Method and device for artificial intelligence or machine learning training

Abstract

Aspects of the present disclosure provide methods and apparatus for artificial intelligence or machine learning (AI / ML) model training in a wireless communication network to overcome technical issues related to the transmission of AI / ML model training data, for example, the importance of the data cannot be determined due to the same data type. A first device collectively or separately receives AI / ML model training assistance information and information related to transmission of corresponding AI / ML model training data from a second device. And the first equipment determines data state information (DSI) of the corresponding AI / ML model training data on the basis of the AI / ML model training auxiliary information, and the first equipment determines the data state information (DSI) of the corresponding AI / ML model training data on the basis of the AI / ML model training auxiliary information. The first device transmits the respective AI / ML model training data to the second device based on at least one of the DSI of the respective AI / ML model training data or the information related to the transmission of the respective AI / ML model training data.

Description

Technical Field

[0001] The present invention generally relates to wireless communication, and in particular embodiments, to methods and apparatuses for artificial intelligence or machine learning (AI / ML) training. Background Art

[0002] Artificial intelligence technology can be applied to communication, including communication based on artificial intelligence or machine learning (AI / ML) in the physical layer and / or communication based on AI / ML in the medium access control (MAC) layer. For example, in the physical layer, communication based on AI / ML can aim to optimize component design and / or improve algorithm performance. For the MAC layer, communication based on AI / ML can aim to utilize AI / ML capabilities for learning, prediction, and / or decision-making to solve complex optimization problems with potentially better strategies and / or optimal solutions, such as optimizing functions in the MAC layer.

[0003] In some implementations, the AI / ML architecture in a wireless communication network can involve multiple nodes, where the multiple nodes can be organized in one of two modes: a centralized mode and a distributed mode, and both modes can be deployed in an access network, a core network, an edge computing system, or a third-party network. The centralized training and computing architecture is limited by potentially large communication overhead and strict user data privacy. The distributed training and computing architecture can include several frameworks, such as distributed machine learning and federated learning.

[0004] However, communication in a wireless communication system, including communication associated with AI / ML model training at multiple nodes, typically occurs over non-ideal channels. For example, non-ideal conditions (such as electromagnetic interference, signal degradation, phase delay, fading, and other non-ideal conditions) may cause communication signals to attenuate and / or distort, or otherwise interfere with or reduce the communication capabilities of the system.

[0005] Traditional AI / ML model training processes typically rely on hybrid automatic repeat request (HARQ) feedback and retransmission processes to attempt to ensure that data transmitted between devices participating in AI / ML model training is successfully received. However, the communication overhead and latency associated with such retransmissions can be problematic.

[0006] In addition, there may be significant differences in the processing capacity and / or availability of training data for the AI / ML training process among different nodes / devices, which means that there may be significant differences in the ability of different nodes to effectively participate in the AI / ML model training process. In practice, such differences usually mean that the training latency of an AI / ML model training process involving multiple nodes / devices (e.g., an AI / ML model training process based on distributed learning or federated learning) is dominated by the node / device with the maximum latency (due to communication latency and / or computational latency).

[0007] Therefore, one way to reduce the training latency of the AI / ML model training process can be to minimize the communication latency and / or computational latency. These latencies can be reduced by using only important data, e.g., by transmitting only important data and / or by using only important data to perform the AI / ML model training process.

[0008] In existing wireless communication systems, e.g., in 5G network systems, the importance of data can be determined based on the quality of service (QoS) defined at a higher layer. For example, the priority in the 5G QoS identifier (5G QoS identifier, 5QI) can be used to indicate the importance of data. For uplink data scheduling, the priority of the 5QI is mapped to the priority of the MAC layer logical channel. When a user equipment (UE) performs processes such as MAC power distribution unit (PDU) multiplexing and assembly, all selected logical channels are served in descending order of priority. In other words, a logical channel with a higher priority has a higher transmission opportunity and thus will be considered to have a higher importance level.

[0009] The importance of data defined at a higher layer can be associated with a certain data type. Each data type can be regarded as having a certain data importance level. In other words, each data type is associated with a corresponding data importance level, i.e., different data types indicate different data importance levels. Therefore, the importance of data can be determined based on the data type of the data. However, it is not clear how to determine the data importance when the data types are the same. This may be a problem, especially for intelligent communication systems deploying AI / ML models, such as 6G networks.

[0010] Due to these and other reasons, new methods and devices are needed so that new AI-enabled applications and processes can be realized while minimizing the signaling, communication overhead, and latency associated with existing AI / ML model training processes. SUMMARY OF THE INVENTION

[0011] There are limitations in the training process of existing artificial intelligence or machine learning (AI / ML) models. For example, as mentioned above, when the data types are the same, it is not clear how to determine the data importance, which may be a problem, especially for intelligent communication systems that deploy AI / ML models, such as 6G wireless networks. In a 6G network, a user equipment (UE) can collect AI / ML model training data samples and report the collected samples to the network (e.g., a base station (BS)). The AI / ML model training data samples can include reference signals or measurements of sensors (e.g., cameras). The source of the AI / ML model training data may be stable. For example, the reference signals are periodically sent from the BS. However, the content of the AI / ML training data samples may vary at different time instances. This means that the importance of the AI / ML training data samples may vary at different time instances. For example, the importance of the AI / ML training data samples collected at one time instance may be higher than that of other AI / ML training data samples collected at another time instance. However, it is not necessarily possible to determine the importance of the AI / ML training data samples based on the data type because each AI / ML training data sample may be of the same type of data.

[0012] In addition, in a 6G intelligent communication system, the performance of an AI / ML model can be determined not only based on the inference accuracy but also based on the transmission overhead and latency. If the UE sends all the collected AI / ML model training data samples without considering their data importance levels, a large amount of network resources may be required to send the AI / ML training data, thus potentially degrading the overall performance of the AI / ML model and the communication system.

[0013] In existing communication systems (e.g., 5G networks), there are some problems in determining or evaluating data importance. For example, in 5G, the data importance can be determined or evaluated only based on the priority in the 5G QoS identifier (5QI) determined at a higher layer. However, in 5G, the importance of data cannot be determined at the physical layer. In addition, as mentioned above, for data with the same data type, the concept of data importance is lacking. If the data importance is not determined, for example, due to the huge transmission overhead, the overall performance of the AI / ML model and the intelligent communication system in 6G may decline.

[0014] Aspects of the present invention provide solutions to overcome at least some of the above limitations, such as specific methods and devices for artificial intelligence / machine learning (AI / ML) model training.

[0015] According to a first general aspect of the present invention, there is provided a method for supporting AI / ML model training in a wireless communication network. The method according to the first general aspect of the present invention may include: a first device receiving AI / ML model training assistance information from a second device. The method according to the first general aspect of the present invention may further include: the first device determining data state information (DSI) of corresponding AI / ML model training data based on the AI / ML model training assistance information. The method according to the first general aspect of the present invention may further include: the first device receiving information related to the transmission of the corresponding AI / ML model training data from the second device. The method according to the first general aspect of the present invention may further include: the first device transmitting the corresponding AI / ML model training data to the second device based on at least one of the DSI of the corresponding AI / ML model training data or the information related to the transmission of the corresponding AI / ML model training data.

[0016] In some embodiments of the method according to the first general aspect of the present invention, the corresponding AI / ML model training data is selectively transmitted, and the information related to the transmission of the corresponding AI / ML model training data includes a DSI threshold, wherein the processor-executable instructions further include processor-executable instructions that, when executed, cause the processor to perform the following operations: determining whether the corresponding AI / ML model training data will be transmitted to the second device based on the DSI threshold and the DSI of the corresponding AI / ML model training data.

[0017] In some embodiments of the method according to the first general aspect of the present invention, the corresponding AI / ML model training data is selectively transmitted, and the information related to the transmission of the corresponding AI / ML model training data includes information indicating whether the corresponding AI / ML model training data will be transmitted to the second device, wherein the processor-executable instructions further include processor-executable instructions that, when executed, cause the processor to perform the following operations: transmitting at least one of the DSI of the corresponding AI / ML model training data or the AI / ML model training dataset information to the second device; receiving information indicating whether the corresponding AI / ML model training data will be transmitted to the second device from the second device.

[0018] In some embodiments of the method according to the first general aspect of the present invention, the AI / ML model training dataset information includes at least one of the following: the AI / ML model training dataset size, or the DSI distribution information of the AI / ML model training dataset.

[0019] In some embodiments of the method according to the first general aspect of the present invention, the AI / ML model training dataset information is sent using a buffer status report (BSR) or a scheduling request (SR).

[0020] In some embodiments of the method according to the first general aspect of the present invention, the corresponding AI / ML model training data is sent according to a corresponding report format, and the corresponding report format is determined based on at least one of the DSI of the corresponding AI / ML model training data or the information related to the transmission of the corresponding AI / ML model training data.

[0021] In some embodiments of the method according to the first general aspect of the present invention, the corresponding report format indicates the corresponding transmission accuracy, and the level of the corresponding transmission accuracy is determined based on the level of the DSI of the corresponding AI / ML model training data.

[0022] In some embodiments of the method according to the first general aspect of the present invention, the corresponding report format indicates the configuration for the transmission of the corresponding AI / ML model training data.

[0023] In some embodiments of the method according to the first general aspect of the present invention, the configuration for the transmission of the corresponding AI / ML model training data indicates at least one of the resources for the transmission of the corresponding AI / ML model training data or the quantization granularity for the transmission of the corresponding AI / ML model training data.

[0024] In some embodiments of the method according to the first general aspect of the present invention, the corresponding report format indicates whether the corresponding AI / ML model training data includes channel state information (CSI) or raw channel information.

[0025] In some embodiments of the method according to the first general aspect of the present invention, the relationship between the DSI of the corresponding AI / ML model training data and the corresponding report format is configured by a second device.

[0026] In some embodiments of the method according to the first general aspect of the present invention, the AI / ML model training auxiliary information includes at least one of information about a reference AI / ML model or at least one reference input data value.

[0027] In some embodiments of the method according to the first broad aspect of the present invention, the information about the reference AI / ML model includes at least one of the following: reference AI / ML model type, reference AI / ML model structure, one or more reference AI / ML model parameters, reference AI / ML model gradients, reference AI / ML model activation functions, reference AI / ML model input data types, reference AI / ML model output data types, reference AI / ML model input data dimensions, or reference AI / ML model output data dimensions.

[0028] In some embodiments of the method according to the first broad aspect of the present invention, determining the DSI of the corresponding AI / ML model training data includes inputting the corresponding AI / ML model training data into the reference AI / ML model and determining the DSI based on the output of the reference AI / ML model.

[0029] In some embodiments of the method according to the first broad aspect of the present invention, the corresponding AI / ML model training data input into the reference AI / ML model replaces at least one reference input data value.

[0030] In some embodiments of the method according to the first broad aspect of the present invention, the processor-executable instructions further include processor-executable instructions that, when executed, cause the processor to perform the following operation: receive updated AI / ML model training assistance information from a second device.

[0031] In some embodiments of the method according to the first broad aspect of the present invention, the DSI of the corresponding AI / ML model training data includes information indicating at least one of the following: data uncertainty of the corresponding AI / ML model training data; data importance of the corresponding AI / ML model training data; degree of need of the corresponding AI / ML model training data for AI / ML model training; or data diversity of the corresponding AI / ML model training data.

[0032] In some embodiments of the method according to the first broad aspect of the present invention, the data uncertainty of the corresponding AI / ML model training data is determined based on at least one of the following: entropy, minimum confidence, marginal sampling, or generalization error.

[0033] In some embodiments of the method according to the first broad aspect of the present invention, the AI / ML model training is at least partially performed by a second device.

[0034] In some embodiments of the method according to the first broad aspect of the present invention, the corresponding AI / ML model training data includes at least one of the local AI / ML model training data of the local AI / ML model of the first device or the local gradients associated with the local AI / ML model.

[0035] In some embodiments of the method according to the first broad aspect of the present invention, a first device and a second device cooperate to perform AI / ML model training, wherein the processor-executable instructions further include processor-executable instructions that, when executed, cause the processor to perform the following operations: perform a part of the AI / ML model training before determining the DSI of the corresponding AI / ML model training data, wherein the corresponding AI / ML model training data includes the corresponding output of the part of the AI / ML model training.

[0036] According to a second broad aspect of the present invention, a method for performing AI / ML model training in a wireless communication network is provided herein. The method according to the second broad aspect of the present invention may include: a first device sending AI / ML model training assistance information for determining data state information (DSI) of corresponding AI / ML model training data to a second device. The method according to the second broad aspect of the present invention may further include: the first device sending information related to the transmission of the corresponding AI / ML model training data to the second device. The method according to the second broad aspect of the present invention may further include: the first device receiving the corresponding AI / ML model training data from the second device, where the corresponding AI / ML model training data is transmitted based on at least one of the DSI of the corresponding AI / ML model training data or the information related to the transmission of the corresponding AI / ML model training data. The method according to the second broad aspect of the present invention may further include: the first device performing AI / ML model training using the corresponding AI / ML model training data.

[0037] In some embodiments of the method according to the second broad aspect of the present invention, the corresponding AI / ML model training data is selectively transmitted, and the information related to the transmission of the corresponding AI / ML model training data includes a DSI threshold, wherein the processor-executable instructions further include processor-executable instructions that, when executed, cause the processor to perform the following operations: configure a DSI threshold for determining whether the corresponding AI / ML model training data will be transmitted to the first device.

[0038] In some embodiments of the method according to the second broad aspect of the present invention, the corresponding AI / ML model training data is selectively sent, and the information related to the sending of the corresponding AI / ML model training data includes information indicating whether the corresponding AI / ML model training data will be sent to the first device. Wherein, the processor-executable instructions further include processor-executable instructions that cause the processor to perform the following operations when executed: receive at least one of the DSI of the corresponding AI / ML model training data or the AI / ML model training dataset information from the second device; determine whether the corresponding AI / ML model training data will be sent to the first device using at least one of the DSI of the corresponding AI / ML model training data or the AI / ML model training dataset information; send information indicating whether the corresponding AI / ML model training data will be sent to the first device to the second device.

[0039] In some embodiments of the method according to the second broad aspect of the present invention, the AI / ML model training dataset information includes at least one of the AI / ML model training dataset size or the DSI distribution information of the AI / ML model training dataset.

[0040] In some embodiments of the method according to the second broad aspect of the present invention, the AI / ML model training dataset information is sent using a buffer status report (BSR) or a scheduling request (SR).

[0041] In some embodiments of the method according to the second broad aspect of the present invention, the corresponding AI / ML model training data is sent according to a corresponding report format, and the corresponding report format is determined based on at least one of the DSI of the corresponding AI / ML model training data or the information related to the sending of the corresponding AI / ML model training data.

[0042] In some embodiments of the method according to the second broad aspect of the present invention, the corresponding report format indicates the corresponding transmission accuracy, and the level of the corresponding transmission accuracy is determined based on the level of the DSI of the corresponding AI / ML model training data.

[0043] In some embodiments of the method according to the second broad aspect of the present invention, the corresponding report format indicates the configuration for the sending of the corresponding AI / ML model training data.

[0044] In some embodiments of the method according to the second broad aspect of the present invention, the configuration for the sending of the corresponding AI / ML model training data indicates at least one of the resources for the sending of the corresponding AI / ML model training data or the quantization granularity for the sending of the corresponding AI / ML model training data.

[0045] In some embodiments of the method according to the second general aspect of the present invention, the corresponding report format indicates whether the corresponding AI / ML model training data includes channel state information (CSI) or raw channel information.

[0046] In some embodiments of the method according to the second general aspect of the present invention, the processor-executable instructions further include processor-executable instructions that, when executed, cause the processor to perform the following operations: configure the relationship between the DSI of the corresponding AI / ML model training data and the corresponding report format.

[0047] In some embodiments of the method according to the second general aspect of the present invention, the AI / ML model training auxiliary information includes at least one of information about a reference AI / ML model or at least one reference input data value.

[0048] In some embodiments of the method according to the second general aspect of the present invention, the information about the reference AI / ML model includes at least one of the following: reference AI / ML model type, reference AI / ML model structure, one or more reference AI / ML model parameters, reference AI / ML model gradient, reference AI / ML model activation function, reference AI / ML model input data type, reference AI / ML model output data type, reference AI / ML model input data dimension, or reference AI / ML model output data dimension.

[0049] In some embodiments of the method according to the second general aspect of the present invention, the processor-executable instructions further include processor-executable instructions that, when executed, cause the processor to perform the following operations: update the AI / ML model training auxiliary information; send the updated AI / ML model training auxiliary information to a second device.

[0050] In some embodiments of the method according to the second general aspect of the present invention, the DSI of the corresponding AI / ML model training data includes information indicating at least one of the following: data uncertainty of the corresponding AI / ML model training data; data importance of the corresponding AI / ML model training data; degree of need for the corresponding AI / ML model training data for AI / ML model training; or data diversity of the corresponding AI / ML model training data.

[0051] In some embodiments of the method according to the second general aspect of the present invention, the DSI of the corresponding AI / ML model training data is determined based on at least one of the following: entropy, minimum confidence, marginal sampling, or generalization error.

[0052] In some embodiments of the method according to the second broad aspect of the present invention, the corresponding AI / ML model training data includes at least one of the local AI / ML model training data of the second device's local AI / ML model or the local gradient associated with the local AI / ML model.

[0053] In some embodiments of the method according to the second broad aspect of the present invention, the first device and the second device cooperate to perform AI / ML model training such that the first device executes a part of the AI / ML model training before determining the DSI of the corresponding AI / ML model training data, and the corresponding AI / ML model training data includes the corresponding output of the part of the AI / ML model training.

[0054] Corresponding devices for performing the above methods are disclosed.

[0055] For example, according to another aspect of the present invention, there is provided a device comprising a processor and a memory, the memory storing processor-executable instructions which, when executed, cause the processor to: execute the method according to the first broad aspect or the second broad aspect of the present invention as described above.

[0056] According to other aspects of the present invention, there is provided a device comprising one or more units for implementing any method aspect disclosed in the present invention. The term "unit" is used in a broad sense and may be given any of a variety of names, including, for example, module, component, element, member, etc. These units may be implemented using hardware, software, firmware, or any combination thereof.

[0057] Through certain aspects of the present invention, the performance of the AI / ML model is enhanced, and overfitting phenomena during AI / ML model training can be avoided. In addition, since fewer AI / ML model training data samples are transmitted based on the data state information (DSI) of the AI / ML model training data, the transmission overhead (e.g., radio interface overhead) can be reduced. The DSI (e.g., data uncertainty) can be measured at the device (e.g., UE, BS) before the device reports or sends the AI / ML model training data.

[0058] Through some aspects of the present invention, for example, in federated learning, the transmission of AI / ML model training data (e.g., local gradients) that will not contribute to the convergence of the global AI / ML model is avoided. In this way, the signaling overhead in federated learning can be reduced, the performance of the AI / ML model can be improved, and the AI / ML model training can be enhanced.

[0059] Through some aspects of the present invention, rapid convergence of the AI / ML model can be achieved in bilateral AI / ML model training. Since the number of transmissions of the AI / ML model training data set is reduced, additional signaling overhead can be avoided.

[0060] Through some aspects of the present invention, the enhanced performance of the AI / ML model can be balanced and the transmission overhead can be reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] The following are only examples with reference to the drawings showing exemplary embodiments of the present application, wherein:

[0062] Figure 1 is a simplified schematic diagram of a communication system according to an example;

[0063] Figure 2 shows another example of a communication system;

[0064] Figure 3 shows examples of an electronic device (ED), a terrestrial transmit and receive point (T-TRP), and a non-terrestrial transmit and receive point (NT-TRP);

[0065] Figure 4 shows exemplary units or modules in a device;

[0066] Figure 5 shows four EDs communicating with a network device in a communication system according to an embodiment of the present invention;

[0067] Figure 6A shows an example of a neural network with multiple layers of neurons according to an embodiment of the present invention;

[0068] Figure 6B shows an example of a neuron that can be used as a building block of a neural network according to an embodiment of the present invention;

[0069] Figure 7 shows an example of unilateral AI / ML model training at a base station (BS) according to an embodiment of the present invention;

[0070] Figure 8A shows an example of a reference AI / ML model according to an embodiment of the present invention;

[0071] Figure 8B shows an example of a reference AI / ML model with reference AI / ML model input data according to an embodiment of the present invention;

[0072] Figure 8C shows an example of measuring data uncertainty of an AI / ML model using channel information according to an embodiment of the present invention;

[0073] Figure 9A and Figure 9B illustrates an exemplary process of reporting AI / ML model training data from a user equipment (UE) to a BS according to an embodiment of the present invention;

[0074] Figure 10 illustrates an example of performing unilateral AI / ML model training at a UE according to an embodiment of the present invention;

[0075] Figure 11A and Figure 11B illustrates an exemplary process of reporting AI / ML model training data from a BS to a UE according to an embodiment of the present invention;

[0076] Figure 12A illustrates an example of a process of performing AI / ML model training in federated learning;

[0077] Figure 12B illustrates an example of a process of performing AI / ML model training using data state information (DSI) of AI / ML model training data in federated learning according to an embodiment of the present invention;

[0078] Figure 13 illustrates an example of bilateral AI / ML model training according to an embodiment of the present invention;

[0079] Figure 14 illustrates an example of AI / ML model training with data transmission accuracy adaptation according to an embodiment of the present invention;

[0080] Figure 15 is a flowchart of an exemplary process for AI / ML model training according to an embodiment of the present invention;

[0081] Figure 16 is a flowchart of another exemplary process for AI / ML model training according to an embodiment of the present invention.

[0082] The same reference numerals may be used to represent the same components in different drawings. Detailed Description

[0083] In the present invention, an "AI / ML model" refers to a data-driven algorithm that applies AI / ML technology to generate a set of outputs based on a set of inputs.

[0084] In the present invention, "AI / ML model training" refers to a process of training an AI / ML model by learning an input / output relationship in a data-driven manner and obtaining a trained AI / ML model for inference.

[0085] In the present invention, "inference" or "AI / ML inference" refers to the process of using a trained AI / ML model to generate a set of outputs based on a set of inputs.

[0086] In the present invention, "federated learning / federated training" refers to a machine learning technique that trains an AI / ML model across multiple decentralized edge nodes (e.g., UEs, gNBs), where each decentralized edge node performs local model training using local data samples. This technique requires multiple model exchanges but does not require the exchange of local data samples.

[0087] For illustration, specific exemplary embodiments will now be explained in detail in conjunction with the accompanying drawings.

[0088] The embodiments described herein represent information sufficient to practice the claimed subject matter and illustrate methods of practicing such subject matter. After reading the following description with reference to the drawings, those skilled in the art will understand the concepts of the claimed subject matter and will recognize that the application of these concepts is not specifically described herein. It should be understood that these concepts and their applications are within the scope of the present invention and the appended claims.

[0089] In addition, it will be understood that any module, component, or device that executes instructions disclosed herein may include or otherwise access one or more non-transitory computer / processor-readable storage media to store information such as computer / processor-readable instructions, data structures, program modules, and / or other data. A non-exhaustive list of examples of non-transitory computer / processor-readable storage media includes magnetic storage devices such as magnetic tape cartridges, tapes, disk memories, or other magnetic storage devices, compact disc read-only memory (CD-ROM), digital video disc or digital versatile disc (i.e., DVD), Blu-ray Disc TM and other optical discs, volatile and non-volatile, removable and non-removable media implemented in any method or technology, random-access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory, or other storage technologies. Any of these non-transitory computer / processor storage media may be part of the device or may be accessible or connected to the device. The computer / processor-readable / executable instructions for implementing the applications or modules described herein may be stored or otherwise held by such non-transitory computer / processor-readable storage media.

[0090] Examples of communication systems and devices

[0091] Reference Figure 1 , by way of illustrative example and not limitation, a simplified schematic diagram of a communication system is provided. Communication system 100 includes a radio access network 120. The radio access network 120 can be a next-generation (e.g., sixth-generation, "6G" or later) radio access network, or a traditional (e.g., 5G, 4G, 3G, or 2G) radio access network. In the radio access network 120, one or more communication electronic devices (EDs) 110a to 110j (commonly referred to as 110) can be interconnected with each other or connected to one or more network nodes (170a, 170b, commonly referred to as 170). The core network 130 can be part of the communication system and can be dependent on or independent of the radio access technology used in the communication system 100. Additionally, the communication system 100 includes a public switched telephone network (PSTN) 140, the Internet 150, and other networks 160.

[0092] Figure 2 An exemplary communication system 100 is shown. Generally, the communication system 100 is capable of enabling multiple wireless or wired elements to transmit data and other content. The purpose of the communication system 100 can be to provide content such as voice, data, video, and / or text through broadcasting, multicasting, and unicasting, etc. The communication system 100 can operate by sharing resources (e.g., carrier spectral bandwidth) among its constituent elements. The communication system 100 can include a terrestrial communication system and / or a non-terrestrial communication system. The communication system 100 can provide a wide range of communication services and applications (e.g., earth monitoring, remote sensing, passive sensing and positioning, navigation and tracking, autonomous delivery and mobility, etc.). The communication system 100 can provide a high degree of availability and robustness through the joint operation of the terrestrial communication system and the non-terrestrial communication system. For example, integrating a non-terrestrial communication system (or its components) into a terrestrial communication system can enable a heterogeneous network including multiple layers. Compared with traditional communication networks, a heterogeneous network can achieve better overall performance through efficient multi-link joint operation, more flexible function sharing, and faster physical layer link switching between the terrestrial network and the non-terrestrial network.

[0093] A terrestrial communication system and a non-terrestrial communication system can be considered as subsystems of a communication system. In the illustrated example, communication system 100 includes electronic devices (EDs) 110a to 110d (commonly referred to as ED 110), radio access networks (RANs) 120a and 120b, non-terrestrial communication network 120c, core network 130, public switched telephone network (PSTN) 140, Internet 150, and other networks 160. RANs 120a and 120b include corresponding base stations (BSs) 170a and 170b, which can generally be referred to as terrestrial transmit and receive points (T-TRPs) 170a and 170b. Non-terrestrial communication network 120c includes access nodes 120c, which can generally be referred to as non-terrestrial transmit and receive points (NT-TRPs) 172.

[0094] Alternatively or additionally, any ED 110 can be used to access any other T-TRPs 170a and 170b and NT-TRP 172, Internet 150, core network 130, PSTN 140, other networks 160, or any combination of the foregoing, or to be connected to or communicate with the same. In some examples, ED 110a can perform uplink transmission and / or downlink transmission with T-TRP 170a via interface 190a. In some examples, EDs 110a, 110b, and 110d can also communicate directly with each other via one or more sidelink air interfaces 190b. In some examples, ED 110d can perform uplink transmission and / or downlink transmission with NT-TRP 172 via interface 190c.

[0095] The air interfaces 190a and 190b may use similar communication technologies, such as any suitable radio access technology. For example, the communication system 100 may implement one or more channel access methods in the air interfaces 190a and 190b, such as code division multiple access (CDMA), time division multiple access (TDMA), frequency division multiple access (FDMA), orthogonal FDMA (OFDMA), or single-carrier FDMA (SC-FDMA). The air interfaces 190a and 190b may use other higher-dimensional signal spaces, which may involve a combination of orthogonal and / or non-orthogonal dimensions.

[0096] The air interface 190c may implement communication between the ED 110d and one or more NT-TRPs 172 via a wireless link (or simply referred to as a link). For some examples, the link is a dedicated connection for unicast transmission, a connection for broadcast transmission, or a connection between a group of EDs and one or more NT-TRPs for multicast transmission.

[0097] RANs 120a and 120b communicate with the core network 130 to provide various services, such as voice, data, and other services, to the EDs 110a, 110b, and 110c. The RANs 120a and 120b and / or the core network 130 may communicate directly or indirectly with one or more other RANs (not shown), which may (or may not) be directly served by the core network 130 and may (or may not) employ the same radio access technology as the RAN 120a, the RAN 120b, or both. The core network 130 may also serve as a gateway access between (i) the RANs 120a and 120b, or the EDs 110a, 110b, and 110c, or both and (ii) other networks, such as the PSTN 140, the Internet 150, and other networks 160. In addition, some or all of the EDs 110a, 110b, and 110c may include functionality to communicate with different wireless networks over different wireless links using different wireless technologies and / or protocols. Instead of (or in addition to) wireless communication, the EDs 110a, 110b, and 110c may also communicate with a service provider or switch (not shown) and with the Internet 150 over a wired communication channel. The PSTN 140 may include a circuit-switched telephone network for providing plain old telephone service (POTS). The Internet 150 may include computer networks and / or subnets (intranets) and includes protocols such as the Internet Protocol (IP), the Transmission Control Protocol (TCP), and the User Datagram Protocol (UDP). The EDs 110a, 110b, and 110c may be multimode devices capable of operating according to multiple radio access technologies and include multiple transceivers required to support these technologies.

[0098] Figure 3Shows another example of the ED 110 and base stations 170a, 170b, and / or 170c. The ED 110 is used to connect people, objects, machines, etc. The ED 110 can be widely applied in various scenarios, such as cellular communication, device-to-device (D2D), vehicle to everything (V2X), peer-to-peer (P2P), machine-to-machine (M2M), machine-type communication (MTC), internet of things (IoT), virtual reality (VR), augmented reality (AR), industrial control, autonomous driving, telemedicine, smart grid, smart furniture, smart office, smart wearables, smart transportation, smart city, drones, robots, remote sensing, passive sensing, positioning, navigation and tracking, autonomous delivery and mobility, etc.

[0099] Each ED 110 represents any suitable end-user device for wireless operation and can include the following devices (or can be referred to as): user equipment (UE), wireless transmit / receive unit (WTRU), mobile station, fixed or mobile subscriber unit, cellular phone, station (STA), machine type communication (MTC) device, personal digital assistant (PDA), smartphone, laptop, computer, tablet, wireless sensor, consumer electronic device, smartbook, vehicle, car, truck, bus, train, or IoT device, industrial device, or a device in the above devices (such as a communication module, modem, or chip), etc. The next-generation ED 110 can be referred to using other terms. Base stations 170a and 170b are T-TRPs and are hereinafter referred to as T-TRP 170. Also as Figure 3 shown, the NT-TRP is hereinafter referred to as NT-TRP 172. Each ED 110 connected to the T-TRP 170 and / or NT-TRP 172 can be dynamically or semi-statically turned on (i.e., established, activated, or enabled), turned off (i.e., released, deactivated, or disabled), and / or configured in response to one or more of the following: connection availability and connection necessity.

[0100] ED 110 includes a transmitter 201 and a receiver 203 coupled to one or more antennas 204. Only one antenna 204 is shown. Alternatively, one, some, or all of the antennas may be panels. For example, the transmitter 201 and the receiver 203 may be integrated into a transceiver. The transceiver is used to modulate data or other content for transmission by at least one antenna 204 or a network interface controller (NIC). The transceiver is also used to demodulate data or other content received by at least one antenna 204. Each transceiver includes any suitable structure for generating signals for wireless or wired transmission and / or for processing signals received wirelessly or wiredly. Each antenna 204 includes any suitable structure for transmitting and / or receiving wireless or wired signals.

[0101] ED 110 includes at least one memory 208. The memory 208 stores instructions and data used, generated, or collected by ED 110. For example, the memory 208 may store software instructions or modules executed by one or more processing units 210 for implementing some or all of the functions and / or embodiments described herein. Each memory 208 includes one or more any suitable volatile and / or non-volatile storage and retrieval devices. Any suitable type of memory may be used, such as random access memory (RAM), read only memory (ROM), hard disk, optical disk, subscriber identity module (SIM) card, memory stick, secure digital (SD) memory card, on-processor cache, etc.

[0102] ED 110 may also include one or more input / output devices (not shown) or interfaces (such as Figure 1 a wired interface to the Internet 150 in). The input / output devices may interact with users or other devices in the network. Each input / output device includes any suitable structure for providing information to or receiving information from a user, such as a speaker, microphone, keypad, keyboard, display, or touch screen, including network interface communication.

[0103] ED 110 also includes a processor 210 for performing operations including: operations related to preparing a transmission for uplink transmission to NT-TRP 172 and / or T-TRP 170, operations related to processing a downlink transmission received from NT-TRP 172 and / or T-TRP 170, and operations related to processing sidelink transmissions to and from another ED 110. The processing operations related to preparing a transmission for uplink transmission may include operations such as encoding, modulation, transmit beamforming, and generating symbols for transmission. The processing operations related to processing a downlink transmission may include operations such as receive beamforming, demodulation, and decoding received symbols. According to an embodiment, the receiver 203 may receive a downlink transmission (possibly using receive beamforming), and the processor 210 may extract signaling from the downlink transmission (e.g., by detecting and / or decoding the signaling). Examples of signaling may be reference signals sent by NT-TRP 172 and / or T-TRP 170. In some embodiments, the processor 276 implements transmit beamforming and / or receive beamforming based on an indication of a beam direction received from T-TRP 170 (e.g., beam angle information (BAI)). In some embodiments, the processor 210 may perform operations related to network access (e.g., initial access) and / or downlink synchronization, such as operations related to detecting a synchronization sequence, decoding, and acquiring system information. In some embodiments, the processor 210 may perform channel estimation using reference signals received from NT-TRP 172 and / or T-TRP 170, etc.

[0104] Although not shown, the processor 210 may form part of the transmitter 201 and / or the receiver 203. Although not shown, the memory 208 may form part of the processor 210.

[0105] The processor 210, and the processing components of the transmitter 201 and the receiver 203 may each be implemented by the same or different one or more processors for executing instructions stored in a memory (e.g., memory 208). Alternatively, some or all of the processor 210, and the processing components of the transmitter 201 and the receiver 203 may be implemented using dedicated circuitry, such as a programmed field-programmable gate array (FPGA), a graphical processing unit (GPU), or an application-specific integrated circuit (ASIC).

[0106] In some implementations, T-TRP 170 may have other names, such as base station, base transceiver station (BTS), radio base station, network node, network device, network side device, transmission / reception node, Node B, evolved Node B (eNodeB or eNB), home eNodeB, next generation Node B (gNB), transmission point (TP), site controller, access point (AP), wireless router, relay station, remote radio head, terrestrial node, terrestrial network device, terrestrial base station, base band unit (BBU), remote radio unit (RRU), active antenna unit (AAU), remote radio head (RRH), central unit (CU), distribute unit (DU), positioning node, etc. T-TRP 170 may be a macro BS, micro BS, relay node, host node, etc. or a combination thereof. T-TRP 170 may refer to the above devices or the devices (e.g., communication module, modem or chip) in the above devices.

[0107] In some embodiments, parts of T-TRP 170 may be distributed. For example, some modules of T-TRP 170 may be located at a position remote from the device housing the antenna of T-TRP 170 and may be coupled to the device housing the antenna, such as a common public radio interface (CPRI), through a communication link (not shown) sometimes referred to as fronthaul. Thus, in some embodiments, the term T-TRP 170 may also refer to modules on the network side that perform processing operations such as determining the position of ED 110, resource allocation (scheduling), message generation and encoding / decoding, and are not necessarily part of the device housing the antenna of T-TRP 170. These modules may also be coupled to other T-TRPs. In some embodiments, T-TRP 170 may actually be multiple T-TRPs that work together, for example, through coordinated multi-point transmission, to serve ED 110.

[0108] The T-TRP 170 includes at least one transmitter 252 and at least one receiver 254 coupled to one or more antennas 256. Only one antenna 256 is shown. Alternatively, one, some, or all of the antennas may be panels. The transmitter 252 and the receiver 254 may be integrated as a transceiver. The T-TRP 170 further includes a processor 260 for performing operations including operations related to the following: preparing a transmission for downlink transmission to the ED 110; processing an uplink transmission received from the ED 110; preparing a transmission for backhaul transmission to the NT-TRP 172; processing a transmission received from the NT-TRP 172 via the backhaul. The processing operations related to preparing a transmission for downlink transmission or backhaul transmission may include operations such as encoding, modulation, precoding (e.g., MIMO precoding), transmit beamforming, and generating symbols for transmission. The processing operations related to processing a transmission received in the uplink or received via the backhaul may include operations such as receive beamforming, demodulating, and decoding received symbols. The processor 260 may also perform operations related to network access (e.g., initial access) and / or downlink synchronization, such as generating the content of a synchronization signal block (SSB), generating system information, etc. In some embodiments, the processor 260 also generates an indication of a beam direction, e.g., a BAI, which may be scheduled by the scheduler 253 for transmission. The processor 260 performs other network-side processing operations described herein, such as determining the location of the ED 110, determining where to deploy the NT-TRP 172, etc. In some embodiments, the processor 260 may generate signaling, e.g., to configure one or more parameters of the ED 110 and / or one or more parameters of the NT-TRP 172. Any signaling generated by the processor 260 is sent by the transmitter 252. It should be noted that the "signaling" used herein may alternatively be referred to as control signaling. Dynamic signaling may be sent in a control channel (e.g., a physical downlink control channel (PDCCH)), and static or semi-static higher-layer signaling may be included in data packets sent in a data channel (e.g., a physical downlink shared channel (PDSCH)).

[0109] The scheduler 253 can be coupled to the processor 260. The scheduler 253 can be included within or operate separately from the T-TRP 170, and the scheduler can schedule uplink transmissions, downlink transmissions, and / or backhaul transmissions, including issuing scheduling grants and / or configuring grant-free (“configured grant”) resources. The T-TRP 170 also includes a memory 258 for storing information and data. The memory 258 stores instructions and data used, generated, or collected by the T-TRP 170. For example, the memory 258 can store software instructions or modules executed by the processor 260 for implementing some or all of the functions and / or embodiments described herein.

[0110] Although not shown, the processor 260 can form part of the transmitter 252 and / or the receiver 254. Additionally, although not shown, the processor 260 can implement the scheduler 253. Although not shown, the memory 258 can form part of the processor 260.

[0111] The processing components of the processor 260, the scheduler 253, and the transmitter 252 and the receiver 254 can each be implemented by the same or different one or more processors for executing instructions stored in a memory (e.g., the memory 258). Alternatively, some or all of the processing components of the processor 260, the scheduler 253, and the transmitter 252 and the receiver 254 can be implemented using dedicated circuitry (e.g., FPGA, GPU, or ASIC).

[0112] Although NT-TRP 172 is shown only as an example as a drone, NT-TRP 172 can be implemented in any suitable non-ground form. Additionally, NT-TRP 172 may use other names in some implementations, such as non-ground node, non-ground network device, or non-ground base station. NT-TRP 172 includes a transmitter 272 and a receiver 274 coupled to one or more antennas 280. Only one antenna 280 is shown. Alternatively, one, some, or all of the antennas may be panels. The transmitter 272 and the receiver 274 may be integrated as a transceiver. NT-TRP 172 also includes a processor 276 for performing operations, including operations related to the following: preparing a transmission for downlink transmission to ED 110; processing an uplink transmission received from ED 110; preparing a transmission for backhaul transmission to T-TRP 170; processing a transmission received from T-TRP 170 via the backhaul. The processing operations related to preparing a transmission for downlink transmission or backhaul transmission may include operations such as encoding, modulation, precoding (e.g., MIMO precoding), transmit beamforming, and generating symbols for transmission. The processing operations related to processing a transmission received in the uplink or received via the backhaul may include operations such as receive beamforming, demodulation, and decoding received symbols. In some embodiments, the processor 276 implements transmit beamforming and / or receive beamforming based on beam direction information (e.g., BAI) received from T-TRP 170. In some embodiments, the processor 276 may generate signaling, e.g., to configure one or more parameters of ED 110. In some embodiments, NT-TRP 172 implements physical layer processing but does not implement higher layer functions such as functions of the medium access control (MAC) or radio link control (RLC) layers. Since this is only an example, more generally, NT-TRP 172 may also implement higher layer functions in addition to physical layer processing.

[0113] NT-TRP 172 also includes a memory 278 for storing information and data. Although not shown, the processor 276 may form part of the transmitter 272 and / or the receiver 274. Although not shown, the memory 278 may form part of the processor 276.

[0114] The processor 276, and the processing components of the transmitter 272 and the receiver 274 may each be implemented by the same or different one or more processors, the one or more processors being configured to execute instructions stored in a memory (e.g., memory 278). Alternatively, some or all of the processing components of the processor 276 and the transmitter 272 and the receiver 274 may be implemented using dedicated circuitry (e.g., programmed FPGAs, GPUs, or ASICs). In some embodiments, the NT-TRP 172 may actually be multiple NT-TRPs, which work together, e.g., via coordinated multi-point transmission, to serve the ED 110.

[0115] It should be noted that the "TRP" used herein may refer to a T-TRP or an NT-TRP.

[0116] The T-TRP 170, NT-TRP 172, and / or ED 110 may include other components, but for clarity, these components are omitted.

[0117] According to Figure 4 , one or more steps of the example methods provided herein may be performed by corresponding units or modules. Figure 4 Units or modules in a device (e.g., in the ED 110, T-TRP 170, or NT-TRP 172) are shown. For example, a signal may be sent by a transmitting unit or module. For example, a signal may be sent by a transmitting unit or module. A signal may be received by a receiving unit or module. A signal may be processed by a processing unit or module. Other steps may be performed by an artificial intelligence (AI) or machine learning (ML) module. The corresponding units or modules may be implemented using hardware, one or more components or devices that execute software, or a combination thereof. For example, one or more of the units or modules may be integrated circuits, such as programmed FPGAs, GPUs, or ASICs. It should be understood that if the above modules are implemented using software for execution by a processor or the like, these modules may be retrieved in whole or in part by the processor as needed, retrieved individually or collectively for processing, retrieved in one or more instances as needed, and these modules may themselves include instructions for further deployment and instantiation.

[0118] Additional details regarding the ED 110, T-TRP 170, and NT-TRP 172 are known to those of ordinary skill in the art. Thus, these details are omitted here for clarity.

[0119] This document discusses control signaling in some embodiments. Control signaling can sometimes also be referred to as signaling, or control information, or configuration information, or configuration. In some cases, for example, in the physical layer of a control channel, control signaling can be dynamically indicated. An example of dynamically indicated control signaling is the information sent in physical layer control signaling, such as downlink control information (DCI). For example, in RRC signaling or in MAC control elements (CEs), control signaling can sometimes alternatively be semi-statically indicated. Dynamic indication can be an indication in a lower layer, such as physical layer / layer 1 signaling (e.g., in DCI), rather than an indication in a higher layer (e.g., rather than in RRC signaling or in MAC CEs). Semi-static indication can be an indication in semi-static signaling. The semi-static signaling used in this document can refer to non-dynamic signaling, such as higher layer signaling, RRC signaling, and / or MAC CEs. The dynamic signaling used in this document can refer to dynamic signaling, such as physical layer control signaling sent in the physical layer, such as DCI.

[0120] The air interface typically includes multiple components and associated parameters that together specify how transmissions are sent and / or received over a wireless communication link between two or more communication devices. For example, the air interface can include one or more components that define one or more waveforms, one or more frame structures, one or more multiple access schemes, one or more protocols, one or more coding schemes, and / or one or more modulation schemes for transmitting information (e.g., data) over the wireless communication link. The wireless communication link can support a link between a radio access network and a user equipment (e.g., the "Uu" link), and / or the wireless communication link can support a device-to-device link, e.g., a link between two user equipments (e.g., the "sidelink"), and / or the wireless communication link can support a link between a non-terrestrial (NT) communication network and a user equipment (UE). Some examples of the above components are as follows:

[0121] The waveform component can specify the shape and form of the signal being transmitted. Waveform options can include orthogonal multiple access waveforms and non-orthogonal multiple access waveforms. Non-limiting examples of such waveform options include orthogonal frequency division multiplexing (OFDM), filtered OFDM (f-OFDM), time-windowed OFDM, filter bank multicarrier (FBMC), universal filtered multicarrier (UFMC), generalized frequency division multiplexing (GFDM), wavelet packet modulation (WPM), faster than Nyquist (FTN) waveforms, and low peak-to-average power ratio waveforms (low PAPR WF).

[0122] The frame structure component can specify the configuration of a frame or a group of frames. The frame structure component can indicate one or more of the time, frequency, pilot signature, code, or other parameters of the frame or group of frames. More details of the frame structure will be discussed below.

[0123] Multiple access scheme components can specify multiple access technology options, including technologies that define how communication devices share a common physical channel, such as: time division multiple access (TDMA), frequency division multiple access (FDMA), code division multiple access (CDMA), single carrier frequency division multiple access (SC-FDMA), low density signature multicarrier code division multiple access (LDS-MC-CDMA), non-orthogonal multiple access (NOMA), pattern division multiple access (PDMA), lattice partition multiple access (LPMA), resource spread multiple access (RSMA), and sparse code multiple access (SCMA). Additionally, multiple access technology options can include: scheduled access and unscheduled access, also known as license-free access; non-orthogonal multiple access and orthogonal multiple access, e.g., via dedicated channel resources (e.g., not shared among multiple communication devices); contention-based shared channel resources and non-contention-based shared channel resources; cognitive radio-based access.

[0124] Hybrid automatic repeat request (HARQ) protocol components can specify how transmission and / or retransmission are to be performed. Non-limiting examples of transmission and / or retransmission mechanism options include those mechanism options that specify the size of the scheduled data pipeline, the signaling mechanism for transmission and / or retransmission, and the retransmission mechanism.

[0125] Coding and modulation components can specify how the information to be transmitted can be encoded / decoded and modulated / demodulated for transmission / reception. Coding can refer to methods of error detection and forward error correction. Non-limiting examples of coding options include turbo trellis codes, turbo product codes, fountain codes, low density parity check codes, and polar codes. Modulation can simply refer to the constellation (including, e.g., modulation techniques and orders), or more specifically to various types of advanced modulation methods, such as layered modulation and low PAPR modulation.

[0126] In some embodiments, the air interface can be a "one-size-fits-all concept". For example, once the air interface is defined, the components within the air interface cannot be changed or adapted. In some implementations, only limited parameters or modes of the air interface can be configured, such as the cyclic prefix (CP) length or the multiple input multiple output (MIMO) mode. In some embodiments, the air interface design can provide a unified or flexible framework to support frequency bands below 6 GHz and above 6 GHz (e.g., millimeter waves) for licensed and unlicensed access. For example, the flexibility of the configurable air interface provided by scalable system parameters and symbol durations can support optimizing transmission parameters for different spectrum bands and different services / devices. Another example is that the unified air interface can be self-contained in the frequency domain, and the frequency-domain self-contained design can support more flexible radio access network (RAN) slicing by sharing channel resources between different services in frequency and time.

[0127] Frame structure

[0128] The frame structure is a characteristic of the physical layer of wireless communication that defines the transmission structure of the time-domain signal. For example, it is used to support the timing reference and timing adjustment of the basic time-domain transmission unit. Wireless communication between communication devices can be carried out on the time-frequency resources controlled by the frame structure. The frame structure is sometimes referred to as the wireless frame structure.

[0129] According to the frame structure and / or the configuration of the frames in the frame structure, frequency division duplex (FDD) and / or time-division duplex (TDD) and / or full duplex (FD) communication is possible. FDD communication means that transmissions in different directions (e.g., uplink and downlink) occur on different frequency bands. TDD communication means that transmissions in different directions (e.g., uplink and downlink) occur in different durations. FD communication means that sending and receiving occur on the same time-frequency resources, i.e., the device can send and receive on the same frequency resources simultaneously in time.

[0130] An example of a frame structure is the frame structure in Long-Term Evolution (LTE) with the following specifications: The duration of each frame is 10 ms; each frame has 10 subframes, and the duration of each subframe is 1 ms; each subframe includes two time slots, and the duration of each time slot is 0.5 ms; each time slot is used to transmit 7 OFDM symbols (assuming normal CP); each OFDM symbol has a symbol duration and a specific bandwidth (or partial bandwidth or bandwidth partition) related to the number of subcarriers and the subcarrier spacing; the frame structure is based on OFDM waveform parameters, such as subcarrier spacing and CP length (where CP has a fixed length or a finite length option); the switching interval between uplink and downlink in TDD must be an integer multiple of the OFDM symbol duration.

[0131] Another example of a frame structure is the frame structure in New Radio (NR) with the following specifications: Multiple subcarrier spacings are supported, and each subcarrier spacing corresponds to corresponding system parameters; the frame structure depends on the system parameters, but in any case, the frame length is set to 10 ms and consists of 10 subframes, each subframe being 1 ms; a time slot is defined as 14 OFDM symbols, and the time slot length depends on the system parameters. For example, the NR frame structure with a 15 kHz subcarrier spacing for normal CP ("system parameter 1") and the NR frame structure with a 30 kHz subcarrier spacing for normal CP ("system parameter 2") are different. For a 15 kHz subcarrier spacing, the time slot length is 1 ms, and for a 30 kHz subcarrier spacing, the time slot length is 0.5 ms. The NR frame structure may have greater flexibility than the LTE frame structure.

[0132] Another example of a frame structure is an example of a flexible frame structure, such as for 6G networks or later. In a flexible frame structure, a symbol block can be defined as the minimum duration that can be scheduled in the flexible frame structure. A symbol block can be a transmission unit with an optional redundant part (e.g., CP part) and an information (e.g., data) part. An OFDM symbol is an example of a symbol block. Alternatively, a symbol block can be referred to as a symbol. Embodiments of the flexible frame structure include different parameters that can be configurable, such as frame length, subframe length, symbol block length, etc. In some embodiments of the flexible frame structure, a non-exhaustive list of possible configurable parameters includes:

[0133] (1) Frame: The frame length does not need to be limited to 10 ms. The frame length can be configurable and can change over time. In some embodiments, each frame includes one or more downlink synchronization channels and / or one or more downlink broadcast channels, and each synchronization channel and / or broadcast channel can be transmitted in different directions through different beamforming. Among them, the frame length can be more than one possible value and is configured based on the application scenario. For example, an autonomous vehicle may require relatively fast initial access. In this case, the frame length corresponding to the autonomous vehicle application can be set to 5 ms. Another example is that a smart meter on a house may not require fast initial access. In this case, the frame length corresponding to the smart meter application can be set to 20 ms.

[0134] (2) Sub-frame duration: The sub-frame may or may not be defined in a flexible frame structure, depending on the implementation. For example, a frame can be defined to include time slots but not sub-frames. In a frame where the sub-frame is defined, for example, for time domain alignment, the duration of the sub-frame can be configurable. For example, the sub-frame length can be configured to 0.1 ms or 0.2 ms or 0.5 ms or 1 ms or 2 ms or 5 ms, etc. In some embodiments, if the sub-frame is not required in a specific scenario, the sub-frame length can be defined to be the same as the frame length or not defined.

[0135] (3) Time slot configuration: The time slot may or may not be defined in a flexible frame structure, depending on the implementation. In a frame where the time slot is defined, the definition of the time slot (e.g., in terms of duration and / or the number of symbol blocks) can be configurable. In one embodiment, the time slot configuration is common to all UEs or a group of UEs. For this case, the time slot configuration information can be sent to the UEs in the broadcast channel or one or more common control channels. In other embodiments, the time slot configuration can be UE-specific. In this case, the time slot configuration information can be transmitted in a UE-specific control channel. In some embodiments, the time slot configuration signaling can be sent together with the frame configuration signaling and / or the sub-frame configuration signaling. In other embodiments, the time slot configuration can be transmitted independently of the frame configuration signaling and / or the sub-frame configuration signaling. Generally, the time slot configuration can be system-common, base station-common, UE group-common, or UE-specific.

[0136] (4) Subcarrier Spacing (SCS): SCS is a parameter of the scalable system parameters, which allows the SCS to be within a possible range of 15 KHz to 480 KHz. The SCS can vary with the spectral frequency and / or the maximum UE speed to minimize the effects of Doppler shift and phase noise. In some examples, there can be separate transmit and receive frames, and the SCS of the symbols in the receive frame structure can be configured independently of the SCS of the symbols in the transmit frame structure. The SCS in the receive frame can be different from the SCS in the transmit frame. In some examples, the SCS of each transmit frame can be half of the SCS of each receive frame. If the SCS is different between the receive frame and the transmit frame, for example, if an inverse discrete Fourier transform (IDFT) instead of a fast Fourier transform (FFT) is used to achieve a more flexible symbol duration, this difference does not have to be scaled by a factor of 2. Additional examples of frame structures can be used with different SCSs.

[0137] (5) Flexible Transmission Duration of the Basic Transmission Unit: The basic transmission unit can be a symbol block (alternatively referred to as a symbol), which typically includes a redundant part (referred to as the CP) and an information (e.g., data) part, but in some embodiments, the CP can be omitted from the symbol block. The CP length can be flexible and configurable. The CP length can be fixed within a frame or flexible within a frame. The CP length can change with the change of the frame, or with the change of the frame group, or with the change of the subframe, or with the change of the time slot, or dynamically with the change of the scheduling. The information (e.g., data) part can be flexible and configurable. Another possible parameter related to the symbol block that can be defined is the ratio of the CP duration to the information (e.g., data) duration. In some embodiments, the symbol block length can be adjusted according to channel conditions (such as multipath delay, Doppler) and / or latency requirements and / or available duration. For another example, the symbol block length can be adjusted to adapt to the available duration within a frame.

[0138] (6) Flexible Switching Gap: A frame can include a downlink part for downlink transmission from a base station and an uplink part for uplink transmission from a UE. There may be a gap between each uplink part and downlink part, and this gap is called a switching gap. The switching gap length (duration) can be configurable. The switching gap duration can be fixed within a frame or flexible within a frame. The switching gap duration can change with the change of the frame, or with the change of the frame group, or with the change of the subframe, or with the change of the time slot, or dynamically with the change of the scheduling.

[0139] Cell / Carrier / Bandwidth Part (BWP) / Occupied bandwidth

[0140] Devices such as base stations can provide coverage over a cell. Wireless communication with the device can be carried out on one or more carrier frequencies. The carrier frequency can be referred to as a carrier. The carrier can alternatively be referred to as a component carrier (CC). The carrier can be characterized by its bandwidth and reference frequency, such as the center frequency or the lowest frequency or the highest frequency of the carrier. The carrier can be on the licensed spectrum or on the unlicensed spectrum. Wireless communication with the device can also or alternatively be carried out on one or more bandwidth parts (BWPs). For example, a carrier can have one or more BWPs. More generally, wireless communication can be carried out with the device over the spectrum. The spectrum can include one or more carriers and / or one or more BWPs.

[0141] A cell can include one or more downlink resources and optionally one or more uplink resources, or a cell can include one or more uplink resources and optionally one or more downlink resources, or a cell can include one or more downlink resources and one or more uplink resources. For example, a cell can include only one downlink carrier / BWP, or only one uplink carrier / BWP, or include multiple downlink carriers / BWPs, or include multiple uplink carriers / BWPs, or include one downlink carrier / BWP and one uplink carrier / BWP, or include one downlink carrier / BWP and multiple uplink carriers / BWPs, or include multiple downlink carriers / BWPs and one uplink carrier / BWP, or include multiple downlink carriers / BWPs and multiple uplink carriers / BWPs. In some embodiments, a cell can alternatively or additionally include one or more sidelink resources, including sidelink transmission resources and reception resources.

[0142] A BWP is a set of contiguous or non - contiguous frequency sub - carriers on a carrier, or a set of contiguous or non - contiguous frequency sub - carriers on multiple carriers, or a set of non - contiguous or contiguous frequency sub - carriers that can have one or more carriers.

[0143] In some embodiments, a carrier may have one or more BWPs. For example, a carrier may have a bandwidth of 20 MHz and consist of one BWP, or a carrier may have a bandwidth of 80 MHz and consist of two adjacent BWPs, and so on. In other embodiments, a BWP may have one or more carriers. For example, the bandwidth of a BWP may be 40 MHz and consist of two adjacent and consecutive carriers, where the bandwidth of each carrier is 20 MHz. In some embodiments, a BWP may include discontinuous spectrum resources composed of discontinuous multi-carriers. Among them, the first carrier in the discontinuous multi-carriers may be in the mmW band, the second carrier may be in the low-frequency band (such as the 2 GHz band), the third carrier (if any) may be in the THz band, and the fourth carrier (if any) may be in the visible light band. The resources belonging to the BWP in one carrier may be continuous or discontinuous. In some embodiments, the BWP has discontinuous spectrum resources on one carrier.

[0144] Wireless communication can be carried out on the occupied bandwidth. The occupied bandwidth can be defined as the width of the frequency band such that the average transmitted power is equal to a specified percentage β / 2 of the total average transmitted power below the lower frequency limit and above the upper frequency limit. For example, the value of β / 2 is taken as 0.5%.

[0145] The carrier, BWP, or occupied bandwidth can be: signaled dynamically by a network device (e.g., a base station) in physical layer control signaling (e.g., downlink control information (DCI)); or signaled semi-statically by a network device (e.g., a base station) in radio resource control (RRC) signaling or in the medium access control (MAC) layer; or predefined based on the application scenario; or determined by the UE as a function of other parameters known to the UE; or can be fixed by the standard.

[0146] Artificial intelligence (AI) and / or machine learning (ML)

[0147] The number of new devices in future wireless networks is expected to grow exponentially, and the functions of the devices are expected to become increasingly diverse. In addition, many new applications and use cases are expected to emerge, and their quality-of-service requirements will be more diverse than those of 5G applications / use cases. These will require future wireless networks (e.g., 6G networks) to have new key performance indicators (KPIs), which may be extremely challenging. AI technologies, such as ML technologies (e.g., deep learning), have been introduced into telecom applications with the goal of improving system performance and efficiency.

[0148] In addition, continuous progress has been made in terms of antenna and bandwidth capabilities, thus allowing for more and / or better communication over the wireless link. Additionally, for example, through the introduction of general-purpose graphics processing units (GP-GPUs), continuous progress has been made in the field of computer architecture and computing capabilities. Future generations of communication devices may have stronger computing and / or communication capabilities than previous generations, which can allow for the adoption of AI to implement radio access network components. Future generations of networks can also access more accurate and / or new information (compared to previous networks), which can form the input basis for AI models, such as: the physical speed / velocity of device movement, the link budget of the device, the channel conditions of the device, one or more device capabilities and / or the type of service to be supported, sensing information and / or positioning information, etc. To obtain sensing information, the TRP can send a signal to the target object (e.g., a suspected UE), and based on the reflection of the signal, the TRP or another network device calculates the angle (for beamforming of the device), the distance between the device and the TRP, and / or Doppler shift information. Positioning information is sometimes referred to as location and can be obtained in various ways, such as from a positioning report from the UE (e.g., a report of the UE's GPS coordinates), using positioning reference signals (PRSs), using the sensing, tracking, and / or prediction of the device's position described above, etc.

[0149] AI technologies (including ML technologies) can be applied in communications, including AI-based communications in the physical layer and / or AI-based communications in the MAC layer. For the physical layer, AI communications can aim to optimize component design and / or improve algorithm performance. For example, AI can be applied to the following implementations: channel coding, channel modeling, channel estimation, channel decoding, modulation, demodulation, MIMO, waveforms, multiple access, optimization and update of physical layer element parameters, beamforming, tracking, sensing, and / or positioning, etc. For the MAC layer, AI communications can aim to utilize AI capabilities for learning, prediction, and / or decision-making to solve complex optimization problems with potentially better strategies and / or optimal solutions, such as optimizing functions in the MAC layer. For example, AI can be used to implement: intelligent TRP management, intelligent beam management, intelligent channel resource allocation, intelligent power control, intelligent spectrum utilization, intelligent MCS, intelligent HARQ strategies, and / or intelligent transmit / receive mode adaptation, etc.

[0150] In some embodiments, the AI architecture can involve multiple nodes, where the multiple nodes can be organized in one of two modes: centralized and distributed, and both modes can be deployed in the access network, core network, edge computing system, or third-party network. The centralized training and computing architecture can be limited by potentially large communication overhead and strict user data privacy. The distributed training and computing architecture can include several frameworks, such as distributed machine learning and federated learning. In some embodiments, the AI architecture can include an intelligent controller, which can execute based on joint optimization or separate optimization as a single agent or multiple agents. New protocols and signaling mechanisms are needed to enable the corresponding interface links to be personalized with custom parameters to meet specific requirements, while minimizing signaling overhead and maximizing the overall system spectrum efficiency through personalized AI technologies.

[0151] In some embodiments herein, new protocols and signaling mechanisms are provided for operating within and switching between different operating modes for AI training, including switching between the training mode and the normal operating mode, and for measurement and feedback to accommodate different possible measurements and information that may require feedback, depending on the implementation.

[0152] AI Training

[0153] Referring again to Figure 1 and Figure 2 , embodiments of the present invention can be used to implement AI training involving two or more communication devices in the communication system 100. For example, Figure 5FIG. 0 shows four EDs communicating with a network device 452 in a communication system 100 according to one embodiment. The four EDs are respectively shown as corresponding different UEs and will be referred to hereinafter as UE402, UE 404, UE 406, and UE 408. However, an ED does not necessarily have to be a UE.

[0154] The network device 452 is part of a network (e.g., a radio access network 120). The network device 452 may be deployed in an access network, a core network, an edge computing system, or a third-party network, depending on the implementation. The network device 452 may be a T-TRP or a server (or a part thereof). In one example, the network device 452 may be a T-TRP 170 or an NT-TRP 172 (or implemented therein). In another example, the network device 452 may be a T-TRP controller and / or an NT-TRP controller that can manage the T-TRP 170 or the NT-TRP 172. In some embodiments, the components of the network device 452 may be distributed. For example, if the network device 452 is part of a T-TRP serving the UEs 402, 404, 406, and 408, the UEs 402, 404, 406, and 408 may communicate directly with the network device 452. Alternatively, the UEs 402, 404, 406, and 408 may communicate with the network device 452 via one or more intermediate components, such as via a T-TRP and / or via an NT-TRP, etc. For example, via a backhaul link and a radio channel established between the network device 452 and the UEs 402, 404, 406, and 408, the network device 452 may send information (e.g., control signaling, data, training sequences, etc.) to and / or receive information (e.g., control signaling, data, training sequences, etc.) from one or more of the UEs 402, 404, 406, and 408.

[0155] As described above, each of the UEs 402, 404, 406, and 408 includes a corresponding processor 210, a memory 208, a transmitter 201, a receiver 203, and one or more antennas 204 (or alternatively, a panel). For simplicity, only the processor 210, the memory 208, the transmitter 201, the receiver 203, and the antenna 204 of the UE 402 are shown, but the other UEs 404, 406, and 408 also include the same corresponding components.

[0156] For each of the UEs 402, 404, 406, and 408, the communication link between the UE and the corresponding TRP in the network is the air interface. The air interface typically includes multiple components and associated parameters that together specify how to send and / or receive transmissions over the wireless medium.

[0157] Figure 5 The processor 210 of the UE in implements one or more air interface components on the UE side. The air interface components configure and / or implement transmission and / or reception on the air interface. Examples of air interface components are described herein. The air interface components can be in the physical layer, such as a channel encoder (or decoder) that implements the encoding component of the UE's air interface, and / or a modulator (or demodulator) that implements the modulation component of the UE's air interface, and / or a waveform generator that implements the waveform component of the UE's air interface, etc. The air interface components can be in or part of a higher layer such as the MAC layer, such as a module that implements channel prediction / tracking, and / or a module that implements a retransmission protocol (such as a HARQ protocol component that implements the UE's air interface). The processor 210 also directly executes (or controls the UE to execute) the UE-side operations described herein.

[0158] The network device 452 includes a processor 454, a memory 456, and an input / output device 458. The processor 454 implements or instructs other network devices (such as a T-TRP) to implement one or more of the air interface components on the network side. One UE can implement the air interface components differently from another UE on the network side. The processor 454 directly executes (or controls the network components to execute) the network-side operations described herein.

[0159] The processor 454 can be implemented by the same or different one or more processors for executing instructions stored in a memory (such as the memory 456). Alternatively, some or all of the processor 454 can be implemented using dedicated circuitry (such as a programmed FPGA, GPU, or ASIC). The memory 456 can be implemented by volatile and / or non-volatile memory. Any suitable type of memory can be used, such as RAM, ROM, hard disk, optical disk, cache on the processor, etc.

[0160] The input / output device 458 allows interaction with other devices by receiving (inputting) and sending (outputting) information. In some embodiments, the input / output device 458 may be implemented by a transmitter and / or a receiver (or transceiver), and / or one or more interfaces (such as a wired interface, such as a wired interface to an internal network or to the Internet, etc.). In some implementations, the input / output device 458 may be implemented by a network interface, which may be implemented as a network interface card (NIC) and / or a computer port (such as a physical socket to which a plug or cable is connected) and / or a network socket, etc., depending on the implementation.

[0161] The network device 452 and the UE 402 have the ability to implement one or more AI-enabled processes. Specifically, in Figure 5 embodiments, the network device 452 and the UE 402 respectively include ML modules 410 and 460. The ML module 410 is implemented by the processor 210 of the UE 402, and the ML module 460 is implemented by the processor 454 of the network device 452. Therefore, in Figure 5 , the ML module 410 is shown inside the processor 210, and the ML module 460 is shown together with the processor 454. The ML modules 410 and 460 execute one or more AI / ML algorithms to perform one or more AI-enabled processes, such as AI-enabled link adaptation, to optimize the communication link between the network and the UE 402.

[0162] The ML modules 410 and 460 can be implemented using AI models. The term AI model may refer to a computer algorithm that is used to accept defined input data and output defined inference data, where the parameters (e.g., weights) of the algorithm can be updated and optimized through training (e.g., using a training dataset, or using data collected in real life). AI models can be implemented using one or more neural networks (e.g., including deep neural network (DNN), recurrent neural network (RNN), convolutional neural network (CNN), and combinations thereof) and using various neural network architectures (e.g., autoencoders, generative adversarial networks, etc.). Various techniques can be used to train AI models in order to update and optimize their parameters. For example, backpropagation is a commonly used technique for training DNNs, where a loss function is calculated between the inference data generated by the DNN and some target output (e.g., ground truth data). The gradient of the loss function is calculated with respect to the parameters of the DNN, and the calculated gradient is used to update the parameters (e.g., using a gradient descent algorithm) with the goal of minimizing the loss function.

[0163] In some embodiments, the AI model includes a neural network for machine learning. The neural network consists of multiple computing units (which can also be referred to as neurons), and these computing units are arranged in one or more layers. The process of receiving an input at the input layer and generating an output at the output layer can be referred to as forward propagation. In forward propagation, each layer receives an input (which may have any suitable data format, such as a vector, matrix, or multi-dimensional array) and performs calculations to generate an output (which may have a different dimension from the input). The calculations performed by a layer typically involve applying (e.g., multiplying) the input by a set of weights (also referred to as coefficients). Except for the first layer (i.e., the input layer) of the neural network, the input of each layer is the output of the previous layer. The neural network may include one or more layers between the first layer (i.e., the input layer) and the last layer (i.e., the output layer), and these layers can be referred to as inner layers or hidden layers. For example, Figure 6A FIG. 600 shows an example of a neural network including an input layer, an output layer, and two hidden layers. In this example, it can be seen that the output of each of the three neurons in the input layer of the neural network 600 is included in the input vector of each of the three neurons in the first hidden layer. Similarly, the output of each of the three neurons in the first hidden layer is included in the input vector of each of the three neurons in the second hidden layer, and the output of each of the three neurons in the second hidden layer is included in the input vector of each of the two neurons in the output layer. As described above, the basic computing unit in a neural network is a neuron, as Figure 6A shown at 650. Figure 6B FIG. 650 shows an example of a neuron 650 that can be used as a building block of the neural network 600. As Figure 6B shown, in this example, the neuron 650 takes a vector x as input and performs a dot product of the associated vector with the weight w. The final output z of the neuron is the result of the activation function f() of the dot product. Various neural networks can be designed to have various architectures (e.g., various numbers of layers, and each layer performs various functions).

[0164] The neural network is trained to optimize the parameters (e.g., weights) of the neural network. This optimization is performed in an automated manner and can be referred to as machine learning. The training of the neural network includes performing forward propagation on input data samples to generate output values (also referred to as predicted output values or inference output values), and comparing the generated output values with known or desired target values (e.g., ground truth). A loss function is defined to quantitatively represent the difference between the generated output values and the target values, and the goal of training the neural network is to minimize the loss function. Backpropagation is an algorithm used to train the neural network. Backpropagation is used to adjust (also referred to as update) the values of the parameters (e.g., weights) in the neural network such that the computed loss function becomes smaller. Backpropagation involves computing the gradient of the loss function with respect to the parameters to be optimized, and then using a gradient algorithm (e.g., gradient descent) to update the parameters to reduce the loss function. Backpropagation is performed iteratively so that the loss function converges or is minimized over multiple iterations. After the training conditions are met (e.g., the loss function has converged, or a predefined number of training iterations have been performed), the neural network is considered to be trained. The trained neural network can be deployed (or executed) to generate inference output data based on the input data. In some embodiments, even after the neural network has been deployed, the training of the neural network can continue such that the parameters of the neural network can be repeatedly updated with the latest training data.

[0165] Referring again to Figure 5 , in some embodiments, the UE 402 and the network device 452 may exchange information for training purposes. The information exchanged between the UE 402 and the network device 452 is implementation-specific and may not have a human-readable meaning (e.g., it may be intermediate data generated during the execution of an ML algorithm). It may also be possible or, conversely, the exchanged information is not predefined by a standard, e.g., bits may be exchanged, but the bits may not be associated with a predefined meaning. In some embodiments, the network device 452 may provide or indicate to the UE 402 one or more parameters to be used in the ML module 410 implemented at the UE 402. As an example, the network device 452 may send or indicate updated neural network weights to be implemented in the neural network executed by the ML module 410 on the UE side in order to attempt to optimize one or more aspects of the modulation and / or coding for the communication between the UE 402 and the T-TRP or NT-TRP.

[0166] In some embodiments, the UE 402 itself may implement AI, such as performing learning. In other embodiments, the UE 402 itself may not perform learning, but may operate in conjunction with the AI implementation on the network side. For example, by receiving the configuration of an AI model (such as a neural network or other ML algorithm) implemented by the ML module 410 from the network, and / or by providing requested measurement results or observations to assist other devices (such as network devices or other AI-enabled UEs) in training an AI model (such as a neural network or other ML algorithm). For example, in some embodiments, the UE 402 itself may not implement learning or training, but the UE 402 may receive the training configuration information of the ML model determined by the network device 452 and execute the model.

[0167] Although Figure 5 the examples in assume AI / ML capabilities on the network side, it is possible that the network itself does not perform training / learning, and the UE may perform learning / training itself, possibly using dedicated training signals sent from the network. In other embodiments, end-to-end (E2E) learning may be implemented by the UE and the network device 452.

[0168] Using AI, such as by implementing an AI model as described above, various processes (such as link adaptation) can be AI-enabled. Some examples of possible AI / ML training processes and over-the-air information exchange processes between devices during the training phase are described below to facilitate AI-enabled processes according to embodiments of the present invention.

[0169] Referring again to Figure 5 , for wireless federated learning (FL), the network device 452 may initialize a global AI / ML model implemented by the ML module 460, for a group of UEs (such as Figure 5Four UEs (UE 402, UE 404, UE 406, and UE 408) shown in are sampled, and global AI / ML model parameters are broadcast to the UEs. Then, each of UE 402, UE 404, UE 406, and UE 408 can initialize its local AI / ML model with the global AI / ML model parameters and update (train) its local AI / ML model using its own data. Then, each of UE 402, UE 404, UE 406, and UE 408 can report the parameters of its updated local AI / ML model to the network device 452. The network device 452 can then aggregate the updated parameters reported from UE 402, UE 404, UE 406, and UE 408 and update the global AI / ML model. The above process is one iteration of the FL-based AI / ML model training process. The network device 452 and UE 402, UE 404, UE 406, and UE 408 perform multiple iterations until the AI / ML model has converged sufficiently to meet one or more training objectives / criteria, and the AI / ML model is finally determined.

[0170] Aspects of the present invention provide solutions to overcome at least some of the above limitations, such as specific methods and devices for artificial intelligence or machine learning (AI / ML) model training. The methods and devices disclosed in the present invention can overcome technical problems related to transmitting AI / ML model training data, such as the inability to determine the relative importance of data of the same data type.

[0171] As described above, one way to reduce the latency of the AI / ML model training process can be to minimize communication latency and / or computational latency. These latencies can be reduced by sending the corresponding AI / ML model training data based on data importance. The data importance (e.g., the importance of the corresponding AI / ML model training data) can be read or determined based on data state information (DSI), such as data uncertainty. For example, if the uncertainty level of the data is high, the data is more important (i.e., has a higher importance) for the AI / ML model training process.

[0172] DSI (e.g., data uncertainty) can be determined at a device (e.g., but not limited to, a user equipment (UE) or a base station (BS)) based on AI / ML model training auxiliary information. The AI / ML model training auxiliary information can include at least one of information about a reference AI / ML model or at least one reference input data value. For example, the information about the reference AI / ML model can include: reference AI / ML model type, reference AI / ML model structure, one or more reference AI / ML model parameters, reference AI / ML model gradients, reference AI / ML model activation functions, reference AI / ML model input data types, reference AI / ML model output data types, reference AI / ML model input data dimensions, and / or reference AI / ML model output data dimensions. One or more reference input data values can be corresponding reference values to be used as input data for the AI / ML model. In some cases, e.g., when the AI / ML model has only one input data, or when the device (e.g., UE) has all (multiple) inputs of the AI / ML model, the AI / ML model training auxiliary information may not include at least one reference input data value.

[0173] In some embodiments, a DSI threshold (e.g., uncertainty threshold) can be used to determine whether the corresponding AI / ML model training data will be sent. For example, for AI / ML model training, only the AI / ML model training data with an uncertainty level greater than the uncertainty threshold can be reported.

[0174] In some embodiments, the AI / ML model training dataset information can be used to determine whether the corresponding AI / ML model training data will be sent. The AI / ML model training dataset information can include the AI / ML model training dataset size and / or the DSI distribution information of the AI / ML model training dataset.

[0175] A device (e.g., UE, BS) can determine the DSI of the corresponding AI / ML model training data based on the AI / ML model training auxiliary information. Then, the device can send the AI / ML model training data to another device (e.g., a UE or BS that performs AI / ML model training) based on the DSI of the AI / ML model training data and / or information related to the transmission of the AI / ML model training data.

[0176] In some embodiments, a device (e.g., UE, BS) can determine the DSI of the corresponding AI / ML model training data in federated learning. For example, a UE can determine the data uncertainty of its local AI / ML model training data (local AI / ML model training data at the UE) before uploading / sending the local gradients associated with the local AI / ML model to the base station.

[0177] In some embodiments with bilateral AI / ML model training, a device (e.g., UE, BS) may determine the DSI of the corresponding AI / ML model training data after completing at least part of the AI / ML model training. For example, there are embodiments where the UE includes an encoder and a reference decoder, the BS includes a decoder, and the UE and the BS jointly perform AI / ML model training. In these embodiments, after the AI / ML model training at the encoder and the reference decoder is completed, the UE determines the data uncertainty of the AI / ML model training data. The corresponding AI / ML model training data may include the corresponding outputs of parts of the AI / ML model training (e.g., the corresponding outputs of the encoder and the reference decoder at the UE).

[0178] In some embodiments, the AI / ML model training data may be sent based on the DSI of the corresponding AI / ML model training data. The AI / ML model training data may be sent according to a corresponding reporting format that may indicate the corresponding transmission accuracy. In one example, the UE may send AI / ML model training data with a high data uncertainty level and high accuracy to the BS, although this may result in high transmission overhead. On the other hand, the UE may send AI model training data with a low uncertainty level and low accuracy to the BS to reduce the overhead. In this case, the relationship between the DSI of the corresponding AI / ML model training data and the corresponding transmission accuracy may be configured by the BS.

[0179] Unilateral AI / ML Model Training at the BS

[0180] According to some embodiments, the AI / ML model training data may be sent based on the importance of the AI / ML model training data. The importance of the AI / ML model training data may be measured or determined based on the data state information (DSI) of the AI / ML model training data. The DSI may include data uncertainty.

[0181] Taking the data uncertainty of DSI as an example, if the data uncertainty level of the AI / ML model training data is high, the AI / ML model training data can be reported or sent to another device. The AI / ML model training data with a high data uncertainty level can be considered more important than the AI / ML model training data with a low data uncertainty level because the AI / ML model training data with a high data uncertainty level can provide more information (i.e., be more useful) for AI / ML model training. In addition, when performing AI / ML model training using the AI / ML model training data with a high data uncertainty level, overfitting is less likely to occur. On the contrary, the AI / ML model training data with a low data uncertainty level can be considered less important because such AI / ML model training data may not contribute enough to the convergence of the AI / ML model (e.g., contribute little or nothing to the faster convergence of the AI / ML model).

[0182] In some embodiments, AI / ML model training and AI / ML inference can be performed at the BS. For the AI / ML model training and AI / ML inference processes, the BS can receive AI / ML model training data from one or more UEs. The AI / ML model training data can be generated by each UE. The AI / ML model training can be one-sided AI / ML model training because the training is only performed on the BS side.

[0183] Figure 7 An example of one-sided AI / ML model training at the BS 720 in the wireless network 700 according to an embodiment of the present invention is shown. Refer to Figure 7 , there are multiple UEs 710 and BS 720 in the network 700. Each UE 710 is communicatively and operably connected to the BS 720. Each UE 710 can send the AI / ML model training data to the BS 720 based on the importance of the corresponding AI / ML model training data. In some embodiments, each UE 710 can send the AI / ML model training data to the BS 720 in a selective manner based on the importance of the corresponding AI / ML model training data. The importance of the corresponding AI / ML model training data can be determined based on the DSI of the corresponding AI / ML model training data (e.g., the data uncertainty of the corresponding AI / ML model training data).

[0184] To determine the DSI of the corresponding AI / ML model training data, each UE 710 may require some auxiliary information. In some embodiments, the auxiliary information may be AI / ML model training auxiliary information provided by the BS 720. Each UE 710 may receive the AI / ML model training auxiliary information from the BS 720. In some embodiments, the UE 710 may receive the AI / ML model training auxiliary information periodically. However, in some embodiments, the UE 710 may receive the AI / ML model training auxiliary information aperiodically. In other words, before the UE 710 determines the DSI of the corresponding AI / ML model training data, the BS 720 may not need to send the AI / ML model training auxiliary information. The UE 710 may use the received AI / ML model training auxiliary information to determine the DSI of the corresponding AI / ML model training data. The AI / ML model training auxiliary information may include at least one of information about a reference AI / ML model (or query AI / ML model) or at least one reference input data value. The information about the reference AI / ML model may include at least one of the following:

[0185] · The type of the reference AI / ML model (e.g., convolutional neural network (CNN), recurrent neural network (RNN), deep neural network (DNN));

[0186] · The structure of the reference AI / ML model (e.g., the number of layers, the number of neurons in each layer);

[0187] · One or more reference AI / ML model parameters (e.g., weights, coefficients);

[0188] · The reference AI / ML model gradient;

[0189] · The reference AI / ML model activation function (e.g., Sigmoid, rectified linear unit (ReLU), exponential linear unit (ELu), SoftMax);

[0190] · The type of the reference AI / ML model input data;

[0191] · The type of the reference AI / ML model output data;

[0192] · The dimension of the reference AI / ML model input data; or

[0193] · The dimension of the reference AI / ML model output data.

[0194] To determine the DSI of the corresponding AI / ML model training data, a UE (e.g., UE 710) may input the corresponding AI / ML model training data into a reference AI / ML model. The UE may determine the DSI of the corresponding AI / ML model training data based on the output of the reference AI / ML model, and the UE inputs the corresponding AI / ML model training data into the reference AI / ML model.

[0195] When the DSI of the corresponding AI / ML model training data is determined, UE 710 may report or send the corresponding AI / ML model data to BS 720. UE 710 may send the corresponding AI / ML model data based on the DSI of the corresponding AI / ML model training data and / or information related to the transmission of the corresponding AI / ML model training data. For example, the corresponding AI / ML model training data may be selectively sent (e.g., only part of the AI / ML model training data is sent and other AI / ML model training data is not sent) based on the DSI of the corresponding AI / ML model training data and / or information related to the transmission of the corresponding AI / ML model training data. Information related to the transmission of the corresponding AI / ML model training data will be discussed below or elsewhere in the present invention.

[0196] As described above, the AI / ML model training assistance information that can be used to determine the DSI of the corresponding AI / ML model training data may include information about the reference AI / ML model. Figure 8A An example of the reference AI / ML model is shown.

[0197] Figure 8A The reference AI / ML model 800 shown in may be implemented using a DNN. In other words, the type of the reference AI / ML model 800 may be a DNN. The reference AI / ML model input data dimension (i.e., the dimension of the input data of the reference AI / ML model 800) is M, so there are M input data (i.e., Input 1 to Input M ). The reference AI / ML model output data dimension (i.e., the dimension of the output data of the reference AI / ML model 800) is N, so there are N output data (i.e., y 1 to y N ). The reference AI / ML model 800 may include L hidden layers (i.e., the number of hidden layers of the reference AI / ML model 800 is L), and each hidden layer includes K neurons (i.e., the number of neurons in each hidden layer is K). The reference AI / ML model 800 may be trained to optimize one or more reference AI / ML model (weight) parameters (e.g., w 11 , w 1K, w M1 , w MK ). The reference AI / ML model parameters can be determined based on the AI / ML model type (e.g., DNN in this case) and / or the reference AI / ML model structure. The reference AI / ML model activation function can be a predefined function indicated by using a preconfigured function index.

[0198] In some embodiments, information about the reference AI / ML model can be sent from the BS (e.g., BS 720) to the UE (e.g., UE 710) using radio resource control (RRC), medium access control (MAC) control element (MAC-CE), or downlink control information (DCI). The information about the reference AI / ML model can be sent using broadcast signaling, unicast signaling, or multicast signaling. In some embodiments, the information about the reference AI / ML model can be specific to one UE or a group of UEs / devices. In these cases, the information about the reference AI / ML model can be sent using unicast signaling or multicast signaling.

[0199] As described above, the AI / ML model training assistance information that can be used to determine the DSI of the corresponding AI / ML model training data can also include at least one reference input data value or at least one reference AI / ML model input data. There are two ways for the BS (e.g., BS 720) to send the reference AI / ML model input data.

[0200] The first way is that the BS (e.g., BS 720) sends the reference values of all the input data of the reference AI / ML model. The UE (e.g., UE 710) can replace some of the received reference values with the local AI / ML model training data, as Figure 8B shown, Figure 8B shows an example of a reference AI / ML model with reference AI / ML model input data. Specifically, if the UE that determines the DSI of the corresponding AI / ML model training data is UE i (e.g., one of the UEs 710), then the UE i can replace the i-th reference value with its local AI / ML model training data and then calculate the output of the reference AI / ML model. In some embodiments, to simplify this process, some predefined values can be used to indicate the reference AI / ML model input data that will be replaced by the UE i . For example, the reference AI / ML model input data to be replaced can be filled with 0 or 1. The UE iThe position of the reference AI / ML model input data to be replaced can be determined among all the reference data based on the UE index and / or the reference AI / ML model input data dimension (i.e., the dimension of the entire reference AI / ML model input data).

[0201] The second way is that the BS (e.g., BS 720) only sends the reference values of some of the reference AI / ML model input data. The UE (e.g., UE 710) can replace the missing reference values (i.e., the input data for which the reference values are not filled) with its local AI / ML model training data. In other words, the UE (e.g., UE 710) can add its local AI / ML model training data as the input data for the reference AI / ML model (where the input data is not available). UE i The position of the reference AI / ML model input data to be added can be determined based on the UE index and / or the reference AI / ML model input data dimension (i.e., the dimension of the entire reference AI / ML model input data).

[0202] In some embodiments, the DSI of the corresponding AI / ML model training data may include information indicating at least one of the following: the data uncertainty of the corresponding AI / ML model training data; the data importance of the corresponding AI / ML model training data; the degree of need for the corresponding AI / ML model training data for AI / ML model training; or the data diversity of the corresponding AI / ML model training data.

[0203] The data uncertainty of the corresponding AI / ML model training data can be determined based on at least one of the following: entropy, minimum confidence, marginal sampling, or generalization error.

[0204] In some embodiments where the DSI of the corresponding AI / ML model training data includes information indicating the data uncertainty of the corresponding AI / ML model training data, the data uncertainty of the corresponding AI / ML model training data can be determined based on entropy. In this case, the output data of the corresponding AI / ML model training data can be a function of probability. For example, in Figure 8B where the AI / ML model 820 is designed to solve a certain classification problem and the input data is x i (i.e., Input i , P(x i ) is the probability that the input data x i belongs to the category i. The entropy of the output of the AI / ML model 820 can be expressed as follows in Equation (1):

[0205]

[0206] According to the definition of entropy in Equation (1) above, the larger the value of H(x), the higher the data uncertainty.

[0207] In some embodiments where the DSI of the corresponding AI / ML model training data includes information indicating the data uncertainty of the corresponding AI / ML model training data, the data uncertainty of the corresponding AI / ML model training data can be determined based on the minimum confidence. For example, the AI / ML model may not assign a specific class to the corresponding AI / ML model training data because the AI / ML model (e.g., AI / ML model 820) is not confident about class membership. Thus, the AI / ML model (e.g., AI / ML model 820) can be trained to select the most informative and uncertain AI / ML model training data samples. In other words, the AI / ML model training data with a lower prediction confidence will be regarded as the AI / ML model training data with a higher data uncertainty. When selecting the AI / ML model training data with uncertain predictions (e.g., it is difficult to predict the class to which the AI / ML model training data belongs), the margin sampling method can be used. Using the margin sampling method, the AI / ML model training data with the minimum distance from the hyperplane can be the desired AI / ML model training data (e.g., the most uncertain data).

[0208] In some embodiments where the DSI of the corresponding AI / ML model training data includes information indicating the data uncertainty of the corresponding AI / ML model training data, the data uncertainty of the corresponding AI / ML model training data can be determined based on the generalization error (i.e., out-of-sample error). For supervised learning applications in machine learning and statistical learning theory, the generalization error can be used to determine the accuracy with which an algorithm can predict the output values of unprecedented (i.e., previously unseen) or uncertain data.

[0209] Figure 8C An example of measuring data uncertainty using the channel information by the AI / ML model 840 according to an embodiment of the present invention is shown. In Figure 8C the example shown, the data uncertainty is measured based on entropy, and the AI / ML model is implemented using a DNN. The input data of the AI / ML model 840 is the channel information, and the output data of the AI / ML model 840 is the probability of each MCS index. The AI / ML model 840 is designed to provide the probability of each MCS index using the channel information as the input data.

[0210] Assume the UE i ( Figure 8C not shown in the figure) is the UE that determines the data uncertainty of the corresponding AI / ML model training data. When the UE i uses data i 1 as the Input of the AI / ML model 840 i the output probabilities of MCS 5 and MCS 7 are 95% and 5% respectively. If the UEi Taking data i 2 as the Input of the AI / ML model 840 i , the output probabilities of MCS 5 and MCS 7 are 51% and 49% respectively. For data i 1 and data i 2 , the output probabilities of other MCS indices can be ignored (i.e., close to zero). Therefore, in this example, the output probabilities of other MCS indices can be ignored. Since according to formula (1) given above, the entropy of data i 2 is greater than the entropy of data i 1 , so data i 2 provides higher data uncertainty than data i 2 . Therefore, the UE i can report or send data i 2 to the BS to enhance the training of the corresponding AI / ML model at the BS (e.g., improve the training performance of the AI / ML model).

[0211] According to some embodiments, after the DSI of the AI / ML model training data is determined, for example, using one or more methods shown above or elsewhere in the present invention, the UE (e.g., UE 710) or the BS (e.g., BS 720) can determine whether the corresponding AI / ML model training data will be sent to the BS (e.g., BS 720). There can be at least two ways to report the corresponding AI / ML model training data, as Figure 9A and Figure 9B shown.

[0212] Figure 9A shows the first way to report AI / ML model training data in the wireless network 700. In the first way, when the BS 720 sends AI / ML model training assistance information, the BS 720 can also send information related to the transmission of the corresponding AI / ML model training data. For example, the AI / ML model training assistance information and the information related to the transmission of the corresponding AI / ML model training data can be sent together (jointly). The information related to the transmission of the corresponding AI / ML model training data can include a DSI threshold. In some embodiments where the DSI of the AI / ML model training data includes information indicating the data uncertainty of the corresponding AI / ML model training data, the DSI threshold can be a data uncertainty threshold. The BS 720 can configure or pre-configure the data uncertainty threshold for the UE 710 (e.g., for each UE). For each AI / ML model training data, if the data uncertainty of the AI / ML model training data is greater than the data uncertainty threshold received from the BS 720, the UE 710 can send the AI / ML model training data to the BS 720 for AI / ML model training.

[0213] In some embodiments, data uncertainty can be quantified in the form of an uncertainty level (e.g., the uncertainty level ranges from level 0 to level N, where N is a positive integer). In this case, the uncertainty threshold can also be quantified in the form of an uncertainty level from level 0 to level N (e.g., the uncertainty threshold is level 3).

[0214] In the first way of reporting AI / ML model training data, the UE 710 can determine whether the corresponding AI / ML model training data will be sent or reported to the BS 720. As shown above in the example of using the uncertainty threshold, the UE 710 can determine whether the corresponding AI / ML model training data will be sent or reported to the BS 720 based on the DSI threshold and the DSI of the corresponding AI / ML model training data.

[0215] Figure 9B A second way of reporting AI / ML model training data in the wireless network 700 is shown. In the second way, the BS 720 may not send information related to the transmission of the corresponding AI / ML model training data, but instead send AI / ML model training assistance information. After receiving the AI / ML model training assistance information, the UE 710 may send at least one of the DSI of the corresponding AI / ML model training data or the AI / ML model training dataset information. In some embodiments, the DSI of the corresponding AI / ML model training data may include information indicating the data uncertainty (e.g., uncertainty value, uncertainty level) of the corresponding AI / ML model training data.

[0216] When the BS 720 receives the DSI of the corresponding AI / ML model training data (e.g., information indicating the data uncertainty of the corresponding AI / ML model training data), the BS 720 may determine whether the corresponding AI / ML model training data is to be sent from the UE 710. Then, the BS 720 may send information related to the transmission of the corresponding AI / ML model training data. In this case, the information related to the transmission of the corresponding AI / ML model training data may include information indicating whether the corresponding AI / ML model training data is to be sent (reported) to the BS 720. For example, for each AI / ML model training data, if the DSI (e.g., data uncertainty) of the AI / ML model training data meets the requirements of the AI / ML model and the BS 720 determines that the AI / ML model training data will be reported later, the BS 720 may send a permission flag to indicate whether the UE 710 is allowed to send the AI / ML model training data. For example, if the permission flag is set to "1", the UE 710 is allowed to report the AI / ML model training data. Otherwise, the UE 710 is not allowed to report the AI / ML model training data. In some embodiments, the BS 720 may also indicate the dynamic uplink transmission resources to be used for sending / reporting the AI / ML model training data. In some embodiments, the transmission resources for sending / reporting the AI / ML model training data are pre-configured.

[0217] As described above, in the second way of reporting the AI / ML model training data, it can be determined by the BS 720 whether the corresponding AI / ML model training data will be reported. This determination is made using the DSI of the corresponding AI / ML model training data (e.g., the data uncertainty of the corresponding AI / ML model training data) and is indicated to the UE 710 using, for example, a permission flag.

[0218] In some embodiments, the BS 720 may update the AI / ML model training assistance information during the AI / ML model training process. In some cases, after the UE determines the DSI of the corresponding AI / ML model training data, the BS (e.g., the BS 720) may send the updated AI / ML model training assistance information to the UE (e.g., the UE 710). The updated AI / ML model training assistance information may include at least one of the following: information about the updated reference AI / ML model (e.g., updated reference AI / ML model parameters), updated reference input data values, or updated DSI thresholds (e.g., updated uncertainty thresholds). In this way, the evolution of the AI / ML model can be appropriately adapted.

[0219] To reduce transmission overhead, only the updated reference AI / ML model parameters can be sent to the UE, especially when only some of the reference AI / ML model parameters have changed. In some embodiments, if most of the AI / ML model has changed (e.g., the data uncertainty of most of the AI / ML model training data is greater than a preconfigured uncertainty threshold), the entire reference AI / ML model can be sent to the UE.

[0220] In some embodiments, the BS (e.g., BS 720) can use some information received from the UE (e.g., UE 710) to determine whether the corresponding AI / ML model training data will be sent to the BS. For example, the UE can send AI / ML model training dataset information to the BS, and the AI / ML model training dataset information can include at least one of an AI / ML model training dataset and DSI distribution information of the AI / ML model training dataset. The DSI distribution information of the AI / ML model training dataset can include a cumulative distribution function (CDF) and / or a probability density function (PDF). The AI / ML model training dataset information from each UE can be sent together with the corresponding DSI (e.g., data uncertainty level) of the corresponding AI / ML model training data. After receiving at least one of the DSI of the corresponding AI / ML model training data or the AI / ML model training dataset information, the BS can determine whether to report the corresponding AI / ML model training data. The BS can send information (e.g., a permission flag) indicating whether the corresponding AI / ML model training data will be sent to the UE.

[0221] In some embodiments, a buffer status report (BSR) can be used to send AI / ML model training dataset information. For example, the BSR can carry the AI / ML model training dataset information by adding one or more additional fields thereto. To indicate the CDF or PDF, the values of the corresponding AI / ML model training data can be quantized into N data levels (N is a positive integer). The probabilities of the N data levels can be indicated according to the increasing order of the N data levels.

[0222] In some embodiments, the transmission of the AI / ML model training dataset information can be (implicitly) associated with a scheduling request (SR). Among them, the relationship between the SR resource and the AI / ML model training dataset information can be configured or preconfigured by the BS. In this way, when the BS receives an SR on the corresponding resource, it can obtain the AI / ML model training dataset information.

[0223] Unilateral AI / ML Model Training at UE

[0224] In some embodiments, AI / ML model training and AI / ML inference can be performed at the UE. For the AI / ML model training and AI / ML inference processes, the UE can receive AI / ML model training data from the BS. The AI / ML model training data can be generated by the BS. Such AI / ML model training can also be unilateral AI / ML model training because the training is only performed on the UE side.

[0225] Figure 10 An example of performing unilateral AI / ML model training at the UE 710 in a wireless network 700 according to an embodiment of the present invention is shown. Referring to Figure 10 , in network 700, there are a UE 710 and a BS 720 that are communicatively and operatively connected to each other. The BS 720 can send AI / ML model training data to the UE 710 based on the importance of the corresponding AI / ML model training data. In some embodiments, the BS 720 can send the AI / ML model training data to the UE 710 in a selective manner based on the importance of the corresponding AI / ML model training data. The importance of the corresponding AI / ML model training data can be determined based on the DSI of the corresponding AI / ML model training data (e.g., the data uncertainty of the corresponding AI / ML model training data), as shown above or elsewhere in the present invention.

[0226] Compared with the example shown in Figure 7 , in the example shown in Figure 10 , both AI / ML model training and AI / ML inference are performed at the UE 710. The UE 710 can send AI / ML model training assistance information to the BS 720 and receive AI / ML model training data from the BS 720. For example, the BS 720 can use the sounding reference signal sent from the UE 710 to the BS 720 to obtain uplink (UL) channel information that cannot be directly obtained at the UE 710. When an AI / ML model is deployed at the UE 710, the BS 720 can send the obtained UL channel information to the UE 710 as AI / ML model training data.

[0227] In some embodiments of performing unilateral AI / ML model training at a UE (e.g., UE 710), a BS (e.g., BS 720) may determine the DSI (e.g., data uncertainty) of the corresponding AI / ML model training data. When generating the corresponding AI / ML model training data on the BS side, the UE may send relevant AI / ML model training assistance information to the BS using RRC, MAC-CE, DCI, broadcast signaling, unicast signaling, and / or multicast signaling, etc. The AI / ML model training assistance information may include at least one of information about a reference AI / ML model or at least one reference input data value. The BS may determine the DSI of the corresponding AI / ML model training data and send the AI / ML model training data to the UE based on the DSI of the corresponding AI / ML model training data.

[0228] One or more methods for determining the DSI of the corresponding AI / ML model may be substantially the same as the methods in embodiments of performing unilateral AI / ML model training at a BS (e.g., the examples shown above and Figure 9A and Figure 9B ), except that the roles of the BS and the UE are swapped.

[0229] In some embodiments, the BS may selectively send the corresponding AI / ML model training data. For example, the BS may determine whether to report or send the corresponding AI / ML model training data based on a DSI threshold (e.g., uncertainty threshold) received from the UE and the DSI (e.g., data uncertainty) of the corresponding AI / ML model training data. The manner in which the BS determines whether to send the AI / ML model training data to the UE may be substantially similar to the manner in which the UE makes the determination in embodiments of performing unilateral AI / ML model training at a BS (e.g., the examples shown above and Figure 9A and Figure 9B ), except that the roles of the BS and the UE are swapped.

[0230] In some embodiments, the BS may determine whether to report or send the corresponding AI / ML model training data based on information related to the transmission of the corresponding AI / ML model training data received from the UE. The BS may send at least one of the DSI of the corresponding AI / ML model training data or the AI / ML model training dataset information. Then, the UE may determine whether the corresponding AI / ML model training data will be sent to the UE and then send information indicating whether the corresponding AI / ML model training data should be sent to the UE to the BS.

[0231] In some embodiments, e.g., when the UE needs some AI / ML model training data from the BS, the UE may send a request for the AI / ML model training data. After receiving the AI / ML model training data request, the BS may send the AI / ML model training data as shown above or elsewhere in the present invention.

[0232] One or more methods of sending the corresponding AI / ML model training data may be substantially the same as the methods of the embodiments of performing unilateral AI / ML model training at the BS (e.g., the examples shown above and Figure 9A and Figure 9B ), except that the roles of the BS and the UE are swapped. Figure 11A and Figure 11B show examples that illustrate an exemplary process of reporting the corresponding AI / ML model training data from the BS 720 to the UE 710 in the wireless network 700.

[0233] AI / ML Model Training in Federated Learning

[0234] According to some embodiments, federated learning techniques may be used to perform AI / ML model training and AI / ML inference. Federated learning, also known as collaborative learning, is a machine learning technique for training algorithms on multiple decentralized edge devices or servers. Each decentralized edge device or server stores local data samples but cannot exchange them with other devices or servers. Federated learning techniques are contrary to traditional centralized machine learning techniques, in which local data samples are not shared, while in traditional centralized machine learning techniques, all local data sets are uploaded to a single server.

[0235] In a wireless federated learning (FL-based) AI training process, a network node / device / node initializes a global AI model, samples a set of user devices, and broadcasts the global AI model parameters to the user devices. Then, each user device initializes its local AI model using the global AI model parameters and updates (trains) its local AI model using its own data. Then, each user device may report the parameters of its updated local AI model to the network device, and then the network device aggregates the updated parameters reported by the user devices and updates the global AI model. The above process is one iteration of a traditional FL-based AI model training process. The network device and the participating user devices typically perform multiple iterations until the AI model has sufficiently converged to meet one or more training objectives / criteria and the AI model is finally determined.

[0236] Figure 12AAn example of the process of performing AI / ML model training in federated learning in a wireless network 700 is shown. During federated learning, the AI / ML model training can be performed collaboratively or jointly by multiple client devices (e.g., UEs) and a central server device (e.g., BS). For example, federated learning can be performed as described below and Figure 12A as shown. As Figure 12A shown, each UE 710 in the UEs 710 can train its local AI / ML model using local AI / ML training data. When the training is completed, each UE 710 can update the local gradient associated with its local AI / ML model. Then, each UE 710 can send (e.g., upload to the server) the updated local gradient to the BS 720. After receiving the local gradients from each UE 710, the BS 720 can aggregate all the received local gradients and generate one or more global gradients associated with the global AI / ML model. The BS 720 can send (e.g., download to each UE) one or more global gradients to each UE 710. The above process can be repeated until one or more global AI / ML models converge.

[0237] During the federated learning process, there may be frequent data exchanges (e.g., transmission of local gradients and global gradients) between the client (e.g., UE 710) and the server (e.g., BS 720). This may result in additional transmission overhead.

[0238] To keep the transmission overhead within an acceptable level, the transmission of gradients used for AI / ML model training should be controlled. To control the transmission of gradients, the client device (e.g., UE 710, BS 720) can send the local gradient associated with the local AI / ML model to the central server based on the DSI of the corresponding AI / ML model training data. Here, the corresponding AI / ML model training data can include at least one of the local AI / ML model training data of the local AI / ML model of the client device or the local gradient associated with the local AI / ML model.

[0239] For example, the UE 710 may determine the DSI (e.g., data uncertainty) of the local AI / ML model training data of its local AI / ML model. If the DSI of the local AI / ML model training data is less than the DSI threshold (e.g., the data uncertainty of the local AI / ML model training data is less than the uncertainty threshold, which indicates that the local AI / ML model is relatively stable), the UE 710 may not send the local gradient associated with the local AI / ML model (and / or the local AI / ML model training data) because the local gradient will not contribute to the AI / ML model training (e.g., the convergence of the global AI / ML model). In these cases, the UE 710 may not send (e.g., upload) the local gradient associated with the local AI / ML model of the UE 710 in the current iteration.

[0240] Figure 12B An example of a process for training an AI / ML model using data state information (DSI) of AI / ML model training data in federated learning in the wireless network 700 is shown.

[0241] As Figure 12B shown, the UEs 710a and 710b may train their local AI / ML models using their local AI / ML training data. When the training is complete, each of the UEs 710a and 710b may update the local gradient associated with its local AI / ML model. Each of the UEs 710a and 710b may determine the DSI of the corresponding AI / ML model in the manner shown above or elsewhere in the present invention. In Figure 12B this case, the DSI is data uncertainty, and the data uncertainty of the local AI / ML model training data of the UE 710a is greater than the configured or pre-configured uncertainty threshold. Therefore, the UE 710a may send the local gradient associated with its local AI / ML model to the BS 720. On the other hand, the data uncertainty of the local AI / ML model training data of the UE 710b is less than the configured or pre-configured uncertainty threshold. Therefore, the UE 710b may be prohibited from sending the local gradient associated with its local AI / ML model to the BS 720. Optionally, the UE 710b may alternatively send an indication to notify the BS 720 that there is no local AI / ML model update in this iteration. The remaining process may be substantially similar to the process shown above and Figure 12A in.

[0242] Determining the DSI of the corresponding AI / ML model training data (e.g., local AI / ML model training data) during federated learning may be performed in the same manner as above (e.g., in the embodiment of performing unilateral AI / ML model training at the BS) or elsewhere in the present invention.

[0243] Dual AI / ML Model Training

[0244] According to some embodiments, AI / ML model training and AI / ML inference can be performed at both the UE and the BS. In other words, the BS and the UE can perform AI / ML model training and AI / ML inference processes collaboratively. An example of a dual AI / ML model is an autoencoder for channel state information (CSI) compression, as Figure 13 shown.

[0245] Refer to Figure 13 , raw channel data 1311 (e.g., raw channel state information (CSI)) can be generated at the UE 1310. The raw channel data 1311 can be transmitted to the CSI encoder 1312 to obtain compressed channel data 1313 (e.g., compressed CSI). Due to the compression process, some or all of the compressed channel data 1313 can be sent to the BS 1320 with a lower signaling overhead. The BS 1320 can use the decoder 1322 to reconstruct the raw channel data (i.e., the reconstructed channel data 1323) after receiving the compressed channel data 1321 (e.g., compressed CSI). The compressed channel data 1321 input to the decoder 1322 can be the same as the compressed channel data 1313 output from the encoder 1312.

[0246] If the UE 1310 trains the encoder 1312 and the BS 1320 trains the decoder 1322 (i.e., separate training), the UE 1310 can use the reference decoder 1314 to reconstruct the raw channel data and thus generate the reconstructed channel data 1315. In this way, divergence between the output of the encoder 1312 (i.e., the compressed channel data 1313) and the input of the decoder 1322 (i.e., the compressed channel data 1321) can be prevented. The reference decoder 1314 can be configured by the BS 1320 or predefined. In some embodiments, information related to the reference decoder 1314 can be transmitted to the UE 1310. In some embodiments, information related to the reference decoder 1314 or the reference decoder 1314 can be regarded as AI / ML model training assistance information.

[0247] After generating the reconstructed channel data 1315, the UE 1310 can include the output of the encoder 1312 (i.e., the compressed channel data 1313 or V mid ) and the output of the reference decoder 1314 (i.e., the reconstructed channel data 1315 or V out)'s dataset is sent to BS1320. BS1320 can train the decoder 1322 with the received dataset, where V mid and V out are used as labeled data for training the decoder 1322 at BS1320.

[0248] In some embodiments, to support faster AI / ML model training (e.g., faster convergence of the AI / ML model) and avoid additional signaling overhead, when enhancing bilateral AI / ML model training, the DSI (e.g., data uncertainty) of the corresponding AI / ML model training data can be used as a constraint factor. For example, after training is completed at the encoder 1312 and the reference decoder 1314, the UE 1310 can determine the data uncertainty of the dataset including the output of the encoder and the output of the reference decoder. If the data uncertainty level of the dataset is high, e.g., greater than a preconfigured uncertainty threshold, then the UE1310 can send a dataset including the output of the encoder 1312 (i.e., the compressed channel data 1313 or V mid ) and the reconstructed channel data 1315 (i.e., V out ) to BS1320. Conversely, if the data uncertainty level of the dataset is low, e.g., less than a preconfigured uncertainty threshold, then the UE 1310 may not be allowed to send the dataset to BS1320. The UE 1310 can update one or more parameters of the encoder 1312 to obtain a new output (i.e., a new compressed channel data 1313 or a new V mid ) and re-evaluate the data uncertainty of the dataset until the data uncertainty meets the requirements (e.g., greater than a preconfigured uncertainty threshold).

[0249] For embodiments of bilateral AI / ML model training, the determination of the DSI of the corresponding AI / ML model training data (e.g., local AI / ML model training data) can be performed in the same manner as above (e.g., in embodiments of unilateral AI / ML model training at the BS) or elsewhere in the present invention.

[0250] AI / ML Model Training Data Transmission Based on Report Format

[0251] As described above, in some embodiments, corresponding AI / ML model training data can be selectively sent (e.g., some AI / ML model training data is sent while other AI / ML model training data is not). For example, corresponding AI / ML model training data can only be sent when the data uncertainty of the AI / ML model training data is greater than a preconfigured uncertainty threshold. Compared with these embodiments, in some other embodiments, all corresponding AI / ML model training data can be sent regardless of the DSI (e.g., data uncertainty) of the corresponding AI / ML model training data. By sending all AI / ML model training data, more information will be provided for AI / ML model training, thus making it possible to achieve faster convergence of the AI / ML model. However, since less important AI / ML model training data (e.g., AI / ML model training data with low data uncertainty) is transmitted, this data transmission will result in higher transmission overhead.

[0252] To achieve a balance between improving the performance of the AI / ML model and reducing the transmission overhead, AI / ML model training data can be sent according to the corresponding reporting format. In some embodiments, the corresponding reporting format can indicate the corresponding transmission accuracy, and the level of the corresponding transmission accuracy can be determined based on the level of the DSI of the corresponding AI / ML model training data. In some embodiments, the corresponding reporting format can indicate the configuration for sending the corresponding AI / ML model training data.

[0253] As described above, in some embodiments, the corresponding reporting format can indicate the corresponding transmission accuracy. In other words, AI / ML model training data can be sent based on the corresponding transmission accuracy indicated in the corresponding reporting format. If some AI / ML model training data has high data uncertainty (i.e., the AI / ML model training data will make a greater contribution to AI / ML model training), the sending of the AI / ML model training data can be performed with high accuracy to ensure the successful sending of the data. On the other hand, if some AI / ML model training data has low data uncertainty (i.e., the information provided by the AI / ML model training data is limited and thus makes little contribution to AI / ML model training), the sending of the AI / ML model training data can be performed with low accuracy to reduce the transmission overhead.

[0254] In some embodiments, when the corresponding transmission accuracy is greater than a predefined value (e.g., high accuracy), all the corresponding AI / ML model training data is sent. In other words, all the AI / ML model training data can be sent, or the original AI / ML model training data can be sent without preprocessing. For example, if the AI / ML model training data is channel information, when the transmission accuracy is high, the complete channel information including all real and imaginary values can be sent. On the other hand, when the corresponding transmission accuracy is less than the predefined value (e.g., low accuracy), only a part of the corresponding AI / ML model training data or the information extracted from the corresponding AI / ML model training data is sent. For example, if the AI / ML model training data is channel information, when the transmission accuracy is low, traditional CSI such as channel quality information (CQI), rank indicator (RI), layer indicator (LI), reference signal received power (RSRP), precoding matrix indicator (PMI), etc. can be sent instead of the complete channel information.

[0255] In some embodiments, the corresponding report format can indicate whether the corresponding AI / ML model training data includes channel state information (CSI) or the original channel information. For the original channel information, the corresponding report format can indicate, for example, whether to send the channel information of 6 subcarriers in a resource block (RB) or the channel information of 3 subcarriers in the RB.

[0256] As described above, in some embodiments, the corresponding report format can indicate the configuration for the transmission of the corresponding AI / ML model training data. In other words, the AI / ML model training data can be sent based on the configuration for the transmission of the corresponding AI / ML model training data indicated in the corresponding report format. The configuration for the transmission of the corresponding AI / ML model training data can indicate at least one of the resources for the transmission of the corresponding AI / ML model training data or the quantization granularity for the transmission of the corresponding AI / ML model training data.

[0257] In some embodiments, when the corresponding transmission accuracy is greater than a predefined value (e.g., high accuracy), compared with when the corresponding transmission accuracy is less than the predefined value (e.g., low accuracy), a larger number of subcarriers per resource block (RB) or a larger number of bits per RB can be used to transmit the corresponding AI / ML model training data. In other words, when the transmission accuracy is high, more transmission resources can be allocated for the transmission of the corresponding AI / ML model training data or a lower MCS value can be used (i.e., a larger number of subcarriers, RBs, subbands, symbols, and / or mini-slots are used for transmission). On the other hand, when the transmission accuracy is low, fewer transmission resources can be allocated for the transmission of the corresponding AI / ML model training data or a higher MCS value can be used (i.e., a smaller number of subcarriers, RBs, subbands, symbols, and / or mini-slots are used for transmission). For example, due to high transmission accuracy, 6 subcarriers per RB can be used to transmit AI / ML model training data, and the resource utilization density is 1 / 2. However, due to low transmission accuracy, only 2 subcarriers per RB can be used to transmit AI / ML model training data, and the resource utilization density is 1 / 6.

[0258] In some embodiments, when the corresponding transmission accuracy is greater than a predefined value (e.g., high accuracy), a fine quantization granularity (e.g., using more bits to represent the original data) can be used to transmit the corresponding AI / ML model training data. On the other hand, when the corresponding transmission accuracy is less than the predefined value (e.g., low accuracy), a coarse quantization granularity (e.g., using fewer bits to represent the original data) can be used to transmit the corresponding AI / ML model training data. For example, when the transmission accuracy is high, 8 bits can be used to represent a complex value for the transmission of AI / ML model training data (e.g., 4 bits for the real part and 4 bits for the imaginary part). When the transmission accuracy is low, only 4 bits can be used to represent a complex value for the transmission of AI / ML model training data (e.g., 2 bits for the real part and 2 bits for the imaginary part).

[0259] In some embodiments, when the corresponding transmission accuracy is greater than a predefined value (e.g., high accuracy), more transmission resources (e.g., a larger number of subcarriers per RB or a smaller number of bits per RB) and / or a fine quantization granularity can be used to transmit all of the AI / ML model training data. On the other hand, when the corresponding transmission accuracy is less than the predefined value (e.g., low accuracy), only a portion of the corresponding AI / ML model training data or the information extracted from the corresponding AI / ML model training data can be transmitted using fewer transmission resources (e.g., a smaller number of subcarriers per RB or a larger number of bits per RB) and / or a coarse quantization granularity.

[0260] In some embodiments, the relationship between the DSI of the corresponding AI / ML model training data and the corresponding report format can be configured by the BS or device that performs AI / ML training (e.g., the UE that performs unilateral AI / ML model training).

[0261] For example, the BS can configure the relationship between data uncertainty and the corresponding data transmission accuracy for the AI / ML model training data. The BS can configure a mapping table of data uncertainty and the corresponding data transmission accuracy for the AI / ML model training data, as shown in Tables 1 to 4 below. Tables 1 to 4 show how each data uncertainty level is mapped to a data transmission accuracy level. In Tables 1 to 4, the range of the data uncertainty level is from 1 to 8.

[0262] Table 1 shows the mapping between the data uncertainty level and the data transmission accuracy level (represented by all / partial data).

[0263] Data uncertainty level Data transmission accuracy level (represented by all / partial data) 1～2 Extract 20% of all data 3～4 Extract 40% of all data 5～6 Extract 50% of all data 7～8 All data

[0264] Table 1

[0265] Table 2 shows the mapping between the data uncertainty level and the data transmission accuracy level (represented by resource granularity).

[0266] Data uncertainty level Data transmission accuracy level (represented by resource granularity) 1～2 2 subcarriers 3～4 4 subcarriers 5～6 8 subcarriers 7～8 12 subcarriers

[0267] Table 2

[0268] Table 3 shows the mapping between the data uncertainty level and the data transmission accuracy level (represented by quantization).

[0269] Data uncertainty level Data transmission accuracy level (represented by quantization) 1～2 2 bits 3～4 4 bits 5～6 6 bits 7～8 8 bits

[0270] Table 3

[0271] Table 4 is a combination of Tables 1, 2, and 3, and shows the mapping between the data uncertainty level and the data transmission accuracy level through all / partial data, resource granularity, and quantization.

[0272]

[0273] Table 4

[0274] Figure 14 Shows an example of AI / ML model training with data transmission accuracy adaptation according to an embodiment of the present invention. Figure 14 The shown AI / ML model 1400 can be implemented using a DNN. In other words, the type of the AI / ML model 1400 can be a DNN. The AI / ML model input data dimension (i.e., the dimension of the input data of the AI / ML model 1400) is M, so there are M input data (i.e., Input1 To Input M )。The data dimension of the AI / ML model output (i.e., the dimension of the output data of the AI / ML model 1400) is 1, so there is 1 output data (i.e., the optimal MCS at time slot n + k). The AI / ML model 1400 may include L hidden layers (i.e., the number of hidden layers of the AI / ML model 1400 is L), and each hidden layer includes K neurons (i.e., the number of neurons in each hidden layer is K). The goal of the AI / ML model 1400 is to use the input of the historical channel information at time slot n (e.g., Input i ) to predict the optimal MCS at time slot n + k. To provide more information to assist in AI / ML model training, it may be allowed to send all historical channel information. However, to reduce transmission overhead, the corresponding AI / ML model training data may be sent based on the corresponding transmission accuracy indicated in the corresponding reporting format. In other words, different transmission accuracies may be applied to the transmission of the corresponding AI / ML model training data to reduce transmission overhead.

[0275] For embodiments of bilateral AI / ML model training, the determination of the DSI of the corresponding AI / ML model training data (e.g., local AI / ML model training data) may be performed in the same manner as described above (e.g., in embodiments of unilateral AI / ML model training at the BS) or elsewhere in the present invention.

[0276] Figure 15 is a flowchart of an exemplary process for performing AI / ML model training in a wireless communication network according to an embodiment of the present invention. Refer to Figure 15 , in some embodiments, the first device may be a UE and the second device may be a BS. In other embodiments, the first device may be a BS and the second device may be a UE. In other embodiments, the first device and the second device may be UEs. In other embodiments, the first device and the second device may be BSs.

[0277] In step 1510, the first device may receive AI / ML model training assistance information and information related to the transmission of corresponding AI / ML model training data from the second device. In some embodiments, for example, the AI / ML model training assistance information and the information related to the transmission of corresponding AI / ML model training data are sent together using one DCI message or sidelink control information (SCI) message. In some embodiments, the AI / ML model training assistance information and the information related to the transmission of corresponding AI / ML model training data are sent separately. For example, the AI / ML model training assistance information is carried in one DCI or SCI message, and the information related to the transmission of corresponding AI / ML model training data is carried in another DCI or SCI message.

[0278] The AI / ML model training assistance information may include at least one of information about a reference AI / ML model or at least one reference input data value. The information about the reference AI / ML model may include at least one of the following: reference AI / ML model type, reference AI / ML model structure, one or more reference AI / ML model parameters, reference AI / ML model gradient, reference AI / ML model activation function, reference AI / ML model input data type, reference AI / ML model output data type, reference AI / ML model input data dimension, or reference AI / ML model output data dimension. In some embodiments, the AI / ML model training assistance information may be updated by the second device. In these cases, the second device may send the updated AI / ML model training assistance information to the first device.

[0279] The information related to the transmission of corresponding AI / ML model training data may include a DSI threshold. The second device may configure the DSI threshold for determining whether the corresponding AI / ML model training data will be sent to the second device.

[0280] In some embodiments where the first device and the second device cooperate for AI / ML model training, the first device may perform partial AI / ML model training before determining the DSI of the corresponding AI / ML model training data in step 1520, and the corresponding AI / ML model training data may include the corresponding output of the partial AI / ML model training.

[0281] In step 1520, the first device may determine the DSI of the corresponding AI / ML model training data based on the AI / ML model training assistance information.

[0282] In some embodiments, the DSI of the corresponding AI / ML model training data may include information indicating at least one of the following: data uncertainty of the corresponding AI / ML model training data; data importance of the corresponding AI / ML model training data; degree of need for the corresponding AI / ML model training data for AI / ML model training; or data diversity of the corresponding AI / ML model training data. The data uncertainty of the corresponding AI / ML model training data may be determined based on at least one of the following: entropy, minimum confidence, marginal sampling, or generalization error.

[0283] In some embodiments, determining the DSI of the corresponding AI / ML model training data may include inputting the corresponding AI / ML model training data into a reference AI / ML model and determining the DSI based on the output of the reference AI / ML model. The corresponding AI / ML model training data input into the reference AI / ML model may replace at least one reference input data value.

[0284] In step 1530, the first device may determine whether the corresponding AI / ML model training data will be sent to the second device based on the DSI threshold and the DSI of the corresponding AI / ML model training data.

[0285] In step 1540, the first device may send the corresponding AI / ML model training data to the second device based on at least one of the DSI of the corresponding AI / ML model training data or information related to the transmission of the corresponding AI / ML model training data.

[0286] In some embodiments, the corresponding AI / ML model training data may be sent according to a corresponding report format, and the corresponding report format is determined based on at least one of the DSI of the corresponding AI / ML model training data or information related to the transmission of the corresponding AI / ML model training data. In some embodiments, the corresponding report format may indicate the corresponding transmission accuracy, and the level of the corresponding transmission accuracy may be determined based on the level of the DSI of the corresponding AI / ML model training data. In some embodiments, the corresponding report format may indicate the configuration for the transmission of the corresponding AI / ML model training data. The configuration for the transmission of the corresponding AI / ML model training data may indicate at least one of the resources for the transmission of the corresponding AI / ML model training data or the quantization granularity for the transmission of the corresponding AI / ML model training data. In some embodiments, the corresponding report format may indicate whether the corresponding AI / ML model training data includes channel state information (CSI) or raw channel information. In some embodiments, the second device may configure the relationship between the DSI of the corresponding AI / ML model training data and the corresponding report format.

[0287] In some embodiments, the corresponding AI / ML model training data may include at least one of the local AI / ML model training data of the local AI / ML model of the first device or the local gradients associated with the local AI / ML model.

[0288] In step 1550, the second device may perform AI / ML model training using the corresponding AI / ML model training data. As described above, in some embodiments, the AI / ML model training is also partially performed by the first device.

[0289] Figure 16 is a flowchart of another exemplary process for performing AI / ML model training in a wireless communication network according to an embodiment of the present invention. Refer to Figure 16 , in some embodiments, the first device may be a UE and the second device may be a BS. In other embodiments, the first device may be a BS and the second device may be a UE. In other embodiments, the first device and the second device may be UEs. In other embodiments, the first device and the second device may be BSs.

[0290] In step 1610, the first device may receive AI / ML model training assistance information from the second device. The AI / ML model training assistance information may include at least one of information about a reference AI / ML model or at least one reference input data value. The information about the reference AI / ML model may include at least one of the following: reference AI / ML model type, reference AI / ML model structure, one or more reference AI / ML model parameters, reference AI / ML model gradients, reference AI / ML model activation functions, reference AI / ML model input data types, reference AI / ML model output data types, reference AI / ML model input data dimensions, or reference AI / ML model output data dimensions. In some embodiments, the AI / ML model training assistance information may be updated by the second device. In these cases, the second device may send the updated AI / ML model training assistance information to the first device.

[0291] In some embodiments where the first device and the second device cooperate to perform AI / ML model training, before determining the DSI of the corresponding AI / ML model training data in step 1620, the first device may perform partial AI / ML model training, and the corresponding AI / ML model training data may include the corresponding outputs of the partial AI / ML model training.

[0292] In step 1620, the first device may determine the DSI of the corresponding AI / ML model training data based on the AI / ML model training assistance information.

[0293] In some embodiments, the DSI of the corresponding AI / ML model training data may include information indicating at least one of the following: data uncertainty of the corresponding AI / ML model training data; data importance of the corresponding AI / ML model training data; degree of need of the corresponding AI / ML model training data for AI / ML model training; or data diversity of the corresponding AI / ML model training data. The data uncertainty of the corresponding AI / ML model training data may be determined based on at least one of the following: entropy, minimum confidence, marginal sampling, or generalization error.

[0294] In some embodiments, determining the DSI of the corresponding AI / ML model training data may include inputting the corresponding AI / ML model training data into a reference AI / ML model and determining the DSI based on the output of the reference AI / ML model. The corresponding AI / ML model training data input into the reference AI / ML model may replace at least one reference input data value.

[0295] In step 1630, the first device may send at least one of the DSI of the corresponding AI / ML model training data or the AI / ML model training dataset information to the second device. In some embodiments, the AI / ML model training dataset information may include at least one of the AI / ML model training dataset size or the DSI distribution information of the AI / ML model training dataset. In some embodiments, the buffer status report (BSR) or scheduling request (SR) may be used to send the AI / ML model training dataset information. Step 1630 may be an optional step.

[0296] In step 1640, the second device may determine whether the corresponding AI / ML model training data will be sent to the second device using at least one of the DSI of the corresponding AI / ML model training data or the AI / ML model training dataset information. Step 1640 may be an optional step.

[0297] In step 1650, the second device may send information related to the transmission of the corresponding AI / ML model training data to the first device. The information related to the transmission of the corresponding AI / ML model training data may include information indicating whether the corresponding AI / ML model training data will be sent to the second device. Thus, in step 1650, the second device may send information indicating whether the corresponding AI / ML model training data will be sent to the second device to the first device.

[0298] In step 1660, the first device may send the corresponding AI / ML model training data to the second device based on at least one of the DSI of the corresponding AI / ML model training data or the information related to the transmission of the corresponding AI / ML model training data.

[0299] In some embodiments, the corresponding AI / ML model training data may be sent according to a corresponding report format, and the corresponding report format is determined based on at least one of the DSI of the corresponding AI / ML model training data or the information related to the transmission of the corresponding AI / ML model training data. In some embodiments, the corresponding report format may indicate the corresponding transmission accuracy, and the level of the corresponding transmission accuracy may be determined based on the level of the DSI of the corresponding AI / ML model training data. In some embodiments, the corresponding report format may indicate the configuration for the transmission of the corresponding AI / ML model training data. The configuration for the transmission of the corresponding AI / ML model training data may indicate at least one of the resources for the transmission of the corresponding AI / ML model training data or the quantization granularity for the transmission of the corresponding AI / ML model training data. In some embodiments, the corresponding report format may indicate whether the corresponding AI / ML model training data includes channel state information (CSI) or raw channel information. In some embodiments, the second device may configure the relationship between the DSI of the corresponding AI / ML model training data and the corresponding report format.

[0300] In some embodiments, the corresponding AI / ML model training data may include at least one of the local AI / ML model training data of the local AI / ML model of the first device or the local gradient associated with the local AI / ML model.

[0301] In step 1670, the second device may perform AI / ML model training using the corresponding AI / ML model training data. As described above, in some embodiments, the AI / ML model training is also partially performed by the first device.

[0302] The embodiments described above are described in the context of communication between a UE and a BS. However, more generally, devices that communicate wirelessly with each other on time-frequency resources do not necessarily have to be one or more UEs communicating with a BS. For example, two or more UEs can communicate wirelessly with each other via a sidelink using device-to-device (D2D) communication. As another example, two network devices (e.g., a terrestrial base station and a non-terrestrial base station, such as a drone) can communicate wirelessly with each other via a backhaul link. The embodiments are not limited to uplink communication and / or downlink communication. For example, in the above embodiments, the BS can be replaced by another device (e.g., a node or a UE in the network). Uplink / downlink communication can be changed to sidelink communication. Thus, as described above, the first device can be a UE or a network device (e.g., a BS), and the second device can be a UE or a network device (e.g., a BS).

[0303] Through certain aspects of the present invention, the performance of the AI / ML model is enhanced, and overfitting during the AI / ML model training process can be avoided. In addition, since fewer AI / ML model training data samples are transmitted based on data state information (DSI) of the AI / ML model training data, the transmission overhead (e.g., air interface overhead) can be reduced. The DSI (e.g., data uncertainty) can be measured at the device (e.g., a UE, a BS) before the device reports or transmits the AI / ML model training data.

[0304] Through some aspects of the present invention, for example, in federated learning, the transmission of AI / ML model training data (e.g., local gradients) that does not contribute to the convergence of the global AI / ML model is avoided. In this way, the signaling overhead in federated learning can be reduced, the performance of the AI / ML model can be improved, and the AI / ML model training can be enhanced.

[0305] Through some aspects of the present invention, fast convergence of the AI / ML model can be achieved in bilateral AI / ML model training. Since the number of transmissions of the AI / ML model training data set is reduced, additional signaling overhead can be avoided.

[0306] Through some aspects of the present invention, the enhanced performance of the AI / ML model can be balanced and the transmission overhead can be reduced.

[0307] Examples of devices (e.g., an ED or a UE and a TRP or a network device) for performing the various methods described herein are also disclosed.

[0308] For example, the (first) device may include a memory for storing processor-executable instructions and a processor for executing the processor-executable instructions. When the processor executes the processor-executable instructions, the processor can be caused to perform as described herein with respect to, for example Figures 7 to 16Method steps of one or more of the described devices. For example, the processor can enable the device to communicate over the air interface in an operating mode by implementing operations consistent with the operating mode (e.g., performing necessary measurements and generating content from those measurements, such as configured for the operating mode), preparing uplink transmissions, and processing downlink transmissions (e.g., encoding, decoding, etc.), and configuring and / or indicating transmit / receive on one or more RF chains and one or more antennas.

[0309] It should be noted that the expression "at least one of A or B" as used herein can be interchanged with the expression "A and / or B". It refers to a list in which A, or B, or both A and B can be selected. Similarly, "at least one of A, B, or C" as used herein can be interchanged with "A and / or B and / or C" or "A, B, and / or C". It refers to a list in which A or B or C, or both A and B, or both A and C, or both B and C, or all of A, B, and C can be selected. The same principle applies to longer lists of the same format.

[0310] Although the present invention has been described with reference to specific features and embodiments of the present invention, various modifications and combinations can be made without departing from the scope of the present invention. Therefore, the specification and the drawings are to be regarded only as illustrative of some embodiments of the present invention defined by the appended claims, and are contemplated to cover any and all modifications, variations, combinations, or equivalents within the scope of the present invention. Although the present invention and its advantages have been described in detail, various changes, substitutions, and alterations can be made without departing from the present invention defined by the appended claims. In addition, the scope of the present application is not limited to the specific embodiments of the processes, machines, manufactures, compositions of matter, components, methods, and steps described in the specification. From the disclosure of the present invention, those of ordinary skill in the art will readily understand that processes, machines, manufactures, compositions of matter, components, methods, or steps (including those existing currently or developed later) that perform substantially the same functions or achieve substantially the same results as the corresponding embodiments described herein can be used in accordance with the present invention. Therefore, the appended claims are intended to cover such processes, machines, manufactures, compositions of matter, components, methods, or steps within their scope.

[0311] In addition, any module, component, or device that executes instructions illustrated in this document may include or otherwise access one or more non-transitory computer / processor-readable storage media to store information such as computer / processor-readable instructions, data structures, program modules, and / or other data. A non-exhaustive list of examples of non-transitory computer / processor-readable storage media includes magnetic tape cartridges, tapes, disk memories, or other magnetic storage devices, compact disc read-only memory (CD-ROM), digital video disc or digital versatile disc (DVD), Blu-ray Disc TM and other optical discs, or other optical storage, volatile and non-volatile, removable and non-removable media implemented in any method or technology, random-access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory, or other storage technologies. Any of these non-transitory computer / processor storage media may be part of the device or accessible or connected to the device. Any application or module described herein may be implemented using computer / processor-readable / executable instructions that may be stored or otherwise held by these non-transitory computer / processor-readable storage media.

[0312] Definition of Acronyms

[0313] AI Artificial Intelligence

[0314] LTE Long-Term Evolution

[0315] NR New Radio

[0316] BWP Bandwidth Part

[0317] BS Base Station

[0318] CA Carrier Aggregation

[0319] CC Component Carrier

[0320] CG Cell Group

[0321] CSI Channel State Information

[0322] CSI-RS Channel State Information Reference Signal

[0323] DNN Deep Neural Network

[0324] DC Dual Connectivity

[0325] Downlink Control Information (DCI)

[0326] Downlink (DL)

[0327] Downlink Shared Channel (DL-SCH)

[0328] E-UTRAN NR Dual Connectivity (EN-DC), where the Master Cell Group (MCG) uses E-UTRA and the Secondary Cell Group (SCG) uses NR

[0329] Next Generation (or 5G) Node B (gNB)

[0330] Hybrid Automatic Repeat reQuest - ACKnowledgement (HARQ-ACK)

[0331] Master Cell Group (MCG)

[0332] Modulation and Coding Scheme (MCS)

[0333] Medium Access Control - Control Element (MAC-CE)

[0334] Physical Broadcast Channel (PBCH)

[0335] Primary Cell (PCell)

[0336] Physical Downlink Control Channel (PDCCH)

[0337] Physical Downlink Shared Channel (PDSCH)

[0338] Physical Random Access Channel (PRACH)

[0339] Physical Resource Block Group (PRG)

[0340] Primary SCG Cell (PSCell)

[0341] Primary Synchronization Signal (PSS)

[0342] Physical Uplink Control Channel (PUCCH)

[0343] Physical Uplink Shared Channel (PUSCH)

[0344] Random Access Channel (RACH)

[0345] Random Access Preamble IDentifier (RAPID)

[0346] Resource Block (RB)

[0347] Resource Element (RE)

[0348] Radio Resource Management (RRM)

[0349] Remaining System Information (RMSI)

[0350] Reference Signal (RS)

[0351] Reference Signal Receiving Power (RSRP)

[0352] Radio Resource Control (RRC)

[0353] Secondary Cell Group (SCG)

[0354] Sidelink Control Information (SCI)

[0355] System Frame Number (SFN)

[0356] Sidelink (SL)

[0357] Secondary Cell (SCell)

[0358] Semi-Persistent Scheduling (SPS)

[0359] Scheduling Request (SR)

[0360] SRS Resource Indicator (SRI)

[0361] Sounding Reference Signal (SRS)

[0362] Secondary Synchronization Signal (SSS)

[0363] Synchronization Signal Block (SSB)

[0364] Supplementary Uplink (SUL)

[0365] Timing Advance (TA)

[0366] Timing Advance Group (TAG)

[0367] Target UE (TUE)

[0368] Uplink Control Information (UCI)

[0369] User Equipment (UE)

[0370] Uplink (UL)

[0371] Uplink Shared Channel (UL-SCH)

Claims

1. A method for supporting artificial intelligence or machine learning (AI / ML) model training in a wireless communication network, characterized in that, the method comprises: a first device receiving AI / ML model training assistance information from a second device; the first device determining data state information (DSI) of corresponding AI / ML model training data based on the AI / ML model training assistance information; the first device receiving information related to the transmission of the corresponding AI / ML model training data from the second device; the first device transmitting the corresponding AI / ML model training data to the second device based on at least one of the DSI of the corresponding AI / ML model training data or the information related to the transmission of the corresponding AI / ML model training data.

2. The method according to claim 1, characterized in that, the corresponding AI / ML model training data is selectively transmitted, and the information related to the transmission of the corresponding AI / ML model training data includes a DSI threshold, and the method further comprises: the first device determining whether the corresponding AI / ML model training data will be transmitted to the second device based on the DSI threshold and the DSI of the corresponding AI / ML model training data.

3. The method according to claim 1, characterized in that, the corresponding AI / ML model training data is selectively transmitted, and the information related to the transmission of the corresponding AI / ML model training data includes information indicating whether the corresponding AI / ML model training data will be transmitted to the second device, and the method further comprises: the first device transmitting at least one of the DSI of the corresponding AI / ML model training data or AI / ML model training dataset information to the second device; the first device receiving the information indicating whether the corresponding AI / ML model training data will be transmitted to the second device from the second device.

4. The method according to claim 3, characterized in that, the AI / ML model training dataset information includes at least one of the following: the size of the AI / ML model training dataset; or the DSI distribution information of the AI / ML model training dataset.

5. The method according to claim 3 or 4, characterized in that, the AI / ML model training dataset information is transmitted using a buffer status report (BSR) or a scheduling request (SR).

6. The method according to claim 1, characterized in that, The corresponding AI / ML model training data is sent according to a corresponding report format, and the corresponding report format is determined based on at least one of the DSI of the corresponding AI / ML model training data or the information related to the sending of the corresponding AI / ML model training data.

7. The method according to claim 6, wherein, the corresponding report format indicates a corresponding transmission accuracy, and the level of the corresponding transmission accuracy is determined based on the level of the DSI of the corresponding AI / ML model training data.

8. The method according to claim 6 or 7, wherein, the corresponding report format indicates a configuration for the sending of the corresponding AI / ML model training data.

9. The method according to claim 8, wherein, the configuration for the sending of the corresponding AI / ML model training data indicates at least one of the following: resources for the sending of the corresponding AI / ML model training data; or quantization granularity for the sending of the corresponding AI / ML model training data.

10. The method according to any one of claims 6 to 9, wherein, the corresponding report format indicates whether the corresponding AI / ML model training data includes channel state information (CSI) or raw channel information.

11. The method according to any one of claims 6 to 10, wherein, the relationship between the DSI of the corresponding AI / ML model training data and the corresponding report format is configured by the second device.

12. The method according to any one of claims 1 to 11, wherein, the AI / ML model training assistance information includes at least one of information about a reference AI / ML model or at least one reference input data value.

13. The method according to claim 12, wherein, the information about the reference AI / ML model includes at least one of the following: reference AI / ML model type; reference AI / ML model structure; one or more reference AI / ML model parameters; reference AI / ML model gradient; reference AI / ML model activation function; reference AI / ML model input data type; reference AI / ML model output data type; reference AI / ML model input data dimension; or reference AI / ML model output data dimension.

14. The method according to claim 12 or 13, wherein, determining the DSI of the corresponding AI / ML model training data includes: inputting the corresponding AI / ML model training data into the reference AI / ML model; determining the DSI based on the output of the reference AI / ML model.

15. The method according to claim 14, wherein, the corresponding AI / ML model training data input into the reference AI / ML model replaces the at least one reference input data value.

16. The method according to any one of claims 1 to 15, wherein, the method further comprises: the first device receiving updated AI / ML model training assistance information from the second device.

17. The method according to any one of claims 1 to 16, wherein, the DSI of the corresponding AI / ML model training data includes information indicating at least one of the following: the data uncertainty of the corresponding AI / ML model training data; the data importance of the corresponding AI / ML model training data; the degree of need of the corresponding AI / ML model training data for the AI / ML model training; or the data diversity of the corresponding AI / ML model training data.

18. The method according to claim 17, wherein, the data uncertainty of the corresponding AI / ML model training data is determined based on at least one of the following: entropy, minimum confidence, marginal sampling, or generalization error.

19. The method according to any one of claims 1 to 18, wherein, the AI / ML model training is at least partially performed by the second device.

20. The method according to any one of claims 1 to 19, wherein, the corresponding AI / ML model training data includes at least one of the local AI / ML model training data of the local AI / ML model of the first device or the local gradient associated with the local AI / ML model.

21. The method according to any one of claims 1 to 20, wherein, the first device and the second device cooperate to perform the AI / ML model training, and the method further comprises: the first device performing a part of the AI / ML model training before determining the DSI of the corresponding AI / ML model training data, wherein the corresponding AI / ML model training data includes the corresponding output of the part of the AI / ML model training.

22. A device for supporting artificial intelligence or machine learning (AI / ML) model training in a wireless communication network, wherein, the device comprises: a processor; a memory storing processor-executable instructions, which when executed cause the processor to: receive AI / ML model training assistance information from a second device; determine data state information (DSI) of corresponding AI / ML model training data based on the AI / ML model training assistance information; receive information related to the transmission of the corresponding AI / ML model training data from the second device; send the corresponding AI / ML model training data to the second device based on at least one of the DSI of the corresponding AI / ML model training data or the information related to the transmission of the corresponding AI / ML model training data.

23. The device according to claim 22, wherein, the corresponding AI / ML model training data is selectively sent, and the information related to the sending of the corresponding AI / ML model training data includes a DSI threshold, wherein the processor-executable instructions further include processor-executable instructions that, when executed, cause the processor to perform the following operations: Determine whether the corresponding AI / ML model training data will be sent to the second device based on the DSI threshold and the DSI of the corresponding AI / ML model training data.

24. The device according to claim 22, wherein, the corresponding AI / ML model training data is selectively sent, and the information related to the sending of the corresponding AI / ML model training data includes information indicating whether the corresponding AI / ML model training data will be sent to the second device, wherein the processor-executable instructions further include processor-executable instructions that, when executed, cause the processor to perform the following operations: Send at least one of the DSI of the corresponding AI / ML model training data or the AI / ML model training dataset information to the second device; Receive from the second device the information indicating whether the corresponding AI / ML model training data will be sent to the second device.

25. The device according to claim 24, wherein, the AI / ML model training dataset information includes at least one of the following: AI / ML model training dataset size; or DSI distribution information of the AI / ML model training dataset.

26. The device according to claim 24 or 25, wherein, the AI / ML model training dataset information is sent using a buffer status report (BSR) or a scheduling request (SR).

27. The device according to claim 22, wherein, the corresponding AI / ML model training data is sent according to a corresponding report format, and the corresponding report format is determined based on at least one of the DSI of the corresponding AI / ML model training data or the information related to the sending of the corresponding AI / ML model training data.

28. The device according to claim 27, wherein, the corresponding report format indicates a corresponding transmission accuracy, and the level of the corresponding transmission accuracy is determined based on the level of the DSI of the corresponding AI / ML model training data.

29. The device according to claim 27 or 28, wherein, the corresponding report format indicates a configuration for the sending of the corresponding AI / ML model training data.

30. The device according to claim 29, wherein, the configuration for the sending of the corresponding AI / ML model training data indicates at least one of the following: The transmitted resources for the corresponding AI / ML model training data; or The transmitted quantization granularity for the corresponding AI / ML model training data.

31. The apparatus according to any one of claims 27 to 30, wherein, the corresponding report format indicates whether the corresponding AI / ML model training data includes channel state information (CSI) or raw channel information.

32. The apparatus according to any one of claims 27 to 31, wherein, the relationship between the DSI of the corresponding AI / ML model training data and the corresponding report format is configured by the second device.

33. The apparatus according to any one of claims 22 to 32, wherein, the AI / ML model training assistance information includes at least one of information about a reference AI / ML model or at least one reference input data value.

34. The apparatus according to claim 33, wherein, the information about the reference AI / ML model includes at least one of the following: Reference AI / ML model type; Reference AI / ML model structure; One or more reference AI / ML model parameters; Reference AI / ML model gradient; Reference AI / ML model activation function; Reference AI / ML model input data type; Reference AI / ML model output data type; Reference AI / ML model input data dimension; or Reference AI / ML model output data dimension.

35. The apparatus according to claim 33 or 34, wherein, determining the DSI of the corresponding AI / ML model training data includes: Inputting the corresponding AI / ML model training data into the reference AI / ML model; Determining the DSI based on the output of the reference AI / ML model.

36. The apparatus according to claim 35, wherein, the corresponding AI / ML model training data input into the reference AI / ML model replaces the at least one reference input data value.

37. The apparatus according to any one of claims 22 to 36, wherein, the processor-executable instructions further include processor-executable instructions that, when executed, cause the processor to perform the following operations: Receiving updated AI / ML model training assistance information from the second device.

38. The apparatus according to any one of claims 22 to 37, wherein, the DSI of the corresponding AI / ML model training data includes information indicating at least one of the following: Data uncertainty of the corresponding AI / ML model training data; Data importance of the corresponding AI / ML model training data; Degree of need of the corresponding AI / ML model training data for the AI / ML model training; or Data diversity of the corresponding AI / ML model training data.

39. The apparatus according to claim 38, wherein, The data uncertainty of the corresponding AI / ML model training data is determined based on at least one of the following: entropy, minimum confidence, marginal sampling, or generalization error.

40. The apparatus according to any one of claims 22 to 39, wherein, the AI / ML model training is at least partially performed by the second apparatus.

41. The apparatus according to any one of claims 22 to 40, wherein, the corresponding AI / ML model training data includes at least one of local AI / ML model training data of a local AI / ML model of the apparatus or local gradients associated with the local AI / ML model.

42. The apparatus according to any one of claims 22 to 41, wherein, the apparatus and the second apparatus cooperate to perform the AI / ML model training, wherein the processor-executable instructions further include processor-executable instructions that, when executed, cause the processor to perform the following operations: Before determining the DSI of the corresponding AI / ML model training data, perform a part of the AI / ML model training, wherein the corresponding AI / ML model training data includes the corresponding output of the part of the AI / ML model training.

43. An apparatus, wherein, comprises one or more units for performing the method according to any one of claims 1 to 21.

44. A method for performing artificial intelligence or machine learning (AI / ML) model training in a wireless communication network, wherein, the method comprises: A first apparatus sends AI / ML model training assistance information for determining data state information (DSI) of corresponding AI / ML model training data to a second apparatus; The first apparatus sends information related to the transmission of the corresponding AI / ML model training data to the second apparatus; The first apparatus receives the corresponding AI / ML model training data from the second apparatus, and the corresponding AI / ML model training data is sent based on at least one of the DSI of the corresponding AI / ML model training data or the information related to the transmission of the corresponding AI / ML model training data; The first apparatus performs the AI / ML model training using the corresponding AI / ML model training data.

45. The method according to claim 44, wherein, the corresponding AI / ML model training data is selectively sent, and the information related to the transmission of the corresponding AI / ML model training data includes a DSI threshold, and the method further comprises: The first apparatus configures the DSI threshold for determining whether the corresponding AI / ML model training data will be sent to the first apparatus.

46. The method according to claim 44, wherein, The corresponding AI / ML model training data is selectively sent, and the information related to the sending of the corresponding AI / ML model training data includes information indicating whether the corresponding AI / ML model training data will be sent to the first device. The method further includes: The first device receives at least one of the DSI or the AI / ML model training dataset information of the corresponding AI / ML model training data from the second device; The first device determines whether the corresponding AI / ML model training data will be sent to the first device by using at least one of the DSI of the corresponding AI / ML model training data or the AI / ML model training dataset information; The first device sends the information indicating whether the corresponding AI / ML model training data will be sent to the first device to the second device.

47. The method according to claim 46, wherein, The AI / ML model training dataset information includes at least one of the following: The size of the AI / ML model training dataset; or The DSI distribution information of the AI / ML model training dataset.

48. The method according to claim 46 or 47, wherein, The AI / ML model training dataset information is sent by using a buffer status report (BSR) or a scheduling request (SR).

49. The method according to claim 44, wherein, The corresponding AI / ML model training data is sent according to a corresponding report format, and the corresponding report format is determined based on at least one of the DSI of the corresponding AI / ML model training data or the information related to the sending of the corresponding AI / ML model training data.

50. The method according to claim 49, wherein, The corresponding report format indicates a corresponding transmission accuracy, and the level of the corresponding transmission accuracy is determined based on the level of the DSI of the corresponding AI / ML model training data.

51. The method according to claim 49 or 50, wherein, The corresponding report format indicates a configuration for the sending of the corresponding AI / ML model training data.

52. The method according to claim 51, wherein, The configuration for the sending of the corresponding AI / ML model training data indicates at least one of the following: Resources for the sending of the corresponding AI / ML model training data; or The quantization granularity for the sending of the corresponding AI / ML model training data.

53. The method according to any one of claims 49 to 52, wherein, The corresponding report format indicates whether the corresponding AI / ML model training data includes channel state information (CSI) or raw channel information.

54. The method according to any one of claims 49 to 53, wherein, the method further comprises: the first device configures the relationship between the DSI of the corresponding AI / ML model training data and the corresponding report format.

55. The method according to any one of claims 44 to 54, wherein, the AI / ML model training auxiliary information includes at least one of information about a reference AI / ML model or at least one reference input data value.

56. The method according to claim 55, wherein, the information about the reference AI / ML model includes at least one of the following: reference AI / ML model type; reference AI / ML model structure; one or more reference AI / ML model parameters; reference AI / ML model gradient; reference AI / ML model activation function; reference AI / ML model input data type; reference AI / ML model output data type; reference AI / ML model input data dimension; or reference AI / ML model output data dimension.

57. The method according to any one of claims 44 to 56, wherein, the method further comprises: the first device updates the AI / ML model training auxiliary information; the first device sends the updated AI / ML model training auxiliary information to the second device.

58. The method according to any one of claims 44 to 57, wherein, the DSI of the corresponding AI / ML model training data includes information indicating at least one of the following: data uncertainty of the corresponding AI / ML model training data; data importance of the corresponding AI / ML model training data; the degree of need for the corresponding AI / ML model training data for the AI / ML model training; or data diversity of the corresponding AI / ML model training data.

59. The method according to any one of claims 44 to 58, wherein, the DSI of the corresponding AI / ML model training data is determined based on at least one of the following: entropy, minimum confidence, marginal sampling, or generalization error.

60. The method according to any one of claims 44 to 59, wherein, the corresponding AI / ML model training data includes at least one of local AI / ML model training data of the local AI / ML model of the second device or local gradients associated with the local AI / ML model.

61. The method according to any one of claims 44 to 60, wherein, the first device and the second device cooperate to perform the AI / ML model training, such that the first device executes a part of the AI / ML model training before determining the DSI of the corresponding AI / ML model training data, and the corresponding AI / ML model training data includes the corresponding output of the part of the AI / ML model training.

62. A device for artificial intelligence or machine learning (AI / ML) model training in a wireless communication network, characterized in that, the device comprises: a processor; a memory storing processor-executable instructions that, when executed, cause the processor to: send AI / ML model training assistance information for determining data state information (DSI) of corresponding AI / ML model training data to a second device; send information related to the transmission of the corresponding AI / ML model training data to the second device; receive the corresponding AI / ML model training data from the second device, the corresponding AI / ML model training data being transmitted based on at least one of the DSI of the corresponding AI / ML model training data or the information related to the transmission of the corresponding AI / ML model training data; perform the AI / ML model training using the corresponding AI / ML model training data.

63. The device according to claim 62, characterized in that, the corresponding AI / ML model training data is selectively transmitted, and the information related to the transmission of the corresponding AI / ML model training data includes a DSI threshold, wherein the processor-executable instructions further include processor-executable instructions that, when executed, cause the processor to perform the following operations: configure the DSI threshold for determining whether the corresponding AI / ML model training data will be transmitted to the device.

64. The device according to claim 62, characterized in that, the corresponding AI / ML model training data is selectively transmitted, and the information related to the transmission of the corresponding AI / ML model training data includes information indicating whether the corresponding AI / ML model training data will be transmitted to the device, wherein the processor-executable instructions further include processor-executable instructions that, when executed, cause the processor to perform the following operations: receive at least one of the DSI of the corresponding AI / ML model training data or the AI / ML model training dataset information from the second device; determine whether the corresponding AI / ML model training data will be transmitted to the device using at least one of the DSI of the corresponding AI / ML model training data or the AI / ML model training dataset information; send the information indicating whether the corresponding AI / ML model training data will be transmitted to the device to the second device.

65. The device according to claim 64, characterized in that, the AI / ML model training dataset information includes at least one of the following: the size of the AI / ML model training dataset; or the DSI distribution information of the AI / ML model training dataset.

66. The device according to claim 64 or 65, It is characterized in that the AI / ML model training dataset information is sent using a buffer status report (BSR) or a scheduling request (SR).

67. The device according to claim 62, It is characterized in that the corresponding AI / ML model training data is sent according to a corresponding report format, and the corresponding report format is determined based on at least one of the DSI of the corresponding AI / ML model training data or the information related to the transmission of the corresponding AI / ML model training data.

68. The device according to claim 67, It is characterized in that the corresponding report format indicates a corresponding transmission accuracy, and the level of the corresponding transmission accuracy is determined based on the level of the DSI of the corresponding AI / ML model training data.

69. The device according to claim 67 or 68, It is characterized in that the corresponding report format indicates a configuration for the transmission of the corresponding AI / ML model training data.

70. The device according to claim 69, It is characterized in that the configuration for the transmission of the corresponding AI / ML model training data indicates at least one of the following: resources for the transmission of the corresponding AI / ML model training data; or quantization granularity for the transmission of the corresponding AI / ML model training data.

71. The device according to any one of claims 67 to 70, It is characterized in that the corresponding report format indicates whether the corresponding AI / ML model training data includes channel state information (CSI) or raw channel information.

72. The device according to any one of claims 67 to 71, It is characterized in that the processor-executable instructions further include processor-executable instructions that, when executed, cause the processor to perform the following operations: Configure the relationship between the DSI of the corresponding AI / ML model training data and the corresponding report format.

73. The device according to any one of claims 62 to 72, It is characterized in that the AI / ML model training auxiliary information includes at least one of information about a reference AI / ML model or at least one reference input data value.

74. The device according to claim 73, It is characterized in that the information about the reference AI / ML model includes at least one of the following: reference AI / ML model type; reference AI / ML model structure; one or more reference AI / ML model parameters; reference AI / ML model gradient; reference AI / ML model activation function; reference AI / ML model input data type; reference AI / ML model output data type; reference AI / ML model input data dimension; or reference AI / ML model output data dimension.

75. The device according to any one of claims 62 to 74, It is characterized in that The processor-executable instructions further include processor-executable instructions that, when executed, cause the processor to perform the following operations: Update the AI / ML model training assistance information; Send the updated AI / ML model training assistance information to the second device.

76. The device according to any one of claims 62 to 75, wherein, the DSI of the corresponding AI / ML model training data includes information indicating at least one of the following: data uncertainty of the corresponding AI / ML model training data; data importance of the corresponding AI / ML model training data; the degree of need of the corresponding AI / ML model training data for the AI / ML model training; or data diversity of the corresponding AI / ML model training data.

77. The device according to any one of claims 62 to 76, wherein, the DSI of the corresponding AI / ML model training data is determined based on at least one of the following: entropy, minimum confidence, marginal sampling, or generalization error.

78. The device according to any one of claims 62 to 77, wherein, the corresponding AI / ML model training data includes at least one of local AI / ML model training data of the local AI / ML model of the second device or local gradients associated with the local AI / ML model.

79. The device according to any one of claims 62 to 78, wherein, the device and the second device cooperate to perform the AI / ML model training, such that the device performs a part of the AI / ML model training before determining the DSI of the corresponding AI / ML model training data, and the corresponding AI / ML model training data includes the corresponding output of the part of the AI / ML model training.

80. A device, wherein, comprises one or more units for performing the method according to any one of claims 44 to 61.

Images