Method and device for learning artificial intelligence or machine learning model

Abstract

Aspects of the present disclosure provide a method and apparatus for learning an artificial intelligence or machine learning (AI / ML) model on a self-organizing topology, so as to implement a heterogeneous neural network structure in a distributed AI / ML training process, and to solve the problems of data isomerism and the like in the traditional AI / ML training process based on federated learning (based on FL (federated learning, FL-based)). A first node receives a first AI / ML model from a second node and sends the first AI / ML model and a model collection indicator to one or more nodes associated with the first node. The first node receives a report relating to a respective associated AI / ML model of its associated node, and generates a second AI / ML model having a neural network (NN) structure equivalent to a NN structure of the first AI / ML model from the report. However, an AI / ML model of an associated node of the first node may have a different NN structure than NN structures of the first AI / ML model and the second AI / ML model.

Description

Technical Field

[0001] The present disclosure relates generally to wireless communications and, in certain embodiments, to methods and apparatus for learning artificial intelligence or machine learning (AI / ML) models. Background Art

[0002] Artificial intelligence technology can be applied in communications, including AI-based communications in the physical layer and / or AI-based communications in the medium access control (MAC) layer. For example, in the physical layer, AI-based communications may aim to optimize component design and / or improve algorithm performance. For the MAC layer, AI-based communications may aim to use AI capabilities to learn, predict and / or make decisions 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 ​​architecture in a wireless communication network may include multiple nodes, where the multiple nodes may be organized in one of two modes: centralized and distributed, both of which may be deployed in an access network, core network, or edge computing system or a third-party network. Centralized training and computing architectures are subject to potentially large communication overhead and strict user data privacy protection. Distributed training and computing architectures may include several frameworks, such as distributed machine learning and federated learning.

[0004] However, communications in wireless communication systems, including communications associated with AI training at multiple nodes, are often conducted over non-ideal channels. For example, non-ideal conditions such as electromagnetic interference, signal degradation, phase delay, fading, and other non-ideal situations may attenuate and / or distort communication signals, or may otherwise interfere with or degrade the communication capabilities of the system.

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

[0006] Additionally, the processing power and / or training data availability for the AI ​​training process may vary significantly between different nodes / devices, which means that the ability of different nodes to effectively participate in the AI ​​training process may vary significantly. In practice, such differences often mean that the training latency of an AI training process involving multiple nodes / devices (e.g., an AI training process based on distributed learning or federated learning) is dominated by the node / device with the largest latency, which is caused by communication latency and / or computational latency.

[0007] Federated learning, also known as collaborative learning, is a machine learning technique that trains algorithms across multiple distributed edge devices or servers. Each distributed edge device or server holds local data samples, but may not exchange them with other devices or servers. The difference between federated learning and traditional centralized machine learning is that local data samples are not shared in federated learning, while in traditional centralized machine learning, all local data sets are uploaded to one server.

[0008] In a wireless federated learning-based (FL-based) AI training process, the network node / device / node initializes the global AI model, samples a group of user devices, and broadcasts the global AI model parameters to the user devices. Each user device then initializes its local AI model using the global AI model parameters and updates (trains) its local AI model using its own data. Each user device can then report the parameters of its updated local AI model to the network device, which then aggregates the updated parameters reported by the user devices and updates the global AI model. The above process is one iteration of the traditional FL-based AI model training process. The network device and 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 finalized.

[0009] However, different user devices participating in the FL-based AI training process may observe different training data sets, which may not represent the distribution of all training data observed by other user devices participating in the FL-based AI training process, that is, the training data may be non-independently and identically distributed (or non-iid) on the devices participating in the traditional FL-based AI training process. Non-iid training data between devices has been shown to reduce the convergence speed and model accuracy in the traditional FL-based AI training process. Therefore, data heterogeneity is a typical problem in the traditional FL-based AI training process.

[0010] For these and other reasons, new protocols and signaling mechanisms are needed so that new AI-enabled applications and processes can be implemented while minimizing the signaling and communication overhead and latency associated with existing AI training processes. Summary of the invention

[0011] In the traditional federated learning (FL-based) AI training process, there are some limitations. As mentioned above, for example, in the traditional FL-based AI training process, data heterogeneity is a typical problem. In addition, the server node needs to collect a large amount of training data sets (e.g., gradients updated by the client nodes) from multiple associated client nodes. In addition, the traditional FL-based AI training process requires the server node and the associated client nodes to have the same AI / ML model structure. However, since different local nodes may support different AI / ML model structures, data heterogeneity may be a problem in the traditional FL-based AI training process.

[0012] Various aspects of the present disclosure provide solutions to overcome the above limitations, such as specific methods and apparatus for learning artificial intelligence or machine learning (AI / ML) models on self-organizing topologies.

[0013] According to the first broad aspect of the present disclosure, a method for learning an artificial intelligence or machine learning (AI / ML) model on a topology in a wireless communication network is provided herein. The method according to the first broad aspect of the present disclosure may include: a first node receives a first AI / ML model from a second node. After receiving the first AI / ML model, the first node may send the first AI / ML model and a model collection indicator for collecting the AI / ML model from the one or more associated nodes to one or more nodes associated with the first node. The method according to the first broad aspect of the present disclosure may also include: the first node receives a report related to the corresponding associated AI / ML model of the one or more associated nodes from the one or more associated nodes. After receiving such a report, the first node may obtain a second AI / ML model according to the report related to the corresponding associated AI / ML model. The second AI / ML model may have an NN structure equivalent to the neural network (NN) structure of the first AI / ML model. The first node may also send a second AI / ML model to a third node.

[0014] Optionally, the corresponding associated AI / ML model of the one or more associated nodes may have a NN structure different from the NN structure of the first AI / ML model and the second AI / ML model. In other words, the corresponding associated AI / ML model of the one or more associated nodes is not limited to having a NN structure equivalent to the NN structure of the first AI / ML model and the second AI / ML model.

[0015] In some embodiments of the method according to the first broad aspect of the present disclosure, the second AI / ML model is generated using an aggregation algorithm.

[0016] In some embodiments of the method according to the first broad aspect of the present disclosure, the report related to the corresponding associated AI / ML model includes at least one of: a confirmation indicator for transmission of the corresponding associated AI / ML model, information related to the corresponding associated AI / ML model, information related to training data of the corresponding associated AI / ML model, or information related to the performance of the corresponding associated AI / ML model.

[0017] In some embodiments of the method according to the first broad aspect of the present disclosure, the one or more associated nodes generate the report related to the corresponding associated AI / ML model based on the model collection indicator.

[0018] In some embodiments of the method according to the first broad aspect of the present disclosure, the model collection indicator is a dynamic indicator or an event triggered indicator.

[0019] In some embodiments of the method according to the first broad aspect of the present disclosure, the model collection indicator is an event trigger indicator, and the event is triggered when the training of the corresponding associated AI / ML model is completed or the performance of the corresponding associated AI / ML model exceeds a performance metric, and the performance metric is predetermined based on at least one of accuracy or precision.

[0020] In some embodiments of the method according to the first broad aspect of the disclosure, the model collection indicator comprises one of: a distillation indicator; an expansion indicator; or a distillation and expansion indicator.

[0021] In some embodiments, the method according to the first broad aspect of the present disclosure further includes: the first node sending information about the reference AI / ML model to the one or more associated nodes.

[0022] In some embodiments of the method according to the first broad aspect of the present disclosure, the information about the reference AI / ML model includes information indicating the NN structure of the reference AI / ML model, and the information indicating the NN structure of the reference AI / ML model includes at least one of the following: NN algorithm, width, depth, complexity, floating-point operations, total parameters, trainable parameters, or required buffer size.

[0023] In some embodiments of the method according to the first broad aspect of the present disclosure, obtaining the second AI / ML model includes at least one of the following: distilling the corresponding associated AL / ML model into one or more distilled AI / ML models based on the model collection indicator; or expanding the corresponding associated AL / ML model into one or more expanded AI / ML models based on the model collection indicator.

[0024] In some embodiments of the method according to the first broad aspect of the present disclosure, obtaining the second AI / ML model also includes at least one of the following: obtaining the mean of the one or more distilled AI / ML models, or obtaining the mean of the one or more inflated AI / ML models.

[0025] In some embodiments of the method according to the first broad aspect of the present disclosure, obtaining the second AI / ML model also includes obtaining the AI / ML model of the first node, and at least one of the following: obtaining the AI / ML model of the first node and the mean of the one or more distilled AI / ML models, or obtaining the AI / ML model of the first node and the mean of the one or more expanded AI / ML models.

[0026] In some embodiments of the method according to the first broad aspect of the present disclosure, obtaining the AI / ML model of the first node includes: obtaining a preliminary AI / ML model of the first node; and distilling or expanding the preliminary AI / ML model into the AI / ML model of the first node.

[0027] In some embodiments of the method according to the first broad aspect of the present disclosure, the aggregation algorithm includes at least one of model distillation or model expansion based on a model collection indicator.

[0028] In some embodiments of the method according to the first broad aspect of the present disclosure, the second AI / ML model includes one or more AI / ML model parameters that do not include information related to the NN structure of the second AI / ML model.

[0029] In some embodiments of the method according to the first broad aspect of the present disclosure, the NN structure of the second AI / ML model is pre-configured.

[0030] In some embodiments, the method according to the first broad aspect of the present disclosure further includes: the first node sending the second AI / ML model to one or more nodes associated with the third node.

[0031] In some embodiments of the method according to the first broad aspect of the present disclosure, the first node iteratively performs the method.

[0032] In some embodiments of the method according to the first broad aspect of the present disclosure, the first node, the second node, the third node or at least one of the one or more associated nodes is a base station, a user equipment (UE), a relay device, a transmission and reception point (TRP), an edge device, a network device or a network system.

[0033] According to the second broad aspect of the present disclosure, this document provides a method for learning an artificial intelligence or machine learning (AI / ML) model on a topology in a wireless communication network. The method according to the second broad aspect of the present disclosure may include: a first node receives a first AI / ML model and a model collection indicator. After receiving the first AI / ML model and the model collection indicator, the first node may acquire a second AI / ML model. The method according to the second broad aspect of the present disclosure may also include: the first node acquires a report related to the second AI / ML model. The first node may send a report related to the second AI / ML model to the second node based on the model collection indicator.

[0034] Optionally, the second AI / ML model may have a neural network (NN) structure that is different from the NN structure of the first AI / ML model. In other words, the second AI / ML model is not limited to having a NN structure that is equivalent to the NN structure of the first AI / ML model.

[0035] In some embodiments of the method according to the second broad aspect of the present disclosure, the second AI / ML model is generated according to at least one of the first AI / ML model, a local training dataset, or a local AI / ML algorithm.

[0036] In some embodiments of the method according to the second broad aspect of the present disclosure, the report related to the second AI / ML model includes at least one of: a confirmation indicator for transmission of the second AI / ML model, information related to the second AI / ML model, information related to training data of the second AI / ML model, or information related to performance of the second AI / ML model.

[0037] In some embodiments of the method according to the second broad aspect of the present disclosure, the first node further generates a report related to the second AI / ML model based on the model collection indicator.

[0038] In some embodiments of the method according to the second broad aspect of the disclosure, the model collection indicator is a dynamic indicator or an event triggered indicator.

[0039] In some embodiments of the method according to the second broad aspect of the present disclosure, the model collection indicator is an event trigger indicator, and the event is triggered when the training of the second AI / ML model is completed or the performance of the second AI / ML model exceeds a performance metric, and the performance metric is predetermined based on at least one of accuracy or precision.

[0040] In some embodiments of the method according to the second broad aspect of the disclosure, the model collection indicator comprises one of: a distillation indicator; an expansion indicator; or a distillation and expansion indicator.

[0041] In some embodiments, the method according to the second broad aspect of the present disclosure further includes: the first node receiving information about the reference AI / ML model from the second node.

[0042] In some embodiments of the method according to the second broad aspect of the present disclosure, the information about the reference AI / ML model includes information indicating the NN structure of the reference AI / ML model, and the information indicating the NN structure of the reference AI / ML model includes at least one of the following: NN algorithm, width, depth, complexity, floating-point operations, total parameters, trainable parameters, or required buffer size.

[0043] In some embodiments of the method according to the second broad aspect of the present disclosure, the model collection indicator is a distillation indicator, and the generated second AI / ML model is larger than the reference AI / ML model.

[0044] In some embodiments of the method according to the second broad aspect of the present disclosure, the generated second AI / ML model has: a greater number of floating-point operations than the reference model; a greater number of total parameters than the reference model; a greater number of trainable parameters than the reference model; a larger required buffer size than the reference model; a greater width than the reference model; a greater depth than the reference model; or any combination of the foregoing.

[0045] In some embodiments of the method according to the second broad aspect of the present disclosure, the model collection indicator is a bloat indicator and the generated second AI / ML model is smaller than the reference AI / ML model.

[0046] In some embodiments of the method according to the second broad aspect of the present disclosure, the generated second AI / ML model has: fewer floating-point operations than the reference model; fewer total parameters than the reference model; fewer trainable parameters than the reference model; smaller required buffer size than the reference model; smaller width than the reference model; smaller depth than the reference model; or any combination of the foregoing.

[0047] In some embodiments of the method according to the second broad aspect of the present disclosure, the model collection indicators are distillation and expansion indicators, and the first node generates the second AI / ML model without restrictions associated with the reference AI / ML model.

[0048] In some embodiments of the method according to the second broad aspect of the disclosure, the first node receives a first AI / ML model and a model collection indicator from a second node.

[0049] In some embodiments of the method according to the second broad aspect of the present disclosure, the first node receives a model collection indicator from the second node and receives the first AI / ML model from the third node.

[0050] In some embodiments of the method according to the second broad aspect of the present disclosure, the first node iteratively performs the method.

[0051] In some embodiments of the method according to the second broad aspect of the present disclosure, at least one of the first node or the second node is a base station, a user equipment (UE), a relay device, a transmission and reception point (TRP), an edge device, a network device, or a network system.

[0052] Corresponding apparatus and devices for performing the method are disclosed.

[0053] For example, according to another aspect of the present disclosure, a device is provided, which includes a processor and a memory storing processor-executable instructions. When the processor-executable instructions are executed, the processor performs the method according to the first broad aspect or the second broad aspect of the present disclosure mentioned above.

[0054] According to other aspects of the present disclosure, a device is provided, which includes one or more units for implementing any method aspect disclosed in the present disclosure. The term "unit" is used in a broad sense and can be referred to by any of a variety of names, including, for example, modules, components, elements, devices, etc. These units can be implemented using hardware, software, firmware, or any combination of the foregoing.

[0055] According to some aspects of the present disclosure, AI / ML models can be routed in wireless networks. In this way, there is no need to transmit large amounts of data (e.g., training data) between various network devices (network nodes, nodes, base stations, TRPs, etc.) and user devices (e.g., UEs). In addition, heterogeneous AI / ML capabilities can be supported in various network devices and user devices. Any of the various network devices and user devices can operate as an aggregation node.

[0056] According to some aspects of the present disclosure, AI / ML model distillation and / or expansion operations can be performed at the aggregation node. In addition, heterogeneous AI / ML model aggregation on self-organizing topologies can be supported at various network devices (network nodes, nodes, base stations, TRPs, etc.) and user devices (e.g., UEs).

[0057] According to some aspects of the present disclosure, AI / ML model training can be performed dynamically. AI / ML model training can be performed iteratively. Iterations of AI / ML model training can be completed at a desired time. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Reference will now be made, by way of example only, to the accompanying drawings which show exemplary embodiments of the present application, in which:

[0059] Figure 1 is a simplified schematic diagram of a communication system according to an example;

[0060] Figure 2 Another example of a communication system is shown;

[0061] Figure 3 Examples of electronic devices (EDs), terrestrial transmit and receive points (T-TRPs), and non-terrestrial transmit and receive points (NT-TRPs) are shown;

[0062] Figure 4 Exemplary units or modules in the device are shown;

[0063] Figure 5 shows four EDs communicating with a network device in a communication system according to an embodiment of the present disclosure;

[0064] Fig. 6A An example of a neural network with multiple layers of neurons according to an embodiment of the present disclosure is shown;

[0065] Figure 6B shows examples of neurons that may be used as building blocks of a neural network according to embodiments of the present disclosure;

[0066] Figure 7 shows the star topology used in the traditional federated learning (FL) process;

[0067] Figure 8 The difference between the traditional FL-based AI / ML model training process and the AI / ML model learning scheme of the present disclosure is shown;

[0068] Fig. 9 An exemplary self-organizing topology for AI / ML model training according to an embodiment of the present disclosure is shown;

[0069] Fig. 10A and Fig. 10B Two different types of nodes for learning an AI / ML model according to an embodiment of the present disclosure are shown;

[0070] Fig.11A and Fig. 11B An exemplary self-organizing topology according to an embodiment of the present disclosure is shown;

[0071] Fig.12 An exemplary process of AI / ML model aggregation on a self-organizing topology according to an embodiment of the present disclosure is shown;

[0072] Fig.13 An example of heterogeneous AI / ML model aggregation according to an embodiment of the present disclosure is shown;

[0073] Fig.14 An exemplary AI / ML model distillation process in AI / ML model aggregation on a self-organizing topology according to an embodiment of the present disclosure is shown;

[0074] Fig.15 An exemplary AI / ML model expansion process in AI / ML model aggregation on a self-organizing topology according to an embodiment of the present disclosure is shown;

[0075] Fig.16 An exemplary AI / ML model distillation and expansion process in AI / ML model aggregation on a self-organizing topology according to an embodiment of the present disclosure is shown;

[0076] Fig.17 Another exemplary AI / ML model distillation process in AI / ML model aggregation on a self-organizing topology according to an embodiment of the present disclosure is shown;

[0077] Fig.18 Another exemplary AI / ML model expansion process in AI / ML model aggregation on a self-organizing topology according to an embodiment of the present disclosure is shown;

[0078] Fig.19 Another exemplary AI / ML model distillation and expansion process in AI / ML model aggregation on a self-organizing topology according to an embodiment of the present disclosure is shown;

[0079] Fig. 20 Another exemplary process of AI / ML model aggregation on a self-organizing topology according to an embodiment of the present disclosure is shown.

[0080] The same reference numerals may be used in different drawings to denote the same components. DETAILED DESCRIPTION

[0081] In the present disclosure, "data collection" refers to the process of collecting data by a network node, management entity or user equipment (UE) for the purpose of artificial intelligence (AI) / machine learning (ML) model training, data analysis and reasoning.

[0082] In this disclosure, an “AI / ML model” refers to a data-driven algorithm that applies AI / ML techniques to generate a set of outputs based on a set of inputs.

[0083] In the present disclosure, “AI / ML model training” refers to the process of training an AI / ML model by learning input / output relationships in a data-driven manner and obtaining the trained AI / ML model for reasoning.

[0084] In this disclosure, “AI / ML inference” refers to the process of using a trained AI / ML model to produce a set of outputs based on a set of inputs.

[0085] In this disclosure, "online training" refers to situations where the machine learning program is not running and receiving new information in real time.

[0086] In this disclosure, "offline training" refers to the situation where a machine learning program processes incoming data in real time.

[0087] In this disclosure, "on-UE training" refers to online / offline training at the UE.

[0088] In this disclosure, "training on the network" refers to online / offline training on the network.

[0089] In this disclosure, “UE-side (AI / ML) model” refers to an AI / ML model where reasoning is performed entirely at the UE.

[0090] In this disclosure, “network-side (AI / ML) model” refers to an AI / ML model whose reasoning is performed entirely on the network.

[0091] In this disclosure, "model transmission" refers to the transmission of an AI / ML model over an air interface, which may be parameters of a model structure known to the receiving end, or a new model with parameters. The transmission may include a complete model or a partial model.

[0092] In this disclosure, "model download" refers to the transmission of a model from the network to the UE.

[0093] In this disclosure, "model upload" refers to the transmission of a model from a UE to a network.

[0094] In this disclosure, “model deployment” refers to the delivery of a fully developed and tested model runtime image to the target UE / gNodeB (gNB) where inference is to be performed.

[0095] In this disclosure, "federated learning / federated training" refers to a machine learning technique that trains AI / ML models across multiple distributed edge nodes (e.g., UE, gNB), each of which performs local model training using local data samples. This technique requires multiple model exchanges, but does not require the exchange of local data samples.

[0096] In this disclosure, “model monitoring” refers to the process of monitoring the inference performance of an AI / ML model.

[0097] In this disclosure, “model update” refers to retraining or fine-tuning the AI / ML model through online / offline training to improve the model reasoning performance.

[0098] For illustrative purposes, specific exemplary embodiments are explained in detail below with reference to the accompanying drawings.

[0099] The embodiments set forth herein represent information sufficient to practice the subject matter claimed for protection and illustrate methods of practicing such subject matter. After reading the following description in light of the accompanying drawings, those skilled in the art will understand the concepts of the subject matter claimed for protection and will recognize applications of these concepts not specifically mentioned herein. It should be understood that these concepts and applications are within the scope of this disclosure and the appended claims.

[0100] In addition, it should be understood that any module, component, or device for executing instructions disclosed herein may include or otherwise access one or more non-transitory computer / processor readable storage media to store information, for example, to store 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 cassettes, magnetic tapes, disk storage or other magnetic storage devices, compact disc read-only memory (CD-ROM), digital video discs or digital versatile discs (i.e., DVDs), Blu-ray discs, and the like. TMCD-ROMs, 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 technology. Any of these non-transitory computer / processor storage media can be part of a device or can be accessed or connected to a device. Computer / processor readable / executable instructions for implementing the applications or modules described herein can be stored or otherwise held by such non-transitory computer / processor readable storage media.

[0101] Exemplary Communication Systems and Devices

[0102] refer to Figure 1 , a simplified schematic diagram of a communication system is provided as an illustrative example and not a limitation. The communication system 100 includes a radio access network 120. The radio access network 120 may be a next generation (e.g., sixth generation (6G) or higher) 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 (electric devices, ED) 110a to 110j (generally referred to as 110) may be interconnected with each other or connected to one or more network nodes (170a, 170b, generally referred to as 170). The core network 130 may be part of the communication system and may be dependent on or independent of the radio access technology used in the communication system 100. In addition, the communication system 100 includes a public switched telephone network (PSTN) 140, the Internet 150, and other networks 160.

[0103] Figure 2An exemplary communication system 100 is shown. Typically, the communication system 100 enables multiple wireless or wired elements to transmit data and other content. The purpose of the communication system 100 may be to provide content such as voice, data, video and / or text through broadcast, multicast and unicast, etc. The communication system 100 may operate by sharing resources such as carrier spectrum bandwidth between its constituent elements. The communication system 100 may include a terrestrial communication system and / or a non-terrestrial communication system. The communication system 100 may 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 may provide high availability and robustness through the joint operation of a terrestrial communication system and a non-terrestrial communication system. For example, integrating a non-terrestrial communication system (or its components) into a terrestrial communication system may produce a heterogeneous network that may be considered to include multiple layers. Compared with traditional communication networks, heterogeneous networks may achieve better overall performance through efficient multi-link joint operation, more flexible function sharing, and faster physical layer link switching between terrestrial networks and non-terrestrial networks.

[0104] The terrestrial communication system and the non-terrestrial communication system can be considered as subsystems of the communication system. In the example shown, the communication system 100 includes electronic devices (ED) 110a to 110d (generally referred to as ED 110), radio access networks (RAN) 120a and 120b, non-terrestrial communication networks 120c, core networks 130, public switched telephone networks (PSTN) 140, Internet 150, and other networks 160. RAN 120a and 120b include corresponding base stations (BS) 170a and 170b, which can be generally referred to as terrestrial transmission and reception points (T-TRP) 170a and 170b. The non-terrestrial communication network 120c includes an access node 120c, which can be generally referred to as a non-terrestrial transmission and reception point (NT-TRP) 172.

[0105] Alternatively or additionally, any ED 110 may be used to connect, access, or communicate with any other T-TRP 170a and 170b, NT-TRP 172, Internet 150, core network 130, PSTN 140, other network 160, or any combination thereof. In some examples, ED 110a may perform uplink and / or downlink transmissions with T-TRP 170a via interface 190a. In some examples, ED 110a, 110b, and 110d may also communicate directly with each other via one or more sidelink air interfaces 190b. In some examples, ED 110d may perform uplink and / or downlink transmissions with NT-TRP 172 via interface 190c.

[0106] The air interfaces 190a and 190b may use similar communication technologies, such as any suitable wireless 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 utilize other high-dimensional signal spaces that may include a combination of orthogonal dimensions and / or non-orthogonal dimensions.

[0107] The air interface 190c may enable communication between the ED 110d and one or more NT-TRPs 172 via a wireless link or simply, via a link. For some examples, the link is a dedicated connection for unicast transmissions, a connection for broadcast transmissions, or a connection between a group of EDs and one or more NT-TRPs for multicast transmissions.

[0108] The 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 RANs 120a and / or RANs 120b. 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 these RANs and EDs, 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 for communicating with different wireless networks over different wireless links using different wireless technologies and / or protocols. Instead of (or in addition to) wireless communications, the EDs 110a, 110b, and 110c may 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 may include 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 multi-mode devices capable of operating according to a variety of wireless access technologies and include a plurality of transceivers required to support these technologies.

[0109] Figure 3Another example of ED 110 and base stations 170a, 170b and / or 170c is shown. ED 110 is used to connect people, objects, machines, etc. ED 110 can be widely used in various scenarios, for example, 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 wearable devices, smart transportation, smart city, drones, robots, remote sensing, passive sensing, positioning, navigation and tracking, autonomous delivery and mobility, etc.

[0110] Each ED 110 represents any end-user device suitable for wireless operation, which may include the following devices (or may be referred to as): user equipment / device (UE), wireless transmit / receive unit (WTRU), mobile station, fixed or mobile user unit, cellular phone, station (STA), machine type communication (MTC) equipment, personal digital assistant (PDA), smartphone, laptop, computer, tablet, wireless sensor, consumer electronic device, smart book, vehicle, car, truck, bus, train or IoT device, industrial equipment or devices in the aforementioned devices (e.g., communication module, modem or chip), etc. The next generation ED 110 may be referred to using other terms. Base stations 170a and 170b are T-TRPs, which will be referred to as T-TRP 170 hereinafter. Similarly in Figure 3 The NT-TRP shown in will be referred to hereinafter as NT-TRP 172. Each ED 110 connected to T-TRP 170 and / or NT-TRP 172 may be dynamically or semi-statically started (i.e., established, activated, or enabled), shut down (i.e., released, deactivated, or disabled), and / or configured in response to one or more of connection availability and connection necessity.

[0111] ED 110 includes a transmitter 201 and a receiver 203 coupled to one or more antennas 204. Only one antenna 204 is shown in the figure. Alternatively, one, part or all of the antennas can be a panel. Transmitter 201 and receiver 203 can be integrated into, for example, a transceiver. The transceiver is used to modulate data or other content for transmission through at least one antenna 204 or a network interface controller (NIC). The transceiver is also used to demodulate data or other content received by the 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 by wireless or wired means. Each antenna 204 includes: any suitable structure for sending and / or receiving wireless signals or wired signals.

[0112] 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 for implementing some or all of the functions and / or embodiments described herein and executed by the processing unit 210. Each memory 208 includes any suitable volatile and / or non-volatile storage and retrieval device. 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.

[0113] ED 110 may also include one or more input / output devices (not shown) or interfaces (e.g., Figure 1 The input / output devices support interaction with a user or other devices in the network. Each input / output device includes any suitable structure for providing information to a user or receiving information from a user, such as a speaker, a microphone, a keypad, a keyboard, a display or a touch screen, including network interface communications.

[0114] ED 110 also includes a processor 210 for performing operations, including operations related to preparing uplink transmissions to be sent to NT-TRP 172 and / or T-TRP 170, processing downlink transmissions received from NT-TRP 172 and / or T-TRP 170, and processing sidelink transmissions to and from another ED 110. Processing operations related to preparing uplink transmissions to be sent may include operations such as encoding, modulation, transmit beamforming, and generating symbols for transmission. Processing operations related to processing downlink transmissions may include operations such as receive beamforming, demodulating and decoding received symbols. According to an embodiment, the downlink transmission may be received by receiver 203, possibly using receive beamforming, and processor 210 may extract signaling from the downlink transmission (e.g., by detecting and / or decoding the signaling). An example of signaling may be a reference signal sent by NT-TRP 172 and / or T-TRP 170. In some embodiments, the processor 276 implements transmit beamforming and / or receive beamforming according to an indication of a beam direction (e.g., beam angle information (BAI)) received from the T-TRP 170. 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 obtaining system information. In some embodiments, the processor 210 may perform channel estimation, for example, using a reference signal received from the NT-TRP 172 and / or the T-TRP 170.

[0115] 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 constitute part of the processor 210.

[0116] The processor 210 and the processing components of the transmitter 201 and the receiver 203 may be implemented by the same or different one or more processors, respectively, which are used to execute instructions stored in a memory (e.g., the memory 208). Alternatively, the processor 210 and some or all of the processing components of the transmitter 201 and the receiver 203 may be implemented using a programmed field-programmable gate array (FPGA), a graphical processing unit (GPU) or an application-specific integrated circuit (ASIC) or other dedicated circuit.

[0117] T-TRP 170 may use other names in some implementations, such as base station, base transceiver station (BTS), wireless base station, network node, network device, network side device, transmission / reception node, node B, evolved NodeB (eNodeB or eNB), home eNodeB, next generation NodeB (gNB), transmission point (TP), site controller, access point (AP), wireless router, relay station, remote radio head, ground node, ground network device, ground base station, base band unit (BBU), remote radio unit (RRU), active antenna unit (AAU), remote radio head (RRH), centralized unit (CU), distributed unit (DU), positioning node, etc. T-TRP 170 may be a macro BS, a micro BS, a relay node, a host node, etc. or a combination thereof. T-TRP 170 may refer to the above-mentioned device, or refer to a device in the above-mentioned device (eg, a communication module, a modem, or a chip).

[0118] In some embodiments, various parts of T-TRP 170 may be distributed. For example, some modules of T-TRP 170 may be located away from the device that houses the antenna of T-TRP 170, and may be coupled to the device that houses the antenna through a communication link (not shown) sometimes referred to as a fronthaul, such as a common public radio interface (CPRI). Therefore, 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 location of ED 110, resource allocation (scheduling), message generation, and encoding / decoding, and are not necessarily part of the device that houses 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 operate together to serve ED 110 through coordinated multi-point transmission, etc.

[0119] 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 in this figure. Alternatively, one, some or all of the antennas may be panels. The transmitter 252 and the receiver 254 may be integrated into a transceiver. The T-TRP 170 also includes a processor 260 for performing operations, including operations related to preparing downlink transmissions sent to the ED 110, processing uplink transmissions received from the ED 110, preparing backhaul transmissions sent to the NT-TRP 172, and processing transmissions received from the NT-TRP 172 via backhaul. Processing operations related to preparing to send downlink transmissions or backhaul transmissions may include operations such as encoding, modulation, precoding (e.g., MIMO precoding), transmit beamforming, and generating symbols for transmission. Processing operations related to processing received transmissions in the uplink or on 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 a beam direction indication, such as a BAI, that the scheduler 253 may schedule for transmission. The processor 260 performs other network-side processing operations described herein, such as determining the location of the ED 110, determining the location at which the NT-TRP 172 is deployed, etc. In some embodiments, the processor 260 may generate signaling to configure one or more parameters of the ED 110 and / or one or more parameters of the NT-TRP 172, etc. Any signaling generated by the processor 260 is sent by the transmitter 252. Note that, alternatively, the "signaling" used herein may be referred to as control signaling. Dynamic signaling can be sent in a control channel such as a physical downlink control channel (PDCCH), while static or semi-static higher-layer signaling can be included in a data packet sent in a data channel such as a physical downlink shared channel (PDSCH).

[0120] The scheduler 253 may be coupled to the processor 260. The scheduler 253 may be included in or run separately from the T-TRP 170, which may schedule uplink, downlink, and / or backhaul transmissions, including issuing scheduling grants and / or configuring scheduling-free (“configuration grants”) 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 may store software instructions or modules for implementing some or all of the functionality and / or embodiments described herein and executed by the processor 260.

[0121] Although not shown, the processor 260 may form a part of the transmitter 252 and / or the receiver 254. In addition, although not shown, the processor 260 may implement the scheduler 253. Although not shown, the memory 258 may constitute a part of the processor 260.

[0122] The processor 260, the scheduler 253, and the processing components of the transmitter 252 and the receiver 254 may be implemented by the same or different one or more processors, respectively, which are used to execute 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 may be implemented using a dedicated circuit such as an FPGA, a GPU, or an ASIC.

[0123] Although the NT-TRP 172 is shown as a drone only as an example, the NT-TRP 172 can be implemented in any suitable non-ground form. In addition, in some implementations, the NT-TRP 172 can use other names, such as non-ground nodes, non-ground network devices, or non-ground base stations. The NT-TRP 172 includes a transmitter 272 and a receiver 274 coupled to one or more antennas 280. Only one antenna 280 is shown in this figure. Alternatively, one, part, or all of the antennas can be panels. The transmitter 272 and the receiver 274 can be integrated into a transceiver. The NT-TRP 172 also includes a processor 276 for performing operations, including operations related to: preparing downlink transmissions sent to ED 110, processing uplink transmissions received from ED110, preparing backhaul transmissions sent to T-TRP 170, and processing transmissions received from T-TRP 170 via backhaul. Processing operations related to preparing to send a downlink transmission or a backhaul transmission may include operations such as encoding, modulation, precoding (e.g., MIMO precoding), transmit beamforming, and generating symbols for transmission. Processing operations related to processing received transmissions in the uplink or on the backhaul may include operations such as receive beamforming, demodulation and decoding of 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 the T-TRP 170. In some embodiments, the processor 276 may generate signaling to configure one or more parameters of the ED 110, etc. In some embodiments, the NT-TRP 172 implements physical layer processing but does not implement higher-level functions, such as functions of the medium access control (MAC) layer or the radio link control (RLC) layer. Since this is just an example, more generally, the NT-TRP 172 may implement higher-level functions in addition to physical layer processing.

[0124] The 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 constitute part of the processor 276.

[0125] The processor 276 and the processing components of the transmitter 272 and the receiver 274 can be implemented by the same or different one or more processors, which are used to execute instructions stored in a memory (e.g., in the memory 278). Alternatively, some or all of the processing components of the processor 276 and the transmitter 272 and the receiver 274 can be implemented using a programmed FPGA, GPU or ASIC or other dedicated circuit. In some embodiments, the NT-TRP 172 can actually be a plurality of NT-TRPs that operate together to serve the ED 110 through coordinated multi-point transmission or the like.

[0126] Note that "TRP" as used herein may refer to T-TRP or NT-TRP.

[0127] The T-TRP 170, NT-TRP 172, and / or ED 110 may include other components, but these components have been omitted for clarity.

[0128] One or more steps of the exemplary methods provided herein may be performed according to Figure 4 The corresponding unit or module is executed. Figure 4 Units or modules in devices such as ED 110, T-TRP 170, or NT-TRP 172 are shown. For example, a signal can be sent by a sending unit or a sending module. For example, a signal can be sent by a sending unit or a sending module. A signal can be received by a receiving unit or a receiving module. The signal can be processed by a processing unit or a processing module. Other steps can be performed by an artificial intelligence (AI) module or a machine learning (ML) module. The corresponding units or modules can be implemented using hardware, one or more components or devices that execute software, or a combination thereof. For example, one or more units or modules can be programmed integrated circuits such as FPGAs, GPUs, or ASICs. It should be understood that if these modules are implemented, for example, using software executed by a processor, these modules can be retrieved by the processor in whole or in part as needed, retrieved individually or together for processing, retrieved in one or more instances, and these modules themselves can include instructions for further deployment and instantiation.

[0129] Other details about ED 110, T-TRP 170 and NT-TRP 172 are known to those skilled in the art. Therefore, these details are omitted here.

[0130] Control signaling is discussed in some embodiments herein. Control signaling may also sometimes be referred to as signaling, or control information, or configuration information, or configuration. In some cases, control signaling may be indicated dynamically, such as in a physical layer in a control channel. An example of dynamically indicated control signaling is information sent in physical layer control signaling, such as downlink control information (DCI). Alternatively, control signaling may sometimes be semi-statically indicated, such as in RRC signaling or in a MAC control element (CE). A dynamic indication may be an indication in a low layer, such as physical layer / layer 1 signaling (e.g., in DCI), rather than an indication in a high layer (e.g., rather than in RRC signaling or MAC CE). A semi-static indication may be an indication in semi-static signaling. Semi-static signaling used herein may refer to non-dynamic signaling, such as high layer signaling, RRC signaling, and / or MAC CE. Dynamic signaling used herein may refer to dynamic signaling, such as physical layer control signaling sent in the physical layer, such as DCI.

[0131] The air interface typically includes multiple components and associated parameters that together specify how to send and / or receive transmissions over a wireless communication link between two or more communication devices. For example, the air interface may include one or more components that define a waveform, frame structure, multiple access scheme, protocol, coding scheme, and / or modulation scheme for transmitting information (e.g., data) over a wireless communication link. The wireless communication link may support a link between a radio access network and a user equipment (e.g., a "Uu" link), and / or the wireless communication link may support a link between devices, such as a link between two user equipments (e.g., a "sidelink"), and / or the wireless communication link may support a link between a non-terrestrial (NT) communication network and a user equipment (UE). The following are some examples of the above components:

[0132] The waveform component may specify the shape and form of the transmitted signal. Waveform options may 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-domain windowed OFDM, filter bank multicarrier (FBMC), universal filtered multicarrier (UFMC), generalized frequency division multiplexing (GFDM), wavelet packet modulation (WPM), faster than Nyquist (FTN)

[0133] waveform and low peak to average power ratio waveform (PAPR WF).

[0134] ● The frame structure component may specify the configuration of a frame or a group of frames. The frame structure component may indicate one or more of the time, frequency, pilot signature, code, or other parameters of a frame or a group of frames. More details of the frame structure will be discussed below.

[0135] ● The multiple access scheme component 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-

[0136] Multiple access technologies include: scheduled access and non-scheduled access, also known as unlicensed access; non-orthogonal multiple access and orthogonal multiple access, such as using dedicated channel resources (e.g., not shared between multiple communication devices); contention-based shared channel resources and non-contention-based shared channel resources; and cognitive radio-based access.

[0137] ● A hybrid automatic repeat request (HARQ) protocol component may specify how transmission and / or retransmission is performed. Non-limiting examples of transmission and / or retransmission mechanism options include those that specify a scheduled data pipe size, a signaling mechanism for transmission and / or retransmission, and a retransmission mechanism.

[0138] The Coding and Modulation components may specify how the transmitted information may be encoded / decoded and modulated / demodulated for transmission / reception purposes. Coding may 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 may simply refer to a constellation (including, for example, a modulation technique and order), or more specifically to various types of advanced modulation methods, such as hierarchical modulation and low PAPR modulation.

[0139] In some embodiments, the air interface may 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 frequency bands above 6 GHz (e.g., millimeter waves) for authorized and unauthorized access. For example, the flexibility of the configurable air interface provided by extensible system parameters (numerology) and symbol duration can support transmission parameter optimization for different spectrum segments and different services / devices. For another example, a unified air interface can be self-contained in the frequency domain, and the self-contained design in the frequency domain can support more flexible radio access network (RAN) slicing by sharing channel resources between different services in frequency and time.

[0140] Frame structure

[0141] The frame structure is a feature of the physical layer of wireless communication, which defines the time domain signal transmission structure to achieve the timing reference and timing alignment of the basic time domain transmission unit, etc. Wireless communication between communication devices can be carried out on the time and frequency resources controlled by the frame structure. The frame structure is sometimes also called the radio frame structure.

[0142] Depending on the frame structure and / or the frame configuration in the frame structure, it is possible to perform frequency division duplex (FDD) communication and / or time division duplex (TDD) communication and / or full duplex (FD) communication. FDD communication means that transmissions in different directions (such as uplink and downlink) are performed on different frequency bands. TDD communication means that transmissions in different directions (such as uplink and downlink) are performed for different durations. FD communication means that transmission and reception are performed on the same time-frequency resources, that is, the device can send and receive on the same frequency resources at the same time.

[0143] An example of a frame structure is the frame structure in long-term evolution (LTE), which has the following specifications: the duration of each frame is 10 ms; each frame has 10 subframes, each subframe has a duration of 1 ms; each subframe includes two time slots, each time slot has a duration of 0.5 ms; each time slot is used to transmit 7 OFDM symbols (assuming a regular 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 finite length option); in TDD, the switching interval between uplink and downlink needs to be an integer multiple of the OFDM symbol duration.

[0144] Another example of a frame structure is the frame structure in new radio (NR), which has the following specifications: multiple subcarrier spacings are supported, each subcarrier spacing corresponds to a corresponding system parameter; the frame structure depends on the system parameters, but in any case, the frame length is set to 10ms, each frame consists of ten subframes, each subframe is 1ms; the time slot is defined as 14 OFDM symbols; the time slot length depends on the system parameters. For example, the NR frame structure of the conventional CP 15kHz subcarrier spacing ("system parameter 1") and the NR frame structure of the conventional CP 30kHz subcarrier spacing ("system parameter 2") are different. For a subcarrier spacing of 15kHz, the time slot length is 1ms; for a subcarrier spacing of 30kHz, the time slot length is 0.5ms. The NR frame structure can have greater flexibility than the LTE frame structure.

[0145] Another example of a frame structure is an exemplary flexible frame structure, such as for 6G networks or higher. In a flexible frame structure, a symbol block may be defined as a minimum duration that may be scheduled in the flexible frame structure. A symbol block may be a transmission unit having an optional redundant portion (e.g., a CP portion) and an information (e.g., data) portion. An OFDM symbol is an example of a symbol block. Alternatively, a symbol block may be referred to as a symbol. Embodiments of a flexible frame structure include different parameters that may be configurable, such as a frame length, a subframe length, a symbol block length, etc. In some embodiments of the flexible frame structure, a non-exhaustive list of possible configurable parameters includes:

[0146] (1) Frame: The frame length does not need to be limited to 10ms. The frame length can be configurable and vary over time.

[0147] 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. The frame length can be more than one possible value and is configured based on the application scenario. For example, an autonomous driving vehicle may require relatively fast initial access, in which case the frame length corresponding to the autonomous driving vehicle application can be set to 5ms. For another example, a smart meter on a house may not require fast initial access, in which case the frame length corresponding to the smart meter application can be set to 20ms.

[0148] (2) Subframe duration: Subframes may or may not be defined in the flexible frame structure, depending on the implementation. For example, a frame may be defined to include time slots but not subframes. In a frame that defines subframes, the duration of the subframe may be configurable, for example, for time domain alignment. For example, the length of a subframe may be configured to be 0.1 ms or 0.2

[0149] ms or 0.5ms or 1ms or 2ms or 5ms, etc. In some embodiments, if a subframe is not needed in a specific scenario, the subframe length may be defined to be the same as the frame length, or may not be defined.

[0150] (3) Time slot configuration: Time slots may or may not be defined in a flexible frame structure, depending on the implementation. In a frame with time slots defined, the definition of the time slot (for example, in terms of duration and / or number of symbol blocks) may 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 may be sent to the UE in a broadcast channel or one or more common control channels. In other embodiments, the time slot configuration may be UE-specific, in which case the time slot configuration information may be transmitted in a UE-specific control channel. In some embodiments, the time slot configuration signaling may be sent together with the frame configuration signaling and / or the subframe configuration signaling. In other embodiments, the time slot configuration may be sent independently of the frame configuration signaling and / or the subframe configuration signaling. In general, the time slot configuration may be system-wide, base station-wide, UE group-wide, or UE-specific.

[0151] (4) Subcarrier spacing (SCS): SCS is a parameter in the scalable system parameters, which can support a range of SCS between 15KHz and 480KHz. SCS can vary with spectrum frequency and / or maximum UE speed to minimize the impact of Doppler shift and phase noise. In some examples, there can be separate transmit frames and receive frames, and the SCS of symbols in the receive frame structure can be configured independently of the SCS of 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 the SCS of each receive frame. If the SCS between the receive frame and the transmit frame is different, the difference does not have to be scaled by a factor of 2, for example, if an inverse discrete Fourier transform (IDFT) is used instead of a fast Fourier transform (FFT) to achieve a more flexible symbol duration. Other examples of frame structures can be used with different SCSs.

[0152] (5) Flexible transmission duration of basic transmission unit: The basic transmission unit can be a symbol block (also referred to as a symbol), which generally includes a redundant part (referred to as a 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, and the CP length can change with changes in frames, or with changes in frame groups, or with changes in subframes, or with changes in time slots, or dynamically with changes in 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 the following:

[0153] Channel conditions (e.g., multipath delay, Doppler) and / or delay requirements and / or available time. For another example, the symbol block length may be adjusted to fit the available time in a frame.

[0154] (6) Flexible switching gap: A frame may include a downlink part for downlink transmission sent by a base station and an uplink part for uplink transmission sent by a UE. There may be a gap between each uplink part and the downlink part, which is called a switching gap.

[0155] The switching gap length (duration) may be configurable. The switching gap duration may be fixed within a frame or may be flexible within a frame. The switching gap duration may change with changes in frames, changes in frame groups, changes in subframes, changes in time slots, or changes dynamically with changes in scheduling.

[0156] Cell / Carrier / Bandwidth part (BWP) / Occupied bandwidth

[0157] Equipment such as base stations can cover cells. Wireless communication with devices can be carried out on one or more carrier frequencies. A carrier frequency is called a carrier. Alternatively, a carrier can be called a component carrier (CC). A carrier can be characterized by its bandwidth and a reference frequency (e.g., the center frequency, the lowest frequency, or the highest frequency of the carrier). A carrier can be on a licensed spectrum or an unlicensed spectrum. Additionally or alternatively, wireless communication with a device can be carried out on one or more bandwidth parts (BWP). For example, a carrier can have one or more BWPs. More generally, wireless communication with a device can be carried out on a spectrum. A spectrum can include one or more carriers and / or one or more BWPs.

[0158] The cell may include one or more downlink resources and optionally one or more uplink resources, or the cell may include one or more uplink resources and optionally one or more downlink resources, or the cell may include both one or more downlink resources and one or more uplink resources. For example, the cell may include only one downlink carrier / BWP, or only one uplink carrier / BWP, or multiple downlink carriers / BWP, or multiple uplink carriers / BWP, or one downlink carrier / BWP and one uplink carrier / BWP, or one downlink carrier / BWP and multiple uplink carriers / BWP, or multiple downlink carriers / BWP and one uplink carrier / BWP, or multiple downlink carriers / BWP and multiple uplink carriers / BWP. In some embodiments, alternatively or additionally, the cell may include one or more sidelink resources, and the one or more sidelink resources include sidelink transmission and reception resources.

[0159] A BWP is a set of contiguous or non-contiguous frequency subcarriers on a carrier, or a set of contiguous or non-contiguous frequency subcarriers on multiple carriers, or a set of non-contiguous or contiguous frequency subcarriers, and may have one or more carriers.

[0160] 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 continuous BWPs, and so on. In other embodiments, a BWP may have one or more carriers, for example, a BWP may have a bandwidth of 40 MHz and consist of two adjacent continuous carriers, each of which has a bandwidth of 20 MHz. In some embodiments, a BWP may include non-continuous spectrum resources consisting of a non-continuous plurality of carriers, wherein the first carrier of the non-continuous plurality of carriers may be in the mmW frequency band, the second carrier may be in the low frequency band (e.g., the 2 GHz frequency band), the third carrier (if present) may be in the THz frequency band, and the fourth carrier (if present) may be in the visible light frequency band. The resources belonging to a BWP in a carrier may be continuous or non-continuous. In some embodiments, a BWP has non-continuous spectrum resources on a carrier.

[0161] Wireless communication can be performed over an occupied bandwidth. The occupied bandwidth can be defined as the width of the frequency band such that below a lower frequency limit and above an upper frequency limit, the average power transmitted is each equal to a specified percentage β / 2 of the total average transmitted power, for example, β / 2 is 0.5%.

[0162] The carrier, BWP or occupied bandwidth may be signaled dynamically by a network device (e.g., a base station) (e.g., in physical layer control signaling such as downlink control information (DCI)) or semi-statically (e.g., in radio resource control (RRC) signaling or in a medium access control (MAC) layer), or may be predefined based on an application scenario, or determined by the UE as a function of other parameters known to the UE, or may be fixed, for example, by a standard.

[0163] Artificial intelligence (AI) and / or machine learning (ML)

[0164] The number of new devices in future wireless networks is expected to grow exponentially, and the functionality of devices is expected to become increasingly diverse. In addition, many new applications and use cases are expected to emerge, with more diverse quality of service requirements than 5G applications / use cases. These will generate new key performance indicators (KPIs) for future wireless networks (e.g., 6G networks) that may be extremely challenging. AI technologies, such as ML technologies (e.g., deep learning), have been introduced into telecommunication applications with the goal of improving system performance and efficiency.

[0165] In addition, antenna and bandwidth capabilities continue to advance, making it possible to carry out more and / or better communications over wireless links. In addition, the field of computer architecture and computing power continues to advance, for example, through the introduction of general-purpose graphics processing units (GP-GPUs). Future generations of communication devices may have greater computing and / or communication capabilities than previous generations, which may support the use of AI to implement air interface components. Future generations of networks may also obtain more accurate and / or new information (compared to previous generations of networks) that can be the basis of input to AI models, such as the physical speed / rate at which the device moves, 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. In order to obtain sensing information, the TRP may send a signal to a target object (e.g., a potential UE), and 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 based on the reflection of the signal. Positioning information is sometimes referred to as positioning, and can be obtained in a variety of ways, such as positioning reports from the UE (e.g., reports of the UE's GPS coordinates), using a positioning reference signal (PRS), using the above-mentioned sensing, tracking and / or predicting the location of the device.

[0166] AI technology (including ML technology) can be applied to communications, including AI-based communications in the physical layer and / or AI-based communications in the MAC layer. For the physical layer, AI communication can be intended to optimize component design and / or improve algorithm performance. For example, AI can be used to implement the following aspects: channel coding, channel modeling, channel estimation, channel decoding, modulation, demodulation, MIMO, waveforms, multiple access, physical layer element parameter optimization and update, beamforming, tracking, sensing and / or positioning, etc. For the MAC layer, AI communication can be intended to use AI capabilities to learn, predict and / or make decisions, so as to use better strategies and / or optimal solutions to solve complex optimization problems, for example, to optimize functions in the MAC layer. For example, AI can be applied to implement: intelligent TRP management, intelligent beam management, intelligent channel resource allocation, intelligent power control, intelligent spectrum utilization, intelligent MCS, intelligent HARQ strategy and / or intelligent transmission / reception mode adaptation, etc.

[0167] In some embodiments, the AI ​​architecture may include multiple nodes, wherein the multiple nodes may be organized in one of two modes, centralized and distributed, both of which may be deployed in an access network, a core network, or an edge computing system or a third-party network. Centralized training and computing architectures are subject to potentially large communication overhead and strict user data privacy protection. Distributed training and computing architectures may include several frameworks, such as distributed machine learning and federated learning. In some embodiments, the AI ​​architecture may include an intelligent controller that may execute as a single agent or multiple agents based on joint optimization or individual optimization. New protocols and signaling mechanisms are required so that the corresponding interface links can be personalized through custom parameters to meet specific needs, while minimizing signaling overhead and maximizing the overall system spectrum efficiency through personalized AI technology.

[0168] 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 training and normal operating modes, and for measurement and feedback to accommodate different possible measurements and information that may need to be fed back, depending on the implementation.

[0169] AI Training

[0170] Reference again Figure 1 and Figure 2 , the embodiments of the present disclosure may be used to implement AI training involving two or more communication devices in the communication system 100. For example, Figure 5Four EDs are shown in communication with a network device 452 in a communication system 100 according to one embodiment. The four EDs are shown as respective different UEs, hereinafter referred to as UEs 402, 404, 406, and 408. However, an ED does not necessarily have to be a UE.

[0171] Network device 452 is part of a network (e.g., wireless access network 120). Network device 452 can be deployed in an access network, a core network, an edge computing system, or a third-party network, depending on the implementation. Network device 452 can be a T-TRP or a server (or a part thereof). In one example, network device 452 can be a T-TRP 170 or a NT-TRP 172 (or implemented therein). In another example, network device 452 can be a T-TRP controller and / or a NT-TRP controller that can manage T-TRP 170 or NT-TRP 172. In some embodiments, the components of network device 452 can be distributed. For example, if network device 452 is part of a T-TRP serving UE 402, 404, 406, and 408, UE 402, 404, 406, and 408 can communicate directly with network device 452. Alternatively, UEs 402, 404, 406, and 408 may communicate with network device 452 through one or more intermediate components such as T-TRP and / or NT-TRP. For example, network device 452 may send and / or receive information (e.g., control signaling, data, training sequence, etc.) to / from one or more of UEs 402, 404, 406, and 408 through a backhaul link and a wireless channel between network device 452 and UEs 402, 404, 406, and 408.

[0172] As described above, each UE 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 panels). 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.

[0173] For each UE 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 many components and associated parameters that together specify how to send and / or receive transmissions over the wireless medium.

[0174] Figure 5The processor 210 of the UE in the embodiment implements one or more air interface components on the UE side. The air interface component configures and / or implements transmission and / or reception on the air interface. Examples of air interface components are described herein. The air interface component may be located in the physical layer, for example, a channel encoder (or decoder) that implements the encoding component of the air interface of the UE, and / or a modulator (or demodulator) that implements the modulation component of the air interface of the UE, and / or a waveform generator that implements the waveform component of the air interface of the UE, and so on. The air interface component may be in a higher layer such as the MAC layer (or a part thereof), for example, a module that implements channel prediction / tracking, and / or a module that implements a retransmission protocol (for example, a HARQ protocol component that implements the air interface of the UE), and so on. The processor 210 also directly executes (or controls the UE to execute) the UE-side operations described herein.

[0175] 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 (e.g., T-TRP) to implement one or more air interface components on the network side. The implementation of the air interface components on the network side of one UE may be different from that of another UE. The processor 454 directly executes (or controls the network components to execute) the network side operations described herein.

[0176] Processor 454 may be implemented by one or more processors that are the same or different and are used to execute instructions stored in memory (e.g., memory 456). Alternatively, some or all of processor 454 may be implemented using a dedicated circuit such as a programmed FPGA, GPU, or ASIC. Memory 456 may be implemented by volatile and / or non-volatile storage devices. Any suitable type of memory may be used, such as RAM, ROM, hard disk, optical disk, processor cache, etc.

[0177] The input / output device 458 supports interaction with other devices by receiving (input) and sending (output) information. In some embodiments, the input / output device 458 can be implemented by a transmitter and / or a receiver (or transceiver) and / or one or more interfaces (e.g., a wired interface to an internal network or to the Internet, etc.). In some implementations, the input / output device 458 can be implemented by a network interface, which may be implemented as a network interface card (NIC) and / or a computer port (e.g., a physical socket to which a plug or cable is connected) and / or a network socket, etc., depending on the implementation.

[0178] The network device 452 and the UE 402 have the capability to implement one or more AI-enabled processes. Figure 5In the embodiment of the present invention, the network device 452 and the UE 402 include ML modules 410 and 460, respectively. 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. Figure 5 , ML module 410 is shown as being within processor 210, and ML module 460 is shown as being within processor 454. For example, ML modules 410 and 460 execute one or more AI / ML algorithms to perform one or more AI-enabled processes (e.g., AI-enabled link adaptation) to optimize the communication link between the network and UE 402.

[0179] ML modules 410 and 460 may be implemented using an AI model. The term AI model may refer to a computer algorithm for receiving defined input data and outputting defined reasoning data, wherein the parameters of the algorithm (e.g., weights) may be updated and optimized by training (e.g., using a training data set, or using data collected from real life). The AI ​​model may be implemented using one or more neural networks (e.g., including deep neural networks (DNNs), recurrent neural networks (RNNs), convolutional neural networks (CNNs), and combinations thereof) and using various neural network architectures (e.g., autoencoders, generative adversarial networks, etc.). Various techniques may be used to train the AI ​​model in order to update and optimize its parameters. For example, back propagation is a common technique for training DNNs, wherein a loss function is calculated between the reasoning data generated by the DNN and a target output (e.g., a reference truth data). The gradient of the loss function is calculated relative to the parameters of the DNN, and the parameters are updated using the calculated gradient (e.g., using a gradient descent algorithm) to minimize the loss function.

[0180] In some embodiments, the AI ​​model covers a neural network for machine learning. A neural network consists of multiple computing units (which may also be referred to as neurons) arranged in one or more layers. The process of receiving input at an input layer and generating output at an 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 multidimensional array) and performs calculations to generate an output (which may have a different dimension than the input). The calculations performed by the layer typically involve applying (e.g., multiplying) a set of weights (also called coefficients) to the input. 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), which may be referred to as inner layers or hidden layers. For example, Fig. 6AAn example of a neural network 600 including one input layer, one output layer, and two hidden layers is depicted. 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 computational unit in a neural network is a neuron, such as Fig. 6A As shown in 650. Figure 6B An example of a neuron 650 that can be used as a building block of the neural network 600 is shown. Figure 6B As shown, in this example, neuron 650 takes vector x as input and performs a dot product with the associated weight vector w. The final output z of the neuron is the result of the activation function f() on the dot product. Various neural networks can be designed with various architectures (e.g., various numbers of layers, each layer performing various functions).

[0181] Neural networks are 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 a neural network includes forward propagation of input data samples to generate output values ​​(also called predicted output values ​​or inference output values), and comparing the generated output values ​​with known or desired target values ​​(e.g., reference truth values). A loss function is defined to quantitatively represent the difference between the generated output value and the target value, and the goal of training a neural network is to minimize the loss function. Backpropagation is an algorithm for training a neural network. Backpropagation is used to adjust (also called update) the values ​​of parameters (e.g., weights) in a neural network so that the calculated loss function becomes smaller. Backpropagation involves calculating the gradient of the loss function relative to the parameters to be optimized, and updating the parameters using a gradient algorithm (e.g., gradient descent) 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 have been trained. The trained neural network can be deployed (or executed) to generate inference output data from the input data. In some embodiments, training of the neural network may continue even after the neural network has been deployed, such that parameters of the neural network may be repeatedly updated with the latest training data.

[0182] Reference again Figure 5In 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-understandable meaning (e.g., the information may be intermediate data generated during execution of an ML algorithm). Additionally or alternatively, the information exchanged may not be 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 one or more parameters to the UE 402 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 a neural network executed by the ML module 410 on the UE side in order to attempt to optimize one or more aspects of modulation and / or coding for communications between the UE 402 and the T-TRP or NT-TRP.

[0183] In some embodiments, UE 402 itself may implement AI, such as performing learning, while in other embodiments, UE 402 itself may not be able to perform learning, but can operate in conjunction with network-side AI implementation, such as by receiving from the network the configuration of an AI model (e.g., a neural network or other ML algorithm) implemented by ML module 410, and / or assisting other devices (e.g., network devices or other UEs with AI capabilities) in training AI models (e.g., neural networks or other ML algorithms) by providing requested measurements or observations. For example, in some embodiments, UE 402 itself may not perform learning or training, but UE 402 may receive trained configuration information of an ML model determined by network device 452 and execute the model.

[0184] although Figure 5 The example in assumes AI / ML capabilities on the network side, but it is possible that the network itself does not perform training / learning, but the UE itself may perform learning / training using a dedicated training signal sent from the network. In other embodiments, end-to-end (E2E) learning can be implemented by the UE and the network device 452.

[0185] Using AI, for example, by implementing the above-mentioned AI model, various processes (e.g., link adaptation) can be AI-enabled. The following describes some examples of possible AI / ML training processes and over-the-air information exchange processes between devices for facilitating AI-enabled processes during the training phase according to embodiments of the present disclosure.

[0186] Reference again Figure 5For wireless federated learning (FL), the network device 452 can initialize the global AI / ML model implemented by the ML module 460. Figure 5 A group of UEs such as the four UEs 402, 404, 406, and 408 shown are sampled, and the global AI / ML model parameters are broadcast to the UEs. Each of the UEs 402, 404, 406, and 408 can then initialize its local AI / ML model using the global AI / ML model parameters and update (train) its local AI / ML model using its own data. Each of the UEs 402, 404, 406, and 408 can then 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 the UEs 402, 404, 406, and 408 and update the global AI / ML model. The above process is an iteration of the FL-based AI / ML model training process. The network device 452 and the UEs 402, 404, 406, and 408 perform multiple iterations until the AI / ML model has sufficiently converged to meet one or more training objectives / criteria, and the AI / ML model is finally completed.

[0187] Figure 7 An exemplary star topology used in a conventional FL process is shown. Figure 7 In the star topology shown, there is a master node / device 700 (e.g., server, TRP, BS) and several client nodes / devices 701 to 708. The master node 700 initializes the master AI / ML model, samples a group of user client nodes 701 to 708, and distributes the master model (mastermodel, MM) to the client nodes 701 to 708. The client nodes 701 to 708 then initialize the local AI / ML models M1 to M8 of the client nodes using the master AI / ML model, and update (train) their own local AI / ML models M1 to M8 using the client nodes' own data. The client nodes 701 to 708 then report the updated local AI / ML models of the client nodes to the master node 700. The master node 700 aggregates the updated AI / ML models M1 to M8 reported by the client nodes 701 to 708, and updates the master AI / ML model.

[0188] This traditional FL-based AI / ML model training process may have some limitations. In the star topology used in the FL-based AI / ML model training process, the master node 700 needs to collect a large amount of training data sets (e.g., gradients updated by the client nodes) from each of the client nodes 701 to 708. In the traditional FL-based AI training process, the structure of the AI / ML models (e.g., the master AI / ML model and the local AI / ML model) needs to be the same. However, since different client nodes may support different AI / ML model structures, data heterogeneity may be a problem in the traditional FL-based AI training process.

[0189] Various aspects of the present disclosure provide solutions to overcome the above limitations, such as specific methods and devices for learning artificial intelligence or machine learning (AI / ML) models on self-organizing topologies. Through self-organizing topologies, heterogeneous AI / ML model aggregation is supported. The methods and devices shown in the present disclosure can be implemented and deployed with consideration of the computing power of each network node and the potential inward / outward expansion of AI / ML model learning and reasoning.

[0190] The AI / ML model learning scheme shown in this disclosure is different from the traditional FL-based AI / ML model training scheme. Figure 8 The differences between the traditional FL-based AI / ML model training process and the AI / ML model learning scheme of the present disclosure are shown.

[0191] like Figure 8 As shown, the master model 810 does not move in the FL-based AI / ML model training scheme. Alternatively, a large amount of training data set 815 (e.g., gradients updated by client nodes) can be sent from the client node 811a to the master node / device 811, which updates the master AI / ML model 810 based on the collected data. In other words, in the FL-based AI / ML model training scheme, the training data set 815 follows the AI / ML model 810. In contrast, in the AI / ML model learning scheme of the present disclosure, the AI / ML model 820 follows the training data set maintained at the corresponding network edge or node / device 821. In other words, the AI / ML model 820 (e.g., a deep neural model) is routed in the network edge or node / device 821, and the deep learning algorithm is executed locally at each node / device 821.

[0192] In FL-based AI / ML model training, there is only one master node / device 811 that collects AI / ML models and aggregates the collected AI / ML models. Only this master node / device 811 can be called an aggregation node. On the other hand, in the AI / ML model learning scheme of the present disclosure, any node in the topology can collect AI / ML models and aggregate the collected AI / ML models, so there can be several aggregation nodes (e.g., node 821). In fact, unlike FL-based AI / ML model training in which nodes 811 and 811a are arranged in a star topology, multiple nodes / devices 821 are arranged in a self-organizing topology, such as Figure 8 shown.

[0193] Another difference between FL-based AI / ML model training and the AI / ML model learning scheme of the present disclosure relates to the type of data / information exchanged between nodes. In FL-based AI / ML model training, a large amount of training data set 815 (e.g., gradients updated by client nodes) can be exchanged between client nodes 811a and master nodes / devices 811. In the AI / ML model learning scheme of the present disclosure, AI / ML models 820 are exchanged between nodes 821.

[0194] In some embodiments, any node in the network may be a node that collects one or more AI / ML models and aggregates the collected AI / ML models to generate a new or updated AI / ML model (e.g., Figure 8 Node 821 in ). This node may be a type 1 node. A type 1 node is a node for collecting multiple AI / ML models and aggregating the collected AI / ML models to generate a new or updated AI / ML model (e.g., a first type AI / ML model). This node may also be referred to as an aggregate AI / ML node or an aggregate node below or elsewhere in this disclosure. Each aggregate node (or type 1 node) may be communicatively and / or operably connected to or associated with another aggregate node (or type 1 node). Each aggregate node (or type 1 node) may be communicatively and / or operably connected to or associated with one or more other nodes. The one or more other nodes may be type 2 nodes. A type 2 node is a node for training an AI / ML model (e.g., a second type AI / ML model, also referred to as a local AI / ML model) with a set of training data (e.g., local data) without an aggregation operation. A type 2 node may be referred to as a basic AI / ML node or a basic node below or elsewhere in this disclosure. In some embodiments, some aggregation nodes (or type 1 nodes) may not be associated with base nodes (or type 2 nodes) (ie, only with other aggregation nodes (or type 1 nodes)).

[0195] In some embodiments, the AI / ML model can be transmitted between nodes (e.g., BS, TRP, UE) according to the self-organizing topology. In this regard, the training gradients are not transmitted or shared between nodes at the network level (e.g., aggregation nodes and base nodes). This is because the number of training gradients is huge and it is difficult to control the accuracy during the calculation process.

[0196] Fig. 9 An exemplary self-organizing topology 900 for AI / ML model training of an embodiment of the present disclosure is shown. In the self-organizing topology 900, there are multiple nodes 901 to 908 that are communicatively and operably associated with each other. Each node holds AI / ML models M1 to M8 respectively. Any node in the topology 900 can be an aggregation node.

[0197] An aggregation node in topology 900 may perform one or several operations.

[0198] An aggregation node may receive an AI / ML model from another node and send the AI / ML model without performing an aggregation operation. For example, aggregation node 901 may receive an AI / ML model from aggregation node 902 and simply pass the AI / ML model to node 903. The AI / ML model passed to node 903 is not aggregated at node 901.

[0199] The aggregation node may receive one or more AI / ML models from one or more other nodes, and send the AI / ML models with or without aggregation. For example, the aggregation node 901 may receive AI / ML models from nodes 902 and 903, and aggregate the received AI / ML models M2 and M3. The aggregation node 901 then sends the aggregated AI / ML model to another node 904.

[0200] The aggregation node can generate a new or updated AI / ML model and pass the new AI / ML model to another node. For example, the aggregation node 901 generates an updated AI / ML model M1 and passes the updated AI / ML model M1 to the aggregation node 905.

[0201] The aggregation node may aggregate the AI / ML model generated by the aggregation node and another AI / ML model received by the aggregation node from another node. For example, the aggregation node 901 generates its own AI / ML model M1 and also receives the AI / ML model M2 from the node 902. Then, the aggregation node 901 aggregates the AI / ML model M1 and the AI / ML model M2 and sends the aggregated AI / ML model to the node 906.

[0202] In a self-organizing topology, there are multiple nodes that perform communication and computing functions. These nodes can receive at least one computing model (AI / ML model) and send at least one computing model (AI / ML model). The received model and the sent model are not the same entity, but can have the same or different data, information and / or neural network structure. The received and sent AI / ML models can include multiple parameters, such as one or more parameters related to a graph model, a parameter model, a model table, a model algorithm, or a database.

[0203] In terms of learning AI / ML models, nodes in the network can be divided into two types of nodes. Fig. 10A and Fig. 10B Two different types of nodes for learning AI / ML models according to an embodiment of the present disclosure are shown. Fig. 10A A node of a type 2 node or a basic AI / ML node or a basic node is shown. Fig. 10B Another type of node is shown as a Type 2 node or an aggregated AI / ML node or an aggregated node. A base node may also be referred to as a local AI / ML node or a local node.

[0204] The basic AI / ML node can receive one or more general AI / ML models from other nodes (e.g., aggregation nodes) and train its own (customized) AI / ML model. The customized AI / ML model may be referred to as a local AI / ML model. The basic node uses its own AI / ML model training algorithm to train its own local AI / ML model with the assistance of the (current) general AI / ML model (e.g., information related to distillation and / or dilation) received from one or more aggregation nodes. In some embodiments, the neural network (NN) structure of the local AI / ML model may be the same as the NN structure of at least some of the received general AI / ML models. In some embodiments, the NN structure of the local AI / ML model may be different from the NN structure of all received general AI / ML models. When the training of the local AI / ML model is completed, the basic node sends the local AI / ML model to one or more other nodes (e.g., aggregation nodes).

[0205] The aggregation node may receive one or more AI / ML models from other nodes in the network. The received AI / ML models include one or more local AI / ML models from the associated base node and / or one or more common AI / ML models from other aggregation nodes. Fig. 10BThe aggregated AI / ML node shown receives a local AI / ML model, but in some embodiments, the aggregated node may not receive any local AI / ML models from the base node (i.e., only general AI / ML models from other aggregated nodes). In some embodiments, some or all of the received AI / ML models have the same NN structure. In some embodiments, all received AI / ML models have different NN structures. After collecting AI / ML models from other nodes, the aggregated node aggregates the collected AI / ML models (general AI / ML models, local AI / ML models) to generate a new or updated general AI / ML model. The aggregated node then sends the new / updated AI / ML model to one or more other nodes.

[0206] In some embodiments, the generic AI / ML model is a predefined or preconfigured AI / ML model.

[0207] As mentioned above, the general AI / ML model and the local AI / ML model involved in the AI / ML model learning process have different NN structures. For example, when the aggregation node receives a general AI / ML model and three local AI / ML models, the received AL / ML models may have completely different NN structures. The general AI / ML model can be a 6-layer deep neural network (DNN) model, the first local AI / ML model can be a 4-layer DNN model, the second local AI / ML model can be an 8-layer DNN model, and the third local AI / ML model can be a convolutional neural network (CNN). The NN structure can be regarded as the model structure of the AI / ML model.

[0208] The self-organizing topology of the present disclosure includes a connection between an aggregation node and one or more base nodes and / or a connection between multiple aggregation nodes.

[0209] Any network device / apparatus may operate as an aggregation node in the network. An aggregation node may be communicatively and / or operably connected to one or more base nodes. Such connection may instruct the aggregation node to collect local AI / ML models from the associated base nodes. However, it should be noted that in some embodiments, some aggregation nodes may not be communicatively and / or operably connected to any base node. Such aggregation nodes may not collect local AI / ML models, but only receive general AI / ML models from other aggregation nodes.

[0210] It should be noted that any aggregation node and / or any basic node may be, for example, a UE, a relay device, a base station (BS), a transmission and reception point (TRP), an edge device, an edge computing system or a network system.

[0211] Fig.11A 1 shows an exemplary self-organizing topology 1100 of an embodiment of the present disclosure. Topology 1100 includes connections between aggregation nodes and connections between aggregation nodes and base nodes. With respect to connections between aggregation nodes, each of aggregation nodes 901 to 908 is communicatively and operably connected to their neighboring aggregation nodes. For example, aggregation node 901 may be communicatively and operably connected to aggregation nodes 902 and 908, such as Fig.11A In this example, the connection indicates that the aggregation node 901 can receive the general AI / ML model M from the aggregation node 908. j And send its own general AI / ML model M1 to the aggregation node 902.

[0212] Regarding the connection between the aggregation node and the base node, in this example, each of the aggregation nodes 901 to 908 is communicatively and operably connected to the base node. Specifically, the aggregation node 901 is communicatively and operably connected to the base nodes 901a and 901b. These connections indicate that the aggregation node 901 can receive the general AI / ML model M from the node 908. j901a and 901b, and collect local AI / ML models from base nodes 901a and / or 901b. Similarly, aggregation node 902 is communicatively and operably connected to base nodes 902a and 902b. These connections indicate that aggregation node 902 can send the general AI / ML model M1 received from node 901 to base nodes 902a and 902b, and collect local AI / ML models from base nodes 902a and / or 902b. In addition, aggregation node 903 is communicatively and operably connected to base nodes 903a and 903b. These connections indicate that aggregation node 903 can send the general AI / ML model M2 received from node 902 to base nodes 903a and 903b, and collect local AI / ML models from base nodes 903a and / or 903b. In addition, aggregation node 904 is communicatively and operably connected to base nodes 904a and 904b. These connections indicate that the aggregation node 904 can send the general AI / ML model M3 received from the node 903 to the base nodes 904a and 904b, and collect local AI / ML models from the base nodes 904a and / or 904b. In addition, the aggregation node 905 is communicatively and operably connected to the base nodes 905a and 905b. These connections indicate that the aggregation node 905 can send the general AI / ML model M4 received from the node 904 to the base nodes 905a and 905b, and collect local AI / ML models from the base nodes 905a and / or 905b. In addition, the aggregation node 906 is communicatively and operably connected to the base nodes 906a and 906b. These connections indicate that the aggregation node 906 can send the general AI / ML model M5 received from the node 905 to the base nodes 906a and 906b, and collect local AI / ML models from the base nodes 906a and / or 906b. In addition, the aggregation node 907 is communicatively and operably connected to the base nodes 907a and 907b. These connections indicate that the aggregation node 907 can send the general AI / ML model M6 received from the node 906 to the base nodes 907a and 907b, and collect local AI / ML models from the base nodes 907a and / or 907b. In addition, the aggregation node 908 is communicatively and operably connected to the base nodes 908a and 908b. These connections indicate that the aggregation node 908 can send the general AI / ML model M6 received from the node 906 to the base nodes 907a and 907b. i Send to base nodes 908a and 908b, and collect local AI / ML models from base nodes 908a and / or 908b.

[0213] Although Fig.11A Each base node in the embodiment is communicatively and operably connected to only one aggregation node, but in some embodiments, the base node can be communicatively and operably connected to more than one aggregation node. Fig.11AEach aggregation node in the system is communicatively and operably connected to some base nodes, but in some embodiments, some aggregation nodes may not be communicatively and / or operably connected to any base node. Such aggregation nodes may not collect local AI / ML models, but only receive general AI / ML models from other aggregation nodes.

[0214] Assuming that there is an aggregation node i in the network, the aggregation node i can receive information related to the corresponding AI / ML model from one or more aggregation nodes that are communicatively and operably connected to the aggregation node i. Each aggregation node that sends information related to its AI / ML model can be referred to as a previous aggregation node connected to the aggregation node i. The aggregation node i can send information related to its AI / ML model to one or more other aggregation nodes that are communicatively and operably connected to the aggregation node i. Each aggregation node that receives information related to the AI / ML model of the aggregation node i can be referred to as a next aggregation node connected to the aggregation node i. Fig.11A In the network shown, for aggregation node 902 , the previous aggregation node is aggregation node 901 , and the next aggregation node is aggregation node 903 .

[0215] Although Fig.11A The topology 1100 in FIG. 1 shows that each aggregation node has one previous aggregation node and one next aggregation node, but it should be noted that an aggregation node may have one or more previous aggregation nodes and one or more next aggregation nodes, such as Fig. 11B shown. Fig. 11B is another exemplary self-organizing topology 1150 that illustrates connections between aggregation nodes such that each aggregation node has one or more previous aggregation nodes and one or more next aggregation nodes.

[0216] refer to Fig. 11B , aggregation node 901 sends information related to its AI / ML model M1 to aggregation nodes 902, 903, 904, and 905. Therefore, aggregation nodes 902 to 905 are the next aggregation nodes of aggregation node 901. Aggregation node 901 receives information related to their AI / ML model M1 from aggregation nodes 907 and 908. i and M jTherefore, aggregation nodes 907 and 908 are the previous aggregation nodes of aggregation node 901. Similarly, aggregation nodes 902, 903, 904, and 905 send information related to their AI / ML models M2, M3, M4, and M5 to aggregation nodes 907 and 908. Therefore, aggregation nodes 907 and 908 are the next aggregation nodes of aggregation nodes 902, 903, 904, and 905. Aggregation nodes 902, 903, 904, and 905 receive information related to AI / ML model M1 from aggregation node 901. Therefore, aggregation node 901 is the previous aggregation node of 902, 903, 904, and 905. In addition, aggregation nodes 907 and 908 send information related to AI / ML model M2, M3, M4, and M5 to aggregation nodes 907 and 908. Therefore, aggregation nodes 907 and 908 are the next aggregation nodes of aggregation nodes 902, 903, 904, and 905. Aggregation nodes 902, 903, 904, and 905 receive information related to AI / ML model M1 from aggregation node 901. Therefore, aggregation node 901 is the previous aggregation node of 902, 903, 904, and 905. i and M j The relevant information is sent to the aggregation node 901. Therefore, the aggregation node 901 is the next aggregation node of the aggregation nodes 907 and 908. The aggregation nodes 907 and 908 receive information related to their AI / ML models M2, M3, M4 and M5 from the aggregation nodes 902, 903, 904 and 905. Therefore, the aggregation nodes 902, 903, 904 and 905 are the previous aggregation nodes of 907 and 908.

[0217] As described above, aspects of the present disclosure provide methods for learning artificial intelligence or machine learning (AI / ML) models on self-organizing topologies. Some methods for learning AI / ML models on self-organizing topologies are shown below and elsewhere in the present disclosure.

[0218] Fig.12 An exemplary process of AI / ML model aggregation on a self-organizing topology according to an embodiment of the present disclosure is shown.

[0219] At step 1210, node 1200 i From node 1200 i–1 Receive AI / ML model CM i–1 For example, node 1200 i–1 The AI / ML model CM can be transmitted via broadcast, multicast or unicast signaling (e.g., via broadcast / multicast / unicast RRC, MAC-CE or DCI) or the interface between TRP / BS i–1 . Node 1200 i and 1200 i–1 Can be an aggregation node, AI / ML model CM i–1 Can be a general AI / ML model. Node 1200 i–1 It can be node 1200 i The previous aggregation node.

[0220] In step 1220, upon receiving the AI / ML model CM i–1 Afterwards, node 1200 i AI / ML Model CM i–1 Send to associated node 12011...1201 M For example, node 1200 i The AI / ML model CM can be transmitted via broadcast, multicast or unicast signaling (e.g., via broadcast / multicast / unicast RRC, MAC-CE or DCI) or the interface between TRP / BS i–1 . Node 12011...1201 M May be communicatively and operatively connected to a node 1200 in a self-organizing topology i The basic node. After receiving node 1200 i AI / ML model CM sent i–1 After that, associate nodes 12011...1201 M Train their corresponding AI / ML models. Each of these AI / ML models can be associated with nodes 12011...1201 M A local AI / ML model of an associated node in. Associated nodes 12011...1201 M Each node in the system can use its own training data set, its own AI / ML algorithm (e.g., AI / ML model training algorithm) or a slave node 1200. i Received AI / ML Model CM i–1 In some embodiments, the AI / ML model can be trained based on at least one of the slave nodes 1200. i Received AI / ML Model CM i–1 This training is performed through transfer learning, knowledge distillation, and / or knowledge expansion. AI / ML Model CM i–1 and associated nodes 12011...1201 M The associated AI / ML models can have the same input and / or output types. However, the AI / ML model CM i–1 and associated nodes 12011...1201 M The associated AI / ML model may have a different neural network (NN) structure. In other words, the associated nodes 12011…1201 M The associated AI / ML model does not need to have the same CM as the AI / ML model i–1 The NN structure of is equivalent to the NN structure of .

[0221] At step 1230, node 1200 iSend for associated nodes 12011...1201 M Collect indicators of AI / ML models. For example, node 1200 i The indicator may be transmitted via broadcast, multicast or unicast signaling (e.g., via broadcast / multicast / unicast RRC, MAC-CE or DCI) or an interface between TRP / BS. This indicator may be referred to as a model collection indicator. Node 1200 i To the associated node 12011...1201 M Send model collection indicator to request to associate with node 12011...1201 M Reports related to the corresponding associated AI / ML models.

[0222] In some embodiments, associated nodes 12011 . . . 1201 M The reports related to the associated AI / ML models are associated with nodes 12011…1201 M Based on the model collection indicator. In such embodiments, the model collection indicator can be a dynamic indicator or an event-triggered indicator. In some embodiments where the model collection indicator is a dynamic indicator, node 1200 i At node 1200 i The preconfigured response time is from the associated node 12011...1201 M Receiving and associating nodes 12011...1201 M In some embodiments where the model collection indicator is an event trigger indicator, the associated nodes 12011 . . . 1201 M Send a report related to the corresponding associated AI / ML model when an event is triggered. The event can be associated with nodes 12011...1201 M The training of the corresponding associated AI / ML model is completed or the associated node 12011...1201 M The performance of the corresponding associated AI / ML model of the associated AI / ML model exceeds a specific performance metric. The specific performance metric is predetermined according to at least one of accuracy or precision. If the performance of one of the associated AI / ML models does not exceed the predetermined performance metric (e.g., the accuracy is lower than the predetermined accuracy metric), then the node 1200 may not be generated and / or not sent to the node 1200. i Send reports related to this AI / ML model (e.g., no local AL / ML model report transmission).

[0223] In some embodiments, the associated nodes 12011 . . . 1201 MEach node in step 1235 optionally sends a confirmation indicator for the transmission of its associated AI / ML model. For example, the confirmation indicator can be transmitted via PUCCH or PUSCH, or a sidelink channel, or an interface between TRP / BS. A positive confirmation indicator (e.g., ACK) indicates that a report related to the associated AI / ML model will be sent. A negative confirmation indicator (e.g., NACK) indicates that a report related to the associated AI / ML model will not be sent (e.g., the associated AI / ML model will not be reported). In some embodiments, the confirmation indicator can be sent before sending the report related to the associated AI / ML model. In some other embodiments, the confirmation indicator is included in the report related to the associated AI / ML model.

[0224] In step 1240, the associated nodes 12011 . . . 1201 M To node 1200 i Send and associate nodes 12011...1201 M In some embodiments, associated nodes 12011 . . . 1201 M The report related to the associated AI / ML model may include at least one of the following: M The information related to the corresponding associated AI / ML model and the associated nodes 12011…1201 M The information related to the training data of the corresponding associated AI / ML model, or the information related to the associated nodes 12011…1201 M The information related to the performance of the corresponding associated AI / ML model may include associated nodes 12011...1201 M The NN structure (NN algorithm, width, depth, etc.) and / or one or more AI / ML parameters (weights, biases, activation functions, etc.) of the corresponding associated AI / ML model. The information related to the training data of the corresponding associated AI / ML model may include the amount (size) of the training data and / or information related to the distribution of the training data. The information related to the performance of the corresponding associated AI / ML model may include accuracy, precision, recall, loss information (e.g., average cross entropy loss), i (or other aggregation nodes or BSs) configured with validation / test data sets. It should be noted that in some embodiments, information related to the training data of the corresponding associated AI / ML model and / or information related to the performance of the corresponding associated AI / ML model can assist the aggregation operation of the aggregation node. For example, the training data information and / or the performance information can be used to determine the aggregation weight of the AI / ML model.

[0225] At step 1250, at node 1200 iReceived and associated with node 12011...1201 M After the corresponding AI / ML model-related reports are associated, node 1200 i Generate AI / ML model CM based on received reports i AI / ML Model CM i It can be an updated general AI / ML model. AI / ML model CM i–1 and CM i Can have the same NN structure. In some embodiments, node 1200 i Generate AI / ML model CM using aggregation algorithms (e.g., model distillation, model inflation) i .

[0226] In some embodiments, node 1200 i You can generate your own AI / ML model. The AI / ML model can be a local AI / ML model. The AI / ML model can be generated without being associated with nodes 12011...1201 M The corresponding associated AI / ML model related reports are generated.

[0227] At step 1260, node 1200 i To node 1200 i+1 Send AI / ML model CM i . Node 1200 i and 1200 i+1 Can be an aggregation node, AI / ML model CM i Can be a general AI / ML model. Node 1200 i+1 It can be node 1200 i The next aggregation node of AI / ML model CM i May include one or more AI / ML model parameters (e.g., weights, biases), but does not include any i In other words, the AI / ML model CM i The information related to the NN structure may not be obtained from node 1200. i Send to node 1200 i+1 , because AI / ML model CM i NN structures can be pre-configured and / or AI / ML models CM i The NN structure can be found at node 1200 i+1 In some embodiments, the data is sent to node 1200. i AI / ML Model CM i It can be a complete AI / ML model or a partial AI / ML model.

[0228] As described above, the associated nodes 12011 . . . 1201 M The association of AI / ML models is not limited to having CM with AI / ML models i–1 and AI / ML model CM i In other words, heterogeneous AI / ML model aggregation can be achieved by executing the method for learning AI / ML models shown in the present disclosure.

[0229] Fig.13 An example of heterogeneous AI / ML model aggregation in an embodiment of the present disclosure is shown. The AI / ML model aggregation process can be Fig.12 The AI / ML model aggregation process shown is similar. Fig.13 , node 1200 i From node 1200 i–1 Receive AI / ML model CM i–1 AI / ML Model CM i–1 Can be node 1200 i–1 According to node 1200 i–1 Generated based on the received local model report. AI / ML model CM i–1 A NN with 4 layers may be included.

[0230] Node 1200 i The received AI / ML model CM i–1 Sent to associated nodes 12011, 12012 and 1201 respectively M . Associate nodes 12011, 12012, and 1201 M Can be a basic node. Associated nodes 12011, 12012 and 1201 M Each node in the i–1 Generate your own AI / ML model. Associate nodes 12011, 12012, and 1201 M Each node in the can use its own training data set and its own AI / ML algorithm (e.g., AI / ML model training algorithm) to generate its own local AI / ML model. The local AI / ML model LM1 generated by the associated node 12011 includes a NN with 5 layers, and the local AI / ML model LM2 generated by the associated node 12012 includes a NN with 3 layers. M LM3 including convolutional neural network (CNN) can be generated through transfer learning.

[0231] After receiving reports related to local AI / ML models LM1, LM2, and LM3, node 1200 i Perform aggregation operations (e.g., model distillation, model expansion) to generate an AI / ML model CM from LM1, LM2, and LM3 i Since each of LM1, LM2, and LM3 is important for generating AI / ML model CM i May have different importance (meaning), so when the local AI / ML models are aggregated, each of LM1, LM2, and LM3 can be weighted by W0, W1, and W2, respectively. W0, W1, and W2 can indicate the importance of LM1, LM2, and LM3. Generated AI / ML model CM i It can be an updated general AI / ML model. CM i The NN structure can be combined with AI / ML model CM i–1 In this case, the AI / ML model CM i and AI / ML model CM i–1 Both include a NN with 4 layers.

[0232] As above and Fig.13 As shown, AI / ML model CM i And the local AI / ML models LM1, LM2 and LM3 can have different neural network (NN) structures. The local AI / ML models LM1, LM2 and LM3 do not need to have the same NN structure as the general AI / ML model to generate the AI / ML model CM i In other words, heterogeneous AI / ML model aggregation on a self-organizing topology is achieved.

[0233] In some embodiments, the aggregation node receives one or more heterogeneous AI / ML models from one or more associated base nodes that are communicatively and operably connected to the aggregation node. The aggregation node may distill and / or expand the received heterogeneous AI / ML models. The aggregation may then aggregate the distilled and / or expanded AI / ML models to generate a new or updated general AI / ML model. In some embodiments, the aggregation node may obtain the mean of the distilled AI / ML model and / or the mean of the expanded AI / ML model. After generating the general AI / ML model, the aggregation node may send the new / updated general AI / ML model to one or more other nodes that may be the next aggregation node.

[0234] As described above, if the NN structure of the associated base node of the aggregation node is different from the NN structure of the general AI / ML model of the aggregation node, the aggregation node may distill and / or expand the AI / ML model. Otherwise, if the NN structure of the associated base node of the aggregation node is the same as the NN structure of the general AI / ML model of the aggregation node, the distillation and / or expansion operation may not be performed. Distillation and / or expansion may be part of the aggregation operation of the aggregation node. Distillation may include generating a smaller AI / ML model from the AI / ML model received from the node, and expansion may include generating a larger AI / ML model from the AI / ML model received from the node. Therefore, the distillation of the aggregation node may include generating a general AI / ML by the aggregation node that is smaller than the AI / ML model received from other nodes (e.g., a local AI / ML model received from the associated base node). Similarly, the expansion of the aggregation node may include generating a general AI / ML model by the aggregation node that is larger than the AI / ML model received from other nodes (e.g., a local AI / ML model received from the associated base node).

[0235] To illustrate distillation and / or inflation, when a first AI / ML model is larger than a second AI / ML model, the first AI / ML model may have a greater number of floating-point operations than the second model, a greater number of total parameters than the second model, a greater number of trainable parameters than the second model, a required buffer size greater than the second model, a width greater than the second model, a depth greater than the second model, or any combination of the foregoing. Similarly, to illustrate distillation and / or inflation, when a first AI / ML model is smaller than a second AI / ML model, the first AI / ML model may have a fewer number of floating-point operations than the second model, a fewer number of total parameters than the second model, a fewer number of trainable parameters than the second model, a required buffer size less than the second model, a width less than the second model, a depth less than the second model, or any combination of the foregoing.

[0236] In some embodiments, the model collection indicator sent by the aggregation node (e.g., node 1200 i The model collection indicator sent may be a distillation indicator, an expansion indicator, or a distillation and expansion indicator. With these indicators, the aggregation node may perform operations (eg, aggregation operations) according to its computing capabilities.

[0237] When the aggregation node sends a distillation indicator to one or more associated base nodes, each associated base node may send a confirmation indicator to the aggregation node. If the associated base node has a local AI / ML model that is greater than or equal to the reference AI / ML model, the associated base node may send a positive confirmation indicator (e.g., ACK) and its own local AI / ML model to the aggregation node. On the other hand, if the associated base node has a local AI / ML model that is smaller than the reference AI / ML model, the associated base node may send a negative confirmation indicator (e.g., NACK) to the aggregation node, but not its own local AI / ML model. Here, the reference AI / ML model may be a general AI / ML model or an AI / ML model with a specific NN structure (width, depth). The specific NN structure may be indicated by the aggregation node to each associated base node, or by a BS or network system.

[0238] In a similar manner, when the aggregation node sends an expansion indicator to one or more associated base nodes, each associated base node may send a confirmation indicator to the aggregation node. If the associated base node has a local AI / ML model that is less than or equal to the reference AI / ML model, the associated base node may send a positive confirmation indicator (e.g., ACK) and its own local AI / ML model to the aggregation node. On the other hand, if the associated base node has a local AI / ML model that is greater than the reference AI / ML model, the associated base node may send a negative confirmation indicator (e.g., NACK) to the aggregation node, but not its own local AI / ML model. Here, the reference AI / ML model may be a general AI / ML model or an AI / ML model with a specific NN structure (width, depth). The specific NN structure may be indicated by the aggregation node to each associated base node, or by a BS or network system.

[0239] When the aggregation node sends a distillation and expansion indicator to one or more associated base nodes, the associated base nodes have no restrictions on sending their local AI / ML models. In other words, when receiving the distillation and expansion indicator, all associated base nodes can send their local AI / ML models. Each associated base node can optionally send a positive confirmation indicator (e.g., ACK).

[0240] Fig.14 An exemplary AI / ML model distillation process in AI / ML model aggregation on a self-organizing topology according to an embodiment of the present disclosure is shown. Fig.14 In the AI / ML model distillation process shown, the aggregation node may not generate its own AI / ML model (e.g., a local AI / ML model of the aggregation node).

[0241] At step 1410, node 1200i To node 12011...1201 M Each node in sends a model collection indicator. The model collection indicator is a distillation indicator. Node 1200 i Can be an aggregation node, AI / ML model CM i–1 Can be a general AI / ML model. Node 12011...1201 M May be communicatively and operatively connected to a node 1200 in a self-organizing topology i In some embodiments, node 1200 i The associated nodes 12011 . . . 1201 may be sent via broadcast, multicast or unicast signaling. M Send model collection indicator.

[0242] It should be noted that node 1200 i , 1200 j and 12011……1201 M The nodes 1200 may be communicatively and operably connected in an ad hoc topology by using a side link for device-to-device (D2D) communication, by a network device (e.g., BS, TRP, etc.), or by an interface between network devices. i , 1200 j and 12011……1201 M The information (e.g., model collection indicator, AI / ML model) may be exchanged by using D2D communication, by sidelinks of network devices (e.g., BS, TRP), or by interfaces between network devices. i and associated nodes 12011...1201 M In one example of a UE, these nodes can exchange information (e.g., model collection indicator, AI / ML model) by using a side link of D2D communication or through a network device (e.g., BS, TRP). The nodes can exchange information through the network device so that the aggregation node 1200 i Send information to network devices (e.g., BS, TRP), and the network devices send information to basic nodes 12011...1201 M Send the information. At node 1200 i It is BS and associated with nodes 12011...1201 M In another example of a UE, the node 1200 i and associated nodes 12011...1201 M The exchange information is transmitted as uplink (UL) and / or downlink (DL). iand associated nodes 12011...1201 M In another example where the two aggregation nodes are BSs, the nodes can exchange information via the interface between the BSs. i and 1200 j The information exchange between the two devices can be performed in a similar manner by using a side link of device-to-device (D2D) communication, through a network device (e.g., BS, TRP, etc.) or through an interface between network devices.

[0243] In step 1415, node 1200i optionally sends a message to associated nodes 12011 . . . 1201 M The reference AI / ML model is sent, for example, by broadcast, multicast, or unicast signaling. The reference AI / ML model may include information indicating a NN structure of the reference AI / ML model. The information indicating the NN structure of the reference AI / ML model may include at least one of a NN algorithm, width, depth, complexity, floating point operations, total parameters, trainable parameters, or a required buffer size. In some embodiments, the reference AI / ML model may be predefined and / or preconfigured as a generic AI / ML model.

[0244] In step 1420, the associated basic nodes 12011 . . . 1201 in the self-organizing topology M Send confirmation indicator and / or associate node 12011...1201 M Reports related to the associated AI / ML model. For associated nodes 12011…1201 M For each node in , if the associated node generates a local AI / ML model that is larger than the reference AI / ML model, the associated node sends a signal to node 1200 i Send a positive confirmation indicator (e.g., ACK) and its own local AI / ML model (i.e., AI / ML model M1...M M On the other hand, if the associated node generates a local AI / ML model that is smaller than the reference AI / ML model, the associated node sends a request to node 1200. i Send a negative acknowledgement indicator (e.g., NACK) but not your own local AI / ML model.

[0245] In some embodiments, associated with AI / ML models M1...M M The relevant report may include at least one of the following: M Related information, and corresponding AI / ML models M1…M M Information related to the training data, or the corresponding associated AI / ML model M1…M MInformation related to the performance of the associated nodes 12011...1201 M The associated AI / ML model M1…M M Related reports are further illustrated above or elsewhere in this disclosure.

[0246] In step 1430, at node 1200 i Received and associated with node 12011...1201 M The corresponding associated AI / ML model M1…M M After the relevant report, node 1200 i Associate the corresponding local AI / ML models M1…M M Distill into one or more distilled AI / ML models M C1 ……M CM . Node 1200 i The distillation operation can be performed based on the model collection indicator. One or more distilled AI / ML models M C1 ……M CM It can be a general AI / ML model with a fixed size and predefined input and output parameters.

[0247] It should be noted that only local AI / ML models that are larger than the reference AI / ML model can be associated with nodes 12011...1201 M Send to node 1200 i In other words, node 1200 i Only associated local AI / ML models that are larger than the reference AI / ML model may be distilled into distilled AI / ML models.

[0248] At step 1440, at node 1200 i Get the distilled AI / ML model M C1 ……M CM Afterwards, node 1200 i Get the distilled AI / ML model M C1 ……M CM The mean of the AI / ML model is sent to node 1200. j The average AI / ML model can be considered as node 1200 i Generated general AI / ML model M C . Node 1200 j It can be node 1200 i The next aggregation node.

[0249] Fig.15 An exemplary AI / ML model expansion process in AI / ML model aggregation on a self-organizing topology according to an embodiment of the present disclosure is shown. Fig.15 During the AI / ML model expansion process shown, the aggregation node may not generate its own AI / ML model (e.g., a local AI / ML model of the aggregation node).

[0250] At step 1510, node 1200 i To node 12011...1201 M Each of the nodes sends a model collection indicator. The model collection indicator is an expansion indicator. Node 1200 i Can be an aggregation node, AI / ML model CM i–1 Can be a general AI / ML model. Node 12011...1201 M May be communicatively and operatively connected to a node 1200 in a self-organizing topology i In some embodiments, node 1200 i The associated nodes 12011 . . . 1201 may be sent via broadcast, multicast or unicast signaling. M Send model collection indicator.

[0251] It should be noted that node 1200 i , 1200 j and 12011……1201 M The connections between are shown above or elsewhere in this disclosure, for example, in conjunction with step 1410. It should also be noted that node 1200 i , 1200 j and 12011……1201 M Communications (eg, information exchange) between may be performed in a similar manner as described above or elsewhere in the present disclosure, such as in conjunction with step 1410 .

[0252] At step 1515, node 1200 i Optionally, associate nodes 12011...1201 M The reference AI / ML model is sent, for example, by broadcast, multicast, or unicast signaling. The reference AI / ML model may include information indicating a NN structure of the reference AI / ML model. The information indicating the NN structure of the reference AI / ML model may include at least one of a NN algorithm, width, depth, complexity, floating point operations, total parameters, trainable parameters, or a required buffer size. In some embodiments, the reference AI / ML model may be predefined and / or preconfigured as a generic AI / ML model.

[0253] In step 1520, the associated basic nodes 12011 . . . 1201 in the self-organizing topology M Send confirmation indicator and / or associate node 12011...1201 MReports related to the associated AI / ML model. For associated nodes 12011…1201 M For each node in the associated node, if the associated node generates a local AI / ML model that is smaller than the reference AI / ML model, the associated node sends a signal to node 1200. i Send a positive confirmation indicator (e.g., ACK) and its own local AI / ML model (i.e., AI / ML model M1...M M On the other hand, if the associated node generates a local AI / ML model that is larger than the reference AI / ML model, the associated node sends a request to node 1200. i Send a negative acknowledgement indicator (e.g., NACK) but not your own local AI / ML model.

[0254] In some embodiments, associated with AI / ML models M1...M M The relevant report may include at least one of the following: M Related information, and corresponding AI / ML models M1…M M Information related to the training data, or the corresponding associated AI / ML model M1…M M Information related to the performance of the associated nodes 12011...1201 M The associated AI / ML model M1…M M Related reports are further illustrated above or elsewhere in this disclosure.

[0255] At step 1530, at node 1200 i Received and associated with node 12011...1201 M The corresponding associated AI / ML model M1…M M After the relevant report, node 1200 i Associate the corresponding local AI / ML models M1…M M Inflate to one or more inflated AI / ML models M C1 ……M CM . Node 1200 i The expansion operation may be performed based on the model collection indicator. One or more expanded AI / ML models M C1 ……M CM It can be a general AI / ML model with a fixed size and predefined input and output parameters.

[0256] It should be noted that only local AI / ML models that are larger than the reference AI / ML model can be associated with nodes 12011...1201 M Send to node 1200 iIn other words, node 1200 i Only associated local AI / ML models that are smaller than the reference AI / ML model may be inflated into the inflated AI / ML model.

[0257] At step 1540, at node 1200 i Get the inflated AI / ML model M C1 ……M CM Afterwards, node 1200 i Get the inflated AI / ML model M C1 ……M CM The mean of the AI / ML model is sent to node 1200. j The average AI / ML model can be considered as node 1200 i Generated general AI / ML model M C . Node 1200 j It can be node 1200 i The next aggregation node.

[0258] Fig.16 An exemplary AI / ML model distillation and expansion process in AI / ML model aggregation on a self-organizing topology according to an embodiment of the present disclosure is shown. Fig.16 During the AI / ML model expansion process shown, the aggregation node may not generate its own AI / ML model (e.g., a local AI / ML model of the aggregation node).

[0259] At step 1610, node 1200 i To node 12011...1201 M Each of the nodes sends a model collection indicator. The model collection indicator is a distillation and expansion indicator. Node 1200 i Can be an aggregation node, AI / ML model CM i–1 Can be a general AI / ML model. Node 12011...1201 M May be communicatively and operatively connected to a node 1200 in a self-organizing topology i In some embodiments, node 1200 i The associated nodes 12011 . . . 1201 may be sent via broadcast, multicast or unicast signaling. M Send model collection indicator.

[0260] It should be noted that node 1200 i , 1200 j and 12011……1201 M The connections between are shown above or elsewhere in this disclosure, for example, in conjunction with step 1410. It should also be noted that node 1200 i, 1200 j and 12011……1201 M Communications (eg, information exchange) between may be performed in a similar manner as described above or elsewhere in the present disclosure, such as in conjunction with step 1410 .

[0261] In step 1620, the associated basic nodes 12011 . . . 1201 in the self-organizing topology M Send confirmation indicator and / or associate node 12011...1201 M The report related to the associated AI / ML model. Since the model collection indicator is a distillation and expansion indicator, for the associated basic node 12011...1201 M Local AI / ML model M1…M M Therefore, the associated nodes 12011...1201 M The corresponding local AI / ML model M1…M M The relevant report is sent to node 1200 i . Each associated node 12011...1201 M A positive acknowledgement indicator (eg, ACK) may also be sent along with the report.

[0262] In some embodiments, associated with AI / ML models M1...M M The relevant report may include at least one of the following: M Related information, and corresponding AI / ML models M1…M M Information related to the training data, or the corresponding associated AI / ML model M1…M M Information related to the performance of the associated nodes 12011...1201 M The associated AI / ML model M1…M M Related reports are further illustrated above or elsewhere in this disclosure.

[0263] In step 1630, at node 1200 i Received and associated with node 12011...1201 M The corresponding associated AI / ML model M1…M M After the relevant report, node 1200 i Perform distillation or expansion operations. For example, if you associate a local AI / ML model with a base node (i.e., associate nodes 12011…1201 M AI / ML model M1…M M One) is greater than the reference AI / ML model, then node 1200 iThe associated AL / ML model may be distilled into a distilled AI / ML model (i.e., a distilled AI / ML model M) according to the model collection indicator. C1_蒸馏 ……M CM_蒸馏 On the other hand, if the local AI / ML model of the associated basic node (i.e., the associated nodes 12011...1201 M AI / ML model M1…M M 1) Smaller than the reference AI / ML model, 1200 nodes i The associated AL / ML model (i.e., associated nodes 12011 . . . 1201 M AI / ML model M1…M M 1) is expanded into an expanded AI / ML model (i.e., the expanded AI / ML model M C1_膨胀 ……M CM_膨胀 One). Distilled AI / ML model M C1_蒸馏 ……M CM_蒸馏 and the inflated AI / ML model M C1_膨胀 ……M CM_膨胀 Each model in can be a general AI / ML model with a fixed size and predefined input and output parameters.

[0264] At step 1640, at node 1200 i Get the distilled AI / ML model M for all associated local AI / ML models C1_蒸馏 ……M CM_蒸馏 and / or inflated AI / ML model M C1_膨胀 ……M CM_膨胀 Afterwards, node 1200 i Get the mean of the distilled AI / ML model and the mean of the expanded AI / ML model respectively. Then, node 1200 i To node 1200 j Send the mean distilled AI / ML model and / or the mean inflated AI / ML model. The mean distilled AI / ML model and / or the mean inflated AI / ML model can be considered as the general AI / ML model M C_蒸馏 and the general AI / ML model M C_膨胀 . Node 1200 j It can be node 1200 i The next aggregation node.

[0265] In some embodiments, the aggregation node can generate its own AI / ML model while still performing the various operations shown above (e.g., receiving heterogeneous AI / ML models from associated base nodes, distilling and / or expanding multiple (local) AI / ML models into distilled and / or expanded AI / ML models, averaging the distilled and / or expanded AI / ML models to generate a new general AI / ML model, and sending the aggregated general AI / ML model to the next aggregation node). The aggregation node's own AI / ML model can be a local AI / ML model. The aggregation node's own AI / ML model can be used for the distillation and / or expansion process of the AI / ML model.

[0266] Fig.17 Another exemplary AI / ML model distillation process in the AI / ML model aggregation on the self-organizing topology of the embodiment of the present disclosure is shown. Fig.17 In the AI / ML model distillation process shown, the aggregation node generates its own AI / ML model (e.g., the local AI / ML model of the aggregation node).

[0267] Fig.17 Steps 1710 to 1730 of the exemplary AI / ML model distillation process shown are substantially similar to Fig.14 Steps 1410 to 1430 of an exemplary AI / ML model distillation process are shown.

[0268] At step 1740, node 1200 i Generate your own AI / ML model M i . Node 1200 i Your own AI / ML model M i It can be a local AI / ML model or a general AI / ML model. i The NN structure of your own AI / ML model is different from that of the general AI / ML model (in Fig.12 Step 1210 in the previous aggregation node 1200 i–1 Received Common AI / ML Model CM i–1 ) NN structure, then node 1200 i A distillation or expansion operation is performed at step 1745. For example, if node 1200 i Your own AI / ML model M i If it is larger than the reference AI / ML model, then the node is 1200 i Node 1200 can be collected according to the model indicator i On the other hand, if the node 1200 i If your AI / ML model is smaller than the reference AI / ML model, then node 1200i Node 1200 can be collected according to the model indicator i The AI / ML model itself is expanded into an expanded AI / ML model. The distilled or expanded AI / ML model can be a general AI / ML model M Ci However, if node 1200 i The NN structure of your own AI / ML model is different from the general AI / ML model (in Fig.12 Step 1210 in the previous aggregation node 1200 i–1 Received Common AI / ML Model CM i–1 ) have the same NN structure, then the AI / ML model M i Can be considered as AI / ML model M Ci .

[0269] At step 1750, node 1200 i Get the distilled AI / ML model M C1 ……M CM and the (distilled or expanded) AI / ML model M Ci Then, node 1200 i Send the mean AI / ML model to node 1200 j The average AI / ML model can be considered as node 1200 i Generated general AI / ML model M C . Node 1200 j It can be node 1200 i The next aggregation node.

[0270] Fig.18 Another exemplary AI / ML model expansion process in AI / ML model aggregation on a self-organizing topology according to an embodiment of the present disclosure is shown. Fig.18 In the AI / ML model expansion process shown, the aggregation node can generate its own AI / ML model (e.g., a local AI / ML model of the aggregation node).

[0271] Fig.18 Steps 1810 to 1830 of the exemplary AI / ML model expansion process shown are substantially similar to Fig.15 Steps 1510 to 1530 of an exemplary AI / ML model inflation process are shown.

[0272] At step 1840, node 1200 i Generate your own AI / ML model M i . Node 1200 i Your own AI / ML model M iIt can be a local AI / ML model or a general AI / ML model. i The NN structure of your own AI / ML model is different from that of the general AI / ML model (in Fig.12 Step 1210 in the previous aggregation node 1200 i–1 Received Common AI / ML Model CM i–1 ) NN structure, then node 1200 i A distillation or expansion operation is performed at step 1845. For example, if node 1200 i Your own AI / ML model M i If it is larger than the reference AI / ML model, then the node is 1200 i Node 1200 can be collected according to the model indicator i On the other hand, if the node 1200 i If your AI / ML model is smaller than the reference AI / ML model, then node 1200 i Node 1200 can be collected according to the model indicator i The AI / ML model itself is expanded into an expanded AI / ML model. The distilled or expanded AI / ML model can be a general AI / ML model M Ci However, if node 1200 i The NN structure of your own AI / ML model is different from the general AI / ML model (in Fig.12 Step 1210 in the previous aggregation node 1200 i–1 Received Common AI / ML Model CM i–1 ) have the same NN structure, then the AI / ML model M i Can be considered as AI / ML model M Ci .

[0273] At step 1850, node 1200 i Get the inflated AI / ML model M C1 ……M CM and the (distilled or expanded) AI / ML model M Ci Then, node 1200 i Send the mean AI / ML model to node 1200 j The average AI / ML model can be considered as node 1200 i Generated general AI / ML model M C . Node 1200 j It can be node 1200 i The next aggregation node.

[0274] Fig.19 An exemplary AI / ML model distillation and expansion process in AI / ML model aggregation on a self-organizing topology according to an embodiment of the present disclosure is shown. Fig.19 In the AI / ML model expansion process shown, the aggregation node can generate its own AI / ML model (e.g., a local AI / ML model of the aggregation node).

[0275] Fig.19 Steps 1910 to 1930 of the exemplary AI / ML model expansion process shown are substantially similar to Fig.16 Steps 1610 to 1630 of an exemplary AI / ML model distillation and expansion process are shown.

[0276] At step 1940, node 1200 i Generate your own AI / ML model M i . Node 1200 i Your own AI / ML model M i It can be a local AI / ML model or a general AI / ML model. i The NN structure of your own AI / ML model is different from that of the general AI / ML model (in Fig.12 Step 1210 in the previous aggregation node 1200 i–1 Received Common AI / ML Model CM i–1 ) NN structure, then node 1200 i A distillation or expansion operation is performed at step 1945. For example, if node 1200 i Your own AI / ML model M i If it is larger than the reference AI / ML model, then the node is 1200 i Node 1200 can be collected according to the model indicator i On the other hand, if the node 1200 i If your AI / ML model is smaller than the reference AI / ML model, then node 1200 i Node 1200 can be collected according to the model indicator i The AI / ML model itself is expanded into an expanded AI / ML model. The distilled or expanded AI / ML model can be a general AI / ML model M Ci However, if node 1200 i The NN structure of your own AI / ML model is different from the general AI / ML model (in Fig.12 Step 1210 in the previous aggregation node 1200 i–1 Received Common AI / ML Model CM i–1 ) have the same NN structure, then the AI / ML model Mi Can be considered as AI / ML model M Ci .

[0277] At step 1950, node 1200 i Get the distilled AI / ML model M C1_蒸馏 ……M CM_蒸馏 and the (distilled or expanded) AI / ML model M Ci The mean of the distilled AI / ML model M C1_蒸馏 ……M CM_蒸馏 and the (distilled or expanded) AI / ML model M Ci The mean of can be considered as the general AI / ML model M C_蒸馏 . Node 1200 i Also obtain the inflated AI / ML model M C1_膨胀 ……M CM_膨胀 and the (distilled or expanded) AI / ML model M Ci The mean of the inflated AI / ML model M C1_膨胀 ……M CM_膨胀 and the (distilled or expanded) AI / ML model M Ci The mean of can be considered as the general AI / ML model M C_膨胀 When node 1200 i Get AI / ML model M C_蒸馏 and / or AI / ML model M C_膨胀 When node 1200 i AI / ML model M C_蒸馏 and / or AI / ML model M C_膨胀 Send to node 1200 j . Node 1200 j It can be node 1200 i The next aggregation node.

[0278] refer to Fig.12 and Figures 14 to 19 The AI / ML model aggregation process shown is similar to node 1200. i One or more associated basic nodes 12011...1201 M From node 1200 i Receive general AI / ML model CM i–1 However, in some other embodiments, the node 1200 i One or more associated basic nodes 12011...1201 M From the previous aggregation node 1200 i–1 Receive general AI / ML model CM i–1In such embodiments, one or more base nodes 12011 . . . 1201 M It is not possible to associate with the aggregation node 1200 i Or network devices (e.g., BS, TRP) or network clouds receive the general AI / ML model CM i–1 .

[0279] Fig. 20 Another exemplary process of AI / ML model aggregation on the self-organizing topology of the present disclosure is shown. In this process, nodes 12011...1201 M May be communicatively and operatively connected to a node 1200 in a self-organizing topology i The basic node is similar to Fig.12 However, nodes 12011...1201 M Also communicatively and operatively connected to node 1200 i–1 . Node 1200 i–1 It can be an aggregation node 1200 i The previous aggregation node.

[0280] In step 2010, node 1200 i–1 To the associated node 12011...1201 M One or more of the AI / ML models CM i–1 In some embodiments, node 1200 i–1 Optionally, to node 1200 i Send AI / ML model CM i–1 .

[0281] The remaining operations of step 2010 are similar to those of the above and Fig.12 Step 1220 shown. For example, from node 1200 i–1 Receive AI / ML model CM i–1 After that, nodes 12011...1201 M Each node in the system is based on its own training data set, its own AI / ML algorithm (e.g., AI / ML model training algorithm) or slave node 1200. i–1 Received AI / ML Model CM i–1 to train your own local AI / ML model. Fig.12 The process is similar, associating nodes 12011...1201 M The local AI / ML model does not need to have the same CM as the AI / ML model i–1 The NN structure of is basically similar to the NN structure of .

[0282] Fig. 20Steps 2020 to 2050 of the exemplary AI / ML model expansion process shown are substantially similar to Fig.12 Steps 1230 to 1260 of the exemplary process of the AI / ML model aggregation process are shown. It should be noted that although nodes 12011...1201 M From node 1200 i–1 Receive AI / ML model CM i–1 , but nodes 12011...1201 M From node 1200 i The receiving model collects indicators and sends them to node 1200 i Send its own corresponding local AI / ML model M1…M M .

[0283] In some embodiments, assuming that one AI / ML model training iteration includes receiving a general AI / ML model from a previous aggregation node, receiving a local AI / ML model from a communicatively and operably connected base node, and sending a new / updated general AI / ML model to a next aggregation node, the aggregation node (e.g., Fig.11A Nodes 901 to 908 in the above diagram may join one or more AI / ML model training iterations. In other words, the aggregation node may iteratively execute the method for learning the AI / ML model (e.g., the AI / ML model aggregation process) described in the present disclosure.

[0284] For example, refer to Fig.11A , AI / ML model training is performed at aggregation node 901. Then, AI / ML model training is performed at each of the other aggregation nodes 902 to 908 in a predetermined order. When AI / ML model training is completed at aggregation node 908, each aggregation node in the self-organizing topology 1100 has performed one AI / ML model training iteration.

[0285] In another example, still referring to Fig.11A , AI / ML model training starts at aggregation node 901. Then, AI / ML model training is performed at each of the other aggregation nodes 902 to 908 in a predetermined order. When AI / ML model training is completed at aggregation node 908, aggregation node 908 sends an updated general AI / ML model to aggregation node 901. This is one round of AI / ML model training. The next round of AI / ML model training starts at aggregation node 901 in a similar manner. AI / ML model training can be temporarily stopped or ended at any aggregation node in the self-organizing topology 1100. In this way, each aggregation node can perform one or more AI / ML model training iterations.

[0286] According to some aspects of the present disclosure, AI / ML models can be routed in wireless networks. In this way, there is no need to transmit large amounts of data (e.g., training data) between various network devices (network nodes, nodes, base stations, TRPs, etc.) and user devices (e.g., UEs). In addition, heterogeneous AI / ML capabilities can be supported in various network devices and user devices. Any of the various network devices and user devices can operate as an aggregation node.

[0287] According to some aspects of the present disclosure, AI / ML model distillation and / or expansion operations can be performed at the aggregation node. In addition, heterogeneous AI / ML model aggregation on self-organizing topologies can be supported at various network devices (network nodes, nodes, base stations, TRPs, etc.) and user devices (e.g., UEs).

[0288] According to some aspects of the present disclosure, AI / ML model training can be performed dynamically. AI / ML model training can be performed iteratively. Iterations of AI / ML model training can be completed at a desired time.

[0289] Also disclosed are examples of devices (e.g., ED or UE and TRP or network device) for performing the various methods described herein.

[0290] 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 may execute the instructions described in this document. Fig.12 and Figures 14 to 20 For example, the processor may enable the device to communicate over the air interface in the operating mode by implementing operations consistent with the operating mode, such as performing necessary measurements and generating content from those measurements (as configured for the operating mode), preparing uplink transmissions and processing downlink transmissions (e.g., encoding, decoding, etc.), and configuring and / or instructing transmission / reception on one or more RF chains and one or more antennas.

[0291] Note that the expression "at least one of A or B" as used herein is interchangeable with the expression "A and / or B." This expression refers to a list where one can choose between A or B or both A and B. Similarly, the expression "at least one of A, B, or C" as used herein is interchangeable with "A and / or B and / or C" or "A, B, and / or C." This expression refers to a list where one can choose between: 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. The same principles apply to longer lists with the same format.

[0292] Although the present invention has been described with reference to specific features and embodiments of the present invention, various modifications and combinations of the present invention may be formulated without departing from the present invention. The specification and drawings are therefore only regarded as an illustration of some embodiments of the present invention as defined by the appended claims, and any and all modifications, variants, combinations or equivalents covered within the scope of the present invention are considered. Therefore, although the present invention and its advantages have been described in detail, various changes, substitutions and modifications may be made without departing from the present invention as defined by the appended claims. In addition, the scope of the present application is not limited to the specific embodiments of the processes, machines, manufactured products, material compositions, components, methods and steps described in the specification. According to the disclosure of the present invention, it is easy for a person of ordinary skill in the art to understand that existing or to be developed processes, machines, manufactured products, material compositions, components, methods or steps that perform substantially the same functions as the corresponding embodiments described herein, or that can achieve substantially the same results as the corresponding embodiments described herein, may be used according to the present invention. Therefore, the appended claims are intended to include these processes, machines, manufactured products, material compositions, components, methods or steps within their scope.

[0293] In addition, any module, component, or device that executes instructions as exemplified 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 cassettes, magnetic tapes, disk storage or other magnetic storage devices, compact disc read-only memory (CD-ROM), digital video disc or digital versatile disc (DVD), Blu-ray discs, TM CD-ROMs, 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 can be part of a device or can be accessed or connected to a device. Any application or module described herein can be implemented using computer / processor readable / executable instructions, which can be stored or otherwise held by these non-transitory computer / processor readable storage media.

[0294] Abbreviation Definition

[0295] LTE Long Term Evolution

[0296] NR New Radio

[0297] BWP bandwidth part

[0298] BS Base Station

[0299] CA carrier aggregation

[0300] CC component carrier

[0301] CG cell group

[0302] CSI Channel State Information

[0303] CSI-RS Channel State Information Reference Signal reference signal

[0304] DC dual connectivity

[0305] DCI downlink control information

[0306] DL Downlink

[0307] DL-SCH Downlink shared channel

[0308] EN-DC E-UTRANR dual connectivity (E-UTRA NR dual connectivity), MCG uses E- UTRA, SCG use NR

[0309] gNB Next generation (or 5G) base station

[0310] HARQ-ACK Hybrid automatic repeat request acknowledgment acknowledgement

[0311] MCG master cell group

[0312] MCS modulation and coding scheme

[0313] MAC-CE medium access control-control element element)

[0314] PBCH Physical Broadcast Channel

[0315] Pcell primary cell

[0316] PDCCH physical downlink control channel

[0317] PDSCH physical downlink shared channel

[0318] PRACH physical random access channel

[0319] PRG physical resource block group

[0320] PSCell primary SCG cell

[0321] PSS Primary Synchronization Signal

[0322] PUCCH physical uplink control channel

[0323] PUSCH physical uplink shared channel

[0324] RACH Random access channel

[0325] RAPID Random access preamble identity

[0326] RB resource block

[0327] RE resource element

[0328] RRM Radio Resource Management

[0329] RMSI Remaining system information

[0330] RS reference signal

[0331] RSRP reference signal received power

[0332] RRC Radio Resource Control

[0333] SCG Secondary cell group

[0334] SFN system frame number

[0335] SL sidelink

[0336] Scell ​​secondary cell

[0337] SPS semi-persistent scheduling

[0338] SR Scheduling Request

[0339] SRI SRS resource indicator

[0340] SRS Sounding reference signal

[0341] SSS Secondary synchronization signal

[0342] SSB synchronization signal block

[0343] SUL Supplement uplink

[0344] TA Timing Advance

[0345] TAG timing advance group

[0346] TUE target UE

[0347] UCI Uplink control information

[0348] UE User Equipment

[0349] UL Uplink

[0350] UL-SCH Uplink shared channel

Claims

1. A method for learning an artificial intelligence or machine learning (AI / ML) model on a topology in a wireless communication network, the method comprising: The first node receives a first AI / ML model from the second node; The first node sends the first AI / ML model and a model collection indicator for collecting AI / ML models from the one or more associated nodes to one or more nodes associated with the first node; The first node receives, from the one or more associated nodes, reports related to respective associated AI / ML models of the one or more associated nodes; The first node obtains a second AI / ML model based on the report related to the corresponding associated AI / ML model, the second AI / ML model having a neural network (NN) structure equivalent to that of the first AI / ML model; The first node sends the second AI / ML model to a third node.

2. The method according to claim 1, wherein: The corresponding associated AI / ML models of the one or more associated nodes are not limited to having a NN structure equivalent to the NN structure of the first AI / ML model and the second AI / ML model.

3. The method according to claim 1 or 2, wherein: The report related to the corresponding associated AI / ML model includes at least one of the following: a confirmation indicator for transmission of said corresponding associated AI / ML model, information related to the corresponding associated AI / ML model, Information related to the training data of the corresponding associated AI / ML model, or Information related to the performance of the corresponding associated AI / ML model.

4. The method according to any one of claims 1 to 3, wherein: The model collection indicator is a dynamic indicator or an event-triggered indicator.

5. The method according to claim 4, wherein: The model collection indicator is the event trigger indicator, and the event is triggered when the training of the corresponding associated AI / ML model is completed or the performance of the corresponding associated AI / ML model exceeds a performance metric, and the performance metric is predetermined based on at least one of accuracy or precision.

6. The method according to any one of claims 1 to 5, wherein: The model collection indicator includes one of the following: distillation indicator; expansion indicator; or Distillation and expansion indicators.

7. The method according to claim 6, further comprising: The first node sends information about a reference AI / ML model to the one or more associated nodes.

8. The method according to claim 7, wherein: The information about the reference AI / ML model includes information indicating the NN structure of the reference AI / ML model, and the information indicating the NN structure of the reference AI / ML model includes at least one of the following: NN algorithm, width, depth, complexity, floating-point operations, total parameters, trainable parameters, or required buffer size.

9. The method according to any one of claims 6 to 8, wherein: Acquiring the second AI / ML model includes at least one of the following: Distilling the corresponding associated AL / ML model into one or more distilled AI / ML models according to the model collection indicator; or Inflating the corresponding associated AL / ML model into one or more inflated AI / ML models according to the model collection indicator.

10. A method for learning an artificial intelligence or machine learning (AI / ML) model on a topology in a wireless communication network, the method comprising: The first node receives a first AI / ML model and a model collection indicator; The first node obtains a second AI / ML model; The first node obtains a report related to the second AI / ML model; The first node sends the report related to the second AI / ML model to the second node according to the model collection indicator.

11. The method according to claim 10, wherein: The second AI / ML model is not limited to having a neural network (NN) structure equivalent to that of the first AI / ML model.

12. The method according to claim 10 or 11, wherein: The second AI / ML model is obtained based on at least one of the first AI / ML model, a local training data set, or a local AI / ML algorithm.

13. The method according to any one of claims 10 to 12, wherein: The report related to the second AI / ML model includes at least one of the following: a confirmation indicator for transmission of the second AI / ML model, information related to the second AI / ML model, information related to the training data of the second AI / ML model, or Information related to the performance of the second AI / ML model.

14. The method according to any one of claims 10 to 14, wherein: The model collection indicator is a dynamic indicator or an event-triggered indicator.

15. The method according to claim 14, wherein: The model collection indicator is the event trigger indicator, and the event is triggered when the training of the second AI / ML model is completed or the performance of the second AI / ML model exceeds a performance metric, and the performance metric is predetermined based on at least one of accuracy or precision.

16. The method according to any one of claims 10 to 15, wherein: The model collection indicator includes one of the following: distillation indicator; expansion indicator; or Distillation and expansion indicators.

17. The method according to claim 16, further comprising: The first node receives information about a reference AI / ML model from the second node.

18. The method according to claim 17, wherein: The information about the reference AI / ML model includes information indicating the NN structure of the reference AI / ML model, and the information indicating the NN structure of the reference AI / ML model includes at least one of the following: NN algorithm, width, depth, complexity, floating-point operations, total parameters, trainable parameters, or required buffer size.

19. The method according to claim 17 or 18, wherein: The model collection indicator is the distillation indicator, and the generated second AI / ML model is larger than the reference AI / ML model.

20. The method according to claim 19, wherein: The second AI / ML model obtained has: A greater number of floating point operations than the reference model; A greater number of total parameters than the reference model; A larger number of trainable parameters than the reference model; A required buffer size that is larger than the reference model; Greater than the width of the reference model; greater than the depth of the reference model; or Any combination of the foregoing.

21. An apparatus for a node for learning an artificial intelligence or machine learning (AI / ML) model on a topology in a wireless communication network, the apparatus comprising: processor; a memory storing processor-executable instructions which, when executed, cause the processor to: receiving a first AI / ML model from a second node; sending the first AI / ML model and a model collection indicator for collecting AI / ML models from the one or more associated nodes to one or more nodes associated with the node; receiving, from the associated nodes, reports related to respective associated AI / ML models of the one or more associated nodes; acquiring a second AI / ML model according to the report related to the corresponding associated AI / ML model, the second AI / ML model having a neural network (NN) structure equivalent to the NN structure of the first AI / ML model; Send the second AI / ML model to the third node.

22. The device according to claim 21, wherein The corresponding associated AI / ML models of the one or more associated nodes are not limited to having a NN structure equivalent to the NN structure of the first AI / ML model and the second AI / ML model.

23. The device according to claim 21 or 22, wherein: The report related to the corresponding associated AI / ML model includes at least one of the following: a confirmation indicator for transmission of said corresponding associated AI / ML model, information related to the corresponding associated AI / ML model, Information related to the training data of the corresponding associated AI / ML model, or Information related to the performance of the corresponding associated AI / ML model.

24. The device according to any one of claims 21 to 23, wherein: The model collection indicator is a dynamic indicator or an event-triggered indicator.

25. The device according to claim 24, wherein: The model collection indicator is the event trigger indicator, and the event is triggered when the training of the corresponding associated AI / ML model is completed or the performance of the corresponding associated AI / ML model exceeds a performance metric, and the performance metric is predetermined based on at least one of accuracy or precision.

26. The device according to any one of claims 21 to 25, wherein The model collection indicator includes one of the following: distillation indicator; expansion indicator; or Distillation and expansion indicators.

27. The device according to claim 26, wherein: The processor-executable instructions also include processor-executable instructions that, when executed, cause the processor to: The node sends information about a reference AI / ML model to the one or more associated nodes.

28. The device according to claim 27, wherein The information about the reference AI / ML model includes information indicating the NN structure of the reference AI / ML model, and the information indicating the NN structure of the reference AI / ML model includes at least one of the following: NN algorithm, width, depth, complexity, floating-point operations, total parameters, trainable parameters, or required buffer size.

29. The device according to any one of claims 26 to 28, wherein Generating the second AI / ML model includes at least one of the following: Distilling the corresponding associated AL / ML model into one or more distilled AI / ML models according to the model collection indicator; or Inflating the corresponding associated AL / ML model into one or more inflated AI / ML models according to the model collection indicator.

30. An apparatus comprising one or more units for performing the method according to any one of claims 1 to 9.

31. An apparatus for a node for learning an artificial intelligence or machine learning (AI / ML) model on a topology in a wireless communication network, the apparatus comprising: processor; a memory storing processor-executable instructions which, when executed, cause the processor to: Receiving a first AI / ML model and a model collection indicator; Obtain a second AI / ML model; Obtaining a report related to the second AI / ML model; Sending the report related to the second AI / ML model to a second node associated with the node according to the model collection indicator.

32. The device according to claim 31, wherein The second AI / ML model is not limited to having a neural network (NN) structure equivalent to that of the first AI / ML model.

33. The device according to claim 31 or 32, wherein: The second AI / ML model is obtained based on at least one of the first AI / ML model, a local training data set, or a local AI / ML algorithm.

34. The device according to any one of claims 31 to 33, wherein: The report related to the second AI / ML model includes at least one of the following: a confirmation indicator for transmission of the second AI / ML model, information related to the second AI / ML model, information related to the training data of the second AI / ML model, or Information related to the performance of the second AI / ML model.

35. The device according to any one of claims 31 to 34, wherein: The model collection indicator is a dynamic indicator or an event-triggered indicator.

36. The device according to claim 35, wherein The model collection indicator is the event trigger indicator, and the event is triggered when the training of the second AI / ML model is completed or the performance of the second AI / ML model exceeds a performance metric, and the performance metric is predetermined according to at least one of accuracy or precision.

37. The device according to any one of claims 31 to 36, wherein: The model collection indicator includes one of the following: distillation indicator; expansion indicator; or Distillation and expansion indicators.

38. The device according to claim 37, wherein The processor-executable instructions also include processor-executable instructions that, when executed, cause the processor to: The first node receives information about a reference AI / ML model from the second node.

39. The device according to claim 38, wherein The information about the reference AI / ML model includes information indicating the NN structure of the reference AI / ML model, and the information indicating the NN structure of the reference AI / ML model includes at least one of the following: NN algorithm, width, depth, complexity, floating-point operations, total parameters, trainable parameters, or required buffer size.

40. The device according to claim 38 or 39, wherein: The model collection indicator is the distillation indicator, and the generated second AI / ML model is larger than the reference AI / ML model.

41. The device according to claim 40, wherein The second AI / ML model generated has: A greater number of floating point operations than the reference model; A greater number of total parameters than the reference model; A larger number of trainable parameters than the reference model; A required buffer size that is larger than the reference model; Greater than the width of the reference model; greater than the depth of the reference model; or Any combination of the foregoing.

42. An apparatus comprising one or more units for performing the method according to any one of claims 10 to 20.

Images