Ai / ml model based csi performance monitoring report triggering method and related apparatus

Abstract

Embodiments of the present application provide a CSI performance monitoring report triggering method based on an AI / ML model and related devices, and provide a cumulative sum (CUSUM) algorithm that is adaptive to a channel environment. The algorithm configures different CUSUM parameter sets for different states of a UE. Each CUSUM parameter set includes a relaxation factor k and a decision threshold h, and the values of k and h in different parameter sets are different. The UE can select the most suitable CUSUM parameter according to the current state (or scenario), so that the performance monitoring mechanism of the UE can intelligently balance the detection sensitivity and robustness. Furthermore, the scheme compares whether to trigger a performance degradation report after accumulating the performance difference through the CUSUM algorithm and comparing the performance difference with the decision threshold. The accumulation of the performance difference and the effective filtering of the instantaneous jitter of the channel ensure that the report is triggered only when the performance of the AI / ML model continuously decreases, thereby reducing the uplink signaling overhead caused by false triggering and the resource waste caused by subsequent network responses (such as mode switching).

Description

Technical Field

[0001] This application relates to the field of wireless communication technology, and in particular to a CSI performance monitoring report triggering method and related apparatus based on an AI / ML model. Background Technology

[0002] In wireless communication systems, massively multi-input multiple-output (MIMO) technology can be used to improve system capacity and spectral efficiency. In MIMO, the terminal device needs to measure the channel state information reference signal (CSI-RS) transmitted by the network device and report the channel state information (CSI). The network device performs precoding based on the CSI and then sends the precoded data to the terminal device.

[0003] Utilizing compression algorithms based on artificial intelligence (AI) / machine learning (ML) can further enhance CSI feedback, reducing CSI feedback overhead and improving accuracy. In AI / ML-based compression algorithms, the 3GPP RAN1 working group has reached a consensus that the encoder and decoder based on AI / ML models should be deployed on the terminal device and network device (such as a base station) sides, respectively, referred to as a bilateral AI / ML model. Specifically, the terminal device uses an AI / ML-based CSI encoder to generate CSI feedback information, and the network device uses a corresponding AI / ML-based CSI decoder to reconstruct the CSI based on the received CSI feedback information. In practical deployment, the bilateral AI / ML model requires continuous monitoring of model performance to ensure its reliability and stability in dynamic environments.

[0004] One performance monitoring method is UE-side performance monitoring based on precoded CSI-RS. The core idea of ​​this method is that the network device implicitly transmits the decoded CSI using precoded CSI-RS, and the UE evaluates the performance based on the AI / ML model by comparing the precoded signal with its own original channel measurements. If a report is triggered based on a single measurement result, for example, if a report is submitted whenever the performance based on the AI / ML model is lower than that of the traditional codebook method, this is easily affected by instantaneous channel fluctuations, leading to unnecessary reports and potential mode switching, increasing signaling overhead. Summary of the Invention

[0005] In view of this, this application provides a CSI performance monitoring report triggering method and related apparatus based on AI / ML models to solve at least some of the above-mentioned problems. The disclosed technical solution is as follows:

[0006] Firstly, this application provides a CSI performance monitoring report triggering method based on an AI / ML model, executed by a terminal device. The method includes: obtaining performance difference indicators between the traditional codebook scheme and the AI / ML model at the current monitoring time; determining a target CUSUM parameter set based on the current state of the terminal device and CUSUM parameter set configuration information, where different states of the terminal device correspond to different CUSUM parameter sets, each CUSUM parameter set including a relaxation factor and a decision threshold, with different relaxation factors and decision thresholds in different CUSUM parameter sets; updating the cumulative performance difference at the current monitoring time, where the cumulative performance difference at the current monitoring time is the sum of the cumulative performance difference at the previous monitoring time plus the portion of the performance difference indicator at the current monitoring time exceeding a preset cumulative threshold, where the preset cumulative threshold includes the sum of the target offset and the relaxation factor in the target CUSUM parameter set, and the target offset is a benchmark value for measuring the performance difference indicator; and sending a performance degradation report to the network device if the cumulative performance difference exceeds the decision threshold in the target CUSUM parameter set, where the performance degradation report indicates that the performance of the AI / ML model is lower than that of the traditional codebook scheme.

[0007] As can be seen, this scheme configures different CUSUM parameter sets for different UE states. Each CUSUM parameter set includes a relaxation factor k and a decision threshold h, and the values ​​of k and h are different in different parameter sets. The UE can select the most suitable CUSUM parameter according to the current state (or scenario), so that the UE's performance monitoring mechanism can intelligently balance detection sensitivity and robustness. This scheme accumulates performance differences through the CUSUM algorithm and then compares them with the decision threshold to determine whether to trigger a performance degradation report. By accumulating performance differences and effectively filtering instantaneous channel jitter, it ensures that a report is only triggered when the performance of the AI / ML model shows a sustained decline, thereby reducing uplink signaling overhead caused by accidental touches and resource waste caused by subsequent network responses (such as mode switching).

[0008] In one possible implementation, the method further includes: if the sum of the cumulative performance difference at the previous monitoring time and the first difference is less than or equal to 0, setting the cumulative performance difference at the current monitoring time to 0, where the first difference is the difference between the performance difference index at the current monitoring time and the preset cumulative threshold. This situation indicates that the performance of the AI-ML model at the current monitoring time is better than that at the previous monitoring time; that is, the performance degradation at the previous monitoring time was temporary and not accumulated. Accumulation only applies to situations of continuous deterioration. Therefore, setting the cumulative performance difference at the current monitoring time to 0 and recalculating the cumulative performance difference avoids the impact of temporary performance degradation on the performance degradation judgment.

[0009] In one possible implementation, the cumulative sum of performance differences at the current monitoring moment is updated based on the performance difference index and the relaxation factor in the target CUSUM parameter set, including updating the cumulative sum of performance differences at the current monitoring moment according to the following formula:

[0010]

[0011] Among them, S t S is the cumulative sum of performance differences at the current monitoring time t. t-1 y is the cumulative sum of performance differences corresponding to the previous monitoring time t-1. t Let k be the performance difference metric corresponding to the current monitoring time t. t The relaxation factor used at the current monitoring moment. This is the target offset.

[0012] As can be seen, this scheme only considers the performance difference index y at the current monitoring time. t Greater than the target offset and relaxation factor k t When the sum of these is reached, the cumulative performance difference S t This effectively filters out instantaneous channel jitter, ensuring that reports are only triggered when the performance of the AI / ML model experiences a sustained decline. This reduces uplink signaling overhead caused by accidental touches and resource waste caused by subsequent network responses (such as mode switching).

[0013] In another possible implementation, the method further includes setting the cumulative sum of performance differences at the current monitoring moment to 0 after sending a performance degradation report to the network device. This way, after triggering a performance degradation report, the cumulative sum of performance differences at the current monitoring moment is set to 0, and performance differences are re-accumulated starting from the next monitoring moment. This avoids the continuous impact of accumulated performance differences that trigger reporting, leading to frequent reporting and saving reporting signaling overhead and resources.

[0014] In another possible implementation, before obtaining the performance difference index between the traditional codebook scheme and the AI / ML model at the current monitoring moment, the method further includes: receiving a first message from the network device, the first message including performance-aware configuration information, the performance-aware configuration information including CUSUM parameter set configuration information and parameter set mapping rules, the CUSUM parameter set configuration information including multiple CUSUM parameter sets, and the parameter set mapping rules including the mapping relationship between different state parameters of the terminal device and the CUSUM parameter sets.

[0015] In another possible implementation, the first message is an RRC reconfiguration message. This allows the UE to be configured with the CUSUM parameter set and parameter set mapping rules via the RRC reconfiguration message, eliminating the need for entirely new signaling. This improves signaling efficiency, reduces signaling overhead, and saves resources.

[0016] In another possible implementation, the new information cells in the RRC reconfiguration message are used to carry performance-aware configuration information. It is evident that by using new information cells in the RRC reconfiguration message to carry the corresponding configuration information, there is no need to use entirely new signaling for configuration, thus improving signaling efficiency, reducing signaling overhead, and saving resources.

[0017] In another possible implementation, the terminal device's state parameters include at least one of mobility state, channel quality, service type, and the terminal device's own state. The terminal device's own state includes at least one of remaining battery power, operating temperature, beam management state, and hardware capabilities. In this way, the terminal device can select the most suitable CUSUM parameter set based on the current state of its environment or its own state, allowing the CUSUM algorithm to adapt to either the terminal's current environmental state or its own state.

[0018] In another possible implementation, the mapping relationship between mobility states and CUSUM parameter sets includes: a first CUSUM parameter set corresponding to the stationary state, which includes a first relaxation factor and a first decision threshold; a second CUSUM parameter set corresponding to the low mobility state, which includes a second relaxation factor and a second decision threshold; a third CUSUM parameter set corresponding to the medium mobility state, which includes a third relaxation factor and a third decision threshold; and a fourth CUSUM parameter set corresponding to the high mobility state, which includes a fourth relaxation factor and a fourth decision threshold. The values ​​of the first, second, third, and fourth relaxation factors increase sequentially, as do the values ​​of the first, second, third, and fourth decision thresholds. By configuring different CUSUM parameter sets for different mobility states, the terminal can select the most suitable parameter set for performance monitoring based on its own mobility state, thereby making the sensitivity and robustness of AI / ML model performance monitoring more consistent with the current mobility state.

[0019] In another possible implementation, the mapping relationship between channel quality and parameter sets includes: excellent channel quality corresponds to the fifth CUSUM parameter set, which includes a fifth relaxation factor and a fifth decision threshold; moderate channel quality corresponds to the sixth CUSUM parameter set, which includes a sixth relaxation factor and a sixth decision threshold; and poor channel quality corresponds to the seventh CUSUM parameter set, which includes a seventh relaxation factor and a seventh decision threshold. The values ​​of the fifth, sixth, and seventh relaxation factors increase sequentially; similarly, the values ​​of the fifth, sixth, and seventh decision thresholds also increase sequentially. This allows the terminal to select an appropriate CUSUM parameter set based on the current channel quality, intelligently balancing the sensitivity and robustness of performance monitoring according to the current channel quality.

[0020] In another possible implementation, the mapping relationship between the terminal device's service type and parameter set includes: ultra-reliable low-latency communication corresponds to the ninth CUSUM parameter set, which includes the ninth relaxation factor and the ninth decision threshold; enhanced mobile broadband services correspond to the tenth CUSUM parameter set, which includes the tenth relaxation factor and the tenth decision threshold; and massive machine-type communication services correspond to the eleventh CUSUM parameter set, which includes the eleventh relaxation factor and the eleventh decision threshold. The values ​​of the ninth, tenth, and eleventh relaxation factors increase sequentially, as do the values ​​of the ninth, tenth, and eleventh decision thresholds. This allows the terminal to select an appropriate CUSUM parameter set based on the current service type, intelligently balancing the sensitivity and robustness of performance monitoring according to the current service type.

[0021] In another possible implementation, the performance degradation report includes AI / ML model performance degradation indications and diagnostic information, including performance degradation information when the terminal device triggers the performance degradation report.

[0022] In another possible implementation, the diagnostic information includes at least one of the severity level and the CUSUM parameter set ID used when triggering a performance degradation report, with the severity level indicating the degree of performance degradation of the AI / ML model relative to a traditional codebook scheme.

[0023] In another possible implementation, a performance degradation report is sent to the network device, including by carrying and sending the performance degradation report via a dedicated MAC CE. This provides the network device with more dimensions of performance degradation information with minimal overhead, enabling the network device to understand the severity of the degradation and the scenario in which the UE was located when the problem occurred upon receiving the performance degradation report. This further allows for highly differentiated and intelligent response strategies from the network device.

[0024] Secondly, this application also provides a CSI performance monitoring report triggering method based on an AI / ML model, executed by a network device. The method includes: receiving a performance degradation report from a terminal device, wherein the performance degradation report is sent when the terminal device detects that the cumulative sum of performance differences at the current monitoring time is greater than a decision threshold in a target CUSUM parameter set; wherein the target CUSUM parameter set is determined by the terminal device based on the current state of the terminal device, and different states of the terminal device correspond to different CUSUM parameter sets, each CUSUM parameter set including a relaxation factor and a decision threshold, and the relaxation factor and decision threshold are different in different CUSUM parameter sets; the cumulative sum of performance differences at the current monitoring time is updated by the terminal device based on the performance difference index at the current monitoring time and the relaxation factor in the target CUSUM parameter set, the cumulative sum of performance differences at the current monitoring time is based on the cumulative sum of performance differences corresponding to the previous monitoring time plus the portion of the performance difference index at the current monitoring time exceeding a preset cumulative threshold, the preset cumulative threshold including the relaxation factor in the target CUSUM parameter set.

[0025] Thirdly, this application also provides a communication device, including a module for performing the method as described in any of the first aspects, or a module for performing the method as described in any of the second aspects.

[0026] Fourthly, this application also provides a communication device, comprising: a memory for storing computer instructions; and a processor for executing the computer program or computer instructions stored in the memory, causing the communication device to perform the method as described in any of the first aspects, or the method as described in any of the second aspects.

[0027] Fifthly, this application also provides a computer storage medium for storing a computer program, which, when executed, is used to implement the method described in either the first or second aspect.

[0028] Sixthly, this application also provides a computer program product, wherein the computer program, when run, causes the method described in either the first or second aspect to be performed. Attached Figure Description

[0029] Figure 1 A schematic diagram of the structure of a communication system provided by the present invention;

[0030] Figure 2 A schematic diagram of the structure of an access network device provided by the present invention;

[0031] Figure 3 A flowchart of a traditional AI / ML model performance monitoring method provided by the present invention;

[0032] Figure 4 A schematic diagram of gNB transmission signal in AI / ML model performance monitoring provided by the present invention;

[0033] Figure 5 A flowchart of a performance monitoring method for an AI / ML model provided by the present invention;

[0034] Figure 6 A schematic diagram of the structure of a communication device provided by the present invention;

[0035] Figure 7 A schematic diagram of another communication device provided by the present invention. Detailed Implementation

[0036] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. The terminology used in the following embodiments is for the purpose of describing specific embodiments only and is not intended to be a limitation of this application. As used in the specification and appended claims of this application, the singular expressions "a," "an," "the," "the," "the," and "this" are intended to also include expressions such as "one or more," unless the context clearly indicates otherwise. It should also be understood that in the embodiments of this application, "one or more" refers to one, two, or more; "and / or" describes the relationship between related objects, indicating that three relationships may exist; for example, A and / or B can represent: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship.

[0037] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0038] The "multiple" mentioned in the embodiments of this application refers to two or more. It should be noted that in the description of the embodiments of this application, terms such as "first" and "second" are used only for the purpose of distinguishing descriptions and should not be construed as indicating or implying relative importance, nor should they be construed as indicating or implying order.

[0039] The technical solutions provided in this application can be applied to communication systems, which may include, but are not limited to, the following systems: second-generation (2G) communication systems, third-generation (3G) communication systems, long-term evolution (LTE) systems, universal mobile telecommunication system (UMTS), worldwide interoperability for microwave access (WiMAX) communication systems, fifth-generation (5G) systems or new radio (NR) systems, 5.5G systems or sixth-generation (6G) systems, and future mobile communication systems, vehicle-to-X (V2X); V2X may include vehicle-to-network (V2N), vehicle-to-vehicle (V2V), vehicle-to-infrastructure (V2I), vehicle-to-pedestrian (V2P), long-term evolution-vehicle (LTE-V) technology, vehicle-to-everything (V2X), and machine-type communications. Communication, MTC, Internet of Things (IoT), Ambient Internet of Things (AIOT), Long Term Evolution of Machine (LTE-M), Machine to Machine (M2M), etc.

[0040] The communication system can be applied to scenarios including: terrestrial cellular communication, non-terrestrial network (NTN), satellite communication, high altitude platform station (HAPS) communication, vehicle-to-everything (V2X) communication, integrated access and backhaul (IAB) communication, and reconfigurable intelligent surface (RIS) communication, etc.

[0041] For example, Figure 1 A schematic diagram of the architecture of a communication system provided in an embodiment of this application is shown.

[0042] like Figure 1 As shown, the communication system includes network device 101 and terminal device 102.

[0043] Network device 101 can be a device on the network side used to provide network communication functions. In some cases, it is also called a network element. Network device can usually be a base station (including functional units of the base station, or a combination of functional units of the base station) or a core network unit. The core network unit can be a functional unit in the core network, including but not limited to access and mobility management function (AMF) unit or session management function (SMF) unit.

[0044] In this embodiment, the base station can be any device with wireless transceiver capabilities, including but not limited to: evolved Node B (NodeB, eNB, or e-NodeB) in Long Term Evolution (LTE), base station (gNodeB or gNB) or transmission receiving point (TRP) in New Radio (NR), base stations in subsequent 3GPP evolutions, access nodes, wireless relay nodes, and wireless backhaul nodes in Wi-Fi systems. The base station can be: macro base station, micro base station, pico base station, small cell, relay station, or balloon station, etc. The base station can include one or more co-located or non-co-located transmission receiving points (TRPs). The base station can also be a radio controller, centralized unit (CU), and / or distributed unit (DU) in a cloud radioaccess network (CRAN) scenario. The base station can communicate with the terminal device 102, or it can communicate with the terminal device 102 through a relay station. Terminal devices can communicate with multiple base stations using different technologies. For example, a terminal device can communicate with a base station that supports LTE networks, or with a base station that supports 5G networks, or even have dual connections with both LTE and 5G base stations.

[0045] In practical applications, when network devices function as access network devices, multiple network devices can collaborate to assist terminal devices in achieving wireless access, with different network devices each implementing some of the functions of a base station. For example, network devices can be central units (CUs), distributed units (DUs), CU-control plane (CPs), CU-user plane (UPs), or radio units (RUs), etc. CUs and DUs can be set up separately or included in the same network element, such as a baseband unit (BBU). RUs can be included in radio frequency devices or radio frequency units, such as remote radio units (RRUs), active antenna units (AAUs), or remote radio heads (RRHs).

[0046] In different systems, CU (or CU-CP and CU-UP), DU, or RU may have different names, but those skilled in the art will understand their meaning. For example, in an ORAN system, CU can also be called O-CU (Open CU), DU can also be called O-DU, CU-CP can also be called O-CU-CP, CU-UP can also be called O-CU-UP, and RU can also be called O-RU. Any of the units among CU (or CU-CP, CU-UP), DU, and RU in this application can be implemented through software modules, hardware modules, or a combination of software and hardware modules. CU (or CU-CP and CU-UP), DU, and RU can implement different protocol layer functions.

[0047] Figure 2 This is a schematic diagram of the structure of an access network device. As an implementation example, such as... Figure 2As shown, an access network device may include at least one CU and at least one DU. This design can be referred to as CU and DU separation. A CU can be connected to one or more DUs. CU and DU can be separated according to the protocol layer of the wireless network: for example, the functions of the Packet Data Convergence Protocol (PDCP) layer and above (such as the Radio Resource Control (RRC) layer and the Service Data Adaptation Protocol (SDAP) layer) are set in the CU, while the functions of the protocol layers below the PDCP layer (such as the Radio Link Control (RLC) layer, the Media Access Control (MAC) layer, and the Physical (PHY) layer) are set in the DU; or, for example, the functions of the protocol layers above the PDCP layer are set in the CU, while the functions of the protocol layers below the PDCP layer are set in the DU, without restriction. When the CU includes CU-CP and CU-UP, CU-CP is used to implement the control plane functions of the CU, and CU-UP is used to implement the user plane functions of the CU. For example, when a CU is configured to implement the functions of the PDCP, RRC, and SDAP layers, CU-CP is used to implement the RRC layer functions and the control plane functions of the PDCP layer, while CU-UP is used to implement the SDAP layer functions and the user plane functions of the PDCP layer. This application does not limit the names of CU and DU. The above division of CU and DU processing functions according to protocol layers is merely an example; other division methods are also possible.

[0048] The CU can be connected to the core network. Optionally, the CU can have some of the functions of the core network.

[0049] Furthermore, some functions of the DU can be separated and configured. For example... Figure 2As shown, this functionality can be implemented by a radio unit (RU). The RU can have radio frequency (RF) capabilities. This application does not limit the name of the RU. The DU and RU can be split or separated within the PHY layer. For example, the DU can implement higher-level functions in the PHY layer, and the RU can implement lower-level functions in the PHY layer, or implement both lower-level and RF functions. Higher-level functions in the PHY layer include functions closer to the MAC layer, and lower-level functions in the PHY layer include functions closer to the RF layer. For example, higher-level functions in the PHY layer include one or more of the following: forward error correction (FEC) encoding / decoding, scrambling, or modulation / demodulation. Lower-level functions in the PHY layer include one or more of the following: fast Fourier transform (FFT) / inverse fast Fourier transform (IFFT), beamforming, or extraction and filtering of the physical random access channel (PRACH), etc. The RU can communicate with the terminal device via the air interface using RF signals. The pre-coding function of the PHY layer code can be located in the DU or the RU. The separation between the DU and RU can be done in various ways without restriction. An interface exists between the DU and RU. For example, depending on the separation method, the interface between the DU and RU can be a Common Public Radio Interface (CPRI) interface or an Enhanced Common Public Radio Interface (eCPRI) interface.

[0050] Optionally, any one of CU, CU-CP, CU-UP, DU, and RU can be a software module, a hardware structure, or a combination of software and hardware structures, without limitation. The different entities can exist in the same or different forms. For example, CU, CU-CP, CU-UP, and DU are software modules, and RU is a hardware structure. For the sake of brevity, all possible combinations are not listed here. These modules and the methods they execute are also within the protection scope of the embodiments of this application. For example, when the method of the embodiments of this application is executed by an access network device, it can be specifically executed by at least one of CU, CU-CP, CU-UP, DU, or RU.

[0051] For example, the receiving circuit of terminal device 102 may include a main radio (MR) and a low-power radio (LR).

[0052] In the embodiments of this application, the terminal device can be of various forms, such as a mobile phone, a tablet computer, a computer with wireless transceiver capabilities, a virtual reality (VR) terminal device, an augmented reality (AR) terminal device, a wireless terminal in industrial control, a vehicle-mounted terminal device, a wireless terminal in self-driving technology, a wireless terminal in remote medical care, a wireless terminal in a smart grid, a wireless terminal in transportation safety, a wireless terminal in a smart city, a wireless terminal in a smart home, a wearable terminal device, etc. A terminal may also be referred to as a terminal device, user equipment (UE), access terminal device, vehicle-mounted terminal, industrial control terminal, UE unit, UE station, mobile station, mobile station, remote station, remote terminal device, mobile device, UE terminal device, wireless communication device, UE agent, or UE device, etc. A terminal can also be a fixed terminal or a mobile terminal.

[0053] In some embodiments, the communication system may also include other devices that communicate with network devices and / or interrupt devices, which is not a limitation of this application.

[0054] like Figure 3 As shown, the basic process of the UE-side performance monitoring scheme based on precoding CSI-RS is as follows:

[0055] S1, gNB (base station) sends downlink CSI-RS signal.

[0056] S2, the UE measures the downlink CSI-RS signal and obtains CSI feedback information based on the AI / ML model.

[0057] S3, the UE sends CSI feedback information to the gNB.

[0058] The UE measures the downlink CSI-RS signal transmitted by the gNB at time T0 to obtain the original channel matrix H. The channel matrix H is then subjected to singular value decomposition (SVD) to obtain the precoding matrix V. The precoding matrix V is then processed by the encoder in the UE's AI / ML model for feature extraction, compression coding, and quantization to obtain the bitstream, which is the CSI feedback information. The UE reports this CSI feedback information model to the gNB.

[0059] Compressed precoding matrices (also known as implicit CSI feedback) can save feedback overhead and are consistent with the traditional codebook-based CSI feedback framework. Therefore, AI / ML model-based CSI feedback has attracted much attention in 3GPP research.

[0060] S4, gNB sends standard CSI-RS and precoded CSI-RS to UE.

[0061] The standard CSI-RS is not pre-coded; the UE uses this CSI-RS to measure complete channel information.

[0062] The pre-coded CSI-RS is obtained by precoding based on CSI feedback information.

[0063] Precoding is a signal processing technique performed at the gNB (gear-by-wire) transmitter. The gNB uses its acquired CSI (Content Sign-In) information to weight and phase-adjust the signal to be transmitted, aiming to more accurately align the signal energy with the UE (User Equipment) and overcome fading and interference during wireless channel transmission. Mathematically, precoding is equivalent to multiplying the vector of data symbols to be transmitted by a precoding matrix, which is calculated based on the CSI.

[0064] The gNB uses the precoding matrix indicated by the first CSI feedback information reported by the UE at time T0 to perform precoding to obtain the first precoded CSI-RS. Simultaneously, the gNB uses the decoder in the AI / ML model to decode the bitstream reported by the UE at time T0 (i.e., the second CSI feedback information) to obtain the reconstructed precoding matrix. Using the reconstructed precoding matrix The CSI-RS signal is precoded to obtain the second precoded CSI-RS.

[0065] At time T1, the gNB transmits the first precoded CSI-RS and the second precoded CSI-RS.

[0066] like Figure 4 As shown, at time T0, the gNB transmits a CSI-RS resource set for measurement to the UE. The UE measures the CSI-RS signal to obtain CSI and feeds back two types of CSI to the gNB: one is CSI feedback information based on the traditional codebook scheme, and the other is CSI feedback information encoded and compressed based on the AI / ML model. At time T1, the gNB transmits two CSI-RS resource sets to the UE: one is a CSI-RS for channel inference, which focuses on the real-time information of the current channel; the other is a CSI-RS for performance monitoring, which uses the precoding matrix obtained by the gNB decoding the second CSI feedback information to precode the reference signal. It focuses on the quality after processing by the AI / ML model at the past time T0 and performs performance evaluation at the current time T1.

[0067] S5, the UE calculates the first performance index of the traditional codebook scheme based on the measurement results of the standard CSI-RS, and calculates the second performance index of the AI / ML model based on the precoding CSI-RS.

[0068] At time T1, the UE receives two types of CSI-RS resource sets: the standard CSI-RS resource set and the precoded CSI-RS.

[0069] In performance monitoring scenarios, the UE uses the standard CSI-RS to measure complete channel information and calculates the first performance metric of the traditional codebook scheme based on this, such as the equivalent signal-to-interference-plus-noise ratio (SINR) or the hypothetical block error rate (BLER). The SINR or BLER represents the performance corresponding to the traditional codebook scheme, and the first performance metric corresponding to the traditional codebook scheme serves as a benchmark for comparison with the actual performance of the AI / ML model scheme.

[0070] The legacy codebook is a set of predefined precoding matrices in the 3GPP standard. The process of calculating the performance metrics of the legacy codebook scheme is as follows: The UE obtains channel information, i.e., the channel matrix H, by measuring downlink CSI-RS. Further, the UE iterates through all candidate precoding matrices in the legacy codebook set. For each candidate precoding matrix W, the UE calculates the corresponding performance metric based on the channel matrix H. This performance metric is typically the precoded effective signal-to-interference-plus-noise ratio (SINR), with the goal of finding the precoding matrix W_legacy that maximizes the effective SINR. The UE uses the selected optimal precoding matrix W_legacy and the channel matrix H to calculate the effective SINR, which characterizes the performance of the legacy codebook scheme.

[0071] Furthermore, the UE can further derive the hypothetical block error rate (BLER) based on the calculated equivalent signal-to-interference-plus-noise ratio (SINR). The UE can predefine a mapping table (or curve) based on network configuration, which describes the mapping relationship between the SINR and the hypothetical BLER under a specific modulation and coding scheme. The UE obtains the BLER value corresponding to the SINR obtained in the previous step by querying the SINR-BLER mapping table. This BLER is not obtained through actual data transmission and decoding failure statistics, but is theoretically calculated based on channel quality; therefore, it is also called the hypothetical block error rate (BLER). This parameter represents the expected performance of a traditional codebook scheme under these channel conditions.

[0072] Simultaneously, the UE can use the second precoding CSI-RS to calculate a second performance metric for the AI / ML model scheme, such as the equivalent signal-to-interference-plus-noise ratio (SINR) or the assumed block error rate (BIR). By comparing the second performance metric with the first performance metric, the UE can determine whether the AI / ML model-based scheme is superior to the traditional codebook scheme. The UE can obtain the precoding matrix output by the gNB based on the second precoding CSI-RS. The equivalent signal-to-interference-plus-noise ratio (SINNR) of the AI / ML model is calculated using the original channel matrix H obtained from the standard CSI-RS received at time T1. The calculation method for the assumed block error rate (BIR) of the AI / ML model is the same as that of the traditional codebook scheme, and will not be repeated here.

[0073] S6, the UE determines whether the second performance indicator is lower than the first performance indicator; if so, it executes S7; otherwise, it continues to return to execute S1~S5.

[0074] If the second performance metric is better than the first performance metric—for example, if the equivalent signal-to-interference-plus-noise ratio (SNR) of the traditional codebook scheme is lower than that of the AI / ML model-based scheme, or if the assumed block error rate (APR) of the traditional codebook scheme is higher than that of the AI / ML model-based scheme—then it indicates that the performance of using the AI / ML model is superior to that of the traditional codebook scheme. Conversely, it indicates that the performance of using the AI / ML model is inferior to that of the traditional codebook scheme.

[0075] If the performance of the AI / ML model is better than that of the traditional codebook solution, then continue to monitor the performance of the AI / ML model.

[0076] S7, the UE sends an AI-CSI performance degradation report (hereinafter referred to as the performance degradation report) to the gNB.

[0077] If the performance of the AI / ML model is inferior to that of the traditional codebook solution, the UE triggers a performance degradation report to the gNB, which indicates that the performance of the AI / ML model is lower than that of the traditional codebook solution.

[0078] This monitoring scheme is based on a single measurement result triggering the report. That is, as long as the performance of the AI / ML model is lower than that of the traditional codebook scheme, the gNB will be notified of the AI / ML model performance degradation. This performance degradation reporting triggering mechanism is too sensitive and easily affected by instantaneous channel fluctuations, resulting in frequent reporting and potential mode switching, which in turn increases signaling overhead.

[0079] While precoded CSI-RS-based schemes have lower overhead, they lack a clear and effective triggering criterion. Simply reporting whenever a performance metric is low (i.e., initiating a report whenever the detected performance metric falls below a threshold) leads to high and unstable system signaling overhead.

[0080] At the 3GPP RAN1#120 meeting, Apple proposed exploring a mechanism similar to Radio Link Failure (RLF) or Beam Failure Detection (BFD) for UE-side performance monitoring, with the UE initiating a report. While the proposal pointed in the direction of "similar to RLF / BFD," it did not provide specific technical solutions for how to implement this mechanism.

[0081] In addition, at the 3GPP RAN1#120 meeting, ViVo proposed two triggering logics for performance monitoring based on precoded CSI-RS:

[0082] One triggering logic is to use the probability that the actual KPI is lower than the threshold as the UE-side monitoring KPI. When the probability exceeds the set threshold, two types of operations can be triggered: one is to directly trigger a report, and the other is to trigger network-side monitoring / monitoring decision.

[0083] Another triggering logic is to use the actual KPI being lower than a threshold as the triggering condition for a UE-side monitoring event. The reporting of this event can trigger network-side monitoring and monitoring decisions.

[0084] Event triggering is essentially still based on the instantaneous result of a single measurement, which cannot effectively solve the problem of false triggering caused by instantaneous channel fluctuations, nor can it respond to continuous performance degradation.

[0085] To address the aforementioned issues, this invention provides a performance monitoring report triggering method based on an AI / ML model. This method, building upon precoded CSI-RS for performance monitoring, designs an adaptive Cumulative Sum (CUSUM) algorithm for the channel environment. This algorithm can capture performance degradation trends without being overly sensitive to outliers, thus avoiding frequent reporting. Specifically, the gNB can configure different CUSUM parameter sets based on different UE states. Each CUSUM parameter set includes different relaxation factors k and decision threshold h. The UE can dynamically select the CUSUM parameter set best suited to its current state, thus implementing an adaptive CUSUM algorithm for CSI performance monitoring.

[0086] The performance monitoring report triggering method for AI / ML models provided by this invention will be described in detail below with reference to the accompanying drawings.

[0087] Please see Figure 5This document illustrates a flowchart of a CSI performance monitoring report triggering method based on an AI / ML model, as provided in an embodiment of the present invention. This embodiment uses a gNB as the network device and a UE as the terminal device for illustration. In practical applications, the network device can be any device on the network side capable of providing network communication functions, and the terminal device can be of various forms; this application does not limit its scope. Figure 5 As shown, the method may include the following steps:

[0088] S101, the gNB sends the first message to the UE. The first message carries performance-aware configuration information.

[0089] For example, the first message could be an RRC reconfiguration message. For instance, performance-aware configuration information can be configured to the UE using information elements (IEs) in the RRC reconfiguration message.

[0090] In other embodiments, the existing IEs in the RRC reconfiguration message can also be used to carry performance-aware configuration information. In this scenario, it is necessary to identify whether these IEs are currently carrying the information that the IE originally carried or the newly added performance-aware configuration information.

[0091] This performance-aware configuration information is used to configure the parameter configuration information required for UE to perform performance monitoring based on AI / ML models, such as precoding CSI-RS configuration information, traditional codebook baseline configuration information, CUSUM parameter set list configuration information, and parameter mapping rule configuration information.

[0092] Configuration information for precoded CSI-RS: Used to configure the precoded CSI-RS resources used to carry the gNB side, for example, the precoded CSI-RS obtained by precoding the CSI-RS using the precoded matrix output by the decoder of the AI / ML model.

[0093] Traditional codebook baseline configuration information: This specifies the traditional codebook configuration (e.g., a specific eType2 codebook subset or parameters) that the UE should refer to when calculating the performance metrics of the traditional codebook scheme.

[0094] The CUSUM parameter set list configuration information is used to configure a list containing multiple parameter sets (cusumParameterSetList). Each CUSUM parameter set includes: CUSUM parameter set ID (parameterSetId), relaxation factor k, and decision threshold h.

[0095] CUSUM parameter set ID: Used to uniquely identify the CUSUM parameter set.

[0096] Relaxation factor k: Used to absorb normal fluctuations in the performance differences between AI / ML models and traditional codebook solutions, avoiding the accumulation of small, transient performance differences.

[0097] Decision threshold h: This represents the threshold at which the UE decides whether to trigger a performance degradation report. If the cumulative sum of performance differences at the current monitoring time is greater than this decision threshold, a performance degradation report will be triggered.

[0098] Different CUSUM parameter sets have different values ​​for k and h, corresponding to different UE states. The UE state here can include the state information of the environment in which the UE is located, and / or the state of the UE itself. In other words, the UE can select the CUSUM parameter set that best matches its current state, thereby realizing a channel environment-adaptive CUSUM parameter set mechanism. For example, in service scenarios requiring low latency and high reliability, the UE can choose a parameter set with smaller values ​​for k and h to achieve rapid and sensitive detection of minor performance degradation; while in scenarios with drastic channel changes, such as high-speed movement, the UE can choose a parameter set with larger values ​​for k and h to filter out noise caused by rapid channel fading and avoid frequent false alarms.

[0099] Configuration information for parameter set mapping rules: This information informs the UE how to select which parameter set based on the UE's state. This configuration information includes mapping parameters and a mapping relationship table.

[0100] Mapping Criteria: These specify the basis for mapping UE parameters, such as mobility state, average channel quality (CQI Level), service type, or the UE's own state. The gNB can be configured to use common mapping parameters for all UEs within its service range, or it can be configured to use different mapping parameters for UEs within its service range; this invention does not limit this.

[0101] The mapping table provides the mapping relationship between mapping parameters and CUSUM parameter sets. For example, the relaxation factor k and decision threshold h in the CUSUM parameter set differ depending on the UE's mobility state; different CUSUM parameter sets are used depending on the average channel quality of the channel the UE is in; different CUSUM parameter sets are used depending on the UE's service type; and different CUSUM parameter sets are also used depending on the UE's own state. Of course, it is also possible to configure the mapping relationship between at least two combinations of mapping parameters and different CUSUM parameter sets.

[0102] Optionally, the gNB can determine whether to send the first message to the UE based on the UE type and UE performance. For example, if a terminal in the Internet of Things does not have the ability to deploy AI / ML models, the gNB will not send the first message to such a terminal.

[0103] S102, at each monitoring time t, gNB sends standard CSI-RS resources and precoded CSI-RS resources.

[0104] Standard CSI-RS resources refer to the physical resources (such as time-frequency resources) that provide the UE with the CSI-RS required for channel inference; precoding CSI-RS resources refer to the periodic or semi-persistent CSI-RS resources that provide the UE with for performance monitoring.

[0105] The standard CSI-RS is a conventional reference signal transmitted by the base station. The UE calculates the channel information, i.e., the channel matrix H, by receiving and processing the standard CSI-RS. In order for the UE to accurately measure the true characteristics of the channel (such as multipath, fading, etc.), the standard CSI-RS itself must be transmitted in a predefined manner without carrying additional dynamic information. That is, the standard CSI-RS is an uncoded reference signal.

[0106] Precoded CSI-RS is a special reference signal specifically used for monitoring the performance of AI / ML models.

[0107] The gNB uses the encoded and compressed CSI reported by the UE at the previous monitoring time to decode and reconstruct the output CSI as a precoding matrix, and then precodes the CSI-RS. In other words, the gNB implicitly transmits the decoded CSI (i.e., the reconstructed precoding matrix) through the precoded CSI-RS. After receiving the precoded CSI-RS, the UE can analyze it to obtain the quality of the CSI on the gNB side, thereby evaluating the performance of the entire AI / ML model.

[0108] S103, the UE calculates the first performance index of the codebook scheme based on the standard CSI-RS, calculates the second performance index of the AI / ML model scheme based on the pre-coded CSI-RS, and obtains the performance difference index y at the current monitoring time. t .

[0109] At each monitoring time t, the UE obtains channel information (i.e., channel matrix H) through the measurement standard CSI-RS, and further calculates the first performance index of the traditional codebook scheme under this channel using the configured traditional codebook benchmark, such as equivalent SINR or assumed BLER. The calculation process can be found in [link to relevant documentation]. Figure 3 The relevant content in S5. In this embodiment, the first performance index is taken as an example using the equivalent SINR, denoted as SINR. Legacy,t .

[0110] Simultaneously, at each monitoring time t, the UE can obtain the output CSI (i.e., the reconstructed precoding matrix) from the AI / ML model decoding and reconstruction on the gNB side by transmitting the signal through precoded CSI-RS. ), based on the reconstructed precoding matrix The second performance metric of the AI / ML model scheme, such as the equivalent SINR or the assumed BLER, is calculated from the original channel matrix H. The calculation process is similar to that of the performance metric calculation in the traditional codebook scheme and will not be repeated here. This embodiment uses the equivalent SINR as an example for the second performance metric, denoted as SINR. AI,t .

[0111] The performance difference index for a single monitoring session at monitoring time t is the difference between the performance index of the traditional codebook solution and the performance index of the AI / ML model solution. The calculation formula is as follows:

[0112] (1)

[0113] y t A value greater than 0 indicates that at monitoring time t, the traditional codebook solution outperforms the AI / ML model solution. The purpose is to detect whether a persistent positive shift has occurred, meaning that the performance of the AI / ML model solution is consistently inferior to that of the traditional codebook solution.

[0114] Formula (1) takes the equivalent signal-to-interference-plus-noise ratio (SINR) as the performance index. In other embodiments of the present invention, the performance difference index can also be the difference between the assumed BLER of the traditional codebook scheme and the assumed BLER of the AI / ML model scheme, i.e. .

[0115] Theoretically, the performance of AI / ML model solutions is expected to be no worse than that of traditional codebook solutions, i.e., y t <0, therefore a target offset for the performance difference metric can be set. If a certain advantage is expected in the AI / ML model solution, such as a GdB improvement over traditional codebook solutions, then the following setting can be made: This invention is based on Let's take an example to illustrate. Target offset. Used as a benchmark value.

[0116] S104, the UE determines the target CUSUM parameter set that matches the current state.

[0117] The UE queries the UE's status information corresponding to the mapping parameters and mapping relationship table configured in the gNB, and further queries the mapping relationship table to obtain the CUSUM parameter set corresponding to this status information, i.e., the target CUSUM parameter set, including the relaxation factor k. t and decision threshold h t .

[0118] S105, the UE utilizes the target CUSUM parameter set and performance difference index y tUpdate the cumulative performance difference and S at the current monitoring time. t .

[0119] For example, at each monitoring time t, the cumulative performance difference S is updated according to the following formula. t :

[0120] (2)

[0121] Where S0=0, this formula indicates that only the performance difference index y at the current monitoring time is considered. t Greater than the target offset and relaxation factor k t When the sum of these is reached, the cumulative performance difference S t Only then will it increase. When When the performance difference is zero, it indicates that the performance of the AI-ML model has improved compared to the previous monitoring time. In other words, the performance degradation at the previous monitoring time was temporary. This temporary performance degradation is not accumulated, but only for the situation of continuous deterioration. Therefore, the cumulative sum of performance differences at the current monitoring time is set to 0, and the cumulative sum of performance differences is recalculated to avoid the impact of temporary performance degradation on the performance degradation judgment.

[0122] Relevant statistical theories and practices have proven that when the relaxation factor k is set to half of the target offset, the CUSUM plot has the highest efficiency for detecting offsets of this specific size, i.e., the shortest average detection time. Therefore, in this invention, the normal value of the relaxation factor k (the value of k under normal circumstances) is usually set to the target offset. Half of, for example, if the target offset is If the value is 3dB, then k = 1.5dB.

[0123] S106, UE comparison performance difference cumulative sum S t Is it greater than the decision threshold h corresponding to the current monitoring time t? t If so, then execute S107.

[0124] When S t >h t If the triggering conditions for reporting are met, a performance degradation report is reported to the gNB. If S t ≤h t Then, continue monitoring the performance differences at the next monitoring time.

[0125] S107, the UE sends a performance degradation report to the gNB.

[0126] The UE can build a dedicated media access control (MAC) layer control element (CE) bearer performance degradation report.

[0127] For example, the performance degradation report may include not only trigger indications, but also key diagnostic information fields such as severity level and CUSUM parameter set ID, which can provide gNB with more diagnostic information.

[0128] Trigger indicator: Used to inform gNB that a performance degradation event has been triggered, and can occupy 1 bit.

[0129] Severity Level: This quantifies the degree of performance degradation of the AI / ML model solution relative to the traditional codebook solution when a report is triggered, occupying 2 to 3 bits. For example, 2 bits are used to encode 4 levels: 00: mild warning; 01: moderate warning; 10: severe warning; 11: reserved or other special cases.

[0130] CUSUM parameter set ID: Optional field, can occupy 1~2 bits. If the gNB is configured with multiple CUSUM parameter sets, the UE can report the CUSUM parameter set ID used when triggering, so that the gNB can know the UE's status.

[0131] For example, when the UE obtains an uplink transmission opportunity, it sends the constructed dedicated MAC CE to the gNB. For instance, the dedicated MAC CE may be transmitted to the gNB along with other data or control information, or the dedicated MAC CE may be transmitted through a specifically allocated PUSCH resource.

[0132] S108, after the UE successfully sends the performance degradation report, it will... t Reset to 0.

[0133] After the performance degradation report was successfully submitted, S t =0, which means the cumulative sum of performance differences is recalculated.

[0134] S109, gNB resolves performance degradation reports and executes corresponding response procedures based on the resolution results and response strategies.

[0135] After receiving a performance degradation report, the gNB parses it to obtain the triggering event and severity level, and optionally, the CUSUM parameter set ID used at the time of triggering. Based on the parsed information, known information on the gNB side (such as the historical performance of the AI / ML model, the UE's location, configuration, etc.), and the response strategy, the gNB executes the corresponding response.

[0136] For example, the response strategy on the gNB side may include:

[0137] Low severity: Record and observe. The accumulated data from the records is used for long-term performance analysis and model evaluation.

[0138] Medium Severity: Send a handover command (e.g., downlink DCI or MAC CE) to the UE to temporarily switch to traditional codebook CSI feedback to ensure link stability. Further, consider adjusting the UE's CUSUM parameter. If the parameter is determined to be overly sensitive (e.g., frequently triggering low-severity performance degradation reports, which in turn triggers medium-severity reports), the CUSUM parameter corresponding to this context in the mapping table can be updated, such as increasing h or k.

[0139] High severity: Forces a switch to traditional codebook CSI feedback, triggering AI / ML model lifecycle management. For example, it can perform lifecycle management processes such as online updates, retraining, or version rollback for AI / ML models in relevant regions or scenarios.

[0140] This embodiment provides a CSI performance monitoring report triggering method based on an AI / ML model, offering a CUSUM algorithm that adapts to the channel environment. This algorithm configures different CUSUM parameter sets for different UE states. Each CUSUM parameter set includes a relaxation factor k and a decision threshold h, with different values ​​for k and h in different parameter sets. The UE can select the most suitable CUSUM parameters based on its current state (or scenario), enabling the UE's performance monitoring mechanism to intelligently balance detection sensitivity and robustness. This approach resolves the inherent contradiction of fixed threshold or counting schemes failing to meet different needs in a variable wireless environment, significantly improving the reliability and adaptability of the AI / ML model. Furthermore, this scheme accumulates performance differences using the CUSUM algorithm and then compares them with a decision threshold to determine whether a performance degradation report is triggered. By accumulating performance differences and effectively filtering instantaneous channel jitter, it ensures that a report is only triggered when the AI / ML model's performance shows a sustained decline, thereby reducing uplink signaling overhead caused by accidental touches and resource waste from subsequent network responses (such as mode switching).

[0141] Moreover, the performance degradation report reported by this scheme can also carry the severity level and the CUSUM parameter set ID used when it is triggered. In this way, the gNB is provided with more dimensions of performance degradation information with minimal additional overhead. This allows the gNB to obtain the severity of the degradation and the scenario in which the UE is located when the problem occurs after receiving the performance degradation report, which further enables the gNB's response strategy to be highly differentiated and intelligent.

[0142] The following section will introduce the mapping parameters and their corresponding mapping relationships. As mentioned earlier, the mapping parameters may include mobility status, average channel quality, service type, and UE's own status.

[0143] 1) Mobility Status

[0144] In 3GPP communication standards, a UE's mobility state is typically determined based on the number of cell reselections / handovers the UE experiences within a certain timeframe. For example, four mobility states can be defined:

[0145] Stationary: If the UE does not undergo cell reselection or handover within a period of T_cr_max (e.g., 10s), it indicates that the UE's location is fixed. In this state, the channel state is relatively the most stable and most sensitive to small but continuous degradation of AI / ML model performance. A small k value can be used to allow small differences to accumulate, and a small h value can be used to ensure rapid triggering.

[0146] Low Mobility: The number of cell reselection / handovers performed by the UE within a certain period (e.g., 10 seconds) is low, below the moderate mobility threshold N_cr_M (e.g., 2 times). For example, a UE walking or in a slowly moving vehicle. In this state, channel changes are slow and may fluctuate. Appropriately increasing the k value absorbs normal channel fluctuation noise, while increasing the h value avoids false alarms caused by momentary jitter due to slow movement.

[0147] Medium Mobility: The number of cell reselection / handovers performed by the UE within a certain period is greater than or equal to the medium mobility threshold N_cr_M but less than the high-speed mobility threshold N_cr_H (e.g., 5 times). For example, a UE in a vehicle traveling on a regular road. In this state, channel changes are rapid, requiring a larger k value to smooth out the performance difference index y. t At the same time, a higher h value is used to ensure that reports are only made when there is a significant and sustained downward trend in performance.

[0148] High Mobility: The number of cell reselection / handovers experienced by the UE within a certain period of time is greater than or equal to the high-speed mobility threshold N_cr_H. For example, a UE on a highway or high-speed train. In this state, the channel changes drastically and fading is rapid, with the largest k and h values, to achieve the strongest robustness. This filters out pseudo-performance degradation caused by rapid fading or frequent handovers, focusing only on truly serious failures of the AI / ML model.

[0149] For example, the gNB can configure counting threshold values ​​(such as N_cr_M, N_cr_H) and timing duration (T_cr_max) in the UE via RRC signaling. The gNB can flexibly adjust the criteria for judging mobility status by updating the technical threshold values ​​and timing duration according to the actual scenario (such as dense urban centers, open highways, etc.).

[0150] In one example, with a timing duration T_cr_max = 10s, N_cr_M = 2, and N_cr_H = 5, the UE determines its own mobility status in the following ways:

[0151] If the UE performs 0 cell handovers within 10 seconds, it is determined to be in a stationary state; if the UE performs 1 cell handover within 10 seconds, it is determined to be in a low mobility state; if the UE performs 2 to 4 cell handovers within 10 seconds, it is determined to be in a medium mobility state; if the UE performs 5 or more cell handovers within 10 seconds, it is determined to be in a high mobility state.

[0152] The mapping relationship between UE mobility status and CUSUM parameter set is shown in the table below:

[0153] Table 1

[0154]

[0155] 2) Average channel quality

[0156] The UE can calculate the average channel quality indicator (CQI) over a period of time as a measure of channel quality. For example, three average CQI levels can be defined as follows:

[0157] High: 12 ≤ average CQI ≤ 15, indicating good channel quality. Performance fluctuations in this state mainly stem from the AIML model itself. A high-sensitivity configuration, i.e., small k-values ​​and small h-values, should be used to quickly detect model problems.

[0158] Medium: 7 ≤ average CQI ≤ 11, indicating that the channel quality fluctuates to some extent. This is a relatively common scenario, and a medium k and h value can be used to balance detection sensitivity and false alarm rate.

[0159] Low: 1 ≤ average CQI ≤ 6, indicating very poor channel quality and large performance jitter. In this state, robust parameter configurations should be used, such as large k and h values, to avoid misinterpreting channel degradation as a performance decline in the AI / ML model and reduce false alarms.

[0160] The mapping relationship between the UE's average CQI level and the CUSUM parameter set is shown in Table 2:

[0161] Table 2

[0162]

[0163] 3) Business Type

[0164] The gNB can learn the service type and Quality of Service (QoS) requirements currently serving the UE. It can configure corresponding CUSUM parameter sets based on the QoS requirements of different service types. For example, for services with extremely high reliability and latency requirements (such as ultra-reliability and ultra-low-latency communication (URLLC) services), sensitive parameter configurations (such as small k and h values) should be used to achieve real-time detection and immediate reporting of performance degradation. For services focused on long-term throughput (such as services in enhanced mobile broadband (eMBB) scenarios), which have a certain tolerance for short-term performance fluctuations, standard parameter configurations (such as medium k and h values) can be used. For services with small data packets and low latency sensitivity (such as services in massive machine-type communication (mMTC) scenarios), to avoid excessive performance reporting storms from a large number of UEs, insensitive parameter configurations (such as large k and h values) should be used, reporting only severe and continuously deteriorating performance issues.

[0165] The mapping relationship table corresponding to the service types of UE is shown in Table 3:

[0166] Table 3

[0167]

[0168] Tables 1 to 3 above are merely examples. The relaxation factor k and decision threshold h in the CUSUM parameter set can be configured according to the actual application. This invention does not impose any special restrictions on the values ​​of k and h in the CUSUM parameter set.

[0169] 4) UE's own state

[0170] The UE's own state encompasses all internal states originating from the UE itself that may affect AI / ML model performance or UE reporting strategies, thereby making the performance monitoring mechanism more intelligent and more aligned with real-world scenarios. Some possible UE states may include:

[0171] ① Power / Battery State

[0172] The UE's remaining battery power or current power consumption mode. When the UE's battery is too low, it may enter a power-saving mode, actively reducing the performance of its processor, which may affect the real-time performance and accuracy of AI / ML model inference.

[0173] For example, when the UE has sufficient power, performance monitoring can be performed using the standard parameter set. When the UE's power is below 20%, a less sensitive CUSUM parameter set (such as a larger relaxation factor k and decision threshold h) can be selected to reduce reporting triggered by performance fluctuations (which may be caused by power consumption optimization), thereby saving power consumption for reporting transmission.

[0174] ②Temperature Control / Heat Dissipation Status

[0175] When performing complex calculations (AI / ML model inference), the UE may generate a lot of heat. When the temperature is too high, the UE will trigger thermal protection, forcibly reducing the processor's processing frequency (i.e., frequency reduction), which will directly affect the performance of the AI / ML model.

[0176] For example, when the UE is running within the normal temperature range, it can use the standard parameter set. If the UE detects that the processor temperature is too high and enters a frequency reduction mode, in this state, the UE can choose to use a parameter set with higher tolerance, namely a larger relaxation factor k and decision threshold h, to avoid the temporary performance degradation caused by frequency reduction being misjudged as a continuous deterioration of the AI / ML model's performance, thereby avoiding unnecessary rollback, i.e., reverting to the traditional codebook scheme.

[0177] ③ Beam Management State

[0178] In 5G (especially millimeter wave) systems, beam stability is crucial for ensuring communication quality. The beam management status of a UE directly reflects the stability of the connection, such as whether beam switching occurs frequently or whether beam failure (BFD) has been experienced.

[0179] For example, in a beam-stabilized state, the UE is locked onto one or a few stable strong beams. At this time, the channel is relatively stable, and a more sensitive parameter set can be used to quickly detect subtle performance degradation of the AI / ML model.

[0180] In beam switching / beam instability mode, the UE is experiencing frequent beam switching or has just recovered from a beam failure. At this time, the channel changes drastically, and a more robust (less sensitive) parameter set should be selected (such as a larger relaxation factor k and decision threshold h) to filter out the drastic performance jitter caused by beam changes.

[0181] ④ Hardware Capability / Processing Load State

[0182] This status includes the capability level of the UE's own AI processing unit (such as the neural network processing unit, NPU) and the current processor load.

[0183] For example, under low load conditions, the UE is not currently running other applications that consume a lot of computing resources (such as large games or video rendering), the NPU resources are sufficient, the AI / ML model performance is stable, and the normal parameter set can be used.

[0184] Under high load, the UE is running multiple high-load applications, and the computing resources allocated to CSI processing may be preempted, causing performance fluctuations in the AI / ML model. In this situation, the UE can switch to a more tolerant parameter set (such as a larger relaxation factor k and decision threshold h) to avoid false reporting caused by performance fluctuations due to resource contention.

[0185] In summary, the UE's own state is a multi-dimensional concept that enables the UE to more intelligently diagnose itself. Before reporting performance degradation to the network side, it can first determine whether the performance decline is caused by the AI / ML model or by special circumstances due to the UE's own state. In this way, the accuracy of reports can be greatly improved, signaling overhead can be reduced, and more refined network operation and maintenance can be achieved.

[0186] In addition, multiple state combinations can be configured to map to the corresponding CUSUM parameter sets. For example, the gNB can configure the mapping between the UE's mobility state and UE's own state as two state combinations and the CUSUM parameter set. The gNB can also configure the mapping between the UE's service type and average channel quality combination and the CUSUM parameter set, which will not be detailed here.

[0187] Figure 6 This is a schematic block diagram of a communication device provided in an embodiment of this application.

[0188] like Figure 6 As shown, the communication device may include a processing module 201 and a communication module 202. The processing module 201 can implement corresponding processing functions; optionally, the processing module 201 may also be referred to as a processing unit. The communication module 202 can implement corresponding communication functions, which can be internal communication functions of the communication device or communication functions between the communication device and other devices. Optionally, the communication module 202 may also be referred to as a communication interface, transceiver module, or transceiver unit.

[0189] Optionally, the communication device further includes a storage module, which can be used to store instructions and / or data; the processing module 201 can read the instructions and / or data in the storage module so that the communication device can implement the aforementioned method embodiments.

[0190] In one possible design, the communication device may correspond to the terminal device in the above method embodiments, or to a component (such as a circuit, chip, or chip system) configured in the terminal device. The communication device can be used to execute the steps or processes performed by the terminal device in any of the above method embodiments.

[0191] In one possible implementation, the processing module 201 is used to obtain the performance difference index between the traditional codebook scheme and the AI / ML model at the current monitoring time; and to determine the target CUSUM parameter set based on the current state of the terminal device and the CUSUM parameter set configuration information, wherein different states of the terminal device correspond to different CUSUM parameter sets, and each CUSUM parameter set includes a relaxation factor and a decision threshold, and the relaxation factor and decision threshold are different in different CUSUM parameter sets; further, based on the performance difference index and the relaxation factor in the target CUSUM parameter set, the cumulative sum of performance differences at the current monitoring time is updated, wherein the cumulative sum of performance differences at the current monitoring time is the sum of performance differences corresponding to the previous monitoring time plus the portion of the performance difference index at the current monitoring time that exceeds a preset cumulative threshold, and the preset cumulative threshold includes the relaxation factor in the target CUSUM parameter set.

[0192] The communication module 202 is used to send a performance degradation report to the network device when the processing module 201 determines that the cumulative performance difference is greater than the decision threshold in the target CUSUM parameter set. The performance degradation report is used to indicate that the performance of the AI / ML model is lower than that of the traditional codebook scheme.

[0193] In another possible implementation, the cumulative sum of performance differences at the current monitoring moment is updated according to the following formula:

[0194]

[0195] Among them, S t S is the cumulative sum of performance differences at the current monitoring time t. t-1 y is the cumulative sum of performance differences corresponding to the previous monitoring time t-1. t Let k be the performance difference metric corresponding to the current monitoring time t. t The relaxation factor used at the current monitoring moment. This is the target offset.

[0196] In another possible implementation, the processing module 201 is also configured to set the cumulative sum of performance differences at the current monitoring moment to 0 after sending a performance degradation report to the network device.

[0197] In another possible implementation, the communication module 202 is also used to receive a first message from the network device. The first message includes performance-aware configuration information, which includes CUSUM parameter set list information and parameter set mapping rules. The CUSUM parameter set list information includes CUSUM parameter sets, and the parameter set mapping rules include the mapping relationship between different state parameters of the terminal device and the CUSUM parameter sets.

[0198] In another possible implementation, the first message is an RRC reconfiguration message.

[0199] In another possible implementation, performance-aware configuration information is carried by adding new cells in the RRC reconfiguration message.

[0200] In another possible implementation, the status parameters of the terminal device include at least one of mobility status, channel quality, service type, and the terminal device's own status, which includes at least one of remaining battery power, operating temperature, beam management status, and hardware capabilities.

[0201] In another possible implementation, the mapping between mobility states and the CUSUM parameter set includes:

[0202] The first CUSUM parameter set corresponding to the static state includes the first relaxation factor and the first decision threshold.

[0203] The low mobility state corresponds to the second CUSUM parameter set, which includes a second relaxation factor and a second decision threshold.

[0204] The mobility state corresponds to the third CUSUM parameter set, which includes the third relaxation factor and the third decision threshold.

[0205] The high mobility state corresponds to the fourth CUSUM parameter set, which includes the fourth relaxation factor and the fourth decision threshold;

[0206] The values ​​of the first relaxation factor, the second relaxation factor, the third relaxation factor, and the fourth relaxation factor increase sequentially, as do the values ​​of the first decision threshold, the second decision threshold, the third decision threshold, and the fourth decision threshold.

[0207] In another possible implementation, the mapping relationship between channel quality and parameter set includes:

[0208] Excellent channel quality corresponds to the fifth CUSUM parameter set, which includes the fifth relaxation factor and the fifth decision threshold.

[0209] The channel quality is moderate, which corresponds to the sixth CUSUM parameter set. The sixth CUSUM parameter set includes the sixth relaxation factor and the sixth decision threshold.

[0210] Poor channel quality corresponds to the seventh CUSUM parameter set, which includes the seventh relaxation factor and the seventh decision threshold.

[0211] The values ​​of the fifth, sixth, and seventh relaxation factors increase sequentially; the values ​​of the fifth, sixth, and seventh decision thresholds also increase sequentially.

[0212] In another possible implementation, the mapping relationship between the service type of the terminal device and the parameter set includes:

[0213] The ultra-reliable low-latency communication type corresponds to the ninth CUSUM parameter set, which includes the ninth relaxation factor and the ninth decision threshold.

[0214] Enhanced mobile broadband services correspond to the tenth CUSUM parameter set, which includes the tenth relaxation factor and the tenth decision threshold.

[0215] Massive machine-type communication services correspond to the eleventh CUSUM parameter set, which includes the eleventh relaxation factor and the eleventh decision threshold.

[0216] The values ​​of the ninth, tenth, and eleventh relaxation factors increase sequentially, as do the values ​​of the ninth, tenth, and eleventh decision thresholds.

[0217] In another possible implementation, the processing module 201 is specifically used to: determine the current state of the terminal device; determine the target CUSUM parameter set ID that matches the current state based on the parameter set mapping rules; and read the CUSUM parameter set corresponding to the target CUSUM parameter set ID.

[0218] In another possible implementation, the performance degradation report includes AI / ML model performance degradation indications and diagnostic information, including performance degradation information when the terminal device triggers the performance degradation report.

[0219] In another possible implementation, the diagnostic information includes at least one of the severity level and the CUSUM parameter set ID used when triggering a performance degradation report, with the severity level indicating the degree of performance degradation of the AI / ML model relative to a traditional codebook scheme.

[0220] In another possible implementation, the communication module 202 is specifically used to carry and send performance degradation reports via a dedicated MAC CE.

[0221] In another possible implementation, the communication module 202 is specifically used to transmit performance degradation reports via uplink shared channel resources or uplink control channel resources.

[0222] The above are merely examples; for detailed steps or procedures, please refer to the descriptions in the foregoing embodiments.

[0223] In one possible design, the communication device may correspond to the network device (which may be a RAN, a core network device, or a functional unit within a core network device) in the above method embodiments, or a component (such as a circuit, chip, or chip system) configured within a network device. The communication device can be used to execute the steps or processes performed by the network device in any of the above method embodiments.

[0224] In one possible implementation, the communication module 202 is configured to: receive a performance degradation report from a terminal device, the performance degradation report being sent when the terminal device detects that the cumulative sum of performance differences at the current monitoring time exceeds a decision threshold in a target CUSUM parameter set; wherein, the target CUSUM parameter set is determined by the terminal device based on the current state of the terminal device, different states of the terminal device correspond to different CUSUM parameter sets, each CUSUM parameter set includes a relaxation factor and a decision threshold, and the relaxation factor and decision threshold are different in different CUSUM parameter sets; the cumulative sum of performance differences at the current monitoring time is updated by the terminal device based on the performance difference index at the current monitoring time and the relaxation factor in the target CUSUM parameter set, the cumulative sum of performance differences at the current monitoring time is the sum of performance differences corresponding to the previous monitoring time plus the portion of the performance difference index at the current monitoring time exceeding a preset cumulative threshold, the preset cumulative threshold including the relaxation factor in the target CUSUM parameter set.

[0225] In another possible implementation, the communication module 202 is further configured to: send a first message, the first message including performance-aware configuration information, the performance-aware configuration information including CUSUM parameter set list information and parameter set mapping rules, the CUSUM parameter set list information including the CUSUM parameter set, and the parameter set mapping rules including the mapping relationship between different state parameters of the terminal device and the CUSUM parameter set.

[0226] In another possible implementation, the first message is an RRC reconfiguration message.

[0227] In another possible implementation, the new information cell in the RRC reconfiguration message carries the performance-aware configuration information.

[0228] The above are merely examples; for detailed steps or procedures, please refer to the descriptions in the foregoing embodiments.

[0229] Figure 7 This is another schematic block diagram of the communication device provided in the embodiments of this application.

[0230] The communication device can be a terminal device, a network device, a chip, chip system, or processor that implements the above methods. This communication device can be used to implement the methods described in the above method embodiments; for details, please refer to the descriptions in the above method embodiments.

[0231] like Figure 7 As shown, the communication device may include one or more processors 301, which may also be referred to as processing units or processing modules, and can implement certain control functions. The processor 301 may be a general-purpose processor or a dedicated processor, such as a baseband processor or a central processing unit. The baseband processor can be used to process communication protocols and communication data, while the central processing unit can be used to control the communication device (e.g., base station, baseband chip, user, user chip), execute software programs, and process data from the software programs.

[0232] In an alternative design, the processor 301 may also store instructions and / or data, which can be executed by the processor 301 to cause the communication device to perform the methods described in the above method embodiments.

[0233] In another alternative design, the communication device may include a communication interface 302 for implementing receiving and transmitting functions. For example, the communication interface 302 may be a transceiver circuit, interface, interface circuit, or transceiver. The transceiver circuit, interface, interface circuit, or transceiver for implementing receiving and transmitting functions may be separate or integrated. The aforementioned transceiver circuit, interface, interface circuit, or transceiver may be used for reading and writing code / data, or it may be used for transmitting or relaying signals.

[0234] Optionally, the communication device may include one or more memories 303, which may store instructions that can be executed on the processor 301 to cause the communication device to perform the methods described in the above method embodiments. Optionally, the memories 303 may also store data. Optionally, the processor 301 may also store instructions and / or data. The processor 301 and the memories 303 may be provided separately or integrated together.

[0235] It should be understood that, in one possible design, the steps in the method embodiments provided in this application can be implemented by integrated logic circuits in the processor's hardware or by instructions in software form. The steps of the methods disclosed in the embodiments of this application can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules in the processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, detailed descriptions are not provided here.

[0236] In one implementation, the communication device can correspond to the UE in the above method embodiments and can be used to execute the various steps and / or processes executed by the UE in the above method embodiments. The processor 301 can be used to execute instructions stored in the memory 303, and when the processor 301 executes the instructions stored in the memory, the processor 301 is used to execute the various steps and / or processes of the above method embodiments corresponding to the terminal device.

[0237] In another implementation, the communication device can correspond to the network device in the above method embodiments and can be used to execute the various steps and / or processes executed by the network device in the above method embodiments. The processor 301 can be used to execute instructions stored in the memory 303, and when the processor 301 executes the instructions stored in the memory, the processor 301 is used to execute the various steps and / or processes of the above method embodiments corresponding to the network device.

[0238] It is understood that the aforementioned processor can be one or more chips. For example, the processor can be a field-programmable gate array (FPGA), an application-specific integrated circuit (ASIC), a system-on-chip (SoC), a central processor unit (CPU), a network processor (NP), a digital signal processor (DSP), a microcontroller unit (MCU), a programmable logic device (PLD), or other integrated chips.

[0239] It is understood that the memory in the embodiments of this application can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM). It should be noted that the memory used in the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0240] This application also provides a computer-readable storage medium storing instructions that, when executed on one or more computing devices, cause the one or more computing devices to perform the data transmission method described in the above embodiments.

[0241] Computer-readable storage media can be non-transitory computer-readable storage media, such as read-only memory (ROM), random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage devices.

[0242] This application also provides a computer program product. When executed by one or more computing devices, the computer program product allows the computing devices to execute any of the aforementioned data transmission methods. The computer program product can be a software installation package. When any of the aforementioned data transmission methods is required, the computer program product can be downloaded and executed on a computer.

[0243] This application also provides a processor, including: an input circuit, an output circuit, and a processing circuit. The processing circuit receives signals through the input circuit and transmits signals through the output circuit, causing the processor to execute the data transmission method described in the above embodiments.

[0244] In specific implementation, the processor can be one or more chips, the input circuit can be input pins, the output circuit can be output pins, and the processing circuit can be transistors, gate circuits, flip-flops, and various logic circuits. The input signal received by the input circuit can be received and input by, for example, but not limited to, a receiver, and the signal output by the output circuit can be output to, for example, but not limited to, a transmitter and transmitted by the transmitter. Furthermore, the input circuit and the output circuit can be the same circuit, which is used as the input circuit and the output circuit at different times. This application does not limit the specific implementation of the processor and various circuits.

[0245] This application also provides a chip system including one or more processors for calling and executing instructions stored in memory, thereby executing the data transmission method described in the above embodiments. The chip system may be composed of a chip or may include chips and other discrete devices. The chip system may include input circuitry or interfaces for transmitting information or data, and output circuitry or interfaces for receiving information or data.

[0246] In the embodiments of this application, the terms and English abbreviations are exemplary examples given for ease of description and should not be construed as limiting the application in any way. This application does not preclude the possibility of defining other terms that can achieve the same or similar functions in existing or future agreements.

[0247] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When these computer instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated.

[0248] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0249] It should be understood that in the various embodiments of this application, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0250] In summary, the above description is merely a preferred embodiment of the technical solution of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for triggering CSI performance monitoring reports based on AI / ML models, characterized in that, The method, executed by a terminal device, includes: Obtain performance difference metrics between traditional codebook solutions and AI / ML models at the current monitoring time; Based on the current state of the terminal device and the CUSUM parameter set configuration information, a target CUSUM parameter set is determined. Different states of the terminal device correspond to different CUSUM parameter sets. Each CUSUM parameter set includes a relaxation factor and a decision threshold. The relaxation factor and decision threshold are different in different CUSUM parameter sets. Update the cumulative sum of performance differences at the current monitoring time. The cumulative sum of performance differences at the current monitoring time is the sum of performance differences at the previous monitoring time plus the portion of the performance difference index at the current monitoring time that exceeds a preset cumulative threshold. The preset cumulative threshold includes the sum of the target offset and the relaxation factor in the target CUSUM parameter set. The target offset is the benchmark value for measuring the performance difference index. If the cumulative sum of the performance differences exceeds the decision threshold in the target CUSUM parameter set, a performance degradation report is sent to the network device. The performance degradation report is used to indicate that the performance of the AI / ML model is lower than that of the traditional codebook scheme.

2. The method according to claim 1, characterized in that, The method further includes: If the sum of the cumulative performance difference at the previous monitoring time and the first difference is less than or equal to 0, the cumulative performance difference at the current monitoring time is set to 0, and the first difference is the difference between the performance difference index at the current monitoring time and the preset cumulative threshold.

3. The method according to claim 1 or 2, characterized in that, The step of updating the cumulative sum of performance differences at the current monitoring moment based on the performance difference index and the relaxation factor in the target CUSUM parameter set includes: Update the cumulative sum of performance differences at the current monitoring moment according to the following formula: Among them, S t S is the cumulative sum of performance differences at the current monitoring time t. t-1 y is the cumulative sum of performance differences corresponding to the previous monitoring time t-1. t Let k be the performance difference metric corresponding to the current monitoring time t. t The relaxation factor used at the current monitoring moment. This is the target offset.

4. The method according to claim 1 or 2, characterized in that, The metrics for obtaining the performance difference between the traditional codebook solution and the AI / ML model at the current monitoring time include: Obtain the first performance index of the traditional codebook scheme at the current monitoring time, the first performance index being calculated based on the standard CSI-RS received at the current monitoring time; Obtain the second performance index of the AI / ML model at the current monitoring time. The second performance index is calculated based on the precoded CSI-RS received at the current monitoring time. Both the first performance index and the second performance index are equivalent signal-to-interference-plus-noise ratio or assumed block error rate. The performance difference index is obtained by calculating the difference between the first performance index and the second performance index.

5. The method according to claim 1 or 2, characterized in that, The method further includes: after sending a performance degradation report to the network device, setting the cumulative sum of performance differences corresponding to the current monitoring time to 0.

6. The method according to claim 1 or 2, characterized in that, Before obtaining the performance difference metric between the traditional codebook scheme and the AI / ML model at the current monitoring time, the method further includes: The system receives a first message from a network device. The first message includes performance-aware configuration information, which includes CUSUM parameter set configuration information and parameter set mapping rules. The CUSUM parameter set configuration information includes multiple CUSUM parameter sets, and the parameter set mapping rules include the mapping relationship between different state parameters of the terminal device and the CUSUM parameter sets.

7. The method according to claim 6, characterized in that, The first message is an RRC reconfiguration message.

8. The method according to claim 7, characterized in that, The newly added information element in the RRC reconfiguration message is used to carry the performance-aware configuration information.

9. The method according to claim 6, characterized in that, The status parameters of the terminal device include at least one of mobility status, channel quality, service type, and the terminal device's own status. The terminal device's own status includes at least one of remaining battery power, operating temperature, beam management status, and hardware capabilities.

10. The method according to claim 9, characterized in that, The mapping relationship between mobility status and the CUSUM parameter set includes: The first CUSUM parameter set corresponding to the static state, the first CUSUM parameter set includes a first relaxation factor and a first decision threshold; The low mobility state corresponds to the second CUSUM parameter set, which includes a second relaxation factor and a second decision threshold. The mobility state corresponds to the third CUSUM parameter set, which includes a third relaxation factor and a third decision threshold. The high mobility state corresponds to the fourth CUSUM parameter set, which includes a fourth relaxation factor and a fourth decision threshold; The values ​​of the first relaxation factor, the second relaxation factor, the third relaxation factor, and the fourth relaxation factor increase sequentially, as do the values ​​of the first decision threshold, the second decision threshold, the third decision threshold, and the fourth decision threshold.

11. The method according to claim 9, characterized in that, The mapping relationship between channel quality and parameter set includes: Excellent channel quality corresponds to the fifth CUSUM parameter set, which includes a fifth relaxation factor and a fifth decision threshold. The channel quality is moderate, which corresponds to the sixth CUSUM parameter set, which includes the sixth relaxation factor and the sixth decision threshold. The channel quality difference corresponds to the seventh CUSUM parameter set, which includes the seventh relaxation factor and the seventh decision threshold. The values ​​of the fifth relaxation factor, the sixth relaxation factor, and the seventh relaxation factor increase sequentially; the values ​​of the fifth decision threshold, the sixth decision threshold, and the seventh decision threshold also increase sequentially.

12. The method according to claim 9, characterized in that, The mapping relationship between the service type and parameter set of the terminal device includes: The ultra-reliable low-latency communication type corresponds to the ninth CUSUM parameter set, which includes the ninth relaxation factor and the ninth decision threshold; The enhanced mobile broadband service corresponds to the tenth CUSUM parameter set, which includes the tenth relaxation factor and the tenth decision threshold. Massive machine-type communication services correspond to the eleventh CUSUM parameter set, which includes the eleventh relaxation factor and the eleventh decision threshold. The values ​​of the ninth relaxation factor, the tenth relaxation factor, and the eleventh relaxation factor increase sequentially, as do the values ​​of the ninth decision threshold, the tenth decision threshold, and the eleventh decision threshold.

13. The method according to claim 1 or 2, characterized in that, The determination of the target CUSUM parameter set based on the current state of the terminal device and the CUSUM parameter set configuration information includes: Determine the current state of the terminal device; Determine the target CUSUM parameter set ID that matches the current state based on the parameter set mapping rules; Read the CUSUM parameter set corresponding to the target CUSUM parameter set ID.

14. The method according to claim 1 or 2, characterized in that, The performance degradation report includes AI / ML model performance degradation indicators and diagnostic information, and the diagnostic information includes performance degradation information when the terminal device triggers the performance degradation report.

15. The method according to claim 14, characterized in that, The diagnostic information includes at least one of the severity level and the CUSUM parameter set ID used when triggering the performance degradation report, wherein the severity level indicates the degree of performance degradation of the AI / ML model relative to a traditional codebook scheme.

16. The method according to claim 1 or 2, characterized in that, Send a performance degradation report to the network device, including: The performance degradation report is carried and sent via a dedicated MAC CE.

17. The method according to claim 1 or 2, characterized in that, Send a performance degradation report to the network device, including: The performance degradation report is transmitted via uplink shared channel resources or uplink control channel resources.

18. A method for triggering CSI performance monitoring reports based on AI / ML models, characterized in that, Performed by a network device, the method includes: Receive a performance degradation report from a terminal device, which is sent when the terminal device detects that the cumulative performance difference at the current monitoring time is greater than a decision threshold in the target CUSUM parameter set; The target CUSUM parameter set is determined by the terminal device based on the current state of the terminal device. Different states of the terminal device correspond to different CUSUM parameter sets. Each CUSUM parameter set includes a relaxation factor and a decision threshold. The relaxation factor and decision threshold are different in different CUSUM parameter sets. The cumulative sum of performance differences at the current monitoring time is obtained by the terminal device based on the performance difference index at the current monitoring time and the relaxation factor in the target CUSUM parameter set. The cumulative sum of performance differences at the current monitoring time is the sum of performance differences at the previous monitoring time plus the portion of the performance difference index at the current monitoring time that exceeds a preset cumulative threshold. The preset cumulative threshold includes the sum of the target offset and the relaxation factor in the target CUSUM parameter set. The target offset is a benchmark value for measuring the performance difference index.

19. The method according to claim 18, characterized in that, Before receiving the performance degradation report from the terminal device, the method further includes: Send a first message, which includes performance-aware configuration information. The performance-aware configuration information includes a CUSUM parameter set list and parameter set mapping rules. The CUSUM parameter set list includes the CUSUM parameter set, and the parameter set mapping rules include the mapping relationship between different state parameters of the terminal device and the CUSUM parameter set.

20. The method according to claim 19, characterized in that, The first message is an RRC reconfiguration message.

21. The method according to claim 20, characterized in that, The newly added information cell in the RRC reconfiguration message carries the performance-aware configuration information.

22. A communication device, characterized in that, It includes a module for performing the method as described in any one of claims 1 to 17, or a module for performing the method as described in any one of claims 18 to 21.

23. A communication device, characterized in that, include: Memory, used to store computer instructions; A processor for executing a computer program or computer instructions stored in the memory, causing the communication device to perform the method as described in any one of claims 1 to 17, or the method as described in any one of claims 18 to 21.

Images