 depends on the reporting capability of the user equipment.

[0552] Option 4: For AI-based CSI prediction, since the reporting of predicted CSI can reuse the existing Rel-18 Doppler codebook during the model inference stage, the main difference lies in the algorithm used to calculate the predicted CSI. Therefore, this option presents the CPU usage of AI-based CSI prediction during the model inference stage. At least one of the following methods can be used to determine the CPU usage during this stage:

[0553] Method 1: Considering that the algorithm complexity of AI-based CSI prediction is not significantly different from that of traditional CSI prediction, a portion of CPU usage can be added to the traditional Rel-18 Doppler codebook CPU usage rules. For channel measurement resources that are aperiodic CSI-RS resources, and where there are K resources, then O CPU =Y1·K+X, where Y1∈{2 / 3,1,2,3}, the specific value depends on the reporting capability of the user equipment, and X∈{0,1 / 2,1 / 4,2 / 3,1,2,3,4}, the specific value depends on the reporting capability of the user equipment. For channel measurement resources that are periodic or semi-persistent CSI-RS resources, when N4=1, O CPU =4+X, when N4>1, O CPU =Y2·N4+X≥4+X, where X∈at least one of {0,1 / 2,1 / 4,2 / 3,1,2,3,4}, and Y2∈at least one of {1 / 2,1 / 4,2 / 3,1,2,3,4}. The specific value of Y2 depends on the user equipment's reporting capability.

[0554] Method 2: Considering that the algorithm complexity of AI-based CSI prediction is lower or higher than that of traditional CSI prediction, and for channel measurement resources that are aperiodic CSI-RS resources, and the number of resources is K, then O CPU =Y1·K, where Y1∈{1 / 2,1 / 3,1 / 4,2 / 3,1,3 / 2,2,5 / 2,3,4}, and its specific value depends on the reporting capability of the user equipment. For channel measurement resources that are periodic or semi-persistent CSI-RS resources, when N4=1, O CPU =1,2,3,4,5 or6, when N4>1, O ...

Claims

A Channel State Information (CSI) prediction method, executed in a user equipment, includes: The system receives measurement configuration information sent by the base station, which includes at least one of the following: measurement reference signal configuration information and CSI reporting configuration information. Receive the measurement reference signal sent by the base station, obtain the predicted CSI based on the measurement reference signal, and compress and report the predicted CSI. as well as The monitoring results are reported to the base station, and the monitoring results are used to indicate the performance monitoring results of at least one predicted CSI relative to the actual measured CSI. According to the CSI prediction method of claim 1, wherein, The performance monitoring results include the Normalized Mean Square Error (NMSE) between the at least one predicted CSI and the actual measured CSI, the Squared Generalized Cosine Similarity (SGCS) result, and / or the Generalized Cosine Similarity (GCS). The CSI prediction method according to claim 1 or 2, wherein, The reporting of monitoring results can be either non-event-triggered or event-triggered. According to claim 3, the CSI prediction method, wherein, The non-event-triggered monitoring result report includes at least one of the following: the quantified value of the performance monitoring result within the monitoring window; The monitoring window contains a bitmap showing the state of the performance monitoring results relative to the threshold value, a predicted CSI index, and a combination number indication. According to claim 3, the CSI prediction method, wherein, The event-triggered monitoring result report includes at least one of the following information: the status information of the event occurring within the monitoring window, the event index within the monitoring window and / or the quantized value of the performance monitoring result within the monitoring window, the monitoring result, and the predicted CSI. The CSI prediction method according to claim 4 or 5, wherein, The non-event-triggered monitoring results and / or the event-triggered monitoring results are reported together with the predicted CSI in a single reporting instance. The CSI prediction method according to any one of claims 1 to 6 further includes: The configuration supporting AI-based CSI prediction function is determined. The configuration of the AI-based CSI prediction function includes at least one of the following information: the definition of performance monitoring events, the reporting priority information of the content reported by the AI ​​model at different lifecycle management (LCM) stages, and the indication information of the CSI processing time corresponding to the CSI reporting at different stages of the AI ​​model, as well as the occupancy information of the CSI processing unit. According to the CSI prediction method of claim 7, wherein, The performance monitoring event is defined as including at least one of the following: the performance monitoring result of the first predicted CSI within the monitoring window is lower than the threshold value; the mean, median and / or variance of the predicted CSI within the monitoring window is lower than the threshold value; or the first predicted CSI is lower than the threshold value within multiple consecutive monitoring windows. The CSI prediction method according to claim 7 or 8, wherein, The reporting priority information of the AI ​​model at different LCM stages includes at least one of the following: the reporting priority of the performance monitoring results is between the reporting of the bearer layer-1 reference signal received power L1-RSRP or the layer-1 signal-to-interference-plus-noise ratio L1-SINR and the traditional CSI prediction reporting or AI-based CSI prediction reporting; the reporting priority of the performance monitoring results corresponding to CSI reporting under different time domain behaviors is higher than that of the corresponding CSI reporting. The CSI prediction method according to any one of claims 1 to 9 further includes: The system reports to the base station its ability to support AI-based CSI prediction, which includes information on the number of CSIs predicted within the supported prediction window. The CSI prediction method according to any one of claims 1 to 10 further includes: The system receives monitoring indication information sent by the base station, which is used to instruct the user equipment to start performance monitoring. The CSI prediction method according to any one of claims 1 to 10 further includes: Send a data collection request message to the base station. The data collection request message includes at least one of the following: a data collection request message for model training, a data collection request message for model inference, and a data collection request message for model monitoring. According to the CSI prediction method of claim 12, wherein, The data collection request message carries at least 1 bit of indication information via at least one dedicated scheduling request (SR) or uplink control information (UCI). The CSI prediction method according to any one of claims 1 to 13 further includes: The system receives monitoring and measurement signals sent by the base station, which are used for CSI measurement. The CSI prediction method according to any one of claims 1 to 14 further includes: The user equipment receives LCM indication information sent by the base station, which is used to instruct the user equipment to perform corresponding LCM operations. According to the CSI prediction method of claim 15, wherein, The LCM indication information includes function rollback operation indication information. According to the CSI prediction method of claim 16, wherein, The fallback operation indication information of the function indicates the cessation of AI-based CSI prediction reporting and / or the activation of conventional Doppler codebook CSI reporting. The fallback operation indication information of the function indicates the fallback operation of the function through the Media Access Control Element (MAC CE) or Downlink Control Information (DCI). The CSI prediction method according to any one of claims 1 to 17 further includes: Receive downlink data information sent by the base station. The CSI prediction method according to any one of claims 1 to 18, wherein, The measurement resource set of the measurement configuration information includes at least one Channel State Information Reference Signal (CSI-RS) resource for CSI measurement, wherein at least a first resource is used for CSI measurement corresponding to the measurement window during CSI prediction model training, and at least a second resource is used for CSI measurement corresponding to the prediction window during CSI prediction model training. The CSI prediction method according to any one of claims 1 to 18, wherein, The channel measurement resources in the first set of measurement resources in the measurement configuration information are used for CSI measurements corresponding to the measurement window during the CSI prediction model training process, and the channel measurement resources in the second set of measurement resources in the measurement configuration information are used for CSI measurements corresponding to the prediction window during the CSI prediction model training process. The CSI prediction method according to any one of claims 1 to 18, wherein, The measurement resources in the first resource set subset of the measurement configuration information are used for CSI measurements corresponding to the measurement window during the CSI prediction model training process, and the measurement resources in the second resource set subset of the measurement configuration information are used for CSI measurements corresponding to the prediction window during the CSI prediction model training process. The CSI prediction method according to any one of claims 1 to 21, wherein, The user equipment determines whether the predicted CSI is used for AI-based CSI prediction by using the CSI reporting configuration field. The CSI prediction method according to any one of claims 1 to 21, wherein, The user equipment determines the predicted CSI as an AI-based CSI prediction through the codebook configuration field under the CSI reporting configuration field. The CSI prediction method according to any one of claims 1 to 21, wherein, The user equipment determines, through the reporting quantity field under the CSI reporting configuration field, whether the CSI prediction used for model training is an AI-based CSI prediction and / or whether the reported CSI is obtained through AI model inference, and / or determines the default reporting resource for periodic or semi-continuous reporting through the reporting configuration type field under the CSI reporting configuration field. The CSI prediction method according to any one of claims 1 to 21, wherein, The user equipment determines, through the information element IE, the predicted CSI is the reporting of CSI information based on AI-based CSI prediction and / or traditional CSI prediction. The CSI prediction method according to any one of claims 1 to 21, wherein, The user equipment determines various reporting quantities through the CSI reporting configuration field or the reporting quantity field under the AI-based CSI reporting configuration field. The CSI prediction method according to any one of claims 1 to 26, wherein, The measurement resource set of the measurement configuration information includes at least one non-periodic CSI-RS measurement resource, and a portion of the measurement resources in the measurement resource set are used for CSI measurements under the measurement window during the CSI prediction process. The CSI prediction method according to any one of claims 1 to 26, wherein, The measurement resource set of the measurement configuration information includes at least one non-periodic CSI-RS measurement resource. The first measurement resource set includes M non-periodic CSI-RS measurement resources for CSI measurement under the measurement window during the CSI prediction process. The second measurement resource set includes N non-periodic CSI-RS measurement resources for CSI measurement under the prediction window during the CSI prediction process, where M and N are greater than or equal to 1. The CSI prediction method according to any one of claims 1 to 28, wherein, The performance monitoring results are quantized, and the number of bits quantized is Y, where Y is greater than 1. According to the CSI prediction method of claim 29, wherein, When the performance monitoring result is less than the threshold value, the first quantization interval is used; When the performance monitoring result is greater than the threshold value, a second quantization interval is used, wherein the first quantization interval is greater than the second quantization interval. The CSI prediction method according to any one of claims 1 to 28, wherein, The performance monitoring results include, within the monitoring window, the performance monitoring results between at least one predicted CSI and the associated actual measured CSI, and the performance monitoring results include information on all layers and / or all subbands corresponding to the CSI. The CSI prediction method according to any one of claims 1 to 28, wherein, The performance monitoring results include, within a monitoring window, the performance monitoring results between the precoding vectors of at least one layer contained in a predicted CSI and the corresponding layer of its associated true measured CSI, the performance monitoring results containing information on all subbands of the corresponding layer in the CSI. The CSI prediction method according to any one of claims 1 to 32, wherein, For periodic or semi-continuous predicted CSI reporting, the performance monitoring results are reported in a periodic or semi-continuous manner, and the predicted CSI is reported jointly with the monitored performance results. The CSI prediction method according to any one of claims 1 to 32, wherein, For periodic or semi-persistent predicted CSI reporting, the reporting of performance monitoring results is periodic, semi-persistent, or non-periodic, and the predicted CSI is reported independently of the monitored performance results. The CSI prediction method according to any one of claims 1 to 32, wherein, For non-periodic predicted CSI reporting, the reporting of performance monitoring results is non-periodic, and the reporting of predicted CSI and the reporting of monitoring results are triggered by the same trigger instance. The CSI prediction method according to any one of claims 1 to 32, wherein, Within the monitoring window, the performance monitoring result between at least one predicted CSI and the associated actual measured CSI is compared with a threshold value to determine whether the performance monitoring result exceeds the threshold value. The CSI prediction method according to any one of claims 11 to 36, wherein, During model training, model inference, and / or model monitoring, the user equipment sends the data collection request message to the base station to trigger data collection, and / or receives reference signals sent by the base station for training, inference, and / or monitoring. The CSI prediction method according to any one of claims 1 to 37, wherein, If the predicted CSI and the actual measured CSI are in the same time slot, then the predicted CSI is associated with the actual measured CSI. The CSI prediction method according to any one of claims 1 to 37, wherein, If the predicted CSI and the actual measured CSI are located in different time slots, the predicted CSI is associated with the actual measured CSI that has the smallest absolute deviation from the predicted CSI in the time domain. The CSI prediction method according to any one of claims 1 to 37, wherein, If the predicted CSI and the actual measured CSI are located in different time slots, the predicted CSI is associated with the actual measured CSI that is later in the time domain and has the smallest time domain deviation from the predicted CSI. The CSI prediction method according to any one of claims 1 to 37, wherein, If the predicted CSI and the actual measured CSI are located in different time slots, the predicted CSI is associated with the actual measured CSI that is earlier in the time domain and has the smallest time domain deviation from the predicted CSI. The CSI prediction method according to any one of claims 1 to 41, wherein, The performance monitoring results are triggered by at least one of the following methods: Within a single or multiple consecutive monitoring windows, the predicted CSI and the associated actual measured CSI satisfy certain constraints. Certain constraints must be satisfied between a single or consecutive predicted CSI and the associated actual measured CSI. The CSI prediction method according to any one of claims 1 to 42, wherein, The user device reports an event index, reports the status of the event, and / or reports the corresponding performance monitoring results. The CSI prediction method according to any one of claims 1 to 42, wherein, Based on the event type reported by the performance monitoring results, determine whether to revert from AI-based predictive CSI reporting to traditional Doppler codebook CSI reporting. The CSI prediction method according to any one of claims 1 to 44, wherein, For AI-based CSI prediction, if the reported predicted CSI is periodic, non-periodic, or semi-persistent, based on performance monitoring results, it is determined whether to revert from AI-based predicted CSI reporting to traditional Doppler codebook CSI reporting. The revert operation of the AI-based CSI prediction is performed by receiving MAC CE or DCI sent by the base station. The CSI prediction method according to any one of claims 1 to 44, wherein, For AI-based CSI prediction, if the number of predicted CSIs generated by the AI ​​model inference is N4, and the first X predicted CSIs within the prediction window have first-class performance, while the next N4-X predicted CSIs have second-class performance, then the number of predicted CSIs output should be reduced. The CSI prediction method according to any one of claims 1 to 44, wherein, For AI-based CSI prediction, if the reported predicted CSI is non-periodic, based on performance monitoring results, it is determined to fall back from AI-based predicted CSI reporting to traditional Doppler codebook CSI reporting. This is done via a DCI activation message, where the activated codebook type is configured to the CSI reporting configuration corresponding to the traditional Doppler codebook CSI reporting. The CSI prediction method according to any one of claims 1 to 47, wherein, The monitoring results are event-triggered monitoring results, and the uplink resources used to report the event-triggered monitoring results are indicated or requested through scheduling requests (SR) or uplink control information (UCI). The CSI prediction method according to any one of claims 1 to 47, wherein, In the reporting of predicted CSI or monitoring results, the training, inference, and / or performance monitoring of the model meet the requirements of a certain CSI processing timeline. The CSI prediction method according to any one of claims 1 to 47, wherein, In the reporting of predicted CSI or monitoring results, the training, inference, and / or performance monitoring of the model meet the requirements of a certain CSI processing unit. A Channel State Information (CSI) prediction method, executed at a base station, includes: The system receives and sends measurement configuration information to the user equipment, the measurement configuration information including at least one of the following: measurement reference signal configuration information and CSI reporting configuration information; A measurement reference signal is sent to the user equipment, the measurement reference signal being used to obtain the predicted CSI; as well as The monitoring results reported by the user equipment are received, and the monitoring results are used to indicate at least one performance monitoring result of the predicted CSI relative to the actual measured CSI. According to the CSI prediction method of claim 51, wherein, in, The performance monitoring results include the Normalized Mean Square Error (NMSE) between the at least one predicted CSI and the actual measured CSI, the Squared Generalized Cosine Similarity (SGCS) result, and / or the Generalized Cosine Similarity (GCS). The CSI prediction method according to claim 51 or 52, wherein, The monitoring results are either non-event-triggered monitoring results or event-triggered monitoring results. According to the CSI prediction method of claim 53, wherein, The non-event-triggered monitoring results include at least one of the following: the quantized value of the performance monitoring result within the monitoring window; the bitmap, predicted CSI index, and combination number indication information corresponding to the state of the performance monitoring result relative to the threshold value within the monitoring window. According to the CSI prediction method of claim 53, wherein, The monitoring results triggered by the event include at least one of the following: the status information of the event occurring within the monitoring window, the event index within the monitoring window and / or the quantized value of the performance monitoring results within the monitoring window, the monitoring results, and the predicted CSI. The CSI prediction method according to claim 54 or 55, wherein, The non-event-triggered monitoring results and / or the event-triggered monitoring results are included in a single reporting instance along with the predicted CSI. The CSI prediction method according to any one of claims 51 to 56 further includes: The configuration supporting AI-based CSI prediction function is determined. The configuration of the AI-based CSI prediction function includes at least one of the following information: the definition of performance monitoring events, the reporting priority information of the content reported by the AI ​​model at different lifecycle management (LCM) stages, and the indication information of the CSI processing time corresponding to the CSI reporting at different stages of the AI ​​model, as well as the occupancy information of the CSI processing unit. The CSI prediction method according to claim 57, wherein, The performance monitoring event is defined as including at least one of the following: the performance monitoring result of the first predicted CSI within the monitoring window is lower than the threshold value; the mean, median and / or variance of the predicted CSI within the monitoring window is lower than the threshold value; or the first predicted CSI is lower than the threshold value within multiple consecutive monitoring windows. The CSI prediction method according to claim 57 or 58, wherein, The reporting priority information of the AI ​​model at different LCM stages includes at least one of the following: the reporting priority of the performance monitoring results is between the reporting of the bearer layer-1 reference signal received power L1-RSRP or the layer-1 signal-to-interference-plus-noise ratio L1-SINR and the traditional CSI prediction reporting or AI-based CSI prediction reporting; the reporting priority of the performance monitoring results corresponding to CSI reporting under different time domain behaviors is higher than that of the corresponding CSI reporting. The CSI prediction method according to any one of claims 51 to 59 further includes: The system receives a report from the user equipment indicating its ability to support AI-based CSI prediction, wherein the AI-based CSI prediction capability includes information on the number of CSIs predicted within the supported prediction window. The CSI prediction method according to any one of claims 51 to 60 further includes: Send monitoring instruction information to the user equipment, the monitoring instruction information being used to instruct the user equipment to start performance monitoring. The CSI prediction method according to any one of claims 51 to 60 further includes: The system receives a data collection request message sent by the user equipment, the data collection request message including at least one of the following: a data collection request message for model training, a data collection request message for model inference, and a data collection request message for model monitoring. According to the CSI prediction method of claim 62, wherein, The data collection request message carries at least 1 bit of indication information via at least one dedicated scheduling request (SR) or uplink control information (UCI). The CSI prediction method according to any one of claims 51 to 63 further includes: A monitoring and measurement signal is sent to the user equipment, the monitoring and measurement signal being used for CSI measurement. The CSI prediction method according to any one of claims 51 to 64 further includes: The LCM instruction information is sent to the user equipment, which instructs the user equipment to perform the corresponding LCM operation. According to the CSI prediction method of claim 65, wherein, The LCM indication information includes function rollback operation indication information. According to the CSI prediction method of claim 66, wherein, The fallback operation indication information of the function indicates the cessation of AI-based CSI prediction reporting and / or the activation of conventional Doppler codebook CSI reporting. The fallback operation indication information of the function indicates the fallback operation of the function through the Media Access Control Element (MAC CE) or Downlink Control Information (DCI). The CSI prediction method according to any one of claims 51 to 67 further includes: Send downlink data information to the user equipment. The CSI prediction method according to any one of claims 51 to 68, wherein, The measurement resource set of the measurement configuration information includes at least one Channel State Information Reference Signal (CSI-RS) resource for CSI measurement, wherein at least a first resource is used for CSI measurement corresponding to the measurement window during CSI prediction model training, and at least a second resource is used for CSI measurement corresponding to the prediction window during CSI prediction model training. The CSI prediction method according to any one of claims 51 to 68, wherein, The channel measurement resources in the first set of measurement resources in the measurement configuration information are used for CSI measurements corresponding to the measurement window during the CSI prediction model training process, and the channel measurement resources in the second set of measurement resources in the measurement configuration information are used for CSI measurements corresponding to the prediction window during the CSI prediction model training process. The CSI prediction method according to any one of claims 51 to 68, wherein, The measurement resources in the first resource set subset of the measurement configuration information are used for CSI measurements corresponding to the measurement window during the CSI prediction model training process, and the measurement resources in the second resource set subset of the measurement configuration information are used for CSI measurements corresponding to the prediction window during the CSI prediction model training process. The CSI prediction method according to any one of claims 51 to 71, wherein, The base station determines whether the predicted CSI is used for AI-based CSI prediction by using the CSI reporting configuration field. The CSI prediction method according to any one of claims 51 to 71, wherein, The base station determines the predicted CSI based on AI-based CSI prediction by using the codebook configuration field under the CSI reporting configuration field. The CSI prediction method according to any one of claims 51 to 71, wherein, The base station determines, through the reporting quantity field under the CSI reporting configuration field, whether the CSI prediction used for model training is an AI-based CSI prediction and / or whether the reported CSI is obtained through AI model inference, and / or determines the default reporting resource for periodic or semi-continuous reporting through the reporting configuration type field under the CSI reporting configuration field. The CSI prediction method according to any one of claims 51 to 71, wherein, The base station determines, through the information element IE, the predicted CSI is based on AI-based CSI prediction and / or the reporting of CSI information based on traditional CSI prediction. The CSI prediction method according to any one of claims 51 to 71, wherein, The base station determines various reporting quantities through the CSI reporting configuration field or the reporting quantity field under the AI-based CSI reporting configuration field. The CSI prediction method according to any one of claims 51 to 76, wherein, The measurement resource set of the measurement configuration information includes at least one non-periodic CSI-RS measurement resource, and a portion of the measurement resources in the measurement resource set are used for CSI measurements under the measurement window during the CSI prediction process. The CSI prediction method according to any one of claims 51 to 76, wherein, The measurement resource set of the measurement configuration information includes at least one non-periodic CSI-RS measurement resource. The first measurement resource set includes M non-periodic CSI-RS measurement resources for CSI measurement under the measurement window during the CSI prediction process. The second measurement resource set includes N non-periodic CSI-RS measurement resources for CSI measurement under the prediction window during the CSI prediction process, where M and N are greater than or equal to 1. The CSI prediction method according to any one of claims 51 to 78, wherein, The performance monitoring results are quantized, and the number of bits quantized is Y, where Y is greater than 1. The CSI prediction method according to claim 79, wherein, When the performance monitoring result is less than the threshold value, the first quantization interval is used; When the performance monitoring result is greater than the threshold value, a second quantization interval is used, wherein the first quantization interval is greater than the second quantization interval. The CSI prediction method according to any one of claims 51 to 78, wherein, The performance monitoring results include, within the monitoring window, the performance monitoring results between at least one predicted CSI and the associated actual measured CSI, and the performance monitoring results include information on all layers and / or all subbands corresponding to the CSI. The CSI prediction method according to any one of claims 51 to 78, wherein, The performance monitoring results include, within a monitoring window, the performance monitoring results between the precoding vectors of at least one layer contained in a predicted CSI and the corresponding layer of its associated true measured CSI, the performance monitoring results containing information on all subbands of the corresponding layer in the CSI. The CSI prediction method according to any one of claims 51 to 82, wherein, For periodic or semi-continuous predicted CSI reporting, the performance monitoring results are reported in a periodic or semi-continuous manner, and the predicted CSI is reported jointly with the monitored performance results. The CSI prediction method according to any one of claims 51 to 82, wherein, For periodic or semi-persistent predicted CSI reporting, the reporting of performance monitoring results is periodic, semi-persistent, or non-periodic, and the predicted CSI is reported independently of the monitored performance results. The CSI prediction method according to any one of claims 51 to 82, wherein, For non-periodic predicted CSI reporting, the reporting of performance monitoring results is non-periodic, and the reporting of predicted CSI and the reporting of monitoring results are triggered by the same trigger instance. The CSI prediction method according to any one of claims 51 to 82, wherein, Within the monitoring window, the performance monitoring result between at least one predicted CSI and the associated actual measured CSI is compared with a threshold value to determine whether the performance monitoring result exceeds the threshold value. The CSI prediction method according to any one of claims 61 to 86, wherein, During model training, model inference, and / or model monitoring, the base station receives the data collection request message sent by the user equipment to trigger data collection, and / or sends reference signals to the user equipment for training, inference, and / or monitoring. The CSI prediction method according to any one of claims 51 to 87, wherein, If the predicted CSI and the actual measured CSI are in the same time slot, then the predicted CSI is associated with the actual measured CSI. The CSI prediction method according to any one of claims 51 to 87, wherein, If the predicted CSI and the actual measured CSI are located in different time slots, the predicted CSI is associated with the actual measured CSI that has the smallest absolute deviation from the predicted CSI in the time domain. The CSI prediction method according to any one of claims 51 to 87, wherein, If the predicted CSI and the actual measured CSI are located in different time slots, the predicted CSI is associated with the actual measured CSI that is later in the time domain and has the smallest time domain deviation from the predicted CSI. The CSI prediction method according to any one of claims 51 to 87, wherein, If the predicted CSI and the actual measured CSI are located in different time slots, the predicted CSI is associated with the actual measured CSI that is earlier in the time domain and has the smallest time domain deviation from the predicted CSI. The CSI prediction method according to any one of claims 51 to 91, wherein, The performance monitoring results are triggered by at least one of the following methods: Within a single or multiple consecutive monitoring windows, the predicted CSI and the associated actual measured CSI satisfy certain constraints. Certain constraints must be satisfied between a single or consecutive predicted CSI and the associated actual measured CSI. The CSI prediction method according to any one of claims 51 to 92, wherein, The base station receives the event type, the status of the event, and / or the corresponding performance monitoring results reported by the user equipment. The CSI prediction method according to any one of claims 51 to 92, wherein, Based on the event type reported by the performance monitoring results, determine whether to revert from AI-based predictive CSI reporting to traditional Doppler codebook CSI reporting. The CSI prediction method according to any one of claims 51 to 94, wherein, For AI-based CSI prediction, if the reported predicted CSI is periodic, non-periodic, or semi-persistent, based on performance monitoring results, it is determined whether to revert from AI-based predicted CSI reporting to traditional Doppler codebook CSI reporting. The revert operation is performed by sending a MAC CE or DCI to the user equipment. The CSI prediction method according to any one of claims 51 to 94, wherein, For AI-based CSI prediction, if the number of predicted CSIs generated by the AI ​​model inference is N4, and the first X predicted CSIs within the prediction window have first-class performance, while the next N4-X predicted CSIs have second-class performance, then the number of predicted CSIs output should be reduced. The CSI prediction method according to any one of claims 51 to 94, wherein, For AI-based CSI prediction, if the reported predicted CSI is non-periodic, based on performance monitoring results, it is determined to fall back from AI-based predicted CSI reporting to traditional Doppler codebook CSI reporting. This is done via a DCI activation message, where the activated codebook type is configured to the CSI reporting configuration corresponding to the traditional Doppler codebook CSI reporting. The CSI prediction method according to any one of claims 51 to 97, wherein, The monitoring results are event-triggered monitoring results, and the uplink resources used to report the event-triggered monitoring results are indicated or requested through scheduling requests (SR) or uplink control information (UCI). The CSI prediction method according to any one of claims 51 to 97, wherein, In the monitoring results, the training, inference, and / or performance monitoring of the model meet the requirements of a certain CSI processing timeline. The CSI prediction method according to any one of claims 51 to 97, wherein, In the monitoring results, the training, inference, and / or performance monitoring of the model meet the requirements of a certain CSI processing unit. A wireless communication device, comprising: A processor and a memory, the memory for storing a computer program, the processor for calling and running the computer program stored in the memory to perform the CSI prediction method as described in any one of claims 1 to 100.

Images