Ai / ML-based CSI feedback
By categorizing CSI based on vector diversity and using category-specific models, the method addresses the challenges of high overhead and complexity in CSI feedback, achieving efficient and accurate CSI monitoring with balanced performance and management.
Patent Information
- Application Number
- PCT/CN2024/077273
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2024-02-16
- Publication Date
- 2025-08-21
AI Technical Summary
In massive MIMO systems, the base station faces challenges in monitoring downlink channel state information (CSI) due to high dimensionality and distortion from inaccurate channel estimation and compression, leading to substantial feedback overhead and computational complexity, which existing AI/ML methods struggle to address effectively.
Implement a categorization rule for grouping CSI based on vector diversity values, enabling efficient model training, management, and inference by partitioning CSI datasets into categories, using category-specific models for encoding and decoding, and aligning models on both network and terminal devices.
This approach reduces monitoring overhead and computational complexity while maintaining high lifecycle management efficiency, achieving accurate CSI feedback with balanced reconstruction performance, model complexity, and feedback overhead.
Smart Images

Figure CN2024077273_21082025_PF_FP_ABST
Abstract
Description
AI / ML-BASED CSI FEEDBACKFIELD
[0001] Various example embodiments relate to the field of communication and in particular, to devices, methods, apparatuses and a computer readable storage medium for artificial intelligence (AI) / machine learning (ML) -based channel state information (CSI) feedback.BACKGROUND
[0002] Multiple-Input Multiple-Output (MIMO) system can utilize multiple independent channels on the same time-frequency resources to achieve transmission capabilities several times that of single-antenna system. However, the base station (BS) with massive MIMO needs to monitor the downlink channel state information (CSI) in real-time to fully exploit the spatial multiplexing gain. In Frequency Division Duplex (FDD) systems, the user equipment (UE) needs to estimate and report the downlink CSI to the next-generation NodeB (gNB) through the uplink control channel, which can be substantial due to the high dimension of CSI in massive MIMO systems.
[0003] To reduce the feedback overheads, UEs employ compression and quantization techniques on the CSI. Conversely, the received compressed CSI needs to be reconstructed to its original form at gNB. Obviously, there are two sources of distortion when the gNB acquires the true CSI, one from inaccurate channel estimation, the other from CSI compression. 3GPP approved a new study item (SI) in Release 18 to explore the benefits of augmenting the CSI feedback by AI / ML techniques. Extensive discussions have taken place regarding the appropriate training architectures, quantization schemes, network structures, and other relevant topics.SUMMARY
[0004] In general, example embodiments of the present disclosure provide a solution for AI / ML-based CSI feedback, especially for a monitoring scheme and model-based lifecycle management (LCM) for AI / ML-based CSI feedback. With this solution, the monitoring measurement is generated based on the categorization rule, thereby the solution has low monitoring overhead and low computational complexity while maintaining high LCM efficiency.
[0005] In a first aspect, there is provided a network device. The network device comprises at least one processor and at least one memory storing instructions. The instructions, when executed by the at least one processor, cause the network device at least to: determine, a categorization rule for grouping channel state information (CSI) ; transmit, to a terminal device, information of the categorization rule; and receive, from the terminal device, at least one monitoring measurement for artificial intelligence (AI) / machine learning (ML) based CSI feedback, wherein the at least one monitoring measurement is generated based on the categorization rule.
[0006] In a second aspect, there is provided a terminal device. The terminal device comprises at least one processor and at least one memory storing instructions. The instructions, when executed by the at least one processor, cause the terminal device at least to: receive, from a network device, information of a categorization rule, wherein the categorization rule is for grouping channel state information (CSI) ; and transmit, to the network device, at least one monitoring measurement for artificial intelligence (AI) / machine learning (ML) based CSI feedback, wherein the at least one monitoring measurement is generated based on the categorization rule
[0007] In a third aspect, there is provided a method. The method comprises determining a categorization rule for grouping channel state information (CSI) ; transmitting, to a terminal device, information of the categorization rule; and receiving, from the terminal device, at least one monitoring measurement for artificial intelligence (AI) / machine learning (ML) based CSI feedback, wherein the at least one monitoring measurement is generated based on the categorization rule.
[0008] In a fourth aspect, there is provided a method. The method comprises receiving, from a network device, information of a categorization rule, wherein the categorization rule is for grouping channel state information (CSI) ; and transmitting, to the network device, at least one monitoring measurement for artificial intelligence (AI) / machine learning (ML) based CSI feedback, wherein the at least one monitoring measurement is generated based on the categorization rule.
[0009] In a fifth aspect, there is provided an apparatus. The apparatus comprises means for determining a categorization rule for grouping channel state information (CSI) ; means for transmitting, to a terminal device, information of the categorization rule; and means for receiving, from the terminal device, at least one monitoring measurement for artificial intelligence (AI) / machine learning (ML) based CSI feedback, wherein the at least one monitoring measurement is generated based on the categorization rule.
[0010] In a sixth aspect, there is provided an apparatus. The apparatus comprises means for receiving, from a network device, information of a categorization rule, wherein the categorization rule is for grouping channel state information (CSI) ; and means for transmitting, to the network device, at least one monitoring measurement for artificial intelligence (AI) / machine learning (ML) based CSI feedback, wherein the at least one monitoring measurement is generated based on the categorization rule.
[0011] In a seventh aspect, there is provided a non-transitory computer readable medium comprising program instructions for causing an apparatus to perform at least the method according to any one of the above third to fourth aspects.
[0012] In an eighth aspect, there is provided a computer program comprising instructions, which, when executed by an apparatus, cause the apparatus to perform at least the method according to any one of the above third to fourth aspects.
[0013] In a ninth aspect, there is provided a network device. The network device comprises: determining circuitry configured to determine, a categorization rule for grouping channel state information (CSI) ; transmitting circuitry configured to transmit, to a terminal device, information of the categorization rule; receiving circuitry configured to receive, from the terminal device, at least one monitoring measurement for artificial intelligence (AI) / machine learning (ML) based CSI feedback, wherein the at least one monitoring measurement is generated based on the categorization rule.
[0014] In a tenth aspect, there is provided a terminal device. The terminal device comprises: receiving circuitry configured to receive, from a network device, information of a categorization rule, wherein the categorization rule is for grouping channel state information (CSI) ; transmitting circuitry configured to transmit, to the network device, at least one monitoring measurement for artificial intelligence (AI) / machine learning (ML) based CSI feedback, wherein the at least one monitoring measurement is generated based on the categorization rule.
[0015] It is to be understood that the summary section is not intended to identify key or essential features of embodiments of the present disclosure, nor is it intended to be used to limit the scope of the present disclosure. Other features of the present disclosure will become easily comprehensible through the following description.BRIEF DESCRIPTION OF THE DRAWINGS
[0016] Some example embodiments will now be described with reference to the accompanying drawings, in which:
[0017] FIG. 1A illustrates an example communication network in which embodiments of the present disclosure may be implemented;
[0018] FIG. 1B illustrates an example structure for AI / ML-based CSI feedback according to some embodiments of the present disclosure;
[0019] FIG. 2 illustrates a flowchart illustrating an example of process for AI / ML-based CSI feedback according to some embodiments of the present disclosure;
[0020] FIG. 3 illustrates a flowchart illustrating another example of process for AI / ML-based CSI feedback according to some embodiments of the present disclosure;
[0021] FIG. 4A illustrates a diagram illustrating example of CSI matrix according to some embodiments of the present disclosure
[0022] FIG. 4B illustrates a diagram illustrating example of calculating vector diversity value according to some embodiments of the present disclosure;
[0023] FIG. 5 illustrates a diagram illustrating example of framework for AI / ML-based CSI feedback according to some embodiments of the present disclosure;
[0024] FIG. 6 illustrates a flowchart of a method implemented at a network device according to some embodiments of the present disclosure;
[0025] FIG. 7 illustrates a flowchart of a method implemented at a terminal device according to some other embodiments of the present disclosure;
[0026] FIG. 8 illustrates a simplified block diagram of an apparatus that is suitable for implementing embodiments of the present disclosure; and
[0027] FIG. 9 illustrates a block diagram of an example computer readable medium in accordance with some embodiments of the present disclosure.
[0028] Throughout the drawings, the same or similar reference numerals represent the same or similar element.DETAILED DESCRIPTION
[0029] Principles of the present disclosure will now be described with reference to some example embodiments. It is to be understood that these embodiments are described only for the purpose of illustration and help those skilled in the art to understand and implement the present disclosure, without suggesting any limitation as to the scope of the disclosure. The disclosure described herein can be implemented in various manners other than the ones described below.
[0030] In the following description and claims, unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skills in the art to which this disclosure belongs.
[0031] References in the present disclosure to “one embodiment, ” “an embodiment, ” “an example embodiment, ” and the like indicate that the embodiment described may include a particular feature, structure, or characteristic, but it is not necessary that every embodiment includes the particular feature, structure, or characteristic. Moreover, such phrases are not necessarily referring to the same embodiment. Further, when a particular feature, structure, or characteristic is described in connection with an embodiment, it is submitted that it is within the knowledge of one skilled in the art to affect such feature, structure, or characteristic in connection with other embodiments whether or not explicitly described.
[0032] It shall be understood that although the terms “first” and “second” etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, a first element could be termed a second element, and similarly, a second element could be termed a first element, without departing from the scope of example embodiments. As used herein, the term “and / or” includes any and all combinations of one or more of the listed terms.
[0033] The terminology used herein is for describing particular embodiments only and is not intended to be limiting of example embodiments. As used herein, the singular forms “a” , “an” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms “comprises” , “comprising” , “has” , “having” , “includes” and / or “including” , when used herein, specify the presence of stated features, elements, and / or components etc., but do not preclude the presence or addition of one or more other features, elements, components and / or combinations thereof. As used herein, “at least one of the following: <a list of two or more elements>” and “at least one of <a list of two or more elements>” and similar wording, where the list of two or more elements are joined by “and” or “or” , mean at least any one of the elements, or at least any two or more of the elements, or at least all the elements.
[0034] As used in this application, the term “circuitry” may refer to one or more or all of the following:
[0035] (a) hardware-only circuit implementations (such as implementations in only analog and / or digital circuitry) and
[0036] (b) combinations of hardware circuits and software, such as (as applicable) :
[0037] (i) a combination of analog and / or digital hardware circuit (s) with software / firmware and
[0038] (ii) any portions of hardware processor (s) with software (including digital signal processor (s) ) , software, and memory (ies) that work together to cause an apparatus, such as a mobile phone or server, to perform various functions) and
[0039] (c) hardware circuit (s) and or processor (s) , such as a microprocessor (s) or a portion of a microprocessor (s) , that requires software (e.g., firmware) for operation, but the software may not be present when it is not needed for operation.
[0040] This definition of circuitry applies to all uses of this term in this application, including in any claims. As a further example, as used in this application, the term circuitry also covers an implementation of merely a hardware circuit or processor (or multiple processors) or portion of a hardware circuit or processor and its (or their) accompanying software and / or firmware. The term circuitry also covers, for example and if applicable to the particular claim element, a baseband integrated circuit or processor integrated circuit for a mobile device or a similar integrated circuit in server, a cellular network device, or other computing or network device.
[0041] As used herein, the term “communication network” refers to a network following any suitable communication standards, such as Long Term Evolution (LTE) , LTE-Advanced (LTE-A) , Wideband Code Division Multiple Access (WCDMA) , High-Speed Packet Access (HSPA) , Narrow Band Internet of Things (NB-IoT) and so on. Furthermore, the communications between a terminal device and a network device in the communication network may be performed according to any suitable generation communication protocols, including, but not limited to, the first generation (1G) , the second generation (2G) , 2.5G, 2.75G, the third generation (3G) , the fourth generation (4G) , 4.5G, the future fifth generation (5G) communication protocols, and / or any other protocols either currently known or to be developed in the future. Embodiments of the present disclosure may be applied in various communication systems. Given the rapid development in communications, there will of course also be future type communication technologies and systems with which the present disclosure may be embodied. It should not be seen as limiting the scope of the present disclosure to only the aforementioned system.
[0042] As used herein, the term “network device” and “access network device” refer to a node in a communication network via which a terminal device accesses the network and receives services therefrom. The network device may refer to a base station (BS) or an access point (AP) , for example, a node B (NodeB or NB) , an evolved NodeB (eNodeB or eNB) , a NR NB (also referred to as a gNB) , a Remote Radio Unit (RRU) , a radio header (RH) , a remote radio head (RRH) , a relay, a low power node such as a femto, a pico, and so forth, depending on the applied terminology and technology.
[0043] The term “terminal device” refers to any end device that may be capable of wireless communication. By way of example rather than limitation, a terminal device may also be referred to as a communication device, user equipment (UE) , a Subscriber Station (SS) , a Portable Subscriber Station, a Mobile Station (MS) , or an Access Terminal (AT) . The terminal device may include, but not limited to, a mobile phone, a cellular phone, a smart phone, voice over IP (VoIP) phones, wireless local loop phones, a tablet, a wearable terminal device, a personal digital assistant (PDA) , portable computers, desktop computer, image capture terminal devices such as digital cameras, gaming terminal devices, music storage and playback appliances, vehicle-mounted wireless terminal devices, wireless endpoints, mobile stations, laptop-embedded equipment (LEE) , laptop-mounted equipment (LME) , USB dongles, smart devices, wireless customer-premises equipment (CPE) , an Internet of Things (loT) device, a watch or other wearable, a head-mounted display (HMD) , a vehicle, a drone, a medical device and applications (e.g., remote surgery) , an industrial device and applications (e.g., a robot and / or other wireless devices operating in an industrial and / or an automated processing chain contexts) , a consumer electronics device, a device operating on commercial and / or industrial wireless networks, and the like. In the following description, the terms “terminal device” , “communication device” , “terminal” , “user equipment” and “UE” may be used interchangeably.
[0044] As stated above, 3GPP has initiated a study item on AI / ML for CSI feedback. However, the black-box nature of AI / ML methods makes it challenging to correlate performance improvements with NN model design. The performance of CSI compression is closely tied not only to the design of the AI model but also to the diversity of the channel environment. And dynamic lifecycle management (LCM) is to facilitate model adaptation in response to channel variation. The LCM comprises model training, functionality / model selection, activation, deactivation, switching, and fallback operation, and functionality / model monitoring.
[0045] According to some embodiments of the present disclosure, there is provided a solution for an efficient and easily implementable LCM scheme, encompassing model training, model management, and model inference. Principles and embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings.
[0046] FIG. 1A illustrates a schematic diagram of an example communication network 100A in which some embodiments of the present disclosure can be implemented. As shown in FIG. 1A, the communication network 100A may include terminal device 110 and network device 120. It is to be understood that the number of network devices and terminal devices is only for the purpose of illustration without suggesting any limitations. The system 100A may include any suitable number of network devices and terminal devices adapted for implementing embodiments of the present disclosure.
[0047] Communications in the communication system 100A may be implemented according to any proper communication protocol (s) , comprising, but not limited to, cellular communication protocols of the first generation (1G) , the second generation (2G) , the third generation (3G) , the fourth generation (4G) and the fifth generation (5G) and on the like, wireless local network communication protocols such as Institute for Electrical and Electronics Engineers (IEEE) 802.11 and the like, and / or any other protocols currently known or to be developed in the future. Moreover, the communication may utilize any proper wireless communication technology, comprising but not limited to: Code Division Multiple Access (CDMA) , Frequency Division Multiple Access (FDMA) , Time Division Multiple Access (TDMA) , Frequency Division Duplex (FDD) , Time Division Duplex (TDD) , Multiple-Input Multiple-Output (MIMO) , Orthogonal Frequency Division Multiple (OFDM) , Discrete Fourier Transform spread OFDM (DFT-s-OFDM) and / or any other technologies currently known or to be developed in the future.
[0048] FIG. 1B illustrates an example structure for AI / ML-based CSI feedback according to some embodiments of the present disclosure. As shown in FIG. 1B, the UE-side entity includes an encoder 132 compressing the original CSI data 131 into the latent feature vector and a quantizer 133 transforming the float-type latent vector into fixed-point codeword for feedback. And the codeword is transmitted over the air. The NW-side entity conversely includes a dequantizer 135 transforming the fixed-point codeword into the latent vector and a decoder 136 reconstructing the original CSI data.
[0049] FIG. 2 illustrates a flowchart illustrating an example of process for AI / ML-based CSI feedback according to some embodiments of the present disclosure. For the purpose of discussion, the process 200 will be described with reference to FIG. 1A. The process 200 may involve the terminal devices 110, the network devices 120 as illustrated in FIG. 1A. It would be appreciated that although the process 200 for link has been described in the communication system 100 of FIG. 1, this process may be likewise applied to other communication scenarios where different network devices are jointly deployed to provide respective serving cells.
[0050] In some embodiments, as shown in FIG. 2, the network device 120 determines 201 a categorization rule for grouping CSI. And the network device 120 transmits to 202 terminal device 110, information of the categorization rule 203. Then the network device 120 receives 207 from the terminal device 110, at least one monitoring measurement 206 for AI / ML based CSI feedback. And the at least one monitoring measurement is generated based on the categorization rule. For the solution of the process, it is an efficient monitoring mechanism for AI / ML-based CSI feedback based on the categorization rule.
[0051] FIG. 3 illustrates a flowchart illustrating another example of process for AI / ML-based CSI feedback according to some embodiments of the present disclosure. In FIG. 3, there is detailed signaling workflow to support above mentioned solution. For the purpose of discussion, the process 300 will be described with reference to FIG. 1A. The process 200 may involve the terminal device 110, and the network devices 120 as illustrated in FIG. 1A.
[0052] It would be appreciated that although the process 300 has been described in the communication system 100A of FIG. 1A, this process may be likewise applied to other communication scenarios where different network devices are jointly deployed to provide respective serving cells.
[0053] In some embodiments, as shown in FIG. 3, the data collection 301 is performed for getting the collected CSI dataset. And at 302, the network device 120 determines the CSI categorization rule. And the categorization rule comprises calculating a vector diversity value of a matrix of CSI, and determining a number of categories of CSI and associated thresholds based on the vector diversity value. Specifically, the network device 110 establishes an CSI categorization rule that includes the method for calculating the vector diversity value and the threshold for grouping CSI with this diversity value.
[0054] FIG. 4A illustrates a diagram illustrating example of CSI matrix according to some embodiments of the present disclosure. As shown in FIG. 4A, the CSI matrix is composed of Nsub eigenvectors, each CSI matrix can be regarded as a cluster of Nsub vectors. Nsub represents the number of subbands. And each row of the matrix represents a subband 401. And each element in one row represents the port 402.
[0055] In some embodiments, the calculating vector diversity value comprises normalizing each eigenvector of the matrix of CSI, and each of the eigenvector is for a subband, calculating an average normalized eigenvector over all the subband and determining the vector diversity value based on the average normalized eigenvector. Specifically, for a normalized vector cluster, a straightforward indicator to evaluate its diversity level is to calculate the L1 norm of the average vector formed by taking the mean of each element across all vectors.
[0056] FIG. 4B illustrates a diagram illustrating example of calculating vector diversity value according to some embodiments of the present disclosure. As shown in FIG. 4B, a small L1-norm of the average eigenvector 405 reflects high vector diversity, which results in high model complexity and feedback overhead 406, and vice versa. And a large L1-norm of the average eigenvector 403 reflects low diversity, which results in low model complexity and feedback overhead 404, and vice versa.
[0057] In some embodiments, assuming the CSI matrix is represented by with cl being the eigenvector of the l th Subband, the vector diversity value per CSI can be calculated as follows: normalizing each eigenvector as calculating average normalized eigenvector over all Subbands and apply L1 norm by The range of α is [0, 1] , and selecting -α as the vector diversity value. A larger value indicates higher diversity.
[0058] Therefore, the number of categories and their associated thresholds can be determined based on the vector diversity value. For example, uniformly divide them into three categories and set the ranges as follows: [-1 / 3, 0] , [-2 / 3, -1 / 3] , and [-1, -2 / 3] . In the training stage, the training data can be divided into three categories based on the categorization rule mentioned above, and three models are trained for each category. In the inference phase, determine the category to which the in-field data belongs and select the CSI feedback model accordingly.
[0059] In some embodiments, the network device 120 determines multiple sub-datasets of a collected CSI dataset based on the categorization rule, and transmits to the terminal device 110, at least one AI / ML model for a CSI feedback encoder, a model ID and the associated category ID. The AI / ML model is trained based on at least one sub-dataset among the multiple sub-datasets.
[0060] Specifically, as shown in FIG. 3, at 303, based on the categorization rule stated above, the collected CSI dataset is partitioned into multiple sub-datasets indexed by category ID. For example, the vector diversity value of each CSI in the collected CSI dataset is calculated, and then the collected CSI dataset partitioned into 3 sub-datasets indexed by category IDs #1, #2, #3, corresponding to the vector diversity value ranges [-1 / 3, 0] , [-2 / 3, -1 / 3] , and [-1, -2 / 3] .
[0061] As shown in FIG. 3, at 304 and 308, for each sub-dataset, an individual model is trained, characterized by its model complexity and feedback overhead. Different models result in distinct reconstruction performances at the network device 120. For instance, a sub-dataset with high vector diversity value (e.g., non-line-of-sight (NLOS) scenario) leads to a model with higher complexity and feedback overhead, while the opposite holds for a sub-dataset with low vector diversity value, resulting in a model with lower complexity and feedback overhead. By adopting a per-category training approach, obtain a set of models, each tailored to accommodate different levels of vector diversity.
[0062] Specifically, based on the selected training type (e.g., Type 1, Type 3 defined in 3GPP 38.843) , two model training options are available. And at 304, it is Type 1 joint training. That means each pair of encoder and decoder, associated with a category ID, is trained at the network device 120 using the corresponding sub-dataset. For example, if there are three category IDs #1, #2, #3, three pairs of encoder and decoder are trained at network device 120 using three corresponding sub-dataset.
[0063] As shown in FIG. 3, the encoder and the corresponding assistance information 306 are transmitted to the terminal device 110 for deployment in CSI compression. The assistance information serves not only for aligning the models at both sides but also for aligning the model with the category. Therefore, it includes at least the model ID and the associated category ID. For example, if there are three category IDs #1, #2, #3 and three corresponding models #1, #2, #3, the assistance information comprises the category IDs and corresponding models IDs.
[0064] In some embodiments, the network device 120 transmits to the terminal device 110, a training dataset for training at least one AI / ML model for a CSI feedback encoder and assistance information for two-side model alignment and model-category alignment.
[0065] Specifically, at 308, it is Type 3 separate training. Taking NW-first separate training as an example, the network device 120 firstly comes up with a hypothetical encoder, a quantizer and a decoder, and then trains these components in an end-to-end manner. On completion of the NW-side model training, different from Type 1 training transferring model which incurs vendor property issue, the network device 120 generates and shares with terminal device 110 a training dataset, which consists of massive data pairs {ground-truth CSI, associated codeword} for UE-side model training. In addition to the dataset, some assistance information used for two-sided model alignment and model-category alignment are also shared. For example, the category ID associated with the dataset are included in the assistance information. The network device 120 transmits 309 the dataset along with assistance information including category ID 310 to the terminal device 110.
[0066] At 312, the terminal device 110 trains several category-specific encoders by using the shared dataset along with assistance information to produce the codeword that matches the shared one. By approximating the UE-side encoder output to NW-side encoder output, the collaboration between the UE-side encoder and NW-side decoder can be achieved without disclosing any model information to each other. For example, if there are three category IDs #1, #2, #3 and three corresponding datasets, the UE trains three corresponding models #1, #2, #3.
[0067] In some embodiments, the information of the categorization rule comprises: a pre-processing method for CSI, a vector diversity evaluation metric, the thresholds for grouping CSI, or any combination thereof. Specifically, as shown in FIG. 3, the network device 120 transmits 313 the information of categorization rule 314 to the terminal device 110. The categorization rule includes the CSI pre-processing method, vector diversity evaluation metric and threshold for grouping CSI. Given that each category ID corresponds to a specific model ID, the categorization rule also serves as the monitoring rule to generate monitoring measurement for effective model management. For example, if there are three category IDs #1, #2, #3, corresponding to the vector diversity value ranges [-1 / 3, 0] , [-2 / 3, -1 / 3] , and [-1, -2 / 3] , these ranges are shared with terminal device 110 as the threshold for grouping CSI.
[0068] In some embodiments, the terminal device 110 classifies in-field data samples of CSI based on the categorization rule. Specifically, at 316, according to the categorization rule, the terminal device 110 classify the observed in-field CSI matrices into their corresponding categories. To mitigate fluctuations caused by noise and interference, an observation window length M , representing the number of observed CSI matrices, is defined, resulting in a category ID sequence as the monitoring measurement. For example, if there are three CSI matrices in the observation window, the category ID sequence comprises three IDs corresponding to the three CSI matrices.
[0069] In some embodiments, the terminal device 110 transmits to the network device 120 a request for selecting or switching the AI / ML model in case of a category of first deployment or an observed data category is different from an associated category of in-use model. Specifically, at 317, in the initial deployment stage, the terminal device 110 directly triggers a model selection request to the network device 120. When in the inference stage with CSI feedback model #m in use, if the in-field CSI matrix is classified into a category associated with feedback model #l, where l≠m, the terminal device 110 triggers a model switching request to the network device 120.
[0070] In some embodiments, the at least one monitoring measurement comprises a sequence of category IDs of the categories of CSI. Specifically, the terminal device 110 transmits 318 the monitoring measurements carrying category sequence C derived from observed data samples 319 to the network device 120. Assuming there are K categories, the monitoring overhead number is and M is the number of category IDs in the sequence. For example, if there are four categories and three IDs in the sequence, the monitoring overhead number is
[0071] In some embodiments, the network device 120 determines to select or switch the AI / ML model based on the sequence of category IDs. Specifically, at 321, the network device 120 makes decisions on model selection or switching based on the reported category ID sequence. A simple policy is to select the model corresponding to the category with the highest count. For example, if there are three IDs in the sequence, and two IDs are for the category #1, then the model corresponding to category #1 is selected.
[0072] The network device 120 transmits 322 selected NW-side model ID 323 to the terminal device 110. Specifically, the network device 120 informs the terminal device 110 of the selected CSI decoder so that the terminal device 110 can select the aligned CSI encoder to achieve two-side model collaboration. Then the terminal device 110 activates the designated CSI encoder. And the terminal device 110 sends model activation confirmation message to the network device 120. The network device 120 activates the selected decoder, and then collaborate with terminal device 110 encoder to achieve CSI feedback with updated models. At 329, the inference with updated model is performed.
[0073] FIG. 5 illustrates a diagram illustrating example of framework for AI / ML-based CSI feedback according to some embodiments of the present disclosure. As shown in FIG. 5, the training data collection 501 is performed, and the training data is transmitted to the model training component 502 of NW. The model training component 502 comprises categorization rule 503 and multiple decoder models 504. The categories of multiple decoder models 504 is based on the categorization rule 503. The model training component 502 comprises model management component 506, and the model management component 506 comprises model adjustment decision component 507.
[0074] The model storage 505 stores the trained / updated model. The in-field data is transmitted to the categorization rule component 510 in model inference component 508 of UE as inference data. The categorization rule component 510 aligned with categorization rule component 503. The categorization rule component 510 transmits the monitoring measurements to model adjustment decision component 507. The model inference component 508 comprises multiple encoder models 509. The model adjustment decision component 507 provides the instruction for selecting or switching the encoder model 509 a decoder model 504.
[0075] The simulation results of the effectiveness of the vector-diversity-based model monitoring scheme and corresponding LCM procedures of some embodiments of the present application is provided. The simulation datasets are generated by the system level simulation (SLS) platform, which follows the channel model in 3GPP TR 38.901. The dataset containing 600K samples is drawn from a dense urban scenario (urban macro models) using the parameters shown in Table 1.
[0076] Table 1. SLS configurations for dataset generation
[0077] The dataset is initially divided into a training dataset with 500,000 samples and a test dataset with 100,000 samples. Based on the categorization rule, both training dataset and test dataset are respectively divided into 3 sub-datasets, each corresponding to the vector diversity level [-1 / 3, 0] , [-2 / 3, -1 / 3] , and [-1, -2 / 3] .
[0078] Regarding model training, separate model for each sub-dataset is trained, deriving individual encoder and decoder. The guiding principle for selecting the model structure is to configure high-complexity model and large feedback bits for datasets with large vector diversity level. The specific model structure for each category is listed in Table 2.
[0079] For comparison, the performance of the category-independent models (model #0) is provide, which are trained without categorization, i.e., model trained with all 500,000 mixed samples and test on 100,000 mixed samples.
[0080] Table 2. CSI feedback model configurations
[0081] Table 3 presents the CSI feedback performance of different models with respect to each dataset. The threshold of CSI feedback accuracy is set to be 0.81.
[0082] Table 3. Test results
[0083] Despite achieving an average SGCS of 0.81 without distinguishing categories, Model #0 falls short when delving into each category. It is observed that this model is inadequate for handling data from category #1 and category #2, which together constitute more than 65%of the dataset. The SGCS meeting the threshold primarily arises from exceptional performance in category #3.
[0084] By training model separately for each category, more efficient and accurate feedback can be achieved. Given that CSI in category #1 exhibits the highest vector diversity level, the most complex model with the maximum feedback overhead (300 bits) was employed, reaching the threshold of 0.81 SGCS. Sub-datasets with lower vector diversity levels significantly surpassed the 0.81 SGCS performance by using this model. Similarly, when the vector diversity level is lower, a medium model with medium feedback overhead is sufficient to accommodate CSI in category #2. When the vector diversity level is the lowest, the smallest model with the lowest feedback overhead is used for CSI in category #3.
[0085] The above results demonstrate that the proposed model monitoring scheme along with LCM procedures effectively distinguishes channel complexity, guiding model selection, training, and adjustment. This, in turn, leads to a better balance among reconstruction performance, model complexity, and feedback overhead.
[0086] For the solution of some embodiments of the present application, vector diversity-based CSI categorization is provided. Specifically, some embodiments of the present application provide a rule-based CSI categorization scheme that classifies training and inference CSI data into one of N categories by assessing the dispersion degree among subband eigenvectors in the CSI matrix. This serves as an indicator of the category that guides the selection of the AI / ML model from the model pool. The intensity of frequency selectivity in the experienced channel manifests in the angular differences among eigenvectors of all subbands, subsequently affecting the overall performance of CSI feedback. The measurement of these angular dispersion is termed as the "vector diversity value. "
[0087] Some embodiments of the present application also provide category-specific model training. Specifically, the vector diversity value reflects the correlation among subband eigenvectors in the CSI matrix. A larger vector diversity value indicates richer diversity in the CSI data, necessitating a more complex AI model or / and higher CSI feedback overhead. Therefore, distinct models are trained for each category of data, resulting in a CSI encoder model pool at UE and CSI decoder model pool at NW. The encoder and decoder models are used in pairs, so we assign the same model ID to both models. Each pair of encoder and decoder is distinguished by their model complexity and feedback overhead.
[0088] Some embodiments of the present application also provide category-based model monitoring. Following the categorization scheme, in-field data is classified into one category associated with a pre-trained encoder decoder pair in the model pool. If the category of in-field data differs from current model category, a model adjustment request is triggered. Define the category ID sequence from multiple CSI samples within the observation window as the monitoring measurement. This monitoring measurement is then fed back to the NW for management.
[0089] Some embodiments of the present application also provide category-based model management. The NW, based on monitoring measurement from UE, makes model adjustment decision for NW-side decoder model selection, and delivers model switch instructions to UE to select corresponding encoder model. After achieving model alignment between two sides, inference can be done with updated CSI feedback models.
[0090] Some embodiments of the present application take the advantages that the interpretability is inherent in the underlying principle used to calculate the vector diversity value, and the manual labels, as reference is not required, and the input monitoring achieves model adjustments with low monitoring overhead and low computational complexity while maintaining high LCM efficiency.
[0091] FIG. 6 shows a flowchart of an example method 600 implemented at a network device in accordance with some embodiments of the present disclosure. For the purpose of discussion, the method 600 will be described from the perspective of the access network device 120 with reference to FIG. 1.
[0092] At block 610, the network device 120 determines, a categorization rule for grouping channel state information (CSI) . At block 620, the network device 120 transmits, to a terminal device 110, information of the categorization rule. And at block 630, the network device 120 receives, from the terminal device 110 , at least one monitoring measurement for artificial intelligence (AI) / machine learning (ML) based CSI feedback, wherein the at least one monitoring measurement is generated based on the categorization rule.
[0093] In some embodiments, the categorization rule comprises: calculating a vector diversity value of a matrix of CSI; and determining the number of categories of CSI and associated thresholds based on the vector diversity value.
[0094] In some embodiments, calculating vector diversity value comprises: normalizing each eigenvector of the matrix of CSI, wherein each of the eigenvector is for a subband; calculating an average normalized eigenvector over all the subband; and determining the vector diversity value based on the average normalized eigenvector.
[0095] In some embodiments, the information of the categorization rule comprises at least one of the following: a pre-processing method for CSI; a vector diversity evaluation metric; or the thresholds for grouping CSI.
[0096] In some embodiments, the at least one monitoring measurement comprises a sequence of category IDs of the categories of CSI.
[0097] In some embodiments, the network device 120 determines, multiple sub-datasets of a collected CSI dataset based on the categorization rule; and transmits, to the terminal device, at least one AI / ML model for a CSI feedback encoder, a model ID and the associated category ID, wherein the AI / ML model is trained based on at least one sub-dataset among the multiple sub-datasets.
[0098] In some embodiments, the network device 120 transmits, to the terminal device, a training dataset for training at least one AI / ML model for a CSI feedback encoder and assistance information for two-side model alignment and model-category alignment.
[0099] In some embodiments, the network device 120 receives, from the terminal device, a request for selecting or switching the AI / ML model.
[0100] In some embodiments, the network device 120, determines to select or switch the AI / ML model based on the sequence of category IDs.
[0101] FIG. 7 shows a flowchart of an example method 700 implemented at a terminal device in accordance with some embodiments of the present disclosure. For the purpose of discussion, the method 700 will be described from the perspective of the terminal device 110 with reference to FIG. 1.
[0102] At block 710, the terminal device 110 receives, from the network device 120, information of a categorization rule, wherein the categorization rule is for grouping channel state information (CSI) . At block 720, the terminal device 110 transmits, to the network device 120, at least one monitoring measurement for artificial intelligence (AI) / machine learning (ML) based CSI feedback, wherein the at least one monitoring measurement is generated based on the categorization rule.
[0103] In some embodiments, n the categorization rule comprises: calculating a vector diversity value of a matrix of CSI; and determining the number of categories of CSI and associated thresholds based on the vector diversity value.
[0104] In some embodiments, calculating vector diversity value comprises: normalizing each eigenvector of the matrix of CSI, wherein each of the eigenvector is for a subband; calculating an average normalized eigenvector over all the subband; and determining the vector diversity value based on the average normalized eigenvector.
[0105] In some embodiments, the information of the categorization rule comprises at least one of the following: a pre-processing method for CSI; a vector diversity evaluation metric; or the thresholds for grouping CSI.
[0106] In some embodiments, the at least one monitoring measurement comprises a sequence of category IDs of the categories of CSI.
[0107] In some embodiments, the terminal device 110 receives, from the network device, at least one trained AI / ML model for a CSI feedback encoder, a model ID and the associated category ID.
[0108] In some embodiments, the terminal device 110 receives, from the network device 120, a training dataset for training at least one AI / ML model for a CSI feedback encoder and assistance information for two-side model alignment and model-category alignment.
[0109] In some embodiments, the terminal device 110 classifies in-field data samples of CSI based on the categorization rule.
[0110] In some embodiments, the terminal device 110 transmits, to the network device 120, a request for selecting or switching the AI / ML model in case of a category of first deployment or an observed data category is different from an associated category of in-use model.
[0111] In some embodiments, an apparatus capable of performing any of the method 600 (for example, the network device 120) may comprise means for performing the respective steps of the method 600. The means may be implemented in any suitable form. For example, the means may be implemented in a circuitry or software module.
[0112] In some embodiments, the apparatus comprises means for determining, a categorization rule for grouping channel state information (CSI) , and means for transmitting, to a terminal device 110, information of the categorization rule, and means for receiving, from the terminal device 110, at least one monitoring measurement for artificial intelligence (AI) / machine learning (ML) based CSI feedback, wherein the at least one monitoring measurement is generated based on the categorization rule.
[0113] In some embodiments, the categorization rule comprises: calculating a vector diversity value of a matrix of CSI; and determining the number of categories of CSI and associated thresholds based on the vector diversity value.
[0114] In some embodiments, calculating vector diversity value comprises: normalizing each eigenvector of the matrix of CSI, wherein each of the eigenvector is for a subband; calculating an average normalized eigenvector over all the subband; and determining the vector diversity value based on the average normalized eigenvector.
[0115] In some embodiments, the information of the categorization rule comprises at least one of the following: a pre-processing method for CSI; a vector diversity evaluation metric; or the thresholds for grouping CSI.
[0116] In some embodiments, the at least one monitoring measurement comprises a sequence of category IDs of the categories of CSI.
[0117] In some embodiments, the apparatus further comprises means for determining, multiple sub-datasets of a collected CSI dataset based on the categorization rule; and means for transmitting to the terminal device, at least one AI / ML model for a CSI feedback encoder, a model ID and the associated category ID, wherein the AI / ML model is trained based on at least one sub-dataset among the multiple sub-datasets.
[0118] In some embodiments, the apparatus further comprises means for transmitting, to the terminal device, a training dataset for training at least one AI / ML model for a CSI feedback encoder and assistance information for two-side model alignment and model-category alignment.
[0119] In some embodiments, the apparatus further comprises means for receiving, from the terminal device, a request for selecting or switching the AI / ML model.
[0120] In some embodiments, the apparatus further comprises means for determining to select or switch the AI / ML model based on the sequence of category IDs.
[0121] In some embodiments, the apparatus further comprises means for performing other steps in some embodiments of the method 600. In some embodiments, the means comprises at least one processor and at least one memory including computer program code, the at least one memory and computer program code configured to, with the at least one processor, cause the performance of the apparatus.
[0122] In some embodiments, an apparatus capable of performing any of the method 700 (for example, the terminal device 110) may comprise means for performing the respective steps of the method 700. The means may be implemented in any suitable form. For example, the means may be implemented in a circuitry or software module.
[0123] In some embodiments, the apparatus comprises means for receiving, from the network device 120, information of a categorization rule, wherein the categorization rule is for grouping channel state information (CSI) , and means for transmitting, to the network device 120, at least one monitoring measurement for artificial intelligence (AI) / machine learning (ML) based CSI feedback, wherein the at least one monitoring measurement is generated based on the categorization rule.
[0124] In some embodiments, the categorization rule comprises: calculating a vector diversity value of a matrix of CSI; and determining the number of categories of CSI and associated thresholds based on the vector diversity value.
[0125] In some embodiments, calculating vector diversity value comprises: normalizing each eigenvector of the matrix of CSI, wherein each of the eigenvector is for a subband; calculating an average normalized eigenvector over all the subband; and determining the vector diversity value based on the average normalized eigenvector.
[0126] In some embodiments, the information of the categorization rule comprises at least one of the following: a pre-processing method for CSI; a vector diversity evaluation metric; or the thresholds for grouping CSI.
[0127] In some embodiments, the at least one monitoring measurement comprises a sequence of category IDs of the categories of CSI.
[0128] In some embodiments, the apparatus further comprises means for receiving, from the network device 120, at least one trained AI / ML model for a CSI feedback encoder, a model ID and the associated category ID.
[0129] In some embodiments, the apparatus further comprises means for receiving, from the network device 120, a training dataset for training at least one AI / ML model for a CSI feedback encoder and assistance information for two-side model alignment and model-category alignment.
[0130] In some embodiments, the apparatus further comprises means for classifying in-field data samples of CSI based on the categorization rule.
[0131] In some embodiments, the apparatus further comprises means for transmitting, to the network device 120, a request for selecting or switching the AI / ML model in case of a category of first deployment or an observed data category is different from an associated category of in-use model.
[0132] In some embodiments, the apparatus further comprises means for performing other steps in some embodiments of the method 700. In some embodiments, the means comprises at least one processor and at least one memory including computer program code, the at least one memory and computer program code configured to, with the at least one processor, cause the performance of the apparatus.
[0133] FIG. 8 is a simplified block diagram of a device 800 that is suitable for implementing embodiments of the present disclosure. The device 800 may be provided to implement the communication device, for example the terminal device 110, the network device 120 as shown in FIG. 1. As shown, the device 800 includes one or more processors 810, one or more memories 840 coupled to the processor 810, and one or more communication modules 840 coupled to the processor 810.
[0134] The communication module 840 is for bidirectional communications. The communication module 840 has at least one antenna to facilitate communication. The communication interface may represent any interface that is necessary for communication with other network elements.
[0135] The processor 810 may be of any type suitable to the local technical network and may include one or more of the following: general purpose computers, special purpose computers, microprocessors, digital signal processors (DSPs) and processors based on multicore processor architecture, as non-limiting examples. The device 800 may have multiple processors, such as an application specific integrated circuit chip that is slaved in time to a clock which synchronizes the main processor.
[0136] The memory 820 may include one or more non-volatile memories and one or more volatile memories. Examples of the non-volatile memories include, but are not limited to, a Read Only Memory (ROM) 824, an electrically programmable read only memory (EPROM) , a flash memory, a hard disk, a compact disc (CD) , a digital video disk (DVD) , and other magnetic storage and / or optical storage. Examples of the volatile memories include, but are not limited to, a random access memory (RAM) 822 and other volatile memories that will not last in the power-down duration.
[0137] A computer program 830 includes computer executable instructions that are executed by the associated processor 810. The program 830 may be stored in the ROM 820. The processor 810 may perform any suitable actions and processing by loading the program 830 into the RAM 822.
[0138] The embodiments of the present disclosure may be implemented by means of the program 830 so that the device 800 may perform any process of the disclosure as discussed with reference to FIGS. 2 to 7. The embodiments of the present disclosure may also be implemented by hardware or by a combination of software and hardware.
[0139] In some embodiments, the program 830 may be tangibly contained in a computer readable medium which may be included in the device 800 (such as in the memory 820) or other storage devices that are accessible by the device 800. The device 800 may load the program 830 from the computer readable medium to the RAM 822 for execution. The computer readable medium may include any types of tangible non-volatile storage, such as ROM, EPROM, a flash memory, a hard disk, CD, DVD, and the like. FIG. 9 shows an example of the computer readable medium 900 in form of CD or DVD. The computer readable medium has the program 830 stored thereon.
[0140] Generally, various embodiments of the present disclosure may be implemented in hardware or special purpose circuits, software, logic or any combination thereof. Some aspects may be implemented in hardware, while other aspects may be implemented in firmware or software which may be executed by a controller, microprocessor or other computing device. While various aspects of embodiments of the present disclosure are illustrated and described as block diagrams, flowcharts, or using some other pictorial representations, it is to be understood that the block, apparatus, system, technique or method described herein may be implemented in, as non-limiting examples, hardware, software, firmware, special purpose circuits or logic, general purpose hardware or controller or other computing devices, or some combination thereof.
[0141] The present disclosure also provides at least one computer program product tangibly stored on a non-transitory computer readable storage medium. The computer program product includes computer-executable instructions, such as those included in program modules, being executed in a device on a target real or virtual processor, to carry out the methods 900-1200 as described above with reference to FIGS. 9-12. Generally, program modules include routines, programs, libraries, objects, classes, components, data structures, or the like that perform particular tasks or implement particular abstract data types. The functionality of the program modules may be combined or split between program modules as desired in various embodiments. Machine-executable instructions for program modules may be executed within a local or distributed device. In a distributed device, program modules may be located in both local and remote storage media.
[0142] Program code for carrying out methods of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the program codes, when executed by the processor or controller, cause the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may execute entirely on a machine, partly on the machine, as a stand-alone software package, partly on the machine and partly on a remote machine or entirely on the remote machine or server.
[0143] In the context of the present disclosure, the computer program codes or related data may be carried by any suitable carrier to enable the device, apparatus or processor to perform various processes and operations as described above. Examples of the carrier include a signal, computer readable medium, and the like.
[0144] The computer readable medium may be a computer readable signal medium or a computer readable storage medium. A computer readable medium may include but not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the computer readable storage medium would include an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM) , a read-only memory (ROM) , an erasable programmable read-only memory (EPROM or Flash memory) , an optical fiber, a portable compact disc read-only memory (CD-ROM) , an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. The term “non-transitory, ” as used herein, is a limitation of the medium itself (i.e., tangible, not a signal) as opposed to a limitation on data storage persistency (e.g., RAM vs.ROM) .
[0145] Further, while operations are depicted in a particular order, this should not be understood as requiring that such operations be performed in the particular order shown or in sequential order, or that all illustrated operations be performed, to achieve desirable results. In certain circumstances, multitasking and parallel processing may be advantageous. Likewise, while several specific implementation details are contained in the above discussions, these should not be construed as limitations on the scope of the present disclosure, but rather as descriptions of features that may be specific to particular embodiments. Certain features that are described in the context of separate embodiments may also be implemented in combination in a single embodiment. Conversely, various features that are described in the context of a single embodiment may also be implemented in multiple embodiments separately or in any suitable sub-combination.
[0146] Although the present disclosure has been described in languages specific to structural features and / or methodological acts, it is to be understood that the present disclosure defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are disclosed as example forms of implementing the claims.
Claims
1.A network device comprising:at least one processor; andat least one memory storing instructions that, when executed by the at least one processor, cause the network device at least to:determine, a categorization rule for grouping channel state information (CSI) ;transmit, to a terminal device, information of the categorization rule; andreceive, from the terminal device, at least one monitoring measurement for artificial intelligence (AI) / machine learning (ML) based CSI feedback, wherein the at least one monitoring measurement is generated based on the categorization rule.2.The network device of claim 1, wherein the categorization rule comprises:calculating a vector diversity value of a matrix of CSI; anddetermining the number of categories of CSI and associated thresholds based on the vector diversity value.3.The network device of claim 2, wherein calculating vector diversity value comprises:normalizing each eigenvector of the matrix of CSI, wherein each of the eigenvector is for a subband;calculating an average normalized eigenvector over all the subband; anddetermining the vector diversity value based on the average normalized eigenvector.4.The network device of claim 2 or 3, wherein the information of the categorization rule comprises at least one of the following:a pre-processing method for CSI;a vector diversity evaluation metric; orthe thresholds for grouping CSI.5.The network device of any of claims 1-4, wherein the at least one monitoring measurement comprises a sequence of category IDs of the categories of CSI.6.The network device of any of claims 1-5, wherein the network device is further caused to:determine, multiple sub-datasets of a collected CSI dataset based on the categorization rule; andtransmit, to the terminal device, at least one AI / ML model for a CSI feedback encoder, a model ID and the associated category ID, wherein the AI / ML model is trained based on at least one sub-dataset among the multiple sub-datasets.7.The network device of any of claims 1-5, wherein the network device is further caused to:transmit, to the terminal device, a training dataset for training at least one AI / ML model for a CSI feedback encoder and assistance information for two-side model alignment and model-category alignment.8.The network device of claims 6 or 7, wherein the network device is further caused to:receive, from the terminal device, a request for selecting or switching the AI / ML model.9.The network device of any of claims 6-8, wherein the network device is further caused to:determine to select or switch the AI / ML model based on the sequence of category IDs.10.A terminal device comprising:at least one processor; andat least one memory storing instructions that, when executed by the at least one processor, cause the network device at least to:receive, from a network device, information of a categorization rule, wherein the categorization rule is for grouping channel state information (CSI) ; andtransmit, to the network device, at least one monitoring measurement for artificial intelligence (AI) / machine learning (ML) based CSI feedback, wherein the at least one monitoring measurement is generated based on the categorization rule.11.The terminal device of claim 10, wherein the categorization rule comprises:calculating a vector diversity value of a matrix of CSI; anddetermining the number of categories of CSI and associated thresholds based on the vector diversity value.12.The terminal device of claim 11, wherein calculating vector diversity value comprises:normalizing each eigenvector of the matrix of CSI, wherein each of the eigenvector is for a subband;calculating an average normalized eigenvector over all the subband; anddetermining the vector diversity value based on the average normalized eigenvector.13.The terminal device of claim 11 or 12, wherein the information of the categorization rule comprises at least one of the following:a pre-processing method for CSI;a vector diversity evaluation metric; orthe thresholds for grouping CSI.14.The terminal device of any of claims 10-13, wherein the at least one monitoring measurement comprises a sequence of category IDs of the categories of CSI.15.The terminal device of any of claims 10-14, wherein the terminal device is further caused to:receive, from the network device, at least one trained AI / ML model for a CSI feedback encoder, a model ID and the associated category ID.16.The terminal device of any of claims 10-14, wherein the terminal device is further caused to:receive, from the network device, a training dataset for training at least one AI / ML model for a CSI feedback encoder and assistance information for two-side model alignment and model-category alignment.17.The terminal device of any of claims 10-16, wherein the terminal device is further caused to:classify in-field data samples of CSI based on the categorization rule.18.The terminal device of any of claims 14-17, wherein the terminal device is further caused to:transmit, to the network device, a request for selecting or switching the AI / ML model in case of a category of first deployment or an observed data category is different from an associated category of in-use model.19.A method comprising:determining, a categorization rule for grouping channel state information (CSI) ;transmitting, to a terminal device, information of the categorization rule; andreceiving, from the terminal device, at least one monitoring measurement for artificial intelligence (AI) / machine learning (ML) based CSI feedback, wherein the at least one monitoring measurement is generated based on the categorization rule.20.A method comprising:receiving, from a network device, information of a categorization rule, wherein the categorization rule is for grouping channel state information (CSI) ; andtransmitting, to the network device, at least one monitoring measurement for artificial intelligence (AI) / machine learning (ML) based CSI feedback, wherein the at least one monitoring measurement is generated based on the categorization rule.21.An apparatus comprising:means for determining, a categorization rule for grouping channel state information (CSI) ;means for transmitting, to a terminal device, information of the categorization rule; andmeans for receiving, from the terminal device, at least one monitoring measurement for artificial intelligence (AI) / machine learning (ML) based CSI feedback, wherein the at least one monitoring measurement is generated based on the categorization rule.22.An apparatus comprising:means for receiving, from a network device, information of a categorization rule, wherein the categorization rule is for grouping channel state information (CSI) ; andmeans for transmitting, to the network device, at least one monitoring measurement for artificial intelligence (AI) / machine learning (ML) based CSI feedback, wherein the at least one monitoring measurement is generated based on the categorization rule.23.A non-transitory computer readable medium comprising program instructions that, when executed by an apparatus, cause the apparatus to perform at least the method of any of claims 19 and 20.
Citation Information
Patent Citations
Method and apparatus for supporting machine learning or ai ari techniques for CSI feedback in fdd MIMO systems
CN117044125A
Method for transmitting / receiving channel state information in wireless communication system, and device therefor
US20210143886A1
CSI reports based on ML techniques
WO2024031662A1