Data transmission method and apparatus

Abstract

Provided in the embodiments of the present application are a data transmission method and apparatus. The method comprises: a terminal device receiving data from a network device and auxiliary information related to the data, wherein the data corresponds to a single piece of data or a single data set, or, the data corresponds to an AI / ML model or AI / ML model parameter.

Description

Data transmission method and apparatus Technical Field

[0001] The embodiments of this application relate to the field of communication technology. Background Technology

[0002] With the development of artificial intelligence (AI) and machine learning (ML) technologies, applying AI / ML technologies to wireless communication systems to address the challenges of traditional methods has become a current technological trend. In particular, applying AI / ML technologies or algorithms to signal processing functions related to air interface transmission is a key technological focus in the 5G-Advanced and 6G eras.

[0003] One typical AI / ML model architecture is the single-sided model, where the AI / ML model is deployed on one side of the base station or network equipment to implement a specific air interface processing function. Another typical AI / ML model architecture is the two-sided model, where one network model component, or model reconstructor component, is deployed on the network equipment. This two-sided model can work together with the corresponding terminal model component, or model generator component, deployed on the terminal equipment, forming a paired two-sided model.

[0004] One typical use case for the two-sided model is the measurement and feedback of Channel State Information (CSI). In this use case, CSI is measured on the terminal device side, and CSI feedback information is generated using an AI / ML model (which can be called an encoder, model generator, or UE part model). After being sent to the network side via the air interface, the network side receives the CSI feedback information and reconstructs the original CSI using a corresponding AI / ML model (which can be called a decoder, model reconstructor, or NW part model). In such examples, the AI / ML model can reduce CSI feedback overhead or improve the quality of feedback, thereby improving communication quality.

[0005] Therefore, in a typical AI / ML bilateral model use case, the information from the sending end (network device or terminal device) is processed by the AI / ML model and then sent to the receiving end (terminal device or network device) via the air interface. The receiving end uses the model corresponding to the AI / ML model of the sending end (also called the paired model) to recover the information of the AI / ML model input by the sending end.

[0006] It should be noted that the above introduction to the technical background is only for the purpose of providing a clear and complete explanation of the technical solutions of this application and facilitating understanding by those skilled in the art. It should not be assumed that these technical solutions are known to those skilled in the art simply because they have been described in the background section of this application.

[0007] Summary of the Invention

[0008] However, the inventors discovered that the pairwise nature of the bilateral model necessitates collaboration between network devices and terminal devices on both sides of the air interface for model training. One approach is for network device manufacturers and terminal device manufacturers to jointly develop and train the bilateral model. This predictably leads to highly complex collaborative development issues, especially in cases involving multiple terminal manufacturers or multiple device manufacturers. The complexity arising from one-to-one pairwise collaborative development is extremely high.

[0009] This application embodiment considers a separate training method, where one side first completes the training of the network part model (NW part model) and the terminal part model (UE part model). Then, the dataset generated by the training is transmitted to the other side, which then completes the training of the other part of the model based on it.

[0010] It's important to note that, for example, in the network-first training (NW-first training) process, the network device first performs joint training of the NW part model and the UE part model. The UE part model here is a model assumed by the network side to be used by the terminal device, or a reference model; it can be different from or the same as the UE part model actually used by the terminal device. A similar situation applies to the terminal-first training (UE-first training) process. The party training beforehand cannot assume that it knows the actual implementation information of the other party's device.

[0011] Furthermore, after the NW-first training is completed, a pair of NW part models and a UE part model can be generated. During training, input data samples and output data samples of the UE part model (reference model) can be generated. Multiple such sample data during training can form a sample dataset. Based on its knowledge, the network device combines the corresponding data model input and output data samples into a sample set. The size of the sample data can be determined based on the amount of data required for training the corresponding counterpart model, or the protocol container capacity limit for transmitting the corresponding dataset, etc. Alternatively, other methods can be used to determine the dataset size. The network device sends or transmits the dataset (one or more) to the terminal device or terminal-side device via air interface or other means. The terminal device or terminal-side device receives the dataset and uses it to train the corresponding UE part model or encoder. After training, the corresponding UE part model is expected to be compatible with the NW part model of the network device in the NW-first training and have good performance.

[0012] The aforementioned data transmission from network device to terminal device based on NW-first training can also be applied similarly to UE-first training. The difference is that the data transmission is changed to be transmitted from the terminal device or terminal-side device to the network device or network-side device.

[0013] The challenges of data set transmission include ensuring the accuracy of the dataset's information and its association with the corresponding model or model part. For example, when a network device needs to send multiple datasets to a terminal device, ensuring that these datasets are associated with the expected NW part model is a problem that needs to be solved. Additionally, ensuring that the training performance of a model based on the dataset meets requirements is another issue. For the terminal device, how to implement the pairing process or model recognition process between the model trained on the dataset from the network and the corresponding model of the network device, and how to determine whether the UE part model can be activated after training on the dataset are all problems that need to be solved.

[0014] Another way to address the difficulties of joint training between network devices and terminal devices is through model transmission or model parameter transmission. That is, the network device transmits the model structure and parameters, or only the model parameters, to the terminal device. The terminal device then uses the transmitted model directly or, after further optimization, uses the relevant model or model parameters. Considering that the data volume involved in dataset transmission and model transmission can be quite large, transmission is usually achieved based on various data packets or containers defined by high-level protocols. From this perspective, the two have many similarities, and unless otherwise specified, they can be collectively referred to as data transmission.

[0015] To address one or more of the aforementioned problems, embodiments of this application provide a data transmission method and apparatus.

[0016] According to a first aspect of the embodiments of this application, a data transmission method is provided, the method being applied to a terminal device, the method comprising: the terminal device receiving data from a network device and auxiliary information related to the data, wherein the data corresponds to a single data or dataset, or the data corresponds to an AI / ML model or AI / ML model parameters.

[0017] According to a second aspect of the embodiments of this application, a data transmission method is provided, the method being applied to a network device, the method comprising: the network device sending data and auxiliary information related to the data to a terminal device, wherein the data corresponds to a single data or dataset, or the data corresponds to an AI / ML model or AI / ML model parameters.

[0018] According to a third aspect of the embodiments of this application, a data transmission apparatus is provided, the apparatus being applied to a terminal device, the apparatus comprising: a first receiver, which receives data from a network device and auxiliary information related to the data, wherein the data corresponds to a single data or dataset, or the data corresponds to an AI / ML model or AI / ML model parameters.

[0019] According to a fourth aspect of the embodiments of this application, a data transmission apparatus is provided, the apparatus being applied to a network device, the apparatus comprising: a second transmitter that transmits data and auxiliary information related to the data to a terminal device, wherein the data corresponds to a single data or dataset, or the data corresponds to an AI / ML model or AI / ML model parameters.

[0020] According to a fifth aspect of the present application, a communication system is provided, the communication system including a network device and / or a terminal device, the terminal device including the apparatus described in the third aspect of the present application, and the network device including the apparatus described in the fourth aspect of the present application.

[0021] According to a sixth aspect of the present application, a computer-readable program is provided, wherein when the program is executed in a data transmission apparatus or terminal device, the program causes the data transmission apparatus or terminal device to perform the data transmission method described in the first aspect of the present application.

[0022] According to a seventh aspect of the present application, a computer-readable program is provided, wherein when the program is executed in a data transmission apparatus or network device, the program causes the data transmission apparatus or network device to perform the data transmission method described in the second aspect of the present application.

[0023] According to an eighth aspect of the present application, a storage medium storing a computer-readable program is provided, wherein the computer-readable program causes a data transmission apparatus or terminal device to perform the data transmission method described in the first aspect of the present application.

[0024] According to a ninth aspect of the present application, a storage medium storing a computer-readable program is provided, wherein the computer-readable program causes a data transmission apparatus or network device to perform the data transmission method described in the second aspect of the present application.

[0025] One of the beneficial effects of this application's embodiments is that: the terminal device receives data and related auxiliary information from the network device, thereby enabling the terminal device to use the data and auxiliary information for model training and application. For example, it can associate the received dataset with the expected corresponding network part model; it can also ensure that the training performance meets the requirements; it can also implement a pairing process or model recognition process with the model corresponding to the network device based on the model trained from the dataset from the network device; it can also determine whether the terminal part model can be activated after training with the dataset.

[0026] In particular, for certain specific scenarios or network configurations, transmitting data (datasets, or models / model parameters) corresponding to a specific scenario or network configuration to the terminal device can help the terminal device train or update a model that adapts to the scenario or network configuration, thereby enabling the AI / ML model or function to achieve higher performance gains.

[0027] Specific embodiments of this application are disclosed in detail with reference to the following description and accompanying drawings, indicating how the principles of this application can be adopted. It should be understood that the embodiments of this application are not limited in scope. Within the spirit and scope of the appended claims, embodiments of this application include many changes, modifications, and equivalents.

[0028] Features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, combined with features in other embodiments, or substituted for features in other embodiments.

[0029] It should be emphasized that the term "including / comprises" as used herein refers to the presence of a feature, whole, step, or component, but does not exclude the presence or addition of one or more other features, wholes, steps, or components. Attached Figure Description

[0030] The elements and features described in one drawing or embodiment of this application may be combined with elements and features shown in one or more other drawings or embodiments. Furthermore, in the drawings, similar reference numerals denote corresponding parts in several drawings and can be used to indicate corresponding parts used in more than one embodiment.

[0031] Figure 1 is a schematic diagram of a communication system according to an embodiment of this application;

[0032] Figure 2 is another schematic diagram of the communication system according to an embodiment of this application;

[0033] Figure 3 is an example diagram of a bilateral model of an embodiment of this application;

[0034] Figure 4 is a schematic diagram of a method of using separation training according to an embodiment of this application;

[0035] Figure 5 is a schematic diagram of another method of separate training according to an embodiment of this application;

[0036] Figure 6 is a schematic diagram of the application of the bilateral model in the compressed generation and reporting of CSI;

[0037] Figure 7 is another example diagram of a bilateral model of an embodiment of this application;

[0038] Figure 8 is another example diagram of a bilateral model of an embodiment of this application;

[0039] Figure 9 is a schematic diagram of an embodiment of this application;

[0040] Figure 10 is a schematic diagram of a data transmission method according to an embodiment of this application;

[0041] Figure 11 is an interactive diagram of a data transmission method according to an embodiment of this application;

[0042] Figure 12 is another interactive diagram of the data transmission method according to an embodiment of this application;

[0043] Figure 13 is another interactive diagram of the data transmission method according to an embodiment of this application;

[0044] Figure 14 is a schematic diagram of sending model information corresponding to a dataset to a terminal device according to an embodiment of this application;

[0045] Figure 15 is another interactive diagram of the data transmission method according to an embodiment of this application;

[0046] Figure 16 is another interactive diagram of the data transmission method according to an embodiment of this application;

[0047] Figure 17 is another interactive diagram of the data transmission method according to an embodiment of this application;

[0048] Figure 18 is another interactive diagram of the data transmission method according to an embodiment of this application;

[0049] Figure 19 is another interactive diagram of the data transmission method according to an embodiment of this application;

[0050] Figure 20 is another interactive diagram of the data transmission method according to an embodiment of this application;

[0051] Figure 21 is another interactive diagram of the data transmission method according to an embodiment of this application;

[0052] Figure 22 is another interactive diagram of the data transmission method according to an embodiment of this application;

[0053] Figure 23 is another interactive diagram of the data transmission method according to an embodiment of this application;

[0054] Figure 24 is another schematic diagram of the data transmission method according to an embodiment of this application;

[0055] Figure 25 is an interactive diagram of a model-related information transmission method according to an embodiment of this application;

[0056] Figure 26 is an interactive diagram of a model identification indication method according to an embodiment of this application;

[0057] Figure 27 is another interactive diagram of the model-related information transmission method according to an embodiment of this application;

[0058] Figure 28 is another interactive diagram of the method for transmitting model-related information according to an embodiment of this application;

[0059] Figure 29 is another interactive diagram of the method for transmitting model-related information according to an embodiment of this application;

[0060] Figure 30 is a schematic diagram of a data transmission device according to an embodiment of this application;

[0061] Figure 31 is another schematic diagram of a data transmission device according to an embodiment of this application;

[0062] Figure 32 is a schematic block diagram of the system configuration of a terminal device according to an embodiment of this application;

[0063] Figure 33 is a schematic block diagram of the system configuration of a network device according to an embodiment of this application. Detailed Implementation

[0064] Referring to the accompanying drawings, the foregoing and other features of this application will become apparent from the following description. Specific embodiments of this application are specifically disclosed in the description and drawings, illustrating partial implementations in which the principles of this application may be employed. It should be understood that this application is not limited to the described embodiments; rather, it includes all modifications, variations, and equivalents falling within the scope of the appended claims.

[0065] In the embodiments of this application, the terms "first," "second," etc., are used to distinguish different elements by name, but do not indicate the spatial arrangement or chronological order of these elements, and these elements should not be limited by these terms. The term "and / or" includes any one or more of the terms listed in association and all combinations thereof. The terms "comprising," "including," "having," etc., refer to the presence of the stated features, elements, components, or assemblies, but do not exclude the presence or addition of one or more other features, elements, components, or assemblies.

[0066] In the embodiments of this application, the singular forms "a," "the," etc., including the plural forms, should be broadly understood as "a kind" or "a class" rather than limited to the meaning of "an." Furthermore, the term "the" should be understood to include both the singular and plural forms, unless the context explicitly indicates otherwise. Additionally, the term "according to" should be understood as "at least partially based on…," and the term "based on" should be understood as "at least partially based on…," unless the context explicitly indicates otherwise.

[0067] In the embodiments of this application, the term "communication network" or "wireless communication network" may refer to a network that conforms to any of the following communication standards, such as Long Term Evolution (LTE), LTE-Advanced (LTE-A), Wideband Code Division Multiple Access (WCDMA), High-Speed ​​Packet Access (HSPA), etc.

[0068] Furthermore, communication between devices in a communication system can be carried out according to communication protocols at any stage, including but not limited to the following communication protocols: 1G (generation), 2G, 2.5G, 2.75G, 3G, 4G, 4.5G, as well as 5G, 5G-Advanced, New Radio (NR), 6G, etc., and / or other currently known or future communication protocols.

[0069] In the embodiments of this application, the term "network device" refers, for example, to a device in a communication system that connects a terminal device to a communication network and provides services to that terminal device. Network devices may include, but are not limited to, the following devices: base station (BS), access point (AP), transmission reception point (TRP), broadcast transmitter, mobile management entity (MME), gateway, server, radio network controller (RNC), base station controller (BSC), etc.

[0070] Base stations can include, but are not limited to: NodeBs (or NBs), evolved NodeBs (eNodeBs or eNBs), and 5G base stations (gNBs), Centralized Units (CUs), Distributed Units (DUs), etc. They may also include Remote Radio Heads (RRHs), Remote Radio Units (RRUs), relays, or low-power nodes (e.g., femeto, pico, etc.). The term "base station" can encompass some or all of their functions, and each base station can provide communication coverage to a specific geographic area. The term "cell" can refer to a base station and / or its coverage area, depending on the context in which the term is used.

[0071] In the embodiments of this application, the terms "User Equipment" (UE) or "Terminal Equipment" (TE) refer, for example, to a device that accesses a communication network and receives network services through a network device. A terminal device can be fixed or mobile, and may also be referred to as a mobile station (MS), terminal, subscriber station (SS), access terminal (AT), station, etc.

[0072] Terminal devices may include, but are not limited to, the following devices: cellular phones, personal digital assistants (PDAs), wireless modems, wireless communication devices, handheld devices, machine-type communication devices, laptops, cordless phones, smartphones, smartwatches, digital cameras, etc.

[0073] For example, in scenarios such as the Internet of Things (IoT), terminal devices can also be machines or devices for monitoring or measurement, such as including but not limited to: machine-type communication (MTC) terminals, vehicle communication terminals, device-to-device (D2D) terminals, machine-to-machine (M2M) terminals, and so on.

[0074] Furthermore, the terms "network side" or "network equipment side" refer to one side of the network, which can be a base station or include one or more network devices as described above. The terms "user side," "terminal side," or "terminal equipment side" refer to the side of the user or terminal, which can be a UE or include one or more terminal devices as described above. Unless otherwise specified, "equipment" can refer to either network equipment or terminal equipment.

[0075] The following examples illustrate the scenarios of embodiments of this application, but this application is not limited thereto.

[0076] Figure 1 is a schematic diagram of a communication system according to an embodiment of this application, illustrating the case of a terminal device and a network device as examples. As shown in Figure 1, the communication system 100 may include a network device 101 and terminal devices 102 and 103. For simplicity, Figure 1 only illustrates the case of two terminal devices and one network device, but the embodiments of this application are not limited to this.

[0077] Figure 2 is another schematic diagram of the communication system according to an embodiment of this application, illustrating the communication between the terminal device and the network device, taking the terminal device and the network device as examples.

[0078] As shown in Figure 2, the network-side equipment includes an RU (radio frequency unit) for air interface transmission and reception, and a CU / DU for related radio access (RAN). It also includes core network-related processing units (CT / OAM / related processing units), edge computing-related processing units, and data units related to the core network or RAN side.

[0079] Figure 2 is merely an illustrative example; the relevant equipment units can be deployed in a centralized or distributed manner. Furthermore, the network-side equipment, excluding the RU, considers a hardware-software hybrid device based on a CPU / GPU architecture. The related signal processing extensively utilizes AI / ML models for implementation. The network equipment specifically refers to the equipment corresponding to the air interface-related RAN-side CU / DU units and their connected RU units, forming the traditional base station function, or the network equipment of the new architecture of 6G communication systems. The terminal-side equipment, in addition to the air interface-facing terminal equipment, may also include terminal-related server equipment and related data storage or data management units. The term "terminal-side equipment" here refers to its logical affiliation, not its actual physical location; for example, the terminal-side data unit may also exist within the network-side equipment. Similarly, the terminal-side equipment or the terminal equipment's signal processing unit also extensively utilizes AI / ML models for implementation.

[0080] In the embodiments of this application, higher-layer signaling may be, for example, Radio Resource Control (RRC) signaling; for example, referred to as an RRC message, including MIB, system information, dedicated RRC messages; or referred to as an RRC information element. Higher-layer signaling may also be, for example, Medium Access Control (MAC) signaling; or referred to as a MAC control element. However, this application is not limited to these.

[0081] In the embodiments of this application, one or more AI / ML models can be configured and run in network devices and / or terminal devices. AI / ML models can be used for various signal processing functions of wireless communication, such as CSI measurement, compression and reporting, CSI prediction, beam management and beam prediction, positioning, etc.; this application is not limited thereto.

[0082] Figure 3 is an example diagram of a bilateral model according to an embodiment of this application. As shown in Figure 3, one part of the AI / ML model is at the transmitting end TX, such as the information generator, UE part model, or encoder, and the other part is at the receiving end RX, such as the information reconstruction unit, UE part model, or NW part model. At the transmitting end, source information X is input into the model to generate information Y, and information Y is transmitted to the receiving end. At the receiving end, information Y' transmitted via the channel is input into the model to generate recovered information X'. The above information satisfies the following formula 1:

[0083] Y = f(X)

[0084] Y′=h(Y)+n (Formula 1)

[0085] X′=g(Y′)

[0086] ‖XX′‖→0

[0087] Where f(·) is the transfer function of the neural network on the TX side, X is the source information of the neural network at TX, Y is its output signal, h(·) is the channel response coefficient between TX and RX, n is Gaussian noise, g(·) is the transfer function of the neural network on the RX side, Y' is the input signal of the neural network at RX, and X' is its output signal.

[0088] During the joint iterative training of the two-sided networks, the loss function aims to make X' as close to X as possible, i.e., ‖X-X'‖→0. The loss function can use generalized cosine similarity (GCS), squared generalized cosine similarity (SGCS), Normalized Mean Square Error (NMSE), or Mean Square Error (MSE) as performance metrics. After training, the TX neural network and the RX neural network are deployed on the TX side (e.g., the UE side) and the RX side (e.g., the gNB side), respectively, and communicate with each other via an air interface.

[0089] Due to the difficulties involved in joint training, such as the disclosure of intellectual property rights related to AI / ML implementations by network vendors and terminal vendors, as well as the collaboration on related software and hardware implementations, it is usually difficult to jointly develop the two parts of the bilateral model (NW part model and UE part model), especially in the case of multiple network vendors and multiple terminal vendors. The difficulty and complexity of developing pairwise joint training makes it difficult to support the realization of joint training.

[0090] Figure 4 is a schematic diagram of a method using separate training according to an embodiment of this application. Here, it is assumed that a network-first training (NW-first training) approach is adopted, i.e., a method where the network prioritizes model training. The network device jointly trains the network part model (NW part model) and the UE part model based on its assumptions about the UE part model, or a reference model of the UE part model, where the NW part model is deployed in the network device. The network device can send data used for joint training, such as the input and output data of the NW part model, to the terminal device for training its UE part model. Considering the intellectual property issues of network vendors, directly sharing the input and output data of the NW part model with the terminal may expose implementation-related ownership information. Therefore, a more general approach is for the network device to combine the input and output data of its assumed UE part model into a dataset and send it to the terminal device for training its UE part model. The assumed input and output data format can be given according to predefined rules, or indicated in auxiliary information regarding the dataset's format, or given based on format information obtained from prior sharing of relevant information on the terminal side.

[0091] After receiving the dataset sent by the network device, the terminal device can directly use it to construct and train its UE part model, or it can first train a UE-side assumed NW part model, and then train a matching UE part model based on the NW part model. Here, the training of the UE part model can rely entirely on the dataset from the network device, or it can combine the dataset from the network device with other datasets to train the model.

[0092] Figure 5 is a schematic diagram of another method of separate training according to an embodiment of this application. Here, it is assumed that the UE-first training method is used, that is, the method in which the terminal first trains the model. Its process corresponds to the NW-first training process described above, and will not be repeated here.

[0093] A typical use case of the bilateral model is the compressed generation and reporting of CSI. Figure 6 is a schematic diagram of the application of the bilateral model in the compressed generation and reporting of CSI.

[0094] As shown in Figure 6, the bilateral model consists of a CSI generation unit (also known as a CSI encoder) on the terminal device side and a CSI reconstruction unit (also known as a CSI decoder) on the network device side.

[0095] As shown in Figure 6, the input to the CSI generation unit is channel coefficient information, which can be a channel coefficient vector or matrix generated by channel estimation, a channel eigenvector or singular vector generated by matrix decomposition of the aforementioned channel coefficient matrix, or a signal vector transformed from a channel eigenvector. The output of the CSI generation unit is a CSI bit sequence formed through neural network feature extraction, information compression, and quantization. This bit sequence can be carried as part of uplink control information (UCI) on the uplink control channel or uplink data channel and sent to network devices.

[0096] As shown in Figure 6, the network device extracts the corresponding CSI bit sequence from the received UCI and inputs it into the CSI reconstruction unit. The output of the CSI reconstruction unit is a reconstructed channel coefficient vector or matrix, or a channel eigenvector, singular vector, or transform vector, depending on which one corresponds to the input of the CSI generation unit.

[0097] In addition to the CSI example mentioned above, the bilateral model can also replace traditional signal processing methods and be used in various paired processing modules in communication systems.

[0098] Figure 7 is another example diagram of the bilateral model of the embodiments of this application. As shown in Figure 7, for example, source coding and decoding of the source signal, channel coding and decoding, channel modulation and demodulation, multi-antenna module MIMO / De-MIMO (specifically may include paired signal processing at both ends of the transmitter and receiver for MIMO multiplexing, MIMO space-time coding, MIMO precoding / beamforming (BF), etc., other transmitter network sections from multiple data streams to multiple antenna ports, and corresponding receiver network sections at the receiver), waveform shaping and de-waveforming, etc.

[0099] Therefore, the signal processing procedures or functions of the sending and receiving ends of the communication can be merged, and a neural network model can be introduced to realize the original signal processing.

[0100] Figure 8 is another example diagram of a bilateral model according to an embodiment of this application. As shown in Figure 8, the transmitting and receiving ends can be configured with an information generation unit at the transmitting end and an information reconstruction unit at the receiving end, thereby enabling paired signal processing using the bilateral model.

[0101] Figure 9 is a schematic diagram of an embodiment of this application. As shown in Figure 9, after the joint training is completed, one side of the bilateral model sends the training dataset to the other side for separate training of the other side's model. Specifically, corresponding to NW-first training, the network device sends its assumed dataset corresponding to the UE part model to the terminal device, which may include input and output data sample portions of the assumed or referenced UE part model, as well as auxiliary information related to the data samples and dataset. Here, the auxiliary information may include dataset indication information, dataset corresponding model information, data sample description information, related performance requirement information, etc., to assist the terminal device in training based on the dataset, and in model identification or model pairing by the network device after training.

[0102] The following explanations are provided regarding the terminology and basic descriptive methods used in the embodiments of this application:

[0103] In this embodiment, a dataset refers to a collection of multiple data samples. Since the actual input and output of each AI / ML model are determined by the device implementation, the content of the data samples in this embodiment is as follows: each data sample may contain a portion corresponding to the input of the AI / ML model and a portion corresponding to the output of the AI / ML model. Descriptions of the data sample format, dataset size, and dataset segmentation are described as part of the dataset auxiliary information in this embodiment. This does not preclude the inclusion of this auxiliary information as part of the dataset in some embodiments, particularly in embodiments where the dataset and dataset auxiliary information are sent together.

[0104] In this embodiment of the application, unless otherwise specified, network devices include air-interface-oriented base stations and related core network elements (e.g., network-side data storage devices, computing devices, OAM network elements, LMF network elements, sensing network elements, etc.) related to the AI / ML use cases. Terminal devices include air-interface-oriented terminal devices related to the AI / ML use cases and related server, data storage / management unit, and other terminal-side devices. Unless otherwise specified, network devices and terminal devices refer to the devices on both sides of the air interface corresponding to air interface communication.

[0105] In this application embodiment, for air interface use cases based on AI / ML bilateral models, the UE part model corresponds to the relevant AI / ML neural network deployed in the terminal device, such as the encoder or generator in CSI compression feedback. The NW part model corresponds to the relevant AI / ML neural network deployed in the network device, such as the decoder or reconstructor in CSI compression feedback.

[0106] In this embodiment, the description is mainly based on the scenario where the network device trains first. Therefore, the transmission of the dataset is primarily described as being sent from the network device to the terminal device. For bilateral models, due to their dual structure, the reverse dataset transmission can be described equivalently based on a mirror image. Therefore, the basic scheme of this embodiment, such as the dataset auxiliary information, can also be applied to the scenario where the terminal device trains first, in which case the dataset transmission direction is from the terminal device to the network device.

[0107] In this application embodiment, the embodiment of sending datasets and dataset auxiliary information from network devices to terminal devices may also be used in cases where AI / ML models are only deployed on the terminal side, in which case the UE part model can be replaced with a UE-sided model.

[0108] In the embodiments of this application, data transmission includes sending and / or receiving data, such as a network device sending data to a terminal device, a terminal device receiving data from a network device, a terminal device sending data to a network device, and a network device receiving data from a terminal device.

[0109] In this embodiment, the transmission of datasets or models (model parameters) can be implemented using higher-level protocols as containers. This transmission can be carried through the control plane (C Plane), the user plane (U Plane), or a data plane defined for data transmission. One end of the data transmission is a terminal device or terminal-side device, such as a terminal-side server or storage device. The other end of the data transmission can be a base station or a core network element (e.g., a data network element, OAM network element, LMF network element, UMF network element, NWDAF network element, sensing network element, etc.).

[0110] The solutions of the embodiments of this application are described in detail below.

[0111] First aspect of the embodiments

[0112] This application provides a data transmission method, which will be described from the perspective of a terminal device.

[0113] Figure 10 is a schematic diagram of a data transmission method according to an embodiment of this application. As shown in Figure 10, the method includes:

[0114] 1001: The terminal device receives data from the network device and related auxiliary information.

[0115] Figure 11 is an interactive diagram of a data transmission method according to an embodiment of this application. As shown in Figure 11, the terminal device receives data from the network device and auxiliary information related to the data (which can be simply referred to as dataset auxiliary information).

[0116] In the embodiments of this application, the data corresponds to a single data or a dataset (multiple data), or the data corresponds to an AI / ML model or AI / ML model parameters.

[0117] In other words, network devices can transmit single data or datasets (multiple data sets) to terminal devices. Additionally, network devices can also transmit models or model parameters to terminal devices.

[0118] In this embodiment, the example of a network device transmitting a dataset and related auxiliary information to a terminal device is used for illustration. However, similar methods can be used to transmit single data from a network device to a terminal device, or to transmit a model or model parameters from a network device to a terminal device, and will not be repeated in this embodiment.

[0119] In addition, regarding the model or model parameters, besides the content involved in this embodiment, you can also refer to the relevant content in the embodiment of the fourth aspect below, which will not be repeated here.

[0120] In addition, the model or model parameters in the embodiments of this application may refer to AI / ML models or AI / ML model parameters.

[0121] In some embodiments, the dataset is a collection of multiple data samples.

[0122] Each data sample includes at least one of the following:

[0123] The part corresponding to the input of the target AI / ML model;

[0124] The part corresponding to the output of the target AI / ML model;

[0125] In some embodiments, the auxiliary information associated with the dataset includes at least one of the following:

[0126] Configuration information related to this dataset;

[0127] This dataset contains model indication information.

[0128] This dataset contains relevant information.

[0129] Information regarding the intended use of this dataset;

[0130] Performance information related to this dataset.

[0131] In some embodiments, the data and the auxiliary information are sent together or separately.

[0132] In some embodiments, the auxiliary information is sent all at once, or sent in multiple parts each time.

[0133] In the embodiments of this application, the target AI / ML model refers to the UE part model that the network device assumes corresponds to the terminal device, which can also be called the assumed UE part model or the reference UE part model.

[0134] In some embodiments, as described above, the auxiliary information related to the dataset includes configuration information related to the dataset.

[0135] Figure 12 is another interactive diagram of the data transmission method according to an embodiment of this application. As shown in Figure 12, the terminal device receives a dataset and related configuration information from the network device.

[0136] In some embodiments, the dataset is generated from training data of the network device. As shown in FIG4, in joint training starting from the network, the network device performs joint training based on its NW part model and its assumed UE part model or a reference UE part model. After training is completed, a dataset for training terminal devices can be generated. This dataset consists of multiple data samples, each of which may contain a portion corresponding to the input of the assumed / reference UE part model and a portion corresponding to the output of the assumed / reference UE part model.

[0137] In some embodiments, the configuration information related to the dataset may include configuration information for the data samples and / or configuration information related to the dataset. The configuration information for the data samples can be referenced to the use case of CSI compression feedback described below, but the embodiments of this application are not limited thereto.

[0138] In some embodiments, where an AI / ML model is applied to CSI compressed feedback, each data sample includes at least one of the following:

[0139] The part corresponding to the input of the target AI / ML model;

[0140] The part corresponding to the output of the target AI / ML model,

[0141] The input portion uses high-precision quantization information from CSI, and the output portion uses a predefined format consistent with CSI feedback reporting.

[0142] In some embodiments, the auxiliary information related to the dataset includes configuration information related to the dataset. In the configuration information related to the dataset, for the input part, the configuration information of the high-precision codebook used is provided, and for the output part, the corresponding configuration information based on AI / ML CSI feedback is given.

[0143] For CSI compressed feedback use cases, the sample may contain input information and output information for the corresponding hypothesis / reference encoder.

[0144] For example, the input information can be the eigenvector of the real channel matrix corresponding to the encoder input, or the transform domain vector of the eigenvector, such as the quantized representation of the channel state information (CSI) in the angle delay domain vector. This quantized CSI can be obtained by the terminal side through measurement and reporting of the CSI RS. Typically, high-precision codebook reporting can be used, such as high-precision codebook reporting supported by the extended parameters and codebook formation principles of the Rel-16e Type 2 codebook. The codebook parameter information used during codebook reporting can be sent to the terminal device as data sample configuration information. The data sample configuration information corresponding to the encoder input can include the configuration information of the high-precision codebook used by the terminal device when reporting CSI information, such as the number of beams, amplitude, and angle quantization bit information.

[0145] For example, the output information can be configuration information for the length of the output bit information. If it corresponds to a high rank, it also includes the bit length information for each layer at different ranks. It should be noted that the format of the output information here should be consistent with the predefined conventions for CSI information reporting in the CSI compression feedback use case. Based on the same predefined rules, the terminal device does not need to separately store the understanding rules for receiving and storing separate data samples for the dataset.

[0146] In some embodiments, the target UE part model refers to the UE part model assumed by the network device for the terminal device. When AI / ML is applied to CSI compressed feedback, the input and output of the UE part model can be predefined reference input and output, or they can be input and output formats assumed by the network device. For the latter, additional format specifications are required in the dataset auxiliary information.

[0147] More generally, the configuration information of the data sample includes configuration information related to the input of the target UE part model and the corresponding output of the target UE part model. This includes input-related preprocessing information, output-related post-processing information, input / output-related quantization configuration information, and input / output-related dimensional information. The description of this configuration information is such that the terminal device can understand its content through predefined rules.

[0148] In some embodiments, the configuration information related to the dataset may also include, for example, the number of data samples contained in the dataset, the total number of datasets, and the current sequence number.

[0149] In some embodiments, as described above, the auxiliary information associated with the dataset includes model indication information associated with the dataset.

[0150] Figure 13 is another interactive diagram of the data transmission method according to an embodiment of this application. As shown in Figure 13, the terminal device receives a dataset from the network device and model indication information related to the dataset.

[0151] In some embodiments, the model indication information includes at least one of the following:

[0152] The identifier information of the first model that generated this dataset;

[0153] The identification information of the first model corresponding to the second model that generated this dataset;

[0154] Identification information of the model structure of the first model that generated this dataset;

[0155] Identification information of the model structure that generated the second model for this dataset;

[0156] Parameter identification information of the first model;

[0157] Parameter identification information for the second model;

[0158] The first model version information or model parameter version information that generates this dataset;

[0159] Generate a second model version or model parameter version information for this dataset.

[0160] In some embodiments, the first model corresponds to a UE part model, an encoder, a generator, or a UE sided model; wherein, the UE part model, encoder, and generator are models used on the terminal side in a two-sided model and are essentially the same, and the UE sided model is a one-sided model.

[0161] The second model corresponds to the network part model (NW part model), decoder, reconstructor, or network-sided model (NW sided model). The network part model (NW part model), decoder, and reconstructor are models used on the network side in the two-sided model and are essentially the same. In addition, the network-sided model (NW sided model) is a one-sided model.

[0162] In some embodiments, the terminal device receives one or more datasets from the network device.

[0163] The method further includes: the terminal device using the dataset corresponding to the same model indication information from the multiple datasets to: train a first model used on the same terminal side corresponding to the model indication information; or, using the same second model corresponding to the model indication information to train the first model used on the terminal side.

[0164] The following is a detailed explanation.

[0165] In addition to the dataset, network devices can also transmit the NW part model information corresponding to the dataset to the terminal device, which is used to instruct the terminal device to find the UE part model or NW part model trained with the dataset.

[0166] Figure 14 is a schematic diagram illustrating the transmission of model information corresponding to a dataset to a terminal device according to an embodiment of this application. For one or more datasets generated by NW part model training, the dataset is generated by training a specific NW part model and its paired hypothetical or reference UE part model. For the network device, the NW part model used to generate the transmitted dataset may have corresponding model information, such as the identifier information of the NW part model generating the dataset, the identifier information of the corresponding UE part model, the model identifier information of the bilateral model, the identifier information of the NW part model structure, the identifier information of the UE part model structure, the parameter identifier information of the NW part model or the UE part model, and the model version information, etc. The network device may send one or more of the above-mentioned model-related information as auxiliary information along with the dataset to the terminal device.

[0167] Specifically, in addition to receiving the dataset, the terminal device also receives the model information that generated the dataset. Based on this information, the terminal device can find the UE part model trained using the dataset. This model information is particularly necessary when the terminal has multiple UE part models.

[0168] Since network devices are unaware of the actual UE part model information of the terminal device, it's difficult to guarantee that the datasets they send to the terminal device can complete the terminal's model training. Therefore, network devices typically need to transmit multiple datasets to the terminal device. Of course, considering limitations such as the capacity of the higher-level protocol container carrying the datasets or the terminal device's memory, transmitting multiple datasets to the terminal device to train the UE part model corresponding to the same NW part model is common, as shown in Figure 14. Here, n datasets can be sent to the terminal device by the network device at different times. The auxiliary information of the datasets needs to include model-related information, specifically the model indication information (MI-NW) corresponding to the NW part model used to generate the datasets.

[0169] It should be noted that, due to the pairwise nature of bilateral models, MI-NW can also be the indication information of the assumed / reference UE part model in the NW first training. In this case, MI-NW can be considered as MI-UE. The terminal device receives the dataset and the related model indication information. For multiple datasets of the same model indication information, the terminal device can use multiple datasets for model training of the same UE part model related to the indication information (MI-NW). Alternatively, the multiple datasets can be used to first construct the NW part model related to MI-NW, and then the UE part model can be trained and constructed using this NW part model.

[0170] As mentioned above, the model indication information here may be the indication information of the bilateral model used by the network device to generate the dataset, or the model indication information of the NW part model, or the indication information of its corresponding assumed / reference UE part model.

[0171] The model identification information here can be a model identifier, which can be globally valid or unique within the network. It can also be a local identifier, such as a cell-specific ID unique within the cell corresponding to the network device. In this case, the terminal can also achieve a globally unique model identifier by detecting the Global Cell ID and combining it with the model identifier within the cell.

[0172] After training is completed, the terminal device can use the model identification information to identify or pair models.

[0173] In another scenario, the model indication information may correspond to model structure information, or a model structure information identifier. This model structure and its associated indication information may be predefined or known to both the NW and the UE.

[0174] The model structure information identifier can be a global identifier or a predefined identifier. In this way, the model structure information identifier information and dataset received by the terminal in different cells and at different times can be used to train the UE part model corresponding to the model structure information identifier.

[0175] When the model instruction information received by the terminal device is model structure identification information, the terminal device cannot change the model structure corresponding to the model structure identification information when using the dataset for training; it can only train and update the model parameters.

[0176] In some embodiments, the method further includes:

[0177] The terminal device sends relevant information about the model indication information to the network device. The sending of this information is based on an inquiry from the network device or on an active report from the terminal device.

[0178] Figure 15 is another interactive diagram of the data transmission method according to an embodiment of this application.

[0179] When training is complete, the terminal can report based on the model structure identifier information. To achieve model pairing with network devices, one approach is for the network device to configure model version information in its dataset auxiliary information, in addition to the model structure identifier. After training ends and the training performance meets the requirements, the terminal pairs its model with the network device based on the model structure identifier and model version information.

[0180] Alternatively, the terminal device may only report model structure identification information. If the network device is configured with a test dataset or performance testing requirements, the terminal device can report the model structure identification and model update information after the testing requirements are met. Based on this, the network device can further send model identification information to the terminal device, such as model version identification information and model parameter identification information.

[0181] Alternatively, once the model's performance meets the requirements, the terminal device, based on its model structure identifier and existing model version identifier or model parameter identifier, adds a value to the version or model parameter identifier according to predefined rules. Then, based on the updated identifier value, it reports the model identifier to the network device or performs model pairing. For example, the version number changes from version x to version x+y according to a predefined rule, where y corresponds to an updated value, such as 1.

[0182] Figure 16 is another interactive diagram of the data transmission method according to an embodiment of this application. As shown in Figure 16, the terminal device receives a dataset and related indication information from the network device.

[0183] In some embodiments, the indication information associated with the dataset includes at least one of the following:

[0184] Generate identification information related to the network device configuration corresponding to this dataset;

[0185] Generate time-related information corresponding to this dataset;

[0186] Generate the cell information corresponding to this dataset;

[0187] The dataset identifier specified by the network device.

[0188] In some embodiments, the instruction information sent by the network device to the terminal device regarding the dataset may, in addition to indicating the model information that generated the dataset, also indicate the characteristics of the dataset.

[0189] One scenario is that the identification information related to the network device configuration corresponding to the generation of the dataset is used as the indication information.

[0190] After training on this dataset, if the performance meets the requirements, the terminal device can report the aforementioned network configuration-related identifiers to the network device for the same cell. The network device can then obtain the network configuration parameters corresponding to the dataset based on this information. If the network device wishes to activate the model that generated the dataset, it needs to restore the aforementioned network configuration, thereby activating the NW part model corresponding to the dataset and the terminal-side NW part model. Here, the network configuration, in addition to the standard-defined network parameters, also implicitly includes network parameter configuration instructions related to the network device implementation. However, the relationship between the identifier information and the specific network configuration is determined by the implementation and does not need to be predefined, nor does it need to be known by the terminal device.

[0191] This identifier can be global or local. When the identifier is local, such as a cell-specific identifier, it can be combined with the Global Cell ID. This allows the terminal to report the relevant identifier configured for the network device in that cell when entering the same cell, thereby achieving model recognition and pairing with the network device.

[0192] In another scenario, the dataset uses Global Cell ID and timestamp information as indications.

[0193] In this scenario, the dataset contains timestamp information, such as GMT time, UTC time, and communication synchronization-related times like SFN and slot information. This timestamp can correspond to the time of the first sample in the dataset, or the time of samples from other datasets. Alternatively, network devices can use this timestamp to find the NW part model and corresponding network configuration for generating the dataset.

[0194] For network devices, only the timestamp information corresponding to the dataset needs to be transmitted; the terminal can detect the corresponding GCI itself. Of course, if the network device wants to transmit datasets from other cells to the terminal, it can also include GCI and timestamp information in the instruction information.

[0195] When a terminal device completes model training based on the dataset and enters the GCI cell corresponding to the dataset, the terminal device can use timestamp information or GCI+timestamp information to achieve model recognition or model pairing with the network device.

[0196] In another scenario, the network device provides the dataset with a dataset identifier that the network device can recognize. In this case, the terminal device does not need to understand the meaning of the identifier; the identifier needs to be combined with the GCI. Once the model training is complete and meets performance requirements, the terminal enters the cell corresponding to the dataset to obtain the GCI. The terminal can then use the identifier associated with that GCI to pair models or identify models.

[0197] In some embodiments, the method further includes:

[0198] The terminal device sends relevant information about the dataset indication information to the network device. This information is sent based on a query from the network device or an active report from the terminal device.

[0199] Figure 17 is another interactive diagram of the data transmission method according to an embodiment of this application. As shown in Figure 17, the terminal device sends relevant information of the dataset indication information to the network device.

[0200] In some embodiments, the network device receives relevant information (also referred to as dataset indication information) from the terminal device.

[0201] If the information is related to network configuration, the network device can restore the corresponding network configuration based on it, and then find the model that generates the dataset based on the information.

[0202] If it is timestamp information, the network device can recover the network configuration information corresponding to the timestamp and the model corresponding to the timestamp based on the record.

[0203] If it is dataset identification information, the network device can find the corresponding NW part model and restore the corresponding network configuration based on the correspondence between the identifier, the model, and the network configuration information.

[0204] Furthermore, once the network configuration is restored, the network device can activate the corresponding NW part model and send signaling to activate the UE part model corresponding to that dataset. At this point, the network device needs to utilize the aforementioned dataset indication information to activate the UE part model of the terminal device.

[0205] In addition, indicators based on model information and indicators based on dataset features can be used individually or in combination.

[0206] In some embodiments, the usage information associated with the dataset includes at least one of the following:

[0207] This dataset contains information used for training the model;

[0208] This dataset contains information used for testing the model;

[0209] This dataset contains information used for model validation.

[0210] In some embodiments, the information used by the dataset for testing the model includes at least one of the following:

[0211] Indication information for the test model;

[0212] The dataset indication information corresponding to the test model;

[0213] Performance calculation information of the test model;

[0214] Performance requirements information for the test model.

[0215] The following explanation uses a test as an example.

[0216] Figure 18 is another interactive diagram of the data transmission method according to an embodiment of this application. As shown in Figure 18, the terminal device receives a test dataset and auxiliary information for the test dataset from the network device.

[0217] The network device sends a dataset for performance testing to the terminal device, along with related auxiliary information. The test dataset includes input and output data samples of the UE part model. Note that the input and output samples are given in pairs. The auxiliary information may include test dataset indication information, dataset format information, model indication information, performance calculation information, and performance requirement information.

[0218] After receiving the test dataset and its auxiliary information, the terminal device can determine that the dataset is for testing by referring to the indication information about the test dataset in the auxiliary information. Additionally, the terminal device can identify the UE part model to be tested based on model-related information. When the terminal side already knows the UE part model information for which performance evaluation is required, this information can be omitted from the test dataset auxiliary information.

[0219] In this way, network devices can use input samples from the test dataset as input to the model and compare the model output generated based on this input with the corresponding model output in the test dataset. This comparison can be used for performance evaluation calculations based on metrics such as cosine similarity (GCS), SGCS, or MSE and NMSE. The average of the calculated results can be compared with performance requirement information in auxiliary information to determine whether the UE part model meets the corresponding performance requirements.

[0220] In some embodiments, model indication information is used to indicate which model needs to be tested, and may include one or more of the following: model identification information, model structure identification information, model parameter identification information, model version information, etc. This information enables the terminal device to find the UE part model that needs to be tested.

[0221] In some embodiments, for dataset format information, if the terminal has not yet obtained the relevant information, the auxiliary information may also include the relevant format information of the data sample, such as the preprocessing / postprocessing information of the input data and output data, the configuration information of the input data and output data, the length information and quantization information of the output data, etc.

[0222] In some embodiments, performance calculation information is used to indicate performance metrics, such as SGCS, NMSE, etc.

[0223] In some embodiments, performance requirement information corresponds to specific requirement values, thresholds, or ranges, such as SGCS values.

[0224] Furthermore, the above embodiments can be applied not only to AI / ML use cases with bilateral models but also to AI / ML use cases with unilateral models. In this case, the terminal device model receives model transmissions from the network device, or transmits data and undergoes model retraining or fine-tuning. Since the terminal-side model undergoes relevant model updates, the updated model needs performance evaluation. Considering that performance evaluation based on measurement data requires significant time and overhead, performance testing based on test datasets offers greater validity, comprehensiveness, and accuracy, and does not require the long signaling overhead associated with acquiring test data. Accordingly, the test dataset can be sent from the network device to the terminal device. The dataset sample contains the input portion of the corresponding terminal model and the ground truth information of the corresponding terminal model output. The terminal can use the input portion to generate the model's output and compare it with the ground truth information. The auxiliary information of the dataset includes test dataset indication information, test performance requirement information, configuration information related to the input and output portions of the data sample, preprocessing / postprocessing information, quantization information, etc.

[0225] Auxiliary information may also include model-related information, such as model identification information, model structure identification information, and model version information, to instruct the terminal to find the model information relevant to the performance evaluation requirements. Alternatively, it may include AI / ML-related function identification information, as well as the corresponding model structure identification information and model version information, to instruct the terminal to find the AI / ML function relevant to the performance evaluation requirements, and / or the specific model under that function.

[0226] In some embodiments, the method further includes: the terminal device sending model-related performance information and / or performance indication information to the network device, wherein the sending of the information is based on an inquiry from the network device or on an active report from the terminal device.

[0227] For example, once the test is completed, the terminal device will request a report on the test results.

[0228] Figure 19 is another interactive diagram of the data transmission method according to an embodiment of this application. As shown in Figure 19, the reported content includes relevant model indication information, such as model identification information, model structure identification information, model parameter identification information, model version information, and other information that can uniquely indicate the model being tested in the test dataset, as well as model performance-related information. If the test performance requirements have been configured to the terminal device by the network device, the reported performance information can be relative result information based on the test requirements. If the test performance requirements configured to the terminal device by the network device only include instructions for performance calculation, the terminal device reports the performance calculation results according to the performance calculation instructions.

[0229] When performance information does not meet requirements, network devices can continue to send datasets to terminal devices for further training.

[0230] In some embodiments, the method further includes: the terminal device sending a data transmission request (e.g., a dataset transmission request) to the network device and reporting information related to the received data (e.g., dataset-related information), wherein the reporting of the information is based on an inquiry from the network device or on an active reporting by the terminal device.

[0231] For example, the network device sends a test dataset, previously transmitted training dataset indication information, and performance-related information to the terminal device. The terminal device uses the training dataset indication information to find its corresponding UE part model and then uses the test dataset to perform performance testing.

[0232] Figure 20 is another interactive diagram of the data transmission method according to an embodiment of this application. As shown in Figure 20, after determining that the performance of the UE part model cannot meet the performance requirements, the terminal device requests more datasets from the network device for training. The dataset request information (e.g., dataset request information) may include model indication information for the dataset, or dataset indication information, or other information units within the dataset auxiliary information. Based on these auxiliary information indications, the network device can find or collect the corresponding dataset.

[0233] In some embodiments, the dataset-related information includes dataset indication information, and / or dataset configuration information, and / or other dataset auxiliary information.

[0234] In some embodiments, the method further includes: the terminal device receiving a query from the network device regarding its data receiving capabilities (e.g., dataset receiving capabilities); and the terminal device reporting data-related receiving capability information (e.g., dataset-related receiving capabilities) to the network device.

[0235] Figure 21 is another interactive diagram of the data transmission method according to an embodiment of this application.

[0236] As shown in Figure 21, to support dataset transmission, the network device needs to query the terminal device for its dataset receiving capabilities. Since the receiving capabilities for receiving model transmissions and receiving dataset transmissions are similar, such as terminal-side storage capacity (e.g., related to memory size), this embodiment uses "receiving capability query" and "receiving capability reporting" to describe these capabilities, indicating to the terminal device its ability to receive dataset transmissions or model transmissions. This receiving capability can be represented by metrics such as memory capacity, dataset capacity, and the number of dataset samples that can be received. Capacity information can be reflected through predefined capacity ranges and capacity identifiers.

[0237] Furthermore, the receiving capability can also include the reporting of configuration information for supported datasets, such as whether the input portion of the data sample supports time-frequency domain CSI feature vectors or angular delay domain feature vectors, as well as the quantization method information for the output portion. Of course, it is not limited to this; it can also include the querying and reporting of auxiliary information related to other datasets.

[0238] In addition to the capability reporting of terminal devices, their receiving capabilities also change due to variations in related storage capacities, such as memory. Based on predefined storage capacity ranges and corresponding indicators, multiple capacity range levels and corresponding indicator flags can be established. When the storage capacity available for data or model transmission exceeds the level corresponding to its reported capability—meaning the previous reported receiving capability no longer matches the actual available receiving capability—the terminal device needs to proactively update and report its receiving capability.

[0239] In some embodiments, the method further includes: when the terminal device discovers that its current actual reception capability level is different from the reception capability level in the previously reported reception capability information, it actively reports the current reception capability information.

[0240] Figure 22 is another interactive diagram of the data transmission method according to an embodiment of this application.

[0241] As shown in Figure 22, the specific reporting process can be based on the signaling flow of UAI (UE Assistance Information) reporting. The network device sends reception capability reporting-related configuration information (RRCReconfiguration, OtherConfig) to the terminal device, allowing the UE to proactively report its currently supported reception capability information when its reception capability changes across levels. This information can be carried by UE Assistance Information specifically defined for reporting reception capabilities. Specifically, for terminal devices capable of proactively reporting reception capabilities, the network device can configure the terminal device to report assistance information related to reception capabilities. When the terminal device experiences a cross-level change in reception capability, it can report UE reception capability-related assistance information. To prevent frequent UAI reporting caused by dynamic changes in memory, a relevant timer can be introduced. The terminal device obtains the average value of memory or storage capacity within the timer period and uses this as a basis to determine whether to perform UAI reporting.

[0242] In addition, the active reporting of the aforementioned terminal receiving capabilities can also be carried through RRCReconfiguation or RRCReconfiguationComplete information, or through signaling procedures actively reported by other terminals.

[0243] In some embodiments, the method further includes: the terminal device sending data request information (e.g., dataset request information) to the network device; and the terminal device sending data-related receiving capability information (e.g., dataset-related receiving capability information) to the network device.

[0244] The transmission of datasets can be determined by the terminal device. There are many specific reasons, such as the terminal discovering that the model it is using has poor performance, or that the distribution of the input data it is inputting to the model is different from the distribution of the input data used during training, or that the GCIs of its cell are different from the GCIs corresponding to the dataset used during model training, etc.

[0245] Figure 23 is another interactive diagram of the data transmission method according to an embodiment of this application. As shown in Figure 23, the terminal device actively sends a request for data set transmission to the network device. If the data set receiving capability information has not been reported before, or if the receiving capability information needs to be updated, the terminal device also needs to report or update the data set receiving capability. After receiving the data set receiving capability from the terminal, the network device configures the relevant data set and performs the relevant data set transmission.

[0246] Alternatively, the transmission of datasets can be determined by the network device. The network device can determine, based on model performance, model information reported by the terminal device, or dataset information, that the model corresponding to the terminal device needs a dataset sent from the network device to aid in training. Or, the network may determine that it needs to send a training dataset to the terminal device. Of course, other reasons could also lead the network device to deem it necessary to generate data for the terminal device.

[0247] After determining that a dataset needs to be transmitted to a terminal device, the network device queries the terminal device for dataset receiving capability information. Based on the terminal's reported receiving capability, the network device further configures the dataset and sends the relevant dataset to the terminal device.

[0248] The above embodiments of various application scenarios are merely illustrative examples of the embodiments of this application, but this application is not limited thereto, and appropriate modifications can be made based on the embodiments of the above application scenarios. For example, the embodiments of the above application scenarios can be used alone, or one or more of the embodiments of the above application scenarios can be combined.

[0249] As can be seen from the above embodiments, the terminal device receives data and related auxiliary information from the network device, thereby enabling the terminal device to train and apply the model using the data and auxiliary information. For example, it can associate the received dataset with the expected corresponding network part model; it can also ensure that the training performance meets the requirements; it can also implement a pairing process or model recognition process with the model corresponding to the network device based on the model trained from the dataset from the network device; it can also determine whether the terminal part model can be activated after training with the dataset.

[0250] In particular, for certain specific scenarios or network configurations, transmitting data (datasets, or models / model parameters) corresponding to a specific scenario or network configuration to the terminal device can help the terminal device train or update a model that adapts to the scenario or network configuration, thereby enabling the AI / ML model or function to achieve higher performance gains.

[0251] Second aspect of the embodiments

[0252] This application provides a data transmission method, described from the perspective of a network device. The data transmission method on the network device side is similar to the data transmission method on the terminal device side described in the first aspect's embodiment; related details can be found in the specific description of the first aspect's embodiment.

[0253] Figure 24 is another schematic diagram of the data transmission method according to an embodiment of this application. As shown in Figure 24, the method includes:

[0254] 2401: Network devices send data and related auxiliary information to terminal devices.

[0255] In the embodiments of this application, the data corresponds to a single data or a dataset (multiple data), or the data corresponds to an AI / ML model or AI / ML model parameters.

[0256] In other words, network devices can transmit single data or datasets (multiple data sets) to terminal devices. Additionally, network devices can also transmit models or model parameters to terminal devices.

[0257] In this embodiment, the example of a network device transmitting a dataset and related auxiliary information to a terminal device is used for illustration. However, similar methods can be used to transmit single data from a network device to a terminal device, or to transmit a model or model parameters from a network device to a terminal device, and will not be repeated in this embodiment.

[0258] In addition, regarding the model or model parameters, besides the content involved in this embodiment, you can also refer to the relevant content in the embodiment of the fourth aspect below, which will not be repeated here.

[0259] In some embodiments, the dataset is a collection of multiple data samples, each of which includes at least one of the following:

[0260] The part corresponding to the input of the target AI / ML model;

[0261] The part corresponding to the output of the target AI / ML model;

[0262] In some embodiments, the auxiliary information associated with the dataset includes at least one of the following:

[0263] Configuration information related to this dataset;

[0264] This dataset contains model indication information.

[0265] This dataset contains relevant information.

[0266] Information regarding the intended use of this dataset;

[0267] Performance information related to this dataset.

[0268] In some embodiments, the configuration information associated with the dataset includes the configuration information of the data sample and / or the configuration information of the dataset.

[0269] In some embodiments, the model indication information includes at least one of the following:

[0270] The identifier information of the first model that generated this dataset;

[0271] The identification information of the first model corresponding to the second model that generated this dataset;

[0272] Identification information of the model structure of the first model that generated this dataset;

[0273] Identification information of the model structure that generated the second model for this dataset;

[0274] Parameter identification information of the first model;

[0275] Parameter identification information for the second model;

[0276] The first model version information or model parameter version information that generates this dataset;

[0277] Generate a second model version or model parameter version information for this dataset.

[0278] In some embodiments, the first model corresponds to a UE part model, an encoder, a generator, or a UE sided model.

[0279] The second model corresponds to the network part model (NW part model), decoder, reconstructor, or network side model (NW sided model).

[0280] In some embodiments, the method further includes:

[0281] The network device receives relevant information about the model instruction information sent by the terminal device;

[0282] The network device determines whether to activate or run the first model, and enables the second model corresponding to the first model to work in pairs.

[0283] In some embodiments, the indication information associated with the dataset includes at least one of the following:

[0284] Generate identification information related to the network device configuration corresponding to this dataset;

[0285] Generate time-related information corresponding to this dataset;

[0286] Generate the cell information corresponding to this dataset;

[0287] The dataset identifier specified by the network device.

[0288] In some embodiments, the method further includes:

[0289] The network device receives information related to the dataset indication information sent by the terminal device. The sending of this information is based on an inquiry from the network device or on an active report from the terminal device.

[0290] In some embodiments, the method further includes:

[0291] The network device restores the corresponding network configuration based on the relevant information indicated by the dataset, finds the second model corresponding to the dataset, and then activates the first model of the terminal device through the dataset identifier.

[0292] In some embodiments, the usage information associated with the dataset includes at least one of the following:

[0293] This dataset contains information used for training the model;

[0294] This dataset contains information used for testing the model;

[0295] This dataset contains information used for model validation.

[0296] In some embodiments, the information used by the dataset for testing the model includes at least one of the following:

[0297] Indication information for the test model;

[0298] The dataset indication information corresponding to the test model;

[0299] Performance calculation information of the test model;

[0300] Performance requirements information for the test model.

[0301] In some embodiments, the method further includes:

[0302] The network device receives model-related performance information and / or performance indication information sent by the terminal device. The sending of this information is based on a query from the network device or on an active report from the terminal device.

[0303] In some embodiments, the method further includes:

[0304] The network device receives data transmission requests and related information sent by the terminal device. The sending of this information is based on either a query from the network device or an active report from the terminal device.

[0305] In some embodiments, the data-related information includes data indication information, and / or data configuration information, and / or other data-aided information.

[0306] In some embodiments, the method further includes:

[0307] The network device sends a query to the terminal device regarding its data receiving capabilities;

[0308] The network device receives the receiving capability information related to the dataset reported by the terminal device.

[0309] In some embodiments, the method further includes:

[0310] The network device receives the current reception capability information actively reported by the terminal device.

[0311] In some embodiments, the method further includes:

[0312] The network device receives a dataset request message sent by the terminal device;

[0313] The network device receives data set-related reception capability information sent by the terminal device.

[0314] In some embodiments, the network device sends the data and the auxiliary information together or separately, and / or the network device sends all the auxiliary information at once, or sends a portion of the auxiliary information separately multiple times.

[0315] The above embodiments are merely illustrative examples of embodiments of this application, but this application is not limited thereto, and appropriate modifications can be made based on the above embodiments. For example, the above embodiments can be used alone, or one or more of the above embodiments can be combined.

[0316] As can be seen from the above embodiments, the terminal device receives data and related auxiliary information from the network device, thereby enabling the terminal device to train and apply the model using the data and auxiliary information. For example, it can associate the received dataset with the expected corresponding network part model; it can also ensure that the training performance meets the requirements; it can also implement a pairing process or model recognition process with the model corresponding to the network device based on the model trained from the dataset from the network device; it can also determine whether the terminal part model can be activated after training with the dataset.

[0317] In particular, for certain specific scenarios or network configurations, transmitting data (datasets, or models / model parameters) corresponding to a specific scenario or network configuration to the terminal device can help the terminal device train or update a model that adapts to the scenario or network configuration, thereby enabling the AI / ML model or function to achieve higher performance gains.

[0318] Third aspect of the embodiments

[0319] This application also provides a data transmission method applied to a terminal device. The method includes: the terminal device receiving data from a network device and auxiliary information related to the data, wherein the data is a set composed of multiple data samples.

[0320] This data corresponds to a single dataset or a single data point, or it corresponds to an AI / ML model or AI / ML model parameters.

[0321] For cases where AI / ML models are applied to CSI compressed feedback, each data sample includes at least one of the following:

[0322] The part corresponding to the input of the target AI / ML model;

[0323] The part corresponding to the output of the target AI / ML model,

[0324] The input portion uses high-precision quantization information from CSI, and the output portion uses a predefined format consistent with CSI feedback reporting.

[0325] In some embodiments, the auxiliary information related to the dataset includes configuration information related to the dataset.

[0326] In the configuration information related to this dataset, for the input part, the configuration information of the high-precision codebook used is provided, and for the output part, the corresponding configuration information based on AI / ML CSI feedback is given.

[0327] This application also provides a data transmission method, which is applied to a network device, and the method includes:

[0328] The network device sends a dataset and related auxiliary information to the terminal device. This dataset is a collection of multiple data samples.

[0329] For cases where AI / ML models are applied to CSI compressed feedback, each data sample includes at least one of the following:

[0330] The part corresponding to the input of the target AI / ML model;

[0331] The part corresponding to the output of the target AI / ML model,

[0332] The input portion uses high-precision quantization information from CSI, and the output portion uses a predefined format consistent with CSI feedback reporting.

[0333] In some embodiments, the auxiliary information related to the dataset includes configuration information related to the dataset.

[0334] In the configuration information related to this dataset, for the input part, the configuration information of the high-precision codebook used is provided, and for the output part, the corresponding configuration information based on AI / ML CSI feedback is given.

[0335] Fourth aspect of the embodiment

[0336] This application also provides a method for transmitting model-related information. The method is applied to a terminal device and includes: the terminal device receiving model information and auxiliary information related to the model from a network device.

[0337] Figure 25 is an interactive diagram of a model-related information transmission method according to an embodiment of this application. As shown in Figure 25, the terminal device receives model information and related auxiliary information from the network device.

[0338] In some embodiments, the model information is the model's structure and parameter information, or the model's parameter information.

[0339] In some embodiments, the model-related auxiliary information includes at least one of the following:

[0340] The model's structural indication information;

[0341] Model parameter indication information;

[0342] Model indication information;

[0343] Model version information or model parameter version information.

[0344] In some embodiments, the terminal device sends auxiliary information related to the model based on a query from the network device, or proactively reports auxiliary information related to the model.

[0345] This application also provides a method for transmitting model-related information. The method is applied to a network device and includes: the network device sending model information and auxiliary information related to the model to a terminal device.

[0346] This application also provides a model identifier indication method, which is applied to a network device. The method includes: the network device sending a model identifier to a terminal device. The model identifier consists of a first indication indicating the model structure and a second indication indicating the model parameters.

[0347] Figure 26 is an interactive diagram of a model identifier indication method according to an embodiment of this application. As shown in Figure 26, the network device sends a model identifier to the terminal device, which includes a first indication indicating the model structure and a second indication indicating the model parameters.

[0348] In some embodiments, the mapping relationship between the first indication and the model structure is given by a predefined definition, and / or the model structure is given by a predefined definition.

[0349] In some embodiments, the second indication is indicated by a numerical value, where a higher numerical value corresponds to the updated model parameters.

[0350] This application also provides a method for transmitting model-related information. This method is applied to a terminal device and includes: the terminal device and a network device exchanging model identification information.

[0351] When a terminal device discovers that the version information of its corresponding model parameters is lower than that of the network device for a given model structure, the terminal device sends a model transmission request to the network device and instructs it to transmit the corresponding version of the model parameters.

[0352] Figure 27 is another interactive diagram of the model-related information transmission method according to an embodiment of this application. As shown in Figure 27, the terminal device sends a model transmission request to the network device. Additionally, the terminal device instructs the network device to transmit the corresponding version of the model parameters. This application also provides a model-related information transmission method applied to a network device. The method includes: the network device querying the terminal device for model identification information.

[0353] When a network device finds that the corresponding model parameter version is higher than the model parameter version of the terminal device for a given model structure, the network device transmits the higher version model parameters to the terminal device.

[0354] Figure 28 is another interactive diagram of the model-related information transmission method according to an embodiment of this application. As shown in Figure 28, the network device sends a query about model identification information to the terminal device; in addition, the network device also transmits higher version model parameters to the terminal device.

[0355] Alternatively, when a network device finds that the corresponding model parameters are inconsistent with the configuration used by the network device for a given model structure, the network device will transmit the model parameters that are consistent with its configuration to the terminal device. The network configuration mentioned here may be indicated by its associated identifier.

[0356] The following is a detailed explanation.

[0357] Besides data set transmission, network devices can directly transmit a new model to the terminal device. For bilateral models, the network device can transmit the UE part model to the terminal device. However, considering the terminal's implementation capabilities, transmitting AI / ML models from the network device to the terminal device without any restrictions often raises concerns about whether the model can be supported and run by the terminal device. A more feasible approach is for the network device and terminal device to communicate their supported model structures, with the network device then transmitting the model parameters of a supported model structure to the terminal device.

[0358] The model structures supported by both parties can be predefined. Furthermore, the known model structures and their corresponding indications can also be predefined. Additionally, model parameter information can be indicated using different indications.

[0359] More generally, a model indication method can be applied to various embodiments of the dataset transmission and model transmission of the present invention, wherein a model indication or model identifier can consist of two parts: a first indication for indicating the structure of the model, and a second indication for indicating the parameters of the model, i.e.:

[0360] Model identifier = Model structure identifier + Model parameter identifier

[0361] For example, the model structure identifier is a three-bit information, and the model parameter identifier is a four-bit information.

[0362] The model structure has eight combinations of three bits, from 000 to 111, corresponding to eight predefined model structures.

[0363] The 4-digit information of the model parameter identifier corresponds to 16 versions or 16 associated identifiers. Each identifier corresponds to a scenario, such as LOS-based, NLOS-based, macrocell scenario, small cell scenario, etc., or each identifier corresponds to a network configuration, such as beam angle (horizontal angle, vertical angle, downtilt angle, uptilt angle, etc.), beam width, beam combination pattern, order, etc., as well as other parameter information related to network device antennas, radio frequency, etc.

[0364] The model structures can correspond to various AI / ML models such as Transformer, CNN, RNN, and LSTM. Due to the diversity of models and the existence of various subtypes, it is necessary to predefine the model structures supported by the device or those that can be referenced, based on the air interface use cases. Additionally, indication information for these predefined model structures is also provided. Model parameters can be updated through dataset transfer and training based on the dataset, or updated based on the transfer of model parameters. One method of indicating model parameters is to indicate their version; for example, version n+1 corresponds to a newer version of model parameters than version n. To some extent, a newer version of model parameters has better performance.

[0365] Another possibility is that different parameters can correspond to different application scenarios or different network-side configurations. In this case, there is a one-to-one correspondence between the model parameter indications and the scenario or network-side configuration. There is no relationship that a higher indicator value indicates a newer parameter. For bilateral models, the above one-to-one correspondence needs to be aligned between the definitions on the network device and the terminal device. Possible alignment methods include transmitting the relevant model indication information as auxiliary information along with the dataset to the terminal device during dataset transmission, or transmitting the relevant model indication information as auxiliary information along with the model parameters to the terminal device during model parameter transmission.

[0366] Terminal devices can inform network devices of the scenarios or network configurations they can support by reporting the indicated information, including model structure and model parameters.

[0367] Figure 29 is another interactive diagram of the model-related information transmission method according to an embodiment of this application. As shown in Figure 29, the terminal device can report the model identifier, model structure identifier, or model parameter identifier to the network device.

[0368] For model parameter transmission, the terminal device can initiate the request. Specifically, the terminal and network device can first exchange model identification information. When the terminal finds that the version information of its corresponding model parameters is lower than that of the network device for a certain model structure, the terminal sends a model transmission request to the network device, instructing it to transmit the corresponding version of the model parameters. Alternatively, the network device can initiate the transmission by querying the terminal device for the version information of a certain model structure. If the network device finds that it has a higher version, it can consider initiating the transmission of model parameters and transmitting the higher version of the model parameters to the terminal device.

[0369] When model parameters are not indicated by version information but are associated with scene or network configuration, it can be assumed that the terminal can request the transmission of model parameters if it finds that its corresponding model parameter indication is different from the network's model parameters. Another method is for the network device to initiate the relevant model parameter transmission. The network device instructs the terminal device to report its model parameter indication information for a specific model structure. If the network device finds that the parameter indication information is inconsistent with the model parameter information supported by the network device, it indicates that the terminal cannot support the network device's configuration or the scene corresponding to the network device. In this case, the network device will transmit the corresponding model parameters to the terminal device.

[0370] In another general embodiment of model transmission, the terminal device receives model information and auxiliary information related to the model from the network device, wherein the model information is the structure and parameter information of the model, or the parameter information of the model, or partial parameter information of the model.

[0371] The auxiliary information related to the model may include model structure indication information, model parameter indication information, model indication information, model version information, or model parameter version information.

[0372] If the transmitted information is partial model parameter information, the auxiliary information for the model also includes information indicating the number of neural network layers corresponding to the partial model parameter information (e.g., the last three layers of the neural network model structure, or corresponding indication information), or network topology information corresponding to the partial model parameters (or its corresponding indication information). By receiving this indication information, the terminal device can determine which network parameters need to be updated and which do not.

[0373] After the model parameters are transmitted, the model's performance needs to be tested and evaluated. This can be done using test datasets sent over the network or by the terminal device itself. Once the performance meets the requirements, the terminal device can proactively report its new model information to the network device. The terminal device can use the UAI (User-Agent-Initiated) process to report the new model information, including the model identifier, model structure identifier, and model parameter identifier.

[0374] Alternatively, based on queries from network devices, the relevant new model indication information can be reported only when the network device queries whether the model update is complete. Since the model parameters come from the network device, when reporting a new model, a simple indication of whether the performance meets the requirements can be used to report whether the new model is ready to run.

[0375] Fifth aspect of the embodiment

[0376] This application provides a data transmission apparatus. This apparatus may be, for example, a terminal device, or one or more components or parts configured within a terminal device, corresponding to the embodiments of the first aspect. Content identical to that in the embodiments of the first aspect will not be repeated.

[0377] Figure 30 is a schematic diagram of a data transmission apparatus according to an embodiment of this application. As shown in Figure 30, the data transmission apparatus 3000 includes:

[0378] The first receiver 3001 receives data from the network device and related auxiliary information.

[0379] In the embodiments of this application, the data corresponds to a single data or a dataset (multiple data), or the data corresponds to an AI / ML model or AI / ML model parameters.

[0380] In other words, a network device can transmit a single piece of data or a dataset (multiple data sets) to a terminal device. Additionally, a network device can also transmit a model or model parameters to a terminal device. In some embodiments, the dataset is a collection of multiple data samples, each of which includes at least one of the following:

[0381] The part corresponding to the input of the target AI / ML model;

[0382] The part corresponding to the output of the target AI / ML model;

[0383] In some embodiments, the auxiliary information associated with the dataset includes at least one of the following:

[0384] Configuration information related to this dataset;

[0385] This dataset contains model indication information.

[0386] This dataset contains relevant information.

[0387] Information regarding the intended use of this dataset;

[0388] Performance information related to this dataset.

[0389] In some embodiments, the configuration information associated with the dataset includes the configuration information of the data sample and / or the configuration information of the dataset.

[0390] In some embodiments, the model indication information includes at least one of the following:

[0391] The identifier information of the first model that generated this dataset;

[0392] The identification information of the first model corresponding to the second model that generated this dataset;

[0393] Identification information of the model structure of the first model that generated this dataset;

[0394] Identification information of the model structure that generated the second model for this dataset;

[0395] Parameter identification information of the first model;

[0396] Parameter identification information for the second model;

[0397] The first model version information or model parameter version information that generates this dataset;

[0398] Generate a second model version or model parameter version information for this dataset.

[0399] In some embodiments, the first model corresponds to a UE part model, an encoder, a generator, or a UE sided model.

[0400] The second model corresponds to the network part model (NW part model), decoder, reconstructor, or network side model (NW sided model).

[0401] In some embodiments, the first receiver receives one or more datasets from the network device.

[0402] The device also includes:

[0403] The trainer uses the dataset corresponding to the same model indication information from the multiple datasets to: train a first model used on the same terminal side corresponding to the model indication information; or, use the same second model corresponding to the model indication information to train the first model used on the terminal side.

[0404] In some embodiments, the device further includes:

[0405] The first transmitter 3002 sends relevant information about the model indication information to the network device. The transmission of this information is based on an inquiry from the network device or an active report from the terminal device.

[0406] In some embodiments, the indication information associated with the dataset includes at least one of the following:

[0407] Generate identification information related to the network device configuration corresponding to this dataset;

[0408] Generate time-related information corresponding to this dataset;

[0409] Generate the cell information corresponding to this dataset;

[0410] The dataset identifier specified by the network device.

[0411] In some embodiments, the first transmitter also sends information related to the dataset indication information to the network device, the sending of which is based on an inquiry from the network device or on an active report from the terminal device.

[0412] In some embodiments, the usage information associated with the dataset includes at least one of the following:

[0413] This dataset contains information used for training the model;

[0414] This dataset contains information used for testing the model;

[0415] This dataset contains information used for model validation.

[0416] In some embodiments, the information used by the dataset for testing the model includes at least one of the following:

[0417] Indication information for the test model;

[0418] The dataset indication information corresponding to the test model;

[0419] Performance calculation information of the test model;

[0420] Performance requirements information for the test model.

[0421] In some embodiments, the first transmitter also sends model-related performance information and / or performance indication information to the network device, the transmission of which is based on an inquiry from the network device or on an active report from the terminal device.

[0422] In some embodiments, the first transmitter also sends a data transmission request to the network device and reports information related to the received data. The reporting of this information is based on an inquiry from the network device or on an active report from the terminal device.

[0423] In some embodiments, the data-related information includes data indication information, and / or data configuration information, and / or other data-aided information.

[0424] In some embodiments, the first receiver also receives queries from the network device regarding its data reception capabilities;

[0425] In some embodiments, the first transmitter reports data-related reception capability information to the network device.

[0426] In some embodiments, the first transmitter also proactively reports its current reception capability information when it discovers that its current actual reception capability level is different from the reception capability level in its previously reported reception capability information.

[0427] In some embodiments, the first transmitter sends a data request message to the network device;

[0428] The first transmitter also sends data-related reception capability information to the network device.

[0429] In some embodiments, the data and the auxiliary information are sent together or separately, and / or

[0430] The auxiliary information can be sent all at once, or sent in multiple parts, each time in a portion.

[0431] It is worth noting that the above description only covers the components or modules relevant to this application, but this application is not limited thereto. The data transmission device 3000 may also include other components or modules, and for details regarding these components or modules, please refer to related technologies.

[0432] Furthermore, for simplicity, Figure 30 only illustrates the connection relationships or signal flow between the various components or modules, but those skilled in the art should understand that various related technologies such as bus connections can be used. The aforementioned components or modules can be implemented using hardware facilities such as processors, memory, transmitters, and receivers; this application does not limit this implementation.

[0433] As can be seen from the above embodiments, the terminal device receives data and related auxiliary information from the network device, thereby enabling the terminal device to train and apply the model using the data and auxiliary information. For example, it can associate the received dataset with the expected corresponding network part model; it can also ensure that the training performance meets the requirements; it can also implement a pairing process or model recognition process with the model corresponding to the network device based on the model trained from the dataset from the network device; it can also determine whether the terminal part model can be activated after training with the dataset.

[0434] In particular, for certain specific scenarios or network configurations, transmitting data (datasets, or models / model parameters) corresponding to a specific scenario or network configuration to the terminal device can help the terminal device train or update a model that adapts to the scenario or network configuration, thereby enabling the AI / ML model or function to achieve higher performance gains.

[0435] Implementation of the sixth aspect

[0436] This application provides a data transmission apparatus. This apparatus may be, for example, a network device, or one or more components or parts configured within a network device, corresponding to the embodiments of the second aspect. Content identical to that in the embodiments of the second aspect will not be repeated.

[0437] Figure 31 is another schematic diagram of a data transmission apparatus according to an embodiment of this application. As shown in Figure 31, the data transmission apparatus 3100 includes:

[0438] The second transmitter 3101 sends data and related auxiliary information to the terminal device.

[0439] In the embodiments of this application, the data corresponds to a single data or a dataset (multiple data), or the data corresponds to an AI / ML model or AI / ML model parameters.

[0440] In other words, network devices can transmit single data or datasets (multiple data sets) to terminal devices. Additionally, network devices can also transmit models or model parameters to terminal devices.

[0441] In some embodiments, the dataset is a collection of multiple data samples, each of which includes at least one of the following:

[0442] The part corresponding to the input of the target AI / ML model;

[0443] The part corresponding to the output of the target AI / ML model;

[0444] In some embodiments, the auxiliary information associated with the dataset includes at least one of the following:

[0445] Configuration information related to this dataset;

[0446] This dataset contains model indication information.

[0447] This dataset contains relevant information.

[0448] Information regarding the intended use of this dataset;

[0449] Performance information related to this dataset.

[0450] In some embodiments, the model indication information includes at least one of the following:

[0451] The identifier information of the first model that generated this dataset;

[0452] The identification information of the first model corresponding to the second model that generated this dataset;

[0453] Identification information of the model structure of the first model that generated this dataset;

[0454] Identification information of the model structure that generated the second model for this dataset;

[0455] Parameter identification information of the first model;

[0456] Parameter identification information for the second model;

[0457] The first model version information or model parameter version information that generates this dataset;

[0458] Generate a second model version or model parameter version information for this dataset.

[0459] In some embodiments, the first model corresponds to a UE part model, an encoder, a generator, or a UE sided model.

[0460] The second model corresponds to the network part model (NW part model), decoder, reconstructor, or network side model (NW sided model).

[0461] It is worth noting that the above description only covers the components or modules relevant to this application, but this application is not limited thereto. The data transmission device 3100 may also include other components or modules, and for details regarding these components or modules, please refer to related technologies.

[0462] Furthermore, for simplicity, Figure 31 only illustrates the connection relationships or signal flow between the various components or modules, but those skilled in the art should understand that various related technologies such as bus connections can be used. The aforementioned components or modules can be implemented using hardware facilities such as processors, memory, transmitters, and receivers; this application does not limit this implementation.

[0463] As can be seen from the above embodiments, the terminal device receives data and related auxiliary information from the network device, thereby enabling the terminal device to train and apply the model using the data and auxiliary information. For example, it can associate the received dataset with the expected corresponding network part model; it can also ensure that the training performance meets the requirements; it can also implement a pairing process or model recognition process with the model corresponding to the network device based on the model trained from the dataset from the network device; it can also determine whether the terminal part model can be activated after training with the dataset.

[0464] In particular, for certain specific scenarios or network configurations, transmitting data (datasets, or models / model parameters) corresponding to a specific scenario or network configuration to the terminal device can help the terminal device train or update a model that adapts to the scenario or network configuration, thereby enabling the AI / ML model or function to achieve higher performance gains.

[0465] Seventh aspect of the embodiment

[0466] This application provides a terminal device that includes the data transmission apparatus as described in the fifth aspect of the embodiment.

[0467] Figure 32 is a schematic block diagram of the system configuration of a terminal device according to an embodiment of this application. As shown in Figure 32, the terminal device 3200 may include a processor 3210 and a memory 3220; the memory 3220 is coupled to the processor 3210. It is worth noting that this figure is exemplary; other types of structures may also be used to supplement or replace this structure to implement telecommunications functions or other functions.

[0468] In one embodiment, the functionality of the data transmission device can be integrated into the processor 3210.

[0469] The processor 3210 is configured such that the terminal device receives data from the network device and auxiliary information related to the data, the data corresponding to a single data or dataset, or the data corresponding to an AI / ML model or AI / ML model parameters.

[0470] In another embodiment, the data transmission device can be configured separately from the processor 3210. For example, the data transmission device can be configured as a chip connected to the processor 3210, and the functions of the data transmission device can be realized through the control of the processor 3210.

[0471] As shown in Figure 32, the terminal device 3200 may further include: a communication module 3230, an input unit 3240, a display 3250, and a power supply 3260. It is worth noting that the terminal device 3200 does not necessarily include all the components shown in Figure 32; furthermore, the terminal device 3200 may also include components not shown in Figure 32, which can be found in related technologies.

[0472] As shown in Figure 32, the processor 3210, sometimes also referred to as a controller or operation control, may include a microprocessor or other processor device and / or logic device. The processor 3210 receives input and controls the operation of various components of the terminal device 3200.

[0473] The memory 3220 may be, for example, one or more of a cache, flash memory, hard drive, removable media, volatile memory, non-volatile memory, or other suitable means. It can store various types of data and, in addition, programs for executing related information. The processor 3210 can execute the program stored in the memory 3220 to perform information storage or processing, etc. The functions of other components are similar to those in existing systems and will not be described further here. The various components of the terminal device 3200 can be implemented using dedicated hardware, firmware, software, or a combination thereof, without departing from the scope of the embodiments of this application.

[0474] As can be seen from the above embodiments, the terminal device receives data and related auxiliary information from the network device, thereby enabling the terminal device to train and apply the model using the data and auxiliary information. For example, it can associate the received dataset with the expected corresponding network part model; it can also ensure that the training performance meets the requirements; it can also implement a pairing process or model recognition process with the model corresponding to the network device based on the model trained from the dataset from the network device; it can also determine whether the terminal part model can be activated after training with the dataset.

[0475] In particular, for certain specific scenarios or network configurations, transmitting data (datasets, or models / model parameters) corresponding to a specific scenario or network configuration to the terminal device can help the terminal device train or update a model that adapts to the scenario or network configuration, thereby enabling the AI / ML model or function to achieve higher performance gains.

[0476] Eighth aspect of the embodiment

[0477] This application provides a network device that includes the data transmission apparatus as described in the sixth aspect of the embodiment.

[0478] Figure 33 is a schematic block diagram of the system configuration of a network device according to an embodiment of this application. As shown in Figure 33, the network device 3300 may include a processor 3310 and a memory 3320; the memory 3320 is coupled to the processor 3310. The memory 3320 can store various data; in addition, it also stores an information processing program 3330, and executes the program 3330 under the control of the processor 3310.

[0479] In one embodiment, the function of the data transmission device can be integrated into the processor 3310.

[0480] The processor 3310 can be configured to: send data and related auxiliary information from the network device to the terminal device, the data corresponding to a single data or dataset, or the data corresponding to an AI / ML model or AI / ML model parameters.

[0481] In another embodiment, the data transmission device can be configured separately from the processor 3310. For example, the data transmission device can be configured as a chip connected to the processor 3310, and the functions of the data transmission device can be realized through the control of the processor 3310.

[0482] In addition, as shown in Figure 33, network device 3300 may also include: transceiver 3340 and antenna 3350, etc.; the functions of the above components are similar to those in the prior art, and will not be described in detail here. It is worth noting that network device 3300 does not necessarily need to include all the components shown in Figure 33; in addition, network device 3300 may also include components not shown in Figure 33, which can be referred to in the prior art.

[0483] As can be seen from the above embodiments, the terminal device receives data and related auxiliary information from the network device, thereby enabling the terminal device to train and apply the model using the data and auxiliary information. For example, it can associate the received dataset with the expected corresponding network part model; it can also ensure that the training performance meets the requirements; it can also implement a pairing process or model recognition process with the model corresponding to the network device based on the model trained from the dataset from the network device; it can also determine whether the terminal part model can be activated after training with the dataset.

[0484] In particular, for certain specific scenarios or network configurations, transmitting data (datasets, or models / model parameters) corresponding to a specific scenario or network configuration to the terminal device can help the terminal device train or update a model that adapts to the scenario or network configuration, thereby enabling the AI / ML model or function to achieve higher performance gains.

[0485] Ninth aspect of the embodiment

[0486] This application provides a communication system, including at least one of the terminal device described in the seventh aspect and the network device described in the eighth aspect.

[0487] For example, the structure of this communication system can be seen in Figure 1.

[0488] As shown in Figure 1, the communication system 100 includes network device 101, terminal devices 102 and 103. Terminal devices 102 and 103 are the same as those described in the seventh aspect embodiment, and network device 101 is the same as those described in the eighth aspect embodiment. Repeated content will not be described again.

[0489] The apparatus and methods described above in this application can be implemented in hardware or in combination with software. This application relates to a computer-readable program that, when executed by a logic component, enables the logic component to implement the apparatus or components described above, or to implement the various methods or steps described above. This application also relates to storage media for storing the above programs, such as hard disks, magnetic disks, optical disks, DVDs, flash memory, etc.

[0490] The methods / apparatus described in conjunction with the embodiments of this application can be directly embodied in hardware, software modules executed by a processor, or a combination of both. For example, one or more and / or one or more combinations of functional block diagrams shown in FIG30 can correspond to various software modules in a computer program flow or various hardware modules. These software modules can correspond to the various steps shown in FIG10, respectively. These hardware modules can be implemented, for example, using a field-programmable gate array (FPGA) to embed these software modules.

[0491] The software module can reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art. A storage medium can be coupled to the processor, enabling the processor to read information from and write information to the storage medium; or the storage medium can be an integral part of the processor. The processor and storage medium can reside in an ASIC. The software module can be stored in the memory of a mobile terminal or in a memory card that can be inserted into the mobile terminal. For example, if the device (such as a mobile terminal) uses a high-capacity MEGA-SIM card or a high-capacity flash memory device, the software module can be stored in the MEGA-SIM card or the high-capacity flash memory device.

[0492] One or more and / or one or more combinations of functional blocks described in the accompanying drawings can be implemented as a general-purpose processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or other programmable logic device, discrete gate or transistor logic device, discrete hardware component, or any suitable combination thereof for performing the functions described herein. One or more and / or one or more combinations of functional blocks described in the accompanying drawings can also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors in communication with a DSP, or any other such configuration.

[0493] The present application has been described above with reference to specific embodiments. However, those skilled in the art should understand that these descriptions are exemplary and not intended to limit the scope of protection of the present application. Those skilled in the art can make various modifications and variations to the present application based on its spirit and principles, and these modifications and variations are also within the scope of the present application.

[0494] Regarding the implementation methods including the above embodiments, the following notes are also disclosed:

[0495] 1. A data transmission method, the method being applied to a terminal device, the method comprising:

[0496] The terminal device receives data from the network device, along with related auxiliary information. The dataset is a collection of multiple data samples.

[0497] The data may correspond to a single data set or dataset, or the data may correspond to an AI / ML model or AI / ML model parameters.

[0498] For cases where AI / ML models are applied to CSI compressed feedback, each data sample includes at least one of the following:

[0499] The part corresponding to the input of the target AI / ML model;

[0500] The part corresponding to the output of the target AI / ML model,

[0501] The input portion uses high-precision quantization information from CSI, and the output portion uses a predefined format consistent with CSI feedback reporting.

[0502] 2. According to the method described in Appendix 1, the auxiliary information related to the dataset includes configuration information related to the dataset.

[0503] In the configuration information related to the dataset, for the input part, the configuration information of the high-precision codebook used is provided, and for the output part, the corresponding configuration information based on AI / ML CSI feedback is given.

[0504] 3. A data transmission method, the method being applied to a network device, the method comprising:

[0505] The network device sends a dataset and related auxiliary information to the terminal device. The dataset is a collection of multiple data samples.

[0506] For cases where AI / ML models are applied to CSI compressed feedback, each data sample includes at least one of the following:

[0507] The part corresponding to the input of the target AI / ML model;

[0508] The part corresponding to the output of the target AI / ML model,

[0509] The input portion uses high-precision quantization information from CSI, and the output portion uses a predefined format consistent with CSI feedback reporting. The auxiliary information related to the dataset includes configuration information related to the dataset.

[0510] In the configuration information related to the dataset, for the input part, the configuration information of the high-precision codebook used is provided, and for the output part, the corresponding configuration information based on AI / ML CSI feedback is given.

[0511] 4. A method for transmitting model-related information, the method being applied to a terminal device, the method comprising:

[0512] The terminal device receives model information and auxiliary information related to the model from the network device.

[0513] 5. According to the method described in Appendix 4, wherein,

[0514] The model information refers to the model's structure and parameter information, or the model's parameter information.

[0515] The model-related auxiliary information includes at least one of the following:

[0516] The model's structural indication information;

[0517] Model parameter indication information;

[0518] Model indication information;

[0519] Model version information or model parameter version information,

[0520] The terminal device sends model-related auxiliary information based on the query from the network device, or proactively reports model-related auxiliary information.

[0521] 6. A method for transmitting model-related information, the method being applied to a network device, the method comprising:

[0522] The network device sends model information and related auxiliary information to the terminal device.

[0523] The model information refers to the model's structure and parameter information, or the model's parameter information.

[0524] And / or,

[0525] The model-related auxiliary information includes at least one of the following:

[0526] The model's structural indication information;

[0527] Model parameter indication information;

[0528] Model indication information;

[0529] Model version information or model parameter version information.

[0530] 7. A model identification indication method, the method being applied to a network device, the method comprising:

[0531] The network device sends the model identifier to the terminal device.

[0532] The model identifier consists of a first indicator indicating the model structure and a second indicator indicating the model parameters.

[0533] 8. According to the method described in Appendix 7, wherein,

[0534] The mapping relationship between the first indication and the model structure is given by a predefined definition, and / or,

[0535] The model structure is given by predefinition.

[0536] The second indication is indicated by a numerical value, where a higher value corresponds to the updated model parameters.

[0537] 9. A method for transmitting model-related information, the method being applied to a terminal device, the method comprising:

[0538] Interaction model identification information between terminal devices and network devices

[0539] When a terminal device discovers that the model parameter version information corresponding to a model structure is lower than the model parameter version information of the network device, the terminal device sends a model transmission request to the network device and instructs to transmit the corresponding version of the model parameters.

[0540] 10. A method for transmitting model-related information, the method being applied to a network device, the method comprising:

[0541] The network device queries the terminal device for relevant model identification information.

[0542] When a network device discovers that the model parameter version corresponding to a model structure is higher than the model parameter version of the terminal device, the network device transmits the higher version model parameters to the terminal device.

[0543] Alternatively, when the network device discovers that the model parameter information reported by the terminal model does not match the scene information or network configuration information corresponding to the network device, the network device will transmit the matching model parameters to the terminal device.

Claims

1. A data transmission apparatus, the apparatus being applied to a terminal device, the apparatus comprising: The first receiver receives data from the network device and related auxiliary information. The data may correspond to a single data set or dataset, or the data may correspond to an AI / ML model or AI / ML model parameters.

2. The apparatus according to claim 1, wherein, The dataset is a collection of multiple data samples, each of which includes at least one of the following: The part corresponding to the input of the target AI / ML model; The part corresponding to the output of the target AI / ML model.

3. The apparatus according to claim 1, wherein, The auxiliary information related to the dataset includes at least one of the following: The configuration information related to the dataset; The model indication information related to the dataset; The dataset-related indication information; The dataset's related usage information; The dataset is associated with performance information.

4. The apparatus according to claim 1, wherein, The configuration information related to the dataset includes the configuration information of the data samples and / or the configuration information of the dataset.

5. The apparatus according to claim 1, wherein, The model indication information includes at least one of the following: The identification information of the first model that generates the dataset; The identification information of the first model corresponding to the second model of the dataset is generated; Identification information of the model structure of the first model that generates the dataset; Identification information of the model structure of the second model that generates the dataset; Parameter identification information of the first model; Parameter identification information for the second model; Generate the first model version information or model parameter version information of the dataset; Generate second model version information or model parameter version information for the dataset. in, The first model corresponds to the UE part model, encoder, generator, or UE sided model; The second model corresponds to the network part model (NW part model), decoder, reconstructor, or network side model (NW sided model).

6. The apparatus according to claim 5, wherein, The first receiver receives one or more datasets from the network device. The device further includes: The trainer uses the datasets corresponding to the same model indication information from the multiple datasets to: train a first model used on the same terminal side corresponding to the model indication information; or, train the first model used on the terminal side using the same second model corresponding to the model indication information.

7. The apparatus according to claim 5, wherein, The device further includes: A first transmitter sends information related to the model indication information to the network device, the transmission of which is based on an inquiry from the network device or on an active report from the terminal device.

8. The apparatus according to claim 3, wherein, The dataset-related indication information includes at least one of the following: Generate identification information related to the network device configuration corresponding to the dataset; Generate time-related information corresponding to the dataset; Generate the cell information corresponding to the dataset; The dataset identifier specified by the network device.

9. The apparatus according to claim 3, wherein, The first transmitter also sends information related to the dataset indication information to the network device, the sending of which is based on an inquiry from the network device or on an active report from the terminal device.

10. The apparatus according to claim 3, wherein, The usage information related to the dataset includes at least one of the following: The dataset contains information used for training the model; The dataset contains information used for testing the model; The dataset contains information used for model validation.

11. The apparatus according to claim 10, wherein, The information used in the dataset for testing the model includes at least one of the following: Indication information for the test model; The dataset indication information corresponding to the test model; Performance calculation information of the test model; Performance requirements information for the test model.

12. The apparatus according to claim 10, wherein, The first transmitter also sends model-related performance information and / or performance indication information to the network device, the transmission of which is based on the query of the network device or on the active reporting of the terminal device.

13. The apparatus according to claim 12, wherein, The first transmitter also sends a data transmission request to the network device and reports information related to the received data. The reporting of the information is based on the query of the network device or on the active reporting of the terminal device.

14. The apparatus according to claim 13, wherein, The data-related information includes data indication information, and / or data configuration information, and / or other data auxiliary information.

15. The apparatus according to claim 1, wherein, The first receiver also receives queries from the network device regarding its data receiving capabilities; The device further includes: The first transmitter reports data-related reception capability information to the network device.

16. The apparatus according to claim 15, wherein, The first transmitter also proactively reports its current reception capability information when it discovers that its current actual reception capability level is different from the reception capability level in the previously reported reception capability information.

17. The apparatus according to claim 1, wherein, The device further includes: A first transmitter sends a data request message to the network device; The first transmitter also sends data-related reception capability information to the network device.

18. The apparatus according to claim 1, wherein, The data and the auxiliary information are sent together or separately, and / or The auxiliary information may be sent all at once, or sent in multiple parts each time.

19. A data transmission apparatus, the apparatus being applied to a network device, the apparatus comprising: The second transmitter sends data and related auxiliary information to the terminal device. The data may correspond to a single data set or dataset, or the data may correspond to an AI / ML model or AI / ML model parameters.

20. The apparatus according to claim 19, wherein, The dataset is a collection of multiple data samples, each of which includes at least one of the following: The part corresponding to the input of the target AI / ML model; The part corresponding to the output of the target AI / ML model; And / or, The auxiliary information related to the dataset includes at least one of the following: The configuration information related to the dataset; The model indication information related to the dataset; The dataset-related indication information; The dataset's related usage information; The dataset is associated with performance information.

Images