Method for enhanced sidelink communication with clustering or peer-to-peer federated learning

Abstract

Various aspects are provided that allow a UE to achieve sidelink parameter coordination through various signaling in a clustering FL or a peer-to-peer FL. The UE provides a first ML model information update and a measurement associated with the update. The UE obtains an aggregated ML model information update aggregating the first update with a second ML model information update of a first network node in an FL cluster comprising the UE or a second network node in a second FL cluster. The UE provides a sidelink communication parameter in the SCI to a network node, the parameter being an output of an ML model associated with the aggregated ML model information update. Thus, UEs may derive common SCI parameters from their updated ML model for application in sidelink communications, resulting in a maximized packet reception rate, maximized throughput, or minimized communication latency.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims the benefit of U.S. non-provisional patent application No. 17 / 821,104, filed on August 19, 2022, entitled “METHODS FOR ENHANCED SIDELINK COMMUNICATIONS WITH CLUSTERED OR PEER-TO-PEER FEDERATED LEARNING,” which is expressly incorporated herein by reference in its entirety. Background Art

[0003] The present disclosure relates generally to communication systems, and more particularly to wireless communication systems involving direct communication between wireless devices.

[0004] Wireless communication systems are widely deployed to provide a variety of telecommunication services, such as telephony, video, data, messaging, and broadcast. Typical wireless communication systems may employ multiple access technologies capable of supporting communication with multiple users by sharing available system resources. Examples of such multiple access technologies include code division multiple access (CDMA) systems, time division multiple access (TDMA) systems, frequency division multiple access (FDMA) systems, orthogonal frequency division multiple access (OFDMA) systems, single carrier frequency division multiple access (SC-FDMA) systems, and time division synchronous code division multiple access (TD-SCDMA) systems.

[0005] These multiple access technologies have been adopted in various telecommunication standards to provide a common protocol that enables different wireless devices to communicate at a city, country, region, or even global level. An example telecommunication standard is 5G New Radio (NR). 5G NR is part of the continued mobile broadband evolution promulgated by the Third Generation Partnership Project (3GPP) to meet new requirements associated with latency, reliability, security, scalability (e.g., with the Internet of Things (IoT)) and other requirements. 5G NR includes services associated with enhanced mobile broadband (eMBB), massive machine type communications (mMTC), and ultra-reliable low latency communications (URLLC). Certain aspects of 5G NR may be based on the 4G Long Term Evolution (LTE) standard. Further improvements to 5G NR technology are needed. In addition, these improvements may also be applicable to other multiple access technologies and telecommunication standards that employ these technologies.

[0006] For example, some aspects of wireless communication include direct communication between devices, such as device-to-device (D2D), vehicle-to-everything (V2X), etc. There is a need to further improve such direct communication between devices. Improvements related to direct communication between devices may also be applicable to other multiple access technologies and telecommunication standards that employ these technologies. Summary of the invention

[0007] The following presents a simplified summary of one or more aspects in order to provide a basic understanding of these aspects. This summary is not an extensive overview of all contemplated aspects, and is neither intended to identify key or important elements of all aspects, nor to describe the scope of any or all aspects. Its sole purpose is to present some concepts of one or more aspects in a simplified form as a prelude to a more detailed description presented later.

[0008] In one aspect of the present disclosure, a device for wireless communication is provided. The device includes a processor and a memory coupled to the processor. The processor is configured to provide a first machine learning (ML) model information update and a measurement associated with the first ML model information update. The processor is also configured to obtain an aggregated ML model information update. The aggregated ML model information update includes an aggregation of a first ML model information update and a second ML model information update of one of the following: a first network node in a first federated learning (FL) cluster including the device, or a second network node in a second FL cluster. The processor is also configured to provide a sidelink communication parameter in a sidelink control information (SCI) to the network node. The sidelink communication parameter is an output of an ML model associated with an aggregated ML model information update.

[0009] In one aspect of the present disclosure, a method for wireless communication at a network entity is provided. The method includes: providing a first ML model information update and a measurement associated with the first ML model information update; obtaining an aggregated ML model information update, wherein the aggregated ML model information update includes an aggregation of the first ML model information update and a second ML model information update of one of the following: a first network node in a first FL cluster including the network entity, or a second network node in a second FL cluster; and providing a sidelink communication parameter in an SCI to the network node, wherein the sidelink communication parameter is an output of the ML model associated with the aggregated ML model information update.

[0010] In one aspect of the present disclosure, a device for wireless communication is provided. The device includes: a component for providing a first ML model information update and a measurement associated with the first ML model information update; a component for obtaining an aggregated ML model information update, wherein the aggregated ML model information update includes an aggregation of the first ML model information update and a second ML model information update of one of the following items: a first network node in a first FL cluster including the network entity, or a second network node in a second FL cluster; and wherein the component for providing is also configured to provide a sidelink communication parameter in an SCI to the network node, wherein the sidelink communication parameter is an output of the ML model associated with the aggregated ML model information update.

[0011] In one aspect of the present disclosure, a non-transitory computer-readable medium storing computer executable code is provided. The code, when executed by a processor, causes the processor to: provide a first ML model information update and a measurement associated with the first ML model information update; obtain an aggregated ML model information update, wherein the aggregated ML model information update includes an aggregation of the first ML model information update and a second ML model information update of one of the following: a first network node in a first FL cluster including the network entity, or a second network node in a second FL cluster; and provide a sidelink communication parameter in an SCI to the network node, wherein the sidelink communication parameter is an output of the ML model associated with the aggregated ML model information update.

[0012] In another aspect of the present disclosure, a device for wireless communication is provided. The device includes a processor and a memory coupled to the processor. The processor is configured to obtain a first ML model information update and a measurement associated with the first ML model information update. The processor is also configured to provide an aggregated ML model information update. The aggregated ML model information update includes an aggregation of the first ML model information update and a second ML model information update of one of the following: a first network node in a first FL cluster including the device, or a second network node in a second FL cluster. The sidelink communication parameter in the SCI of the network node is an output of the ML model associated with the aggregated ML model information update.

[0013] In one aspect of the present disclosure, the processor is configured to: provide a first message including first FL information of the device to a first network node; and obtain a second message including second FL information of the first network node, wherein based on the first FL information and the second FL information, the first network node is in a first FL cluster.

[0014] In one aspect of the present disclosure, the first FL information includes at least one of: a machine learning task of the device, information associated with an available sensor coupled to the device for the machine learning task, an available ML model associated with the machine learning task, or available computing resources of the device for the machine learning task.

[0015] In one aspect of the present disclosure, the machine learning task includes at least one of the following: a reinforcement learning task, an object detection task, or a classification task.

[0016] In one aspect of the present disclosure, the processor is configured to obtain a message indicating a FL cluster leader of a first FL cluster including the apparatus.

[0017] In one aspect of the disclosure, the processor is further configured to provide a message indicating that the apparatus is a FL cluster leader of the first FL cluster.

[0018] In one aspect of the disclosure, the first network node and the second network node are the same network node or different network nodes.

[0019] In one aspect of the present disclosure, the first FL cluster and the second FL cluster are the same FL cluster or different FL clusters.

[0020] In one aspect of the present disclosure, the first ML model information update includes at least one of the following: a weight, a gradient, or a scaling factor, the scaling factor being a multiplicand or a summand.

[0021] In one aspect of the present disclosure, a method for wireless communication at a network entity is provided. The method includes: obtaining a first ML model information update and a measurement associated with the first ML model information update; providing an aggregated ML model information update, wherein the aggregated ML model information update includes an aggregation of the first ML model information update and a second ML model information update of one of the following: a first network node in a first FL cluster including the network entity, or a second network node in a second FL cluster; and wherein a sidelink communication parameter in an SCI of the network node is an output of an ML model associated with the aggregated ML model information update.

[0022] In one aspect of the present disclosure, a device for wireless communication is provided. The device includes: a component for obtaining a first ML model information update and a measurement associated with the first ML model information update; a component for providing an aggregated ML model information update, wherein the aggregated ML model information update includes an aggregation of the first ML model information update and a second ML model information update of one of the following items: a first network node in a first FL cluster including the network entity, or a second network node in a second FL cluster; and wherein a sidelink communication parameter in an SCI of the network node is an output of an ML model associated with the aggregated ML model information update.

[0023] In one aspect of the present disclosure, a non-transitory computer-readable medium storing computer executable code is provided. The code, when executed by a processor, causes the processor to: obtain a first ML model information update and a measurement associated with the first ML model information update; provide an aggregated ML model information update, wherein the aggregated ML model information update comprises an aggregation of the first ML model information update and a second ML model information update of one of the following: a first network node in a first FL cluster including the network entity, or a second network node in a second FL cluster; and wherein a sidelink communication parameter in an SCI of the network node is an output of an ML model associated with the aggregated ML model information update.

[0024] To achieve the foregoing and related ends, one or more aspects include the features fully described below and particularly pointed out in the claims. The following description and the accompanying drawings set forth in detail some illustrative features of one or more aspects. However, these features are merely indicative of some of the various ways in which the principles of the various aspects may be employed, and this specification is intended to include all such aspects and their equivalents. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is a diagram illustrating an example of a wireless communication system and an access network.

[0026] Figure 2A is a diagram illustrating an example of a first frame according to various aspects of the present disclosure.

[0027] Figure 2B is a diagram illustrating an example of DL channels within a subframe according to various aspects of the present disclosure.

[0028] Figure 2C is a diagram illustrating an example of a second frame according to various aspects of the present disclosure.

[0029] Figure 2D is a diagram illustrating an example of UL channels within a subframe according to various aspects of the present disclosure.

[0030] Figure 3 Example aspects of a sidelink slot structure are illustrated.

[0031] Figure 4 is a diagram illustrating an example of a first device and a second device involved in wireless communication based on, for example, sidelink communication.

[0032] Figure 5 is a conceptual diagram of an example open radio access network architecture.

[0033] Figure 6 is a diagram illustrating an example of a neural network.

[0034] Figure 7 is a diagram illustrating an example of a graph showing different levels of congestion observed by UEs in an urban environment.

[0035] FIG. 8A to FIG. 8B is a diagram illustrating an example of packet reception rate versus transmitter-receiver distance of a vehicular UE in congested and non-congested urban environments.

[0036] Fig. 9 is a diagram illustrating an example of the FL architecture.

[0037] Fig. 10A and Fig. 10B is a diagram illustrating examples of FL architecture in different applications.

[0038] Fig.11 is a diagram illustrating an example of a clustered FL architecture.

[0039] Fig.12 is a diagram illustrating an example of a peer-to-peer FL architecture.

[0040] Fig.13 is a diagram illustrating an example of cluster formation or peer-to-peer networking in an automotive environment including congested and non-congested environments for clustering or peer-to-peer FLs.

[0041] Fig.14 is a diagram illustrating an example of a signaling procedure for neighbor discovery.

[0042] Fig.15 is a diagram illustrating an example of another signaling procedure for neighbor discovery.

[0043] Fig.16 is a diagram illustrating an example of a node state machine during a cluster formation or leader election process.

[0044] Fig.17 is a diagram illustrating an example of a signaling process for network-assisted cluster formation and leader selection.

[0045] Fig.18is a diagram illustrating an example of a signaling procedure for simultaneous discovery and cluster formation.

[0046] Fig.19 is a diagram illustrating an example of a signaling process for an intra-cluster FL or a peer FL after a leader node is elected or designated.

[0047] Fig. 20 is a diagram illustrating an example of a signaling procedure for inter-cluster FL after one or more iterations of intra-cluster FL.

[0048] Fig.21 is a diagram illustrating an example of a signaling process for synchronizing peer FLs without designating or electing a leader node.

[0049] Fig. 22 is a diagram illustrating an example of a signaling process for an asynchronous peer FL without designating or electing a leader node.

[0050] Fig.23 is a diagram illustrating an example of a signaling procedure for sidelink parameter coordination in response to a clustered or peer FL.

[0051] Fig.24 is a flow chart of a method of wireless communication at a UE or leader node.

[0052] Fig.25 is a flow chart of another method of wireless communication at a leader node or FL parameter network entity.

[0053] Fig.26 is a flow chart of a method of performing wireless communication at a UE.

[0054] Fig. 27 is a flow chart of another method for wireless communication at a leader node.

[0055] Fig.28 is a flow chart of another method for wireless communication at a leader node.

[0056] Fig.29 is a flow chart of a method of performing wireless communications at a FL parameter network entity.

[0057] Fig.30 is a diagram illustrating an example of a hardware implementation for an example apparatus.

[0058] Fig.31 is a diagram illustrating another example of a hardware implementation for another example apparatus. DETAILED DESCRIPTION

[0059] The specific embodiments described below in conjunction with the accompanying drawings are intended as descriptions of various configurations and are not intended to represent the only configurations with which the concepts described herein can be practiced. In order to provide a thorough understanding of the various concepts, the specific embodiments include specific details. However, it is apparent to those skilled in the art that these concepts can be practiced without these specific details. In some instances, well-known structures and components are shown in block diagram form to avoid blurring these concepts.

[0060] In sidelink communication, a transmitting (Tx) UE initially achieves sidelink synchronization with a receiving (Rx) UE. After synchronization, the Tx UE may obtain an allocation of time-frequency resources in which to transmit sidelink data to the Rx UE. According to one or more examples, the resource allocation may be scheduled by the base station in downlink control information (in mode 1 resource allocation), or the resource allocation may be determined by a sensing process performed autonomously by the Tx UE (in mode 2 resource allocation). After determining the resources, the Tx UE may transmit sidelink control information (SCI) including the resource allocation to the Rx UE.

[0061] In one or more examples, multiple SCI parameters may be applied to sidelink communications between different Tx UEs and Rx UEs. While in Mode 1 resource allocation, the base station may select these parameters to coordinate sidelink transmission and reception between different UE pairs, in Mode 2 resource allocation, the UEs themselves independently select the parameters. For example, one Tx UE may select a communication range for multicast negative acknowledgment (NACK) feedback, a reserved resource amount for retransmission, or a modulation and coding scheme (MCS) to optimize an application-related metric (e.g., minimum latency), while another Tx UE may select a different communication range for multicast NACK feedback, a different reserved resource amount for retransmission, or a different MCS to optimize different application-related metrics (e.g., maximized packet reliability). Therefore, in Mode 2 resource allocation, there may be a lack of coordination between UEs when selecting SCI parameters, which may result in suboptimal network performance. For example, a Tx UE that independently selects a significantly large Tx-Rx distance for groupcast NACK feedback may cause significant congestion due to multiple NACK feedbacks from various Rx UEs that are too far away from the Tx UE to successfully decode the sidelink data transmission. This congestion, in turn, may increase the channel busy rate (CBR) or channel occupancy rate (CR) metric measured by other Tx UEs. On the other hand, if the Tx UE selects its Tx-Rx distance for groupcast NACK feedback coordinated with other UEs, these other Tx UEs may effectively increase their MCS, number of retransmissions, or other transmission parameters, resulting in improved system performance. Therefore, providing coordination between UEs in sidelink communications may improve overall network performance (e.g., by allowing UEs to effectively improve MCS or other transmission parameters due to coordinated selection of SCI parameters by other UEs), especially in automated use cases employing artificial intelligence or machine learning (AI / ML) in artificial neural networks (ANNs).

[0062] UE, base station or other network node can be configured to use ANN to perform AI / ML tasks. AI / ML tasks are problems that AI / ML is intended to solve using training data or other input data, which may include tasks such as classification (e.g., classifying information based on labeled data), image recognition (e.g., identifying objects based on labeled images), reinforcement learning (e.g., learning to navigate in an environment using a reward and penalty system). ANN can be used to solve AI / ML tasks. ANN (or simply neural network) is a computational learning system that uses a functional network to convert data input in one form into a desired output that is usually in another form. ANN includes a layered architecture in which the output of one layer of neurons is input to a second layer of neurons (via connections or synapses), the output of the second layer of neurons becomes the input of the third layer of neurons, and so on. These neural networks can be trained to recognize hierarchical structures of features and are therefore increasingly used in object recognition applications. For example, a neural network can employ supervised learning tasks such as classification (which combines ML models such as logistic regression, support vector machines, boosting or other classifiers) to perform object detection and provide bounding boxes of classifications or categories in images. Furthermore, back-propagation or gradient descent can be used to fine-tune these multi-layer architectures to produce more accurate predictions.

[0063] One technique that can be applied to achieve coordination between network nodes is federated learning (FL), which applies AI / ML tasks to obtain sidelink communication parameters. FL refers to a distributed machine learning technique in which multiple decentralized nodes that store local data samples can train a global ML model (e.g., a classifier, a navigation system recommendation system, a digital assistant, a diagnostic system, or other models applied by multiple network nodes) without exchanging data samples themselves between nodes to perform training. The FL framework includes multiple network nodes or entities, namely a centralized aggregation server and participating FL devices (i.e., participants or nodes such as UEs). In one aspect or specific implementation, the FL framework enables FL devices to learn global ML models by allowing messages to be passed between devices through a central aggregation server or coordinator (also referred to as a FL parameter server throughout this disclosure), which can be configured to communicate with various FL devices and coordinate the learning framework. Nodes in the FL environment can process their own data sets and perform local updates to the global ML model, and the central server can aggregate local updates and provide updated global ML models to nodes for further training or prediction.

[0064] However, the FL architecture relies on a centralized server to create, aggregate, and refine global ML models for participating nodes, so updates based on locally trained ML models need to be sent from participating nodes to the server during FL iterations. This centralized approach to FL can have various characteristics. In one example, since the centralized server is the only aggregator of participating nodes, the centralized server can serve the FL system alone. In comparison, if the ML model aggregator role is distributed outside the centralized server, this can facilitate the advancement of the FL process. In another example, participants transmit multiple local model updates to the centralized server. Therefore, distributing the entities to which participants transmit such updates can minimize the communication overhead in the application in which multiple updates are applied. In another example, the FL architecture includes heterogeneity of computing resources that exist for different participants, heterogeneity of training data available for participants, and heterogeneity of training tasks and associated models configured for different participants. In contrast, an FL architecture with homogeneity of such computing resources, training data, training tasks, and associated models can provide statistical simplicity in model training. In another example, even if raw data is not communicated directly between nodes in FL, the exchange of ML model parameters with a distributed ML model aggregator can minimize the leakage of information about the underlying data samples, thereby improving the security and privacy of ML model information.

[0065] To improve the FL architecture, a clustering or hierarchical approach may be applied to FL, where learning nodes working toward a common learning (AI / ML) task are grouped together into clusters. In one example, UEs may be grouped together to form a cluster and a cluster leader (e.g., other UEs) may be designated. The designated cluster leader of the cluster, rather than a direct central aggregation server or coordinator (also referred to as an FL parameter server throughout this disclosure), coordinates learning tasks (AI / ML tasks), including local ML model training and updates within the cluster. According to one example, this coordination is referred to as intra-cluster FL. After the cluster is formed, the centralized FL parameter server may coordinate learning tasks, including global ML model training and updates between clusters. In one aspect, this coordination is referred to as inter-cluster FL. Thus, the cluster leader may act as an intermediary between the learning node and the FL parameter server for coordinating neural network training and optimization between different clusters. Therefore, clustered FL may be used to implement artificial neural network training in a distributed manner, with promoted FL process advancement, minimized communication overhead, statistical simplicity in model training due to homogeneity of computing resources, training data, training tasks, or associated ML models, and improved security and privacy of ML model information.

[0066] FL can also be improved using a peer-to-peer FL architecture. In a peer-to-peer approach, collaborative ML model training between participating nodes can be provided without the need for a designated and centralized server for coordinating training. Instead, participating nodes communicate directly with other participating nodes to coordinate and schedule artificial neural network training, individual model updates, and aggregation of individually updated models. Therefore, artificial neural network training can be similarly implemented in a distributed manner using peer-to-peer FL, with facilitated FL process advancement, minimized communication overhead, statistical simplicity in model training due to homogeneity of computing resources, training data, training tasks, or associated ML models, and improved security and privacy of ML model information.

[0067] A clustered FL architecture or a peer-to-peer FL architecture can assist UEs in coordinating side link parameters derived from ML models. For example, participant nodes in a cluster or peer-to-peer network can respectively apply updated ML models aggregated from information of other participant nodes to derive the required communication range for multicast NACK feedback or other common side link parameters. In addition, participant nodes with similar channel conditions, congestion levels, traffic statistics, or other attributes can be grouped together in a cluster or included in the exchange of updated ML model information. This grouping allows participant nodes with similar data sets (e.g., channel busy rate, channel occupancy, packet reception rate, or other observation metrics) to obtain similar updated ML models, resulting in a higher probability of public SCI parameter output. Therefore, participant nodes in the same cluster or peer-to-peer network can derive public SCI parameters from their updated ML models for application in side link communications, resulting in maximized packet reception rate, maximized throughput, or minimized latency for mode 2 resource allocation.

[0068] As an example, a vehicle UE (VUE) on a street with high congestion may experience one relationship between packet reception rate (PRR) and Tx-Rx distance for different communication ranges between Tx UE and Rx UE, while a VUE on a different street with low or no congestion may experience a different relationship between PRR and Tx-Rx distance for different communication ranges between Tx UE and Rx UE. Here, a higher congested street refers to a street with a higher UE density, while a lower congested street refers to a street with a lower UE density. In addition, PRR may be defined as the percentage of nodes (e.g., Rx UE) among the evaluated Rx UEs that successfully receive a packet from another node (e.g., Tx UE) when sending a packet. Thus, in highly congested streets (e.g., streets with a higher UE density), the VUE may observe that a higher PRR at a large Tx-Rx distance can be achieved using a smaller communication range for multicast NACK feedback, while in low or no congestion streets, a higher PRR at a larger Tx-Rx distance can be achieved using a larger communication range for multicast NACK feedback.

[0069] Therefore, to provide optimized sidelink parameter coordination between UEs, UEs observing similar high congestion levels on one street may be grouped into one cluster or peer network, while UEs observing similar low congestion levels on another street may be grouped into different clusters or peer networks. The respective groups of UEs may use intra-cluster or peer FL to locally train their ML models, and in the case of clustered FL, the FL parameter server, base station, or other network node may use inter-cluster FL to control the dynamic evolution of the learning process. In one example, the respective groups of UEs may apply their ML models to obtain an optimal communication range for multicast NACK feedback common to the UEs in the respective group, and the UEs may include the communication range in the respective SCI for optimizing sidelink data communication.

[0070] In order to perform sidelink parameter optimization using clustered FL or peer FL, one or more signaling processes may be provided that allow a network node (e.g., UE) to discover and form a cluster or peer network with other network nodes, select or elect a leader for a cluster or peer network, or otherwise implement clustered FL or peer FL. These signaling processes may include, for example: (1) a neighbor discovery process, in which a node (e.g., a UE) may discover other nodes (e.g., other UEs or FL parameter servers) that are close to or adjacent to each other and may participate in the ML training task; (2) a signaling process for cluster formation and leader election, in which a node participating in the ML training task may form a cluster with other discovered nodes and elect a cluster leader for the corresponding cluster; (3) a combined signaling process for neighbor discovery and cluster formation and leader election or selection, in which a node participating in the ML training task may join a cluster led by other nodes in response to a corresponding request from the other nodes; (4) a signaling process for cluster formation and leader election, in which nodes participating in the ML training task are grouped together into clusters, in which the other nodes are led by a selected cluster leader of the corresponding cluster; and (5) a signaling process for leader election in a peer-to-peer FL setting, in which a node participating in the ML training task may elect a leader among other discovered nodes. After these signaling processes, network nodes with similar distributions of observed data may be grouped into clusters, thereby maintaining the homogeneity of the data within the corresponding cluster. For example, local ML models are trained separately based on separately collected data (such as time series of CBR measurements, achieved throughput, or other metrics) and UEs affected by similar congestion levels may be grouped with other UEs into corresponding clusters led by a designated cluster leader.

[0071] After completing cluster formation and leader selection or leader election, the network node can implement side link parameter coordination in mode 2 resource allocation configuration by dynamically coordinating one or more signaling processes of clustered FL (intra-cluster or inter-cluster) or peer FL. These signaling processes may include signaling from a participant node (e.g., UE) to a leader (e.g., UE, RSU, edge server, local FL server or other network node), signaling from the leader to a FL parameter network entity (e.g., a base station, FL parameter server or other network node), signaling from the FL parameter network entity to the leader, or signaling from the leader to a participant node. For example, these signaling processes may include: (1) signaling for clustered FL training, in which nodes perform intra-FL training coordinated by a cluster leader within a corresponding cluster through messages passed between learning nodes and the cluster leader; (2) signaling for clustered FL training, in which nodes perform inter-FL training coordinated by a FL parameter server across different clusters through message exchanges between the corresponding cluster leader and the FL parameter server; (3) signaling for peer FL training, in which the leader coordinates FL training between peer nodes; (4) signaling for peer FL training, in which a peer node synchronously coordinates FL training between other peer nodes; and (5) signaling for peer FL training, in which a peer node asynchronously coordinates FL training between other peer nodes.

[0072] Thus, participant nodes may perform FL training within a cluster and across clusters, or within a peer-to-peer network. For example, a cluster or peer-to-peer network may be assigned a leader responsible for coordinating the FL process within its respective cluster or peer-to-peer network. The leader may aggregate local ML models reported by individual participant nodes that observe similar business statistics, respectively, and the cluster leader may communicate this information to the FL parameter network entity along with summary statistics associated with their respective clusters. The FL parameter network entity may dynamically control the evolution of the FL process between clusters. For example, the FL parameter network entity may aggregate ML models across clusters that observe similar data distributions about a certain metric, and the FL parameter network entity may dynamically assign participant nodes to different clusters based on the aggregated information. The FL parameter network entity may dynamically assign participant nodes that are determined by the leader as outliers in a cluster or peer-to-peer network led or utilized by the leader node to different clusters. An outlier UE may be, for example, a UE whose observed traffic statistics have significantly different values ​​or a different distribution of values ​​compared to the values ​​of the summary statistics associated with its corresponding cluster or network (e.g., a CBR value that differs by at least three standard deviations from the overall average CBR across multiple UEs in the cluster or network). The FL parameter network entity may communicate inter-cluster aggregation and dynamic cluster assignments to a cluster leader, which in turn communicates this information to the participant nodes. The leader of a peer-to-peer network may similarly communicate the aggregated ML model information of the participant nodes to each participant node. The participant nodes in a cluster or peer-to-peer network may apply the aggregated ML model to their respective data sets to derive common SCI parameters between the participant nodes in the cluster or peer-to-peer network. Thus, sidelink communication performance may be improved between network nodes having similar congestion levels, traffic statistics, or similar attributes.

[0073] Several aspects of telecommunication systems will now be presented with reference to various apparatus and methods. These apparatus and methods will be described in the following detailed description and illustrated in the accompanying drawings by various blocks, components, circuits, processes, algorithms, etc. (collectively referred to as "elements"). These elements may be implemented using electronic hardware, computer software, or any combination thereof. Whether such elements are implemented as hardware or software depends on the specific application and the design constraints imposed on the overall system.

[0074] As an example, an element, or any part of an element, or any combination of elements can be implemented as a "processing system", which includes one or more processors. The example of a processor includes a microprocessor, a microcontroller, a graphics processing unit (GPU), a central processing unit (CPU), an application processor, a digital signal processor (DSP), a reduced instruction set computing (RISC) processor, a system on a chip (SoC), a baseband processor, a field programmable gate array (FPGA), a programmable logic device (PLD), a state machine, a gate logic piece, a discrete hardware circuit and other suitable hardware configured to perform various functionalities described throughout the present disclosure. One or more processors in a processing system can execute software. Whether it is referred to as software, firmware, middleware, microcode, hardware description language or other names, software should be widely interpreted as meaning instructions, instruction sets, codes, code segments, program codes, programs, subroutines, software components, applications, software applications, software packages, routines, subroutines, objects, executable files, threads of execution, processes, functions, etc.

[0075] Accordingly, in one or more example aspects, the described functions may be implemented with hardware, software, or any combination thereof. If implemented with software, the functions may be stored or encoded on a computer-readable medium as one or more instructions or codes. Computer-readable media include computer storage media. Storage media may be any available medium that can be accessed by a computer. By way of example and not limitation, such computer-readable media may include random access memory (RAM), read-only memory (ROM), electrically erasable programmable ROM (EEPROM), optical disk storage, magnetic disk storage, other magnetic storage devices, a combination of the above-mentioned types of computer-readable media, or any other medium that can be used to store computer executable code in the form of instructions or data structures that can be accessed by a computer.

[0076] Figure 1 1 is a diagram illustrating an example of a wireless communication system and an access network 100. The wireless communication system (also referred to as a wireless wide area network (WWAN)) includes a base station (BS) 102, a user equipment (UE) 104, an evolved packet core (EPC) 160, and another core network 190 (e.g., a 5G core (5GC)). The base station 102 may include a macro cell (a high-power cellular base station) and / or a small cell (a low-power cellular base station). A macro cell includes a base station. A small cell includes a femto cell, a pico cell, and a micro cell.

[0077] The base station 102 configured for 4G Long Term Evolution (LTE) (collectively referred to as Evolved Universal Mobile Telecommunications System (UMTS) Terrestrial Radio Access Network (E-UTRAN)) can interface with the EPC 160 via a first backhaul link 132 (e.g., an S1 interface). The base station 102 configured for 5G New Radio (NR) (collectively referred to as Next Generation RAN (NG-RAN)) can interface with the core network 190 via a second backhaul link 184. Among other functions, the base station 102 can also perform one or more of the following functions: delivery of user data, radio channel encryption and decryption, integrity protection, header compression, mobility control functions (e.g., handover, dual connectivity), inter-cell interference coordination, connection establishment and release, load balancing, distribution of non-access stratum (NAS) messages, NAS node selection, synchronization, radio access network (RAN) sharing, multimedia broadcast multicast service (MBMS), subscriber and equipment tracking, RAN information management (RIM), paging, positioning, and delivery of warning messages. The base stations 102 may communicate with each other directly or indirectly (eg, via the EPC 160 or the core network 190) via the third backhaul link 134 (eg, an X2 interface). The first backhaul link 132, the second backhaul link 184, and the third backhaul link 134 may be wired or wireless.

[0078] Base station 102 can communicate wirelessly with UE 104. Each of base stations 102 can provide communication coverage for a corresponding geographic coverage area 110. There may be overlapping geographic coverage areas 110. For example, a small cell 102' may have a coverage area 110' that overlaps with the coverage area 110 of one or more macro base stations 102. A network including both small cells and macro cells may be referred to as a heterogeneous network. A heterogeneous network may also include a home evolved Node B (eNB) (HeNB), which may provide services to a restricted group referred to as a closed subscriber group (CSG). A communication link 120 between base station 102 and UE 104 may include an uplink (UL) (also referred to as a reverse link) transmission from UE 104 to base station 102 and / or a downlink (DL) (also referred to as a forward link) transmission from base station 102 to UE 104. The communication link 120 may use multiple input multiple output (MIMO) antenna technology, including spatial multiplexing, beamforming and / or transmit diversity. The communication link may be through one or more carriers. For each carrier allocated in the carrier aggregation for up to a total of up to Yx MHz (x component carriers) for transmission in each direction, the base station 102 / UE 104 can use a spectrum with a bandwidth of up to Y megahertz (MHz) (e.g., 5 MHz, 10 MHz, 15 MHz, 20 MHz, 100 MHz, 400 MHz, etc.). The carriers may or may not be adjacent to each other. The allocation of carriers may be asymmetric with respect to DL and UL (e.g., more or fewer carriers may be allocated for DL ​​compared to UL). The component carrier may include a primary component carrier and one or more secondary component carriers. The primary component carrier may be referred to as a primary cell (PCell) and the secondary component carrier may be referred to as a secondary cell (SCell).

[0079] UE 104 can communicate with each other using device-to-device (D2D) communication link 158. D2D communication link 158 can use DL / UL WWAN spectrum. D2D communication link 158 can use one or more side link channels, such as physical side link broadcast channel (PSBCH), physical side link discovery channel (PSDCH), physical side link shared channel (PSSCH) and physical side link control channel (PSCCH). D2D communication can be through various wireless D2D communication systems, such as, for example, WiMedia, Bluetooth, ZigBee, Wi-Fi based on Institute of Electrical and Electronics Engineers (IEEE) 802.11 standard, LTE or NR.

[0080] The wireless communication system may also include a Wi-Fi access point (AP) 150 that communicates with a Wi-Fi station (STA) 152 via a communication link 154 in, for example, a 5 gigahertz (GHz) unlicensed spectrum. When communicating in the unlicensed spectrum, the STA 152 / AP 150 may perform a clear channel assessment (CCA) to determine whether a channel is available prior to communication.

[0081] The small cell 102' may operate in licensed and / or unlicensed spectrum. When operating in an unlicensed spectrum, the small cell 102' may employ NR and use the same unlicensed spectrum (e.g., 5 GHz, etc.) as used by the Wi-Fi AP 150. The small cell 102' employing NR in the unlicensed spectrum may improve the coverage of the access network and / or increase the capacity of the access network.

[0082] The electromagnetic spectrum is typically subdivided into various categories, bands, channels, etc. based on frequency / wavelength. In 5G NR, two initial operating bands have been identified as frequency range designations FR1 (410 MHz-7.125 GHz) and FR2 (24.25 GHz-52.6 GHz). Frequencies between FR1 and FR2 are typically referred to as mid-band frequencies. Although a portion of FR1 is greater than 6 GHz, FR1 is typically (interchangeably) referred to as the "sub-6 GHz" band in various documents and articles. A similar naming issue sometimes occurs with respect to FR2, which is typically (interchangeably) referred to as the "millimeter wave" band in documents and articles, despite being different from the extremely high frequency (EHF) band (30 GHz-300 GHz) identified as the "millimeter wave" band by the International Telecommunication Union (ITU).

[0083] In view of the above aspects, unless otherwise specifically stated, it should be understood that if the term "below 6 GHz" or the like is used in this document, it can be broadly referred to as a frequency that can be less than 6 GHz, can be within FR1, or can include mid-band frequencies. In addition, unless otherwise specifically stated, it should be understood that if the term "millimeter wave" or the like is used in this document, it can be broadly referred to as a frequency that can include mid-band frequencies, can be within FR2, or can be within the EHF band.

[0084] The base station 102, whether a small cell 102' or a large cell (e.g., a macro base station), may include and / or be referred to as an eNB, a gNodeB (gNB), or another type of base station. Some base stations, such as gNB 180, may operate in the traditional sub-6 GHz spectrum, in millimeter wave frequencies, and / or near millimeter wave frequencies to communicate with the UE 104. When the gNB 180 operates in millimeter wave or near millimeter wave frequencies, the gNB 180 may be referred to as a millimeter wave base station. The millimeter wave base station 180 may utilize beamforming 182 with the UE 104 to compensate for path loss and short range. The base station 180 and the UE 104 may each include multiple antennas, such as antenna elements, antenna panels, and / or antenna arrays, to facilitate beamforming.

[0085] The base station 180 may transmit beamformed signals to the UE 104 in one or more transmit directions 182'. UE 104 may receive beamformed signals from base station 180 in one or more receive directions 182″. UE 104 may also send beamformed signals to base station 180 in one or more transmit directions. Base station 180 may receive beamformed signals from UE 104 in one or more receive directions. Base station 180 / UE 104 may perform beam training to determine the best receive direction and transmit direction for each of base station 180 / UE 104. The transmit direction and receive direction of base station 180 may or may not be the same. The transmit direction and receive direction of UE 104 may or may not be the same. Although the beamformed signals are shown as being between UE 104 and base station 102 / 180, UE 104 or RSU 107 may also similarly apply aspects of beamforming to communicate with another UE 104 or RSU 107 (such as based on V2X, V2V, or D2D communication).

[0086] The EPC 160 may include a mobility management entity (MME) 162, other MMEs 164, a serving gateway 166, an MBMS gateway 168, a broadcast multicast service center (BM-SC) 170, and a packet data network (PDN) gateway 172. The MME 162 may communicate with a home subscriber server (HSS) 174. The MME 162 is a control node that handles signaling between the UE 104 and the EPC 160. The MME 162 provides bearer and connection management. All user Internet Protocol (IP) packets are passed through the serving gateway 166, which itself is connected to the PDN gateway 172. The PDN gateway 172 provides UE IP address allocation and other functions. The PDN gateway 172 and the BM-SC 170 are connected to IP services 176. The IP services 176 may include the Internet, an intranet, an IP multimedia subsystem (IMS), a PS streaming service, and / or other IP services. The BM-SC 170 may provide functions for MBMS user service configuration and delivery. BM-SC 170 can be used as an entry point for content provider MBMS transmission, can be used to authorize and initiate MBMS bearer services in a public land mobile network (PLMN), and can be used to schedule MBMS transmission. MBMS gateway 168 can be used to distribute MBMS services to base stations 102 belonging to a multicast broadcast single frequency network (MBSFN) area of ​​a broadcast specific service, and can be responsible for session management (start / stop) and for collecting eMBMS related billing information.

[0087] The core network 190 may include an access and mobility management function (AMF) 192, other AMFs 193, a session management function (SMF) 194, and a user plane function (UPF) 195. The AMF 192 may communicate with a unified data management (UDM) 196. The AMF 192 is a control node that handles signaling between the UE 104 and the core network 190. The AMF 192 provides quality of service (QoS) flow and session management. All user IP packets are delivered through the UPF 195. The UPF 195 provides UE IP address allocation and other functions. The UPF 195 is connected to an IP service 197. The IP service 197 may include the Internet, an intranet, an IMS, a packet switching (PS) streaming media service, and / or other IP services.

[0088] Base stations may include and / or be referred to as gNBs, Node Bs, eNBs, access points, base transceivers, radio base stations, radio transceivers, transceiver functions, basic service sets (BSSs), extended service sets (ESSs), transmit receive points (TRPs), or some other suitable terminology. Base stations 102 provide access points to EPC 160 or core network 190 for UEs 104. Examples of UEs 104 include cellular phones, smart phones, session initiation protocol (SIP) phones, laptops, personal digital assistants (PDAs), satellite radios, global positioning systems, multimedia devices, video devices, digital audio players (e.g., MP3 players), cameras, game consoles, tablets, smart devices, wearable devices, vehicles, electric meters, gas pumps, large or small kitchen appliances, healthcare devices, implants, sensors / actuators, displays, or any other similarly functional devices. Some of the UEs 104 may be referred to as IoT devices (e.g., parking meters, gas pumps, toasters, vehicles, heart monitors, etc.). UE 104 may also be referred to as a station, a mobile station, a subscriber station, a mobile unit, a subscriber unit, a wireless unit, a remote unit, a mobile device, a wireless device, a wireless communication device, a remote device, a mobile subscriber station, an access terminal, a mobile terminal, a wireless terminal, a remote terminal, a handset, a user agent, a mobile client, a client, or some other suitable terminology.

[0089] The deployment of a communication system (such as a 5G NR system) can be arranged in a variety of ways using various components or constituent parts. In a 5G NR system or network, a network node, a network entity, a mobility element of a network, a RAN node, a core network node, a network element, or a network equipment (such as a BS, or one or more units (or one or more components) performing base station functionality) can be implemented in an aggregated or decomposed architecture. For example, a BS (such as a Node B (NB), an eNB, an NR BS, a 5G NB, an access point (AP), a TRP, or a cell, etc.) can be implemented as an aggregated base station (also called an independent BS or a monolithic BS) or a decomposed base station.

[0090] The aggregated base station may be configured to utilize a radio protocol stack physically or logically integrated within a single RAN node. The decomposed base station 181 may be configured to utilize a protocol stack physically or logically distributed between two or more units, such as one or more central units (CUs), one or more distributed units (DUs), or one or more radio units (RUs). In some aspects, the CU 183 may be implemented within a RAN node, and one or more DUs 185 may be co-located with the CU, or alternatively, may be geographically or virtually distributed in one or more other RAN nodes. The DU may be implemented to communicate with one or more RUs 187. Each of the CU, DU, and RU may also be implemented as a virtual unit, i.e., a virtual central unit (VCU), a virtual distributed unit (VDU), or a virtual radio unit (VRU).

[0091] Base station type operations or network designs may take into account the aggregated nature of base station functionality. For example, a disaggregated base station may be used in an integrated access backhaul (IAB) network, an open radio access network (O-RAN (network configurations such as those initiated by the O-RAN Alliance)), or a virtualized radio access network (vRAN, also known as a cloud radio access network (C-RAN)). Decomposition may include distributing functionality across two or more units at various physical locations, as well as virtually distributing functionality of at least one unit, which may enable flexibility in network design. Individual units of a disaggregated base station or disaggregated RAN architecture may be configured for wired or wireless communication with at least one other unit.

[0092] Some wireless communication networks may include vehicle-based communication devices that may communicate from vehicle-to-vehicle (V2V), vehicle-to-infrastructure (V2I) (e.g., from a vehicle-based communication device to a road infrastructure node, such as a roadside unit (RSU)), vehicle-to-network (V2N) (e.g., from a vehicle-based communication device to one or more network nodes, such as a base station), and / or a combination thereof and / or communicate with other devices, which communications may be collectively referred to as vehicle-to-everything (V2X) communications. Again, referring to Figure 1 In some aspects, a UE 104 (e.g., a transmitting vehicle user equipment (VUE) or other UE) may be configured to send a message directly to another UE 104. The communication may be based on V2V / V2X / V2I or other D2D communications, such as proximity services (ProSe), etc. Communications based on V2V, V2X, V2I and / or D2D may also be sent and received by other transmitting and receiving devices (such as a roadside unit (RSU) 107, etc.). Aspects of the communication may be based on PC5 or sidelink communications, for example, as combined with Figure 3 as described in the examples.

[0093] Reference again Figure 1 , a first network entity (e.g., UE 104, RSU 107) may include an ML model information update (MMIU) component 198. The MMIU component 198 is configured to provide a first ML model information update and a measurement associated with the first ML model information update; obtain an aggregated ML model information update, wherein the aggregated ML model information update includes an aggregation of the first ML model information update and a second ML model information update of one of the following: a first network node in a first FL cluster including the network entity, or a second network node in a second FL cluster; and provide a sidelink communication parameter in an SCI to the network node, wherein the sidelink communication parameter is an output of the ML model associated with the aggregated ML model information update. The UE 104 including the MMIU component 198 may be a transmitting device in uplink or sidelink communication, such as a VUE, an IoT device, or other UE, or a receiving device in downlink or sidelink communication, such as another VUE, another IoT device, or other UE. The RSU 107 including the MMIU component 198 may be a transmitting device in uplink or sidelink communication, or a receiving device in downlink or sidelink communication. The first network node and the second network node may be a UE 104, an RSU 107, an edge server, or other network nodes. The network node to which the sidelink communication parameters in the SCI are provided may be the same network node as the first network node and the second network node, or a different network node.

[0094] Still reference Figure 1, the second network entity (e.g., RSU 107 or base station 102 / 180) may include an aggregated ML model information update (AMMIU) component 199. The AMMIU component 199 is configured to obtain a first ML model information update and a measurement associated with the first ML model information update; and provide an aggregated ML model information update, wherein the aggregated ML model information update includes an aggregation of the first ML model information update and a second ML model information update of one of the following: a first network node in a first federated learning (FL) cluster including the device, or a second network node in a second FL cluster. The sidelink communication parameter in the SCI of the network node is an output of the ML model associated with the aggregated ML model information update. The RSU 107 including the AMMIU component 199 can be a transmitting device in an uplink or sidelink communication, or a receiving device in a downlink or sidelink communication. The base station 102 / 180 including the AMMIU component 199 can be a transmitting device in a downlink communication, or a receiving device in an uplink communication. The first network node and the second network node may be UE 104, RSU 107, edge server or other network nodes. The network node associated with the sidelink communication parameters in the SCI may be the same network node as the first network node and the second network node, or a different network node.

[0095] The concepts and various aspects described herein may be applicable to vehicle-to-everything (V2X) or other similar fields, such as D2D communication, IoT communication, industrial IoT (IIoT) communication, and / or other standards / protocols for communication in wireless / access networks. Additionally or alternatively, the concepts and various aspects described herein may be applicable to vehicle-to-pedestrian (V2P) communication, pedestrian-to-vehicle (P2V) communication, vehicle-to-infrastructure (V2I) communication, and / or other frameworks / models for communication in wireless / access networks. In addition, the concepts and various aspects described herein may be applicable to NR or other similar fields, such as LTE, Advanced LTE (LTE-A), Code Division Multiple Access (CDMA), Global System for Mobile Communications (GSM), or other wireless / radio access technologies. In addition, the concepts and various aspects described herein may be applicable to use in aggregated or decomposed base station architectures such as open radio access network (O-RAN) architectures.

[0096] Figure 2A FIG200 is a diagram illustrating an example of a first subframe within a 5G NR frame structure. Figure 2B FIG230 is a diagram illustrating an example of DL channels within a 5G NR subframe. Figure 2C FIG250 is a diagram illustrating an example of a second subframe within a 5G NR frame structure. Figure 2DFIG280 is a diagram illustrating an example of an UL channel within a 5G NR subframe. The 5G NR frame structure may be frequency division duplex (FDD) (wherein for a particular set of subcarriers (carrier system bandwidth), a subframe within the subcarrier set is dedicated to either DL or UL), or may be time division duplex (TDD) (wherein for a particular set of subcarriers (carrier system bandwidth), a subframe within the subcarrier set is dedicated to both DL and UL). Figure 2A , Figure 2C In the example provided, the 5G NR frame structure is assumed to be TDD, where subframe 4 is configured with slot format 28 (most of which are DL), where D is DL, U is UL, and F is flexible between DL / UL, and subframe 3 is configured with slot format 34 (most of which are UL). Although subframes 3 and 4 are shown as having slot formats 34 and 28, respectively, any particular subframe may be configured with any of the various available slot formats 0-61. Slot formats 0 and 1 are both DL and both UL, respectively. Other slot formats 2-61 include a mix of DL, UL, and flexible symbols. The slot format is configured for the UE via a received slot format indicator (SFI) (dynamically configured via DL control information (DCI) or semi-statically / statically configured via radio resource control (RRC) signaling). Note that the following description also applies to the 5G NR frame structure as TDD.

[0097] Other wireless communication technologies may have different frame structures and / or different channels. For example, a 10 millisecond (ms) frame may be divided into 10 equally sized subframes (1 ms). Each subframe may include one or more time slots. A subframe may also include a microslot, which may include 7, 4, or 2 symbols. Each time slot may include 7 or 14 symbols, depending on the time slot configuration. For time slot configuration 0, each time slot may include 14 symbols, and for time slot configuration 1, each time slot may include 7 symbols. The symbol on the DL may be a cyclic prefix (CP) orthogonal frequency division multiplexing (OFDM) (CP-OFDM) symbol. The symbol on the UL may be a CP-OFDM symbol (for high throughput scenarios) or a discrete Fourier transform (DFT) spread OFDM (DFT-s-OFDM) symbol (also known as a single carrier frequency division multiple access (SC-FDMA) symbol) (for power-limited scenarios; limited to single-stream transmission). The number of time slots in a subframe is based on the time slot configuration and parameter set (numerology). For slot configuration 0, different parameter sets μ0 to 4 allow 1, 2, 4, 8, and 16 slots per subframe, respectively. For slot configuration 1, different parameter sets 0 to 2 allow 2, 4, and 8 slots per subframe, respectively. Accordingly, for slot configuration 0 and parameter set μ, there are 14 symbols per slot and 2 per subframe. μtime slots. The subcarrier spacing and symbol length / duration are functions of the parameter set. The subcarrier spacing can be equal to 2 μ *15 kilohertz (kHz), where μ is parameter set 0 to 4. Therefore, the subcarrier spacing for parameter set μ=0 is 15 kHz, and the subcarrier spacing for parameter set μ=4 is 240 kHz. The symbol length / duration is inversely related to the subcarrier spacing. FIG. 2A to FIG. 2D An example is provided for slot configuration 0 with 14 symbols per slot and parameter set μ=2 with 4 slots per subframe. The slot duration is 0.25 ms, the subcarrier spacing is 60 kHz, and the symbol duration is approximately 16.67 μs. Within a frame set, there may be one or more different bandwidth parts (BWPs) that are frequency-division multiplexed (see Figure 2B ). Each BWP can have a specific set of parameters.

[0098] A resource grid may be used to represent the frame structure. Each slot includes a resource block (RB) (also referred to as a physical RB (PRB)) extending over 12 consecutive subcarriers. The resource grid is divided into a plurality of resource elements (REs). The number of bits carried by each RE depends on the modulation scheme.

[0099] like Figure 2A As illustrated, some of the REs carry reference (pilot) signals (RS) for the UE. The RS may include a demodulation RS (DM-RS) for channel estimation at the UE (indicated as R for a particular configuration). x , where 100x is the port number, but other DM-RS configurations are possible) and a channel state information reference signal (CSI-RS). RS may also include a beam measurement RS (BRS), a beam refinement RS (BRRS), and a phase tracking RS (PT-RS).

[0100] Figure 2BExamples of various DL channels within a subframe of a frame are illustrated. The physical downlink control channel (PDCCH) carries DCI within one or more control channel elements (CCEs), each CCE including nine RE groups (REGs), each REG including four consecutive REs in an OFDM symbol. The PDCCH within a BWP may be referred to as a control resource set (CORESET). Additional BWPs may be located at higher and / or lower frequencies on the channel bandwidth. The primary synchronization signal (PSS) may be within symbol 2 of a specific subframe of a frame. The PSS is used by UE 104 to determine subframe / symbol timing and physical layer identification. The secondary synchronization signal (SSS) may be within symbol 4 of a specific subframe of a frame. The SSS is used by the UE to determine the physical layer cell identification group number and radio frame timing. Based on the physical layer identification and the physical layer cell identification group number, the UE may determine the physical cell identifier (PCI). Based on the PCI, the UE may determine the position of the aforementioned DM-RS. The physical broadcast channel (PBCH) carrying the master information block (MIB) can be logically grouped with the PSS and SSS to form a synchronization signal (SS) / PBCH block (also referred to as an SS block (SSB)). The MIB provides the system frame number (SFN) and the number of RBs in the system bandwidth. The physical downlink shared channel (PDSCH) carries user data, broadcast system information not sent over the PBCH such as the system information block (SIB), and paging messages.

[0101] like Figure 2C As illustrated, some of the REs carry DM-RS (indicated as R for a particular configuration, but other DM-RS configurations are possible) for channel estimation at the base station. The UE may send a DM-RS for a physical uplink control channel (PUCCH) and a DM-RS for a physical uplink shared channel (PUSCH). The PUSCH DM-RS may be sent in the first or first two symbols of the PUSCH. Depending on whether a short PUCCH or a long PUCCH is sent and depending on the specific PUCCH format used, the PUCCH DM-RS may be sent in different configurations. The UE may send a sounding reference signal (SRS). The SRS may be sent in the last symbol of a subframe. The SRS may have a comb structure, and the UE may send the SRS on one of the teeth of the comb. The SRS may be used by the base station for channel quality estimation to achieve frequency-dependent scheduling of the UL.

[0102] Figure 2DExamples of various UL channels within a subframe of a frame are illustrated. The PUCCH may be located at a position as indicated in one configuration. The PUCCH carries uplink control information (UCI), such as scheduling requests, channel quality indicator (CQI), precoding matrix indicator (PMI), rank indicator (RI), and hybrid automatic repeat request (HARQ) acknowledgement (ACK) / negative acknowledgement (NACK) feedback. The PUSCH carries data and may additionally be used to carry a buffer status report (BSR), a power headroom report (PHR), and / or UCI.

[0103] Figure 3 Example diagrams 300 and 310 are illustrated, which illustrate example time slot structures that can be used for wireless communication (e.g., for sidelink communication) between UE 104 and UE 104'. The time slot structure can be within a 5G / NR frame structure. Although the following description may focus on 5G NR, the concepts described herein may be applicable to other similar fields, such as LTE, LTE-A, CDMA, GSM, and other wireless technologies. This is just an example, and other wireless communication technologies may have different frame structures and / or different channels. A frame (10ms) can be divided into 10 subframes of the same size (1ms). Each subframe may include one or more time slots. A subframe may also include a microslot, which may include, for example, 7, 4, or 2 symbols. Each time slot may include 7 or 14 symbols, depending on the time slot configuration. For time slot configuration 0, each time slot may include 14 symbols, and for time slot configuration 1, each time slot may include 7 symbols. Diagram 300 illustrates a single time slot transmission, which may correspond to a 0.5 ms transmission time interval (TTI), for example. Diagram 310 illustrates an example double time slot aggregation, for example, an aggregation of two 0.5 ms TTIs. Diagram 300 illustrates a single RB, while diagram 310 illustrates N RBs. In diagram 310, the 10 RBs used for control are only an example. The number of RBs may be different.

[0104] A resource grid may be used to represent the frame structure. Each time slot may include a resource block (RB) (also called a physical RB (PRB)) extending over 12 consecutive subcarriers. The resource grid is divided into a number of resource elements (REs). The number of bits carried by each RE depends on the modulation scheme. Figure 3 As illustrated, some of the REs may include control information, eg, along with a demodulation RS (DMRS). Figure 3 It is also illustrated that the symbol may include CSI-RS. Figure 3The symbol indicated for DMRS or CSI-RS in indicates that the symbol includes DMRS or CSI-RS RE. Such symbols may also include REs containing data. For example, if the number of ports of DMRS or CSI-RS is 1 and the comb-2 pattern is used for DMRS / CSI-RS, half of the REs may include the RS, and the other half of the REs may include data. CSI-RS resources may start at any symbol in the time slot and may occupy 1, 2, or 4 symbols, depending on the number of configured ports. CSI-RS may be periodic, semi-persistent, or aperiodic (e.g., based on DCI triggering). For time / frequency tracking, CSI-RS may be periodic or aperiodic. CSI-RS may be sent in bursts of two or four symbols spread across one or two time slots. Control information may include sidelink control information (SCI). As described herein, at least one symbol may be used for feedback. Symbols before and / or after feedback may be used for transitions between receiving data and sending feedback. Although symbol 12 is illustrated for data, the symbol may be a gap symbol to enable a transition to feedback in symbol 13. Another symbol (e.g., at the end of a time slot) may be used as a gap. The gap enables the device to switch from operating as a transmitting device to preparing to operate as a receiving device, e.g., in a subsequent time slot. As illustrated, data may be sent in the remaining REs. The data may include data messages as described herein. The location of any of the SCI, feedback, and LBT symbols may be different from Figure 3 The examples shown are different. Multiple time slots can be aggregated together. Figure 3 An example aggregation of two time slots is also illustrated. The aggregated number of time slots may also be greater than two. When time slots are aggregated, the symbols and / or slot symbols used for feedback may be different from the symbols and / or slot symbols used for feedback of a single time slot. Although feedback is not illustrated for this aggregation example, symbols in a multi-slot aggregation may also be allocated for feedback, as illustrated in the one time slot example.

[0105] Figure 4 4 is a block diagram of a first wireless communication device 410 communicating with a second wireless communication device 450, for example, via V2V / V2X / D2D communication or in an access network. The device 410 may include a transmitting device that communicates with a receiving device (e.g., device 450) via V2V / V2X / D2D communication. The communication may be based on, for example, a side link, a downlink, or an uplink. The transmitting device 410 may include a UE, a base station, an RSU, etc. The receiving device 450 may include a UE, a base station, an RSU, etc.

[0106] The IP packets from EPC 160 may be provided to controller / processor 475. Controller / processor 475 implements layer 3 and layer 2 functionality. Layer 3 includes a radio resource control (RRC) layer, and layer 2 includes a service data adaptation protocol (SDAP) layer, a packet data convergence protocol (PDCP) layer, a radio link control (RLC) layer, and a medium access control (MAC) layer. The controller / processor 475 provides RRC layer functionality associated with broadcasting of system information (e.g., MIB, SIB), RRC connection control (e.g., RRC connection paging, RRC connection establishment, RRC connection modification, and RRC connection release), inter-radio access technology (RAT) mobility, and measurement configuration for UE measurement reporting; PDCP layer functionality associated with header compression / decompression, security (encryption, decryption, integrity protection, integrity verification), and handover support functions; RLC layer functionality associated with delivery of upper layer packet data units (PDUs), error correction through ARQ, concatenation, segmentation and reassembly of RLC service data units (SDUs), re-segmentation of RLC data PDUs, and reordering of RLC data PDUs; and MAC layer functionality associated with mapping between logical channels and transport channels, multiplexing of MAC SDUs onto transport blocks (TBs), demultiplexing of MAC SDUs from TBs, scheduling information reporting, error correction through HARQ, priority handling, and logical channel prioritization.

[0107] The transmit (TX) processor 416 and the receive (RX) processor 470 implement layer 1 functionality associated with various signal processing functions. Layer 1 (which includes the physical (PHY) layer) may include error detection on the transmission channel, forward error correction (FEC) decoding / decoding of the transmission channel, interleaving, rate matching, mapping to the physical channel, modulation / demodulation of the physical channel, and MIMO antenna processing. The TX processor 416 processes the mapping to the signal constellation based on various modulation schemes (e.g., binary phase shift keying (BPSK), quadrature phase shift keying (QPSK), M-order phase shift keying (M-PSK), M-order quadrature amplitude modulation (M-QAM)). The decoded and modulated symbols can then be divided into parallel streams. Subsequently, each stream can be mapped to an OFDM subcarrier, multiplexed with a reference signal (e.g., a pilot) in the time domain and / or frequency domain, and then combined using an inverse fast Fourier transform (IFFT) to generate a physical channel carrying a time domain OFDM symbol stream. The OFDM stream is spatially pre-decoded to generate multiple spatial streams. Channel estimates from channel estimator 474 can be used to determine the coding and modulation schemes, as well as for spatial processing. Channel estimates can be derived from reference signals and / or channel condition feedback sent by device 450. Each spatial stream can then be provided to a different antenna 420 via a separate transmitter 418TX. Each transmitter 418TX can modulate an RF carrier with a corresponding spatial stream for transmission.

[0108] At the device 450, each receiver 454RX receives a signal through its corresponding antenna 452. Each receiver 454RX recovers the information modulated onto the RF carrier and provides the information to a receive (RX) processor 456. The TX processor 468 and the RX processor 456 implement layer 1 functionality associated with various signal processing functions. The RX processor 456 can perform spatial processing on the information to recover any spatial stream destined for the device 450. If multiple spatial streams are destined for the device 450, the RX processor 456 can combine them into a single OFDM symbol stream. The RX processor 456 then converts the OFDM symbol stream from the time domain to the frequency domain using a fast Fourier transform (FFT). The frequency domain signal includes a separate OFDM symbol stream for each subcarrier of the OFDM signal. By determining the most likely signal constellation point sent by the device 410, the symbols on each subcarrier and the reference signal are recovered and demodulated. These soft decisions can be based on channel estimates calculated by the channel estimator 458. The soft decisions are then decoded and deinterleaved to recover the data and control signals originally sent on the physical channel by device 410. The data and control signals are then provided to controller / processor 459, which implements layer 3 and layer 2 functionality.

[0109] The controller / processor 459 may be associated with a memory 460 that stores program codes and data. The memory 460 may be referred to as a computer readable medium. The controller / processor 459 provides demultiplexing between transport channels and logical channels, packet reassembly, decryption, header decompression, and control signal processing to recover IP packets from the EPC 160. The controller / processor 459 is also responsible for error detection using ACK and / or NACK protocols to support HARQ operations.

[0110] Similar to the functionality described in conjunction with DL transmissions performed by device 410, the controller / processor 459 provides RRC layer functionality associated with system information (e.g., MIB, SIB) acquisition, RRC connection, and measurement reporting; PDCP layer functionality associated with header compression / decompression and security (encryption, decryption, integrity protection, integrity verification); RLC layer functionality associated with delivery of upper layer PDUs, error correction through ARQ, concatenation, segmentation and reassembly of RLC SDUs, resegmentation of RLC data PDUs, and reordering of RLC data PDUs; and MAC layer functionality associated with mapping between logical channels and transport channels, multiplexing of MAC SDUs onto TBs, demultiplexing of MAC SDUs from TBs, scheduling information reporting, error correction through HARQ, priority handling, and logical channel prioritization.

[0111] Channel estimates derived by the channel estimator 458 from a reference signal or feedback transmitted by the device 410 may be used by the TX processor 468 to select appropriate coding and modulation schemes, and to facilitate spatial processing. The spatial streams generated by the TX processor 468 may be provided to different antennas 452 via corresponding transmitters 454TX. Each transmitter 454TX may modulate an RF carrier with a corresponding spatial stream for transmission.

[0112] Transmissions are processed at device 410 in a manner similar to that described in conjunction with the receiver functionality at device 450. Each receiver 418RX receives a signal through its respective antenna 420. Each receiver 418RX recovers information modulated onto an RF carrier and provides the information to a RX processor 470.

[0113] The controller / processor 475 may be associated with a memory 476 that stores program codes and data. The memory 476 may be referred to as a computer readable medium. The controller / processor 475 provides demultiplexing between transport channels and logical channels, packet reassembly, decryption, header decompression, control signal processing to recover IP packets from the device 450. The IP packets from the controller / processor 475 may be provided to the EPC 160. The controller / processor 475 is also responsible for error detection using ACK and / or NACK protocols to support HARQ operations.

[0114] At least one of the TX processors 416, 468, the RX processors 456, 470, and the controller / processors 459, 475 may be configured to combine Figure 1 The MMIU component 198 of the device may perform various aspects. For example, the controller / processor 459, 475 may include an ML model information update (MMIU) component 498, which is configured to provide a first ML model information update and a measurement associated with the first ML model information update; obtain an aggregated ML model information update, wherein the aggregated ML model information update includes an aggregation of the first ML model information update and a second ML model information update of one of the following: a first network node in a first FL cluster including the device, or a second network node in a second FL cluster; and provide a sidelink communication parameter in the SCI to the network node, wherein the sidelink communication parameter is an output of the ML model associated with the aggregated ML model information update.

[0115] In one example, the MMIU component 498 of the controller / processor 475 in the device 410 may provide first ML model information updates and measurements to the device 450 via the TX processor 416, which may send the first ML model information updates and measurements to the device 450 via the antenna 420. The MMIU component 498 of the controller / processor 475 in the device 410 may obtain aggregated ML model information updates from the device 450 via the RX processor 470, which may receive aggregated ML model information updates from the device 450 via the antenna 420. The MMIU component 498 of the controller / processor 475 in the device 410 may provide sidelink communication parameters to the device 450 via the TX processor 416, which may send the sidelink communication parameters to the device 450 via the antenna 420.

[0116] In another example, the MMIU component 498 of the controller / processor 459 in the device 450 may provide first ML model information updates and measurements to the device 410 via the TX processor 468, which may send the first ML model information updates and measurements to the device 410 via the antenna 452. The MMIU component 498 of the controller / processor 459 in the device 450 may obtain aggregated ML model information updates from the device 410 via the RX processor 456, which may receive aggregated ML model information updates from the device 410 via the antenna 452. The MMIU component 498 of the controller / processor 459 in the device 450 may provide sidelink communication parameters to the device 410 via the TX processor 468, which may send the sidelink communication parameters to the device 410 via the antenna 452.

[0117] In addition, at least one of the TX processors 416, 468, the RX processors 456, 470, and the controller / processors 459, 475 may be configured to combine Figure 1 The AMMIU component 199 of the device may be used to perform various aspects. For example, the controller / processor 459, 475 may include an aggregated ML model information update (AMMIU) component 499, which is configured to obtain a first ML model information update and a measurement associated with the first ML model information update; and provide an aggregated ML model information update, wherein the aggregated ML model information update includes an aggregation of the first ML model information update and a second ML model information update of one of the following: a first network node in a first FL cluster including the device, or a second network node in a second FL cluster, wherein the sidelink communication parameter in the SCI of the network node is an output of the ML model associated with the aggregated ML model information update.

[0118] In one example, the AMMIU component 499 of the controller / processor 475 in the device 410 may obtain the first ML model information updates and measurements from the device 450 via the RX processor 470, which may obtain the first ML model information updates and measurements from the device 450 via the antenna 420. The AMMIU component 499 of the controller / processor 475 in the device 410 may provide the aggregated ML model information updates to the device 450 via the TX processor 416, which may send the aggregated ML model information updates to the device 450 via the antenna 420.

[0119] In another example, the AMMIU component 499 of the controller / processor 459 in the device 450 may obtain the first ML model information updates and measurements from the device 410 via the RX processor 456, which may receive the first ML model information updates and measurements from the device 410 via the antenna 452. The AMMIU component 499 of the controller / processor 459 in the device 450 may provide aggregated ML model information updates to the device 410 via the TX processor 468, which may send the aggregated ML model information updates to the device 410 via the antenna 452.

[0120] Figure 5 1 shows a diagram illustrating an example decomposed base station 500 architecture. The decomposed base station 500 architecture may include one or more CUs 510 (e.g., Figure 1CU 183), the one or more CUs can communicate directly with the core network 520 via a backhaul link, or indirectly with the core network 520 through one or more decomposed base station units (such as a near real-time RIC 525 via an E2 link, or a non-real-time RIC 515 associated with a service management and orchestration (SMO) framework 505, or both). CU 510 can communicate with one or more DU 530 (e.g., Figure 1 DU 530 can communicate with one or more RU 540 (e.g., Figure 1 RU 187) for communication. RU 540 can communicate with UE 104 via one or more radio frequency (RF) access links. In some specific implementations, UE 104 can be served by multiple RU 540 simultaneously.

[0121] Each of the units (i.e., CU 510, DU 530, RU 540, and near-RT RIC 525, non-RT RIC 515, and SMO framework 505) may include or be coupled to one or more interfaces configured to receive or send signals, data, or information (collectively referred to as signals) via a wired or wireless transmission medium. Each of the units or an associated processor or controller that provides instructions to the communication interface of these units may be configured to communicate with one or more of the other units via a transmission medium. For example, the unit may include a wired interface configured to receive or send signals to one or more of the other units on a wired transmission medium. Additionally, these units may include a wireless interface that may include a receiver, a transmitter, or a transceiver (such as a radio frequency (RF) transceiver) that is configured to receive or send signals, or both, to one or more of the other units on a wireless transmission medium.

[0122] In some aspects, the CU 510 may host higher-layer control functions. Such control functions may include radio resource control (RRC), packet data convergence protocol (PDCP), service data adaptation protocol (SDAP), etc. Each control function may be implemented using an interface that is configured to communicate signals with other control functions hosted by the CU 510. The CU 510 may be configured to handle user plane functionality (i.e., central unit-user plane (CU-UP)), control plane functionality (i.e., central unit-control plane (CU-CP)), or a combination thereof. In some specific implementations, the CU 510 may be logically split into one or more CU-UP units and one or more CU-CP units. When implemented in an O-RAN configuration, the CU-UP unit may communicate bidirectionally with the CU-CP unit via an interface (such as an E1 interface). As needed, the CU 510 may be implemented to communicate with the DU 530 for network control and signaling.

[0123] DU 530 may correspond to a logical unit that includes one or more base station functions for controlling the operation of one or more RU 540. In some aspects, DU 530 may host one or more of a radio link control (RLC) layer, a medium access control (MAC) layer, and one or more high physical (PHY) layers (such as modules for forward error correction (FEC) encoding and decoding, scrambling, modulation and demodulation, etc.) depending at least in part on functional divisions such as those defined by the 3rd Generation Partnership Project (3GPP). In some aspects, DU 530 may also host one or more low PHY layers. Each layer (or module) may be implemented using an interface that is configured to communicate signals with other layers (and modules) hosted by DU 530 or with control functions hosted by CU 510.

[0124] The lower layer functionality may be implemented by one or more RUs 540. In some deployments, a RU 540 controlled by a DU 530 may correspond to a logical node that hosts RF processing functions or low PHY layer functions (such as performing Fast Fourier Transform (FFT), Inverse FFT (iFFT), digital beamforming, physical random access channel (PRACH) extraction and filtering, etc.), or both, based at least in part on a functional split (such as a lower layer functional split). In such an architecture, the RU 540 may be implemented to handle over-the-air (OTA) communications with one or more UEs 120. In some implementations, real-time and non-real-time aspects of control plane and user plane communications with the RU 540 may be controlled by the corresponding DU 530. In some scenarios, this configuration may enable the implementation of the DU 530 and the CU 510 in a cloud-based RAN architecture (such as a vRAN architecture).

[0125] The SMO framework 505 may be configured to support RAN deployment and provisioning of non-virtualized network elements and virtualized network elements. For non-virtualized network elements, the SMO framework 505 may be configured to support the deployment of dedicated physical resources for RAN coverage requirements, which may be managed via an operation and maintenance interface (such as an O1 interface). For virtualized network elements, the SMO framework 505 may be configured to interact with a cloud computing platform (such as an open cloud (O-Cloud) 590) to perform network element lifecycle management (such as to instantiate virtualized network elements) via a cloud computing platform interface (such as an O2 interface). Such virtualized network elements may include, but are not limited to, CU 510, DU 530, RU 540, and near-RT RIC 525. In some specific implementations, the SMO framework 505 may communicate with hardware aspects of the 4G RAN (such as an open eNB (O-eNB) 511) via the O1 interface. Additionally, in some specific implementations, the SMO framework 505 may communicate directly with one or more RUs 540 via the O1 interface. The SMO framework 505 may also include a non-RT RIC 515 configured to support the functionality of the SMO framework 505 .

[0126] The non-RT RIC 515 may be configured to include logic functions that enable non-real-time control and optimization of RAN elements and resources, artificial intelligence / machine learning (AI / ML) workflows including model training and updating, or policy-based guidance of applications / features in the near-RT RIC 525. The non-RT RIC 515 may be coupled to or communicate with the near-RT RIC 525 (such as via an A1 interface). The near-RT RIC 525 may be configured to include logic functions that enable near-real-time control and optimization of RAN elements and resources via data collection and actions through an interface (such as via an E2 interface) that connects one or more CUs 510, one or more DUs 530, or both, and the O-eNB with the near-RT RIC 525.

[0127] In some implementations, in order to generate an AI / ML model to be deployed in the near-RT RIC 525, the non-RT RIC 515 may receive parameters or external enrichment information from an external server. Such information may be utilized by the near-RT RIC 525 and may be received from a non-network data source or from a network function at the SMO framework 505 or the non-RT RIC 515. In some examples, the non-RT RIC 515 or the near-RT RIC 525 may be configured to tune RAN behavior or performance. For example, the non-RT RIC 515 may monitor long-term trends and patterns of performance and employ AI / ML models to perform corrective actions through the SMO framework 505 (such as via reconfiguration of O1) or via creation of RAN management policies (such as A1 policies).

[0128] The UE 104 including the controller / processor 459, 475 of at least the device 410, 450, the base station 102 / 180 including the aggregation and decomposition base station 181, or other network nodes can be configured to perform AI / ML tasks using an artificial neural network (ANN). ANN (or simply a neural network) is a computational learning system that uses a functional network to convert data input in one form into a desired output, usually in another form. Examples of ANNs include multilayer perceptrons (MLPs), convolutional neural networks (CNNs), deep neural networks (DNNs), deep convolutional networks (DCNs), and recurrent neural networks (RNNs), as well as other neural networks. ANNs include a layered architecture in which the output of one layer of neurons is input to a second layer of neurons (via connections or synapses), the output of the second layer of neurons becomes the input of the third layer of neurons, and so on. These neural networks can be trained to recognize hierarchical structures of features and are therefore increasingly used in object recognition applications. For example, a neural network can employ supervised learning tasks such as classification (which incorporates ML models such as logistic regression, support vector machines, boosting, or other classifiers) to perform object detection and provide bounding boxes for classifications or categories in an image. Furthermore, back-propagation or gradient descent can be used to fine-tune these multi-layer architectures to produce more accurate predictions.

[0129] Figure 6 An example of a neural network 600 (specifically, a CNN) is illustrated. Figure 6 In an example of , the CNN may be designed to detect objects sensed from a camera 602 (e.g., a vehicle-mounted camera) or other sensor. Thus, in this example, the neural network 600 may be applied to image recognition or object detection. In another example, the CNN may be designed to identify ideal actions that a computing agent may take in any given state, such as the direction a vehicle may move in any given location when navigating to a particular destination. Thus, in this other example, the neural network 600 may be applied to reinforcement learning.

[0130] The neural network 600 may initially receive an input 604, e.g., an image, such as a speed limit sign (or other object or size) of size 32×32 pixels. During a forward pass, the input image is initially passed through a convolutional layer 606 including a plurality of convolutional kernels (e.g., six kernels of size 5×5 pixels or some other amount or size) that slide across the image to detect basic patterns or features, such as straight edges and corners. The images output from the convolutional layer 606 (e.g., six images of size 28×28 pixels or some other amount or size) are passed through an activation function (e.g., a rectified linear unit (ReLU)), and then passed as input into a subsampling layer 608, which scales the size of the image down, e.g., by a factor of two (e.g., producing six images of size 14×14 pixels or some other amount or size). These scaled-down images output from the subsampling layer 608 may similarly be passed through an activation function (e.g., ReLU or other function) and similarly passed as input through subsequent convolutional layers, subsampling layers, and activation functions (not shown) to detect more complex features and further scale down the image or kernel size. These outputs are ultimately passed as input to a fully connected layer 610, where each node output from the previous layer is connected to all neurons in the current layer. The output from this layer may similarly be passed through an activation function and potentially as input through one or more other fully connected layers (not shown). Thereafter, the output is passed as input to an output layer 612, which converts the input into an output 614 such as a probability distribution (e.g., using a softmax function). The probability distribution may include a vector of confidence levels or probability estimates that the input image depicts a predicted feature (such as a sign or speed limit value (or other object)).

[0131] During training, ML models (e.g., classifiers) with weights 616 and biases 618 are initially created for different layers of the neural network 600, respectively. For example, when input from a training image (or other source) enters a given layer of an MLP or CNN, the function of the input and weights summed with the bias can be transformed using an activation function before being passed to the next layer. The probability estimates generated from the final output layer can then be applied to a loss function that measures the accuracy of the ANN, such as a cross entropy loss function. Initially, the output of the loss function may be quite large, indicating that the predicted value is far from the true or actual value. In order to reduce the value of the loss function and produce more accurate predictions, gradient descent can be applied.

[0132] In gradient descent, the gradient of the loss function may be calculated with respect to each weight of the ANN using back propagation, where the gradient is calculated for the last layer back to the first layer of the neural network 600. The gradient may then be used to update each weight to reduce the loss function with respect to that weight until a global minimization of the loss function is obtained, for example using stochastic gradient descent. For example, after each weight adjustment, subsequent iterations of the above training process may be performed using the same or new training images, and if the loss function is still large (even if it has been reduced), back propagation may be applied again to identify the gradient of the loss function with respect to each weight. The weights may be updated again, and the process may continue to be repeated until the difference between the predicted and actual values ​​is minimized.

[0133] UE 104 may use a Figure 6 The ANN of the neural network 600 is used to perform AI / ML tasks in various applications, such as autonomous driving (e.g., using self-driving cars) and industrial automation (e.g., using learning-based robots). One such task is reinforcement learning, in which the computing agent of the UE builds an ML model that maps inputs (states in state space) to outputs (actions in action space) by applying a reward and penalty system to maximize a reward function. For example, an autonomous vehicle may apply reinforcement learning to meet speed limits and avoid collisions with other vehicles, while a manufacturing robot may apply reinforcement learning to minimize energy consumption. These applications and other use cases of automation may involve the UE sharing data with other UEs 104 in sidelink communications, such as sensor data or driving intent in V2X communications, to result in improved safety, business efficiency, environmental awareness, or other automation benefits.

[0134] According to one or more examples, in sidelink communication, the first wireless communication device 410 (e.g., Tx UE) initially achieves sidelink synchronization with the second wireless communication device 450 (e.g., Rx UE). After synchronization, the Tx UE may obtain an allocation of time-frequency resources (e.g., one or more time slots, RBs, or subchannels in a resource pool) in which to send sidelink data to the Rx UE. In one example, the resource allocation may be scheduled by the base station in downlink control information (DCI) (in mode 1 resource allocation), or the resource allocation may be determined by a sensing process performed autonomously by the Tx UE (in mode 2 resource allocation). After determining the resources, the Tx UE may transmit sidelink control information (SCI) including the resource allocation to the Rx UE in a physical sidelink control channel (PSCCH).

[0135] The Tx UE may send SCI in two stages, including a first-stage SCI carried on the PSCCH and a second-stage SCI carried on the physical sidelink shared channel (PSSCH). The first-stage SCI may contain information about resource allocation, such as frequency domain resource allocation (FDRA) and time domain resource allocation (TDRA), which indicates the subchannels and time slots selected for carrying the transmission of the sidelink data transmission and reserved for up to two further retransmissions of the sidelink data transmission. The first-stage SCI may also include a resource reservation period for semi-persistently reserved resources, a time mode and number of ports for the PSSCH DMRS, a size and format of the second-stage SCI, or other SCI parameters. The second-stage SCI may carry information for identifying and decoding the sidelink data, such as a source identifier (source ID) and a destination identifier (destination ID) of a sidelink packet carried in the PSSCH, a hybrid automatic repeat request (HARQ) process ID, a new data indicator, a redundancy version, or other SCI parameters. The second stage SCI may also include (eg, in SCI format 2-B) a required communication range for groupcast HARQ feedback and a zone ID of the Tx UE indicating the zone the Tx UE is located in. The Tx UE may send sidelink data in the PSSCH to the Rx UE in the allocated resources.

[0136] After receiving the sidelink transmission, the Rx UE may attempt to blindly decode the PSCCH in all allocated subchannels of the resource pool. If the Rx UE successfully decodes the PSCCH, the UE may also attempt to decode the PSSCH scheduled by the PSCCH for the sidelink data. Based on the decoding result, the Rx UE may provide a hybrid automatic repeat request (HARQ) acknowledgement / negative acknowledgement (ACK / NACK) feedback to the Tx UE in the physical sidelink feedback channel (PSFCH). If the sidelink data transmission is unicast, the Rx UE may provide an ACK to the Tx UE if the sidelink data is successfully decoded, or provide a NACK to the Tx UE if the Rx UE fails to decode the sidelink data. If the sidelink data transmission is multicast, the Rx UE may have two options for transmitting HARQ feedback to the Tx UE: multicast option 1 and multicast option 2. In multicast option 1, if the Rx UE is within the specified communication distance (the required communication range or Tx-Rx distance indicated in the second stage SCI) from the Tx UE, the Rx UE may transmit NACK feedback (but not ACK feedback). In multicast option 2, the Rx UE may transmit ACK / NACK feedback regardless of the Tx-Rx distance. In the unicast or multicast case, if the Tx UE receives a NACK from the Rx UE, the Tx UE may retransmit the sidelink data. Otherwise, if the Tx UE receives an ACK from the Rx UE (or does not receive a reply in multicast option 1), the Tx UE may send new data to the Rx UE, or send data to a different Rx UE.

[0137] Therefore, multiple SCI parameters may be applied to sidelink communications between different Tx UEs and Rx UEs (e.g., FDRA, TDRA, amount of reserved resources, resource reservation period, PSSCH DMRS time pattern and port, second stage SCI size and format, source ID, destination ID, HARQ process ID, new data indicator, redundancy version, required communication range for multicast HARQ feedback, Tx UE zone ID, MCS, etc.). While in mode 1 resource allocation, the base station may select these parameters to coordinate sidelink transmission and reception between different UE pairs, in mode 2 resource allocation, the UEs themselves independently select multiple parameters. For example, one Tx UE may select one communication range for multicast option 1 feedback, one reserved resource amount for retransmission, or one MCS to optimize one application-related metric (e.g., minimum delay), while another Tx UE may select a different communication range for multicast option 1 feedback, a different reserved resource amount for retransmission, or a different MCS to optimize different application-related metrics (e.g., maximized packet reliability). Furthermore, while some UEs may select preconfigured values ​​for these parameters, other UEs may select values ​​for these parameters based on independent measurements such as channel busy rate (CBR), channel occupancy rate (CR), or other metrics for congestion control. For example, a UE may select values ​​for these parameters in a neural network (e.g., Figure 6 AI / ML is employed in a neural network 600 of the UE to output values ​​that the UE can apply to sidelink communication parameters in response to input from an observation data set independent of the UE. For example, the UE can apply reinforcement learning to derive an optimal communication range for multicast option 1 based on a reward and penalty system (e.g., throughput) that can be measured by the UE in response to observed CBR, packet reception rate, or other metrics.

[0138] Therefore, if UEs independently select such parameters in Mode 2 resource allocation, the lack of coordination between UEs may result in suboptimal network performance. For example, a Tx UE that independently selects a significantly large Tx-Rx distance for multicast option 1 may cause significant congestion due to multiple NACK feedbacks from various Rx UEs that are too far away from the Tx UE to successfully decode the sidelink data transmission. This congestion may in turn increase the CBR or CR metrics measured by other Tx UEs, causing these other Tx UEs to inefficiently reduce their MCS, number of retransmissions, or other transmission parameters, resulting in reduced system performance. In addition, since many of these sidelink communication parameters may be applied to automated use cases (such as autonomous driving), the reduced system performance may undesirably affect safety or other aspects of automation. Therefore, it would be helpful to provide coordination between UEs in sidelink communications to improve overall network performance, especially in automated use cases that employ AI / ML in neural networks.

[0139] Figure 7 An example 700 of a UE location map showing different congestion levels observed by a UE 702 in an urban environment is illustrated. Although the distribution of UEs 702 in the illustrated example is particularly applicable to VUEs (UEs in vehicles), the UEs here are not limited to VUEs, but can be, for example, distributed in a manner similar to that of a vehicle. Figure 7 Pedestrian UEs are arranged in different distributions as shown. In this example, the chart shows the xy coordinates of UE 702 relative to the origin (0,0) in the grid representing intersecting streets. As illustrated in the example chart, different streets may have different densities of UE 702 and therefore have different levels of congestion. For example, street 704 may have high congestion, while street 706 may have low congestion. UEs 702 on streets with different levels of congestion may observe different packet reception rates (PRRs), delays, or other metrics (e.g., Tx-Rx distances), and therefore during mode 2 sidelink resource allocation, the UEs may select different sidelink communication parameters (e.g., multicast option 1 communication ranges output from the ML model based on different measured CBRs or CRs). In addition, UEs performing congestion control on the same street as other UEs may select different sidelink communication parameters (e.g., preconfigured based on channel comparison). Therefore, without coordination between UEs 702, network performance may be affected.

[0140] FIG. 8A to FIG. 8B are diagrams of examples 800, 850 illustrating the relationship between PRR 802, 852 and Tx-Rx distance 804, 854 for UEs in congested and non-congested urban environments, respectively, for different communication ranges 806, 856 in multicast option 1. As can be seen in the illustrated example, a multicast Tx UE that selects different communication ranges 806, 856 (e.g., a desired communication range of 20m, 50m, 80m, 100m, 120m, 150m, 180m, or 200m) in which the Rx UE may provide NACK feedback may observe different relationships between PRR 802, 852 and Tx-Rx distance 804, 854 for different congestion levels (e.g., high congestion or low congestion). For example, when the Tx-Rx distance 804, 854 increases, the PRR 802, 852 observed by the Tx UE from the Rx UE may decrease at different rates in environments with different congestion levels according to the desired communication range 806, 856 selected by the Tx UE for NACK feedback. Therefore, for Tx UEs experiencing different levels of congestion, it would be helpful to coordinate the selection of the desired communication range for multicast option 1 to maximize the corresponding PRR and improve network performance.

[0141] One technique that can be applied to achieve coordination between network nodes is federated learning (FL), which applies AI / ML tasks to obtain sidelink communication parameters. Compared with centralized machine learning techniques that upload all local data sets to one server, FL allows the generation of high-quality ML models without aggregating distributed data. Therefore, FL facilitates parallel processing, significantly reduces the costs associated with message exchange, and protects data privacy.

[0142] In the FL framework, nodes such as UEs can learn a global ML model by passing messages between nodes through a central coordinator. For example, nodes can provide weights, biases, gradients, or other ML information to each other (other nodes) through messages exchanged between nodes via a central coordinator (e.g., base station, RSU, edge server, etc.).

[0143] Each node in the FL environment uses a dataset to locally train and update a coordinated global ML model. The dataset can be a local dataset that a node or device can obtain for a certain ML task (e.g., object detection, etc.). The data in a given dataset can be preloaded or accumulated over the life of the device. For example, an accumulated dataset may include recorded data that a node observes from an onboard sensor such as a camera and stores locally at the device. The global ML model can be defined by its model architecture and model weights. An example of a model architecture is a neural network, such as Figure 6 A CNN is described, which may include multiple hidden layers, multiple neurons in each layer, and synapses connecting these neurons together. Model weights are applied to data passed through the various layers of the ML model for processing by the various neurons.

[0144] Fig. 9 An example 900 of a FL architecture is illustrated. The FL parameter server 902 initializes a global ML model 904 ( G ) (e.g., with random or default ML weights such as Figure 6 ) and broadcasts the global ML model 904 to participating devices or learning nodes 906 (e.g., UEs). After receiving the global ML model at the learning node, the iterative process of FL can begin. For example, at a given iteration t, the global ML model 904 (W t G ), the learning node 906 (k nodes 906a, 906b, 906c, 906d) uses its local data set to locally train its corresponding model 908 (model 908a, 908b, 908c, 908d), which can be a preconfigured data set such as a training set and / or sensory data from the environment. After training, the learning node updates its corresponding k-th model W tk (e.g., local weights) until the kth loss function or cost function F for the model is achieved k (W t k ) is minimized or optimized. At this time, due to different update weights, the learning nodes may have different local models W t k .

[0145] Afterwards, the learning node 906 will update the model W corresponding to its local t k The corresponding information is sent to the FL parameter server 902. To reduce the transmission traffic, the corresponding information can be, for example, the modified weights, the weights that have changed beyond a threshold, or the multiplicative or additive increments or percentages of these weights. Upon receiving the corresponding information, the FL parameter server aggregates the corresponding weights of the corresponding models 908 (e.g., by averaging the weights or performing some other calculation on the weights). The FL parameter server 902 can therefore generate an updated global model W including the aggregated weights. t+1 G , after which the aforementioned process may be repeated for subsequent iteration t+1. For example, the FL parameter server 902 may broadcast the updated global model W to the learning node 906 t+1 G , to perform training and local updates again based on the same data set or a different local data set. To reduce transmission traffic, broadcast updates may include sending modified weights, weights that change more than a threshold, or multiplicative or additive increments or percentages of these weights. After the node receives the broadcast update and performs training and local updates, the node may again share its corresponding updated model with the FL parameter server for subsequent aggregation and global model updates. This process may continue to repeat in further iterations and result in updates to the global model W. t G Further update of the global model loss function F G (W t G ) is minimized, or until a predetermined number of iterations has been reached.

[0146] Fig. 10A and Fig. 10B Examples 1000, 1050 of different applications associated with wirelessly connected devices that may benefit from FL are illustrated, including connected vehicles for autonomous driving ( Fig. 10A ) and mobile robots for manufacturing ( Fig. 10B ). For example, in Fig. 10AIn example 1000, network nodes 1002 (nodes 1002a, 1002b, 1002c, e.g., such as Fig. 9 The VUE of the learning node 906 in the example may communicate with the FL parameter server 1004 (e.g., a base station or RSU) to collaboratively train and enhance the accuracy of the neural network 1006 (neural networks 1006a, 1006b, 1006c, 1006d) for detecting objects and object bounding boxes (OBBs) in conjunction with autonomous driving. Similarly, in Fig. 10B In example 1050, network nodes 1052 (nodes 1052a, 1052b, 1052c, such as, for example, Fig. 9 The mobile robot UE of the learning node 906 in the can communicate with the FL parameter server 1054 (e.g., a base station) at the mmWave frequency to collaboratively train the neural network 1056 (neural networks 1056a, 1056b, 1056c, 1056d) associated with the manufacturing-related tasks.

[0147] However, the FL architecture relies on a centralized server (e.g., FL parameter servers 902, 1004, 1054) to create, aggregate, and refine global ML models for participating nodes, and therefore requires that locally trained ML model information be sent from participating nodes to the server during FL iterations. To reduce the sending traffic, the information may include, for example, modified weights, weights that have changed beyond a threshold, or multiplicative or additive increments or percentages of these weights. This centralized approach to FL may have various characteristics. In one example, since the centralized server is the only aggregator of participating nodes, the centralized server may serve the FL system alone. In contrast, if the ML model aggregator role is assigned outside the centralized server, this may facilitate the advancement of the FL process. In another example, participants transmit multiple local model updates to the centralized server. Therefore, distributing the entities to which participants transmit such updates may minimize communication overhead in applications where multiple updates are applied. In another example, the FL architecture includes heterogeneity in computing resources present for different participants, heterogeneity in training data available to participants, and heterogeneity in training tasks and associated models configured for different participants. In contrast, a FL architecture with such homogeneity of computing resources, training data, training tasks, and associated models can provide statistical simplicity in model training. In another example, even if raw data is not directly communicated between nodes in FL, the exchange of ML model parameters with a distributed ML model aggregator can minimize the leakage of information about the underlying data samples, thereby improving the security and privacy of ML model information.

[0148] To improve the FL architecture, a clustering or hierarchical approach may be applied to FL, where learning nodes working towards a common learning task are grouped together into clusters. In clustered FL, multiple clusters may be formed by respective groups of learning nodes (e.g., UEs) that have local datasets and share a common ML task, such as object detection or classification (e.g., detecting OBBs, vehicles or pedestrians on the road, etc.). In one example, UEs may be grouped together to form a cluster and designate a cluster leader (e.g., other UEs) without network assistance (e.g., without base station or RSU participation), but in other examples, cluster formation and leader identification may be configured with network assistance (e.g., via messages that identify clusters and confirm cluster leaders that circulate within the network). For example, a participating node in a cluster may be selected or designated as a leader by other participating nodes in the cluster without network participation, or may be selected or self-designated as a leader with network participation. The cluster leader itself may also be a learning node within the corresponding cluster.

[0149] In addition, in hierarchical FL, the designated cluster leader of each cluster, rather than the FL parameter server, directly coordinates learning tasks, including local ML model training and updates within the cluster. This coordination is referred to as intra-cluster FL. For example, each node within a cluster may pass a message including a local update to the ML model weights to a cluster leader (e.g., a UE designated as the leader of the cluster by an identifier) ​​to aggregate the updated local model and transmit it back to each node within the cluster. After the cluster is formed, a centralized FL parameter server (e.g., an edge server, RSU, or base station) may coordinate learning tasks, including global ML model training and updates between clusters. This coordination is referred to as inter-cluster FL. For example, each cluster leader of the corresponding cluster may pass a message including an aggregated local update to the ML model weights to the FL parameter server to aggregate the updated global ML model and transmit it back to each cluster leader, which in turn passes the updated global model to its corresponding cluster members for further training and updating. Therefore, the cluster leader can act as an intermediary between the learning nodes and the FL parameter server for coordinating neural network training and optimization between different clusters.

[0150] Fig.11 An example 1100 of a clustered FL architecture including a cluster 1102 (clusters 1102a, 1102b) of UEs 1104 (UEs 1104a, 1104b, 1104c, 1104d, e.g., vehicle UEs, pedestrian UEs, etc.) communicating with a FL parameter server 1106 (e.g., a base station, RSU, edge server, or other network entity) is illustrated. UEs 1104 that share a common learning task, model, data set, or computing resource may be grouped into one or more clusters, and multiple clusters 1102 of such UEs may be formed. A cluster leader 1108 (cluster leaders 1108a, 1108b) may be designated for a cluster with or without network assistance. For example, without network assistance, a UE 1104 may elect or subscribe to one of the UEs as the cluster leader 1108 for a given cluster. In an alternative example, with network assistance, the FL parameter server 1106 (e.g., a base station) may designate an RSU, edge server, or other network entity as the cluster leader 1108 for a given cluster (if the network entity is in communication with the UEs 1104 of the corresponding cluster), as well as designating the UEs 1104 that participate in the corresponding cluster. Fig.11 The example of illustrates two UEs communicating with the cluster leader 1108, but it should be understood that the cluster 1102 is not limited to this number of UEs and may have a different number of UEs in other examples.

[0151] UEs 1104 in a given cluster may communicate with Fig. 9The FL process described above is similar to the training process, except that the cluster leader 1108 acts as an aggregator for its corresponding cluster rather than an FL parameter server, and model training is performed at multiple levels (intra-cluster and inter-cluster). In one example, the cluster leader 1108 may act as an aggregator for its corresponding cluster 1102 without participating in model training itself. For example, the cluster leader 1108 of a given cluster may multicast the initialized ML model to the UEs 1104 within the cluster 1102 to perform local training, and these UEs then transmit their individually updated ML models 1110 (models 1110a, 1110b, 1110c, 1110d for different UEs, respectively) back to the cluster leader 1108 to be aggregated into the updated local model 1112 for the cluster. The cluster leader 1108 may then multicast the updated local model 1112 (models 1112a, 1112b for different cluster leaders, respectively) to the UEs 1104 to perform local training again in the next iteration, and the intra-cluster process may be repeated in further iterations. In addition, at certain times or in response to certain events, the cluster leaders 1108 may communicate their respective updated local models to the FL parameter servers 1106 for aggregation into an updated global model 1114 for the cluster. The FL parameter servers 1106 may then transmit these aggregated global models back to the cluster leaders 1108 to circulate within their respective clusters 1102. In another example, the cluster leaders 1108, in addition to acting as an aggregator for their respective clusters 1102, may also participate in model training themselves. In this case, the cluster leaders 1108 may perform the aforementioned functions of the aggregator in addition to performing local training and updates similar to the UE 1104. Therefore, unlike in FL where the learning nodes communicate directly with the FL parameter servers for model optimization, here the cluster leaders communicate directly with the FL parameter servers, and the learning nodes within the cluster communicate directly with the cluster leaders instead. Therefore, clustered FL can be used to implement neural network training in a distributed manner, with facilitated FL process advancement, minimized communication overhead, statistical simplicity in model training due to the homogeneity of computing resources, training data, training tasks, or associated ML models, and improved security and privacy of ML model information.

[0152] It is also possible to use a peer-to-peer FL architecture instead of, for example, Fig.11 1106 to improve FL. In a peer-to-peer approach, collaborative ML model training between participating nodes can be provided without the need for a designated and centralized server for coordinating training (e.g., no FL parameter server 1106). Instead, participating nodes communicate directly with other participating nodes to coordinate and schedule neural network training, individual model updates, and aggregation of individually updated models.

[0153] Fig.12An example 1200 of a peer-to-peer FL architecture including UEs 1202 (UEs 1202a, 1202b, e.g., vehicle UEs, pedestrian UEs, etc.) communicating directly with other UEs for ML model training is illustrated. In one example, the UEs 1202 may elect one of the UEs 1202 as a leader 1204 for a period of time to coordinate ML model training among the UEs 1202. In one example, the leader 1204 may act as an aggregator without participating in model training itself. During training, the leader 1204 may receive and aggregate ML model updates from the UEs 1202, after which the leader may transmit the aggregated ML model updates back to the UEs 1202 for further training and updating. This training process is similar to that described with respect to Fig.11 The intra-cluster method described above is similar to that described above, except that the cluster leader in that example is replaced here by leader 1204. In another example, leader 1204, in addition to acting as an aggregator, may also participate in model training itself, in which case leader 1204 may perform the aforementioned functions of an aggregator in addition to performing model training and updates similar to UE 1202. In an alternative example, no one UE may be leader 1204, and UE 1202 may instead synchronously or asynchronously aggregate ML model updates of other UEs 1202. Thus, neural network training may be similarly implemented in a distributed manner using peer-to-peer FL, with facilitated FL process advancement, minimized communication overhead, statistical simplicity in model training due to homogeneity of computing resources, training data, training tasks, or associated ML models, and improved security and privacy of ML model information. Although Fig.12 The example of illustrates two UEs communicating with the leader 1204, but it should be understood that the peer-to-peer network is not limited to this number of UEs and may have a different number of UEs in other examples.

[0154] Such as about Fig.11 Clustered FL architectures described or such as Fig.12The described peer-to-peer FL architecture can assist UE104, 702, 1104, 1202 or leader 1204 in coordinating the side link parameters derived from the ML model. For example, participant nodes in a cluster or peer network can respectively apply updated ML models aggregated from information of other participant nodes to derive the desired communication range for multicast option 1 or other common side link parameters. In addition, participant nodes with similar channel conditions, congestion levels, traffic statistics, or other attributes can be grouped together in a cluster or included in the exchange of updated ML model information. Such grouping allows participant nodes with similar CBR, CR, PRR or other observed metrics to obtain similar updated ML models, resulting in a higher probability of common SCI parameter output. Therefore, participant nodes in the same cluster or peer network can derive common SCI parameters from their updated ML models for application in side link communications, resulting in maximized PRR or throughput or minimized latency in deployments using mode 2 resource allocation.

[0155] As an example, refer back to Figure 7 , Fig. 8A and Figure 8B , a UE 702 on a street 704 with high congestion may experience Fig. 8A The relationship between PRR 802 and Tx-Rx distance 804 for different communication ranges 806 is illustrated, and a UE 702 on a street 706 with low or no congestion may experience Figure 8B The relationship between PRR 852 and Tx-Rx distance 854 for different communication ranges 856 is illustrated. Fig. 8A In the example of (high congestion), a higher PRR at a large Tx-Rx distance can be achieved using a smaller communication range for multicast option 1, while Figure 8B In the example of (low or no congestion), a higher PRR at a larger Tx-Rx distance can be achieved using a larger communication range for multicast option 1. Therefore, in order to provide optimized sidelink parameter coordination between UEs 702, UEs 702 that observe similar high congestion levels on street 704 can be grouped into one cluster or peer network, while UEs 702 that observe similar low congestion levels on street 706 can be grouped into different clusters or peer networks. The corresponding group of UEs 702 can use intra-cluster or peer-to-peer FL to locally train their ML models, and in the case of clustered FL, the FL parameter server, base station or other network node can use inter-cluster FL to control the dynamic evolution of the learning process. Thereafter, the corresponding group of UEs 702 can apply their ML models to obtain the optimal communication range for multicast option 1 that is common to the UEs in the corresponding group, and the UEs 702 can include the optimal communication range in the corresponding SCI for optimizing sidelink data communication.

[0156] Fig.13 An example 1300 of cluster formation is illustrated, wherein UEs 1302 (UEs 1302a, 1302b, 1302c, 1302d, 1302e, 1302f, 1302g, 1302h, 1302i, 1302j, 1302k) in congested and non-congested environments are grouped into different clusters 1304 (clusters 1304a, 1304b) or peer networks for clustering or peer FL, respectively. Although the illustrated example specifically shows that UEs 1302 are grouped into two clusters or peer networks, in other examples, UEs 1302 may also be grouped into different numbers of clusters or peer networks (e.g., at least one cluster or peer network). In one example, clusters 1304 may be based on zones. For example, UEs 1302 in the same intersection may be in the same zone and therefore may be assigned to the same cluster. With or without assistance from base station 1308, for the corresponding cluster selection (e.g., as described below with respect to Fig.16 described), select (e.g., as described below with respect to Fig.18 described) or specified (e.g., as described below with respect to Fig.17 The cluster leaders 1306 (cluster leaders 1306a, 1306b) of the RSU (described above) may be RSUs or other network nodes covering a specific geographic area, such as an intersection. If the UEs 1302 within the corresponding area or geographic area covered by the RSU (or other network node) experience similar congestion levels or other channel conditions, these UEs may be grouped into the same cluster. For example, in the illustrated example, cluster 1 may have fewer UEs and lower congestion, while cluster 2 may have more UEs and higher congestion. In addition, if the UEs run the same ML task (e.g., reinforcement learning, object detection, classification, etc.), the UEs 1302 may be assigned to the same cluster, allowing these UEs to apply aggregated ML model updates obtained from other UEs in the cluster. Although Fig.13 The example illustrates two ground UEs (UE 1302a, 1302b) and two aviation UEs (UE 1302h, 1302i) communicating with cluster leader 1306 in one cluster in cluster 1304, and five ground UEs (UE 1302c, 1302d, 1302e, 1302f, 1302g) and two aviation UEs (UE 1302j, 1302k) communicating with cluster leader 1306 in another cluster in cluster 1304, but it should be understood that cluster 1304 is not limited to these numbers of UEs and may have different numbers of UEs (ground or aviation) in other examples.

[0157] In order to perform sidelink parameter optimization using clustered FL or peer-to-peer FL, one or more signaling procedures may be provided that allow a network node (e.g., a UE) to discover and form a cluster or peer-to-peer network with other network nodes, to select or elect a leader for a cluster or peer-to-peer network, or to otherwise implement clustered FL or peer-to-peer FL. These signaling procedures may include, for example: (1) a neighbor discovery procedure, in which a node (e.g., a UE) may discover other nodes (e.g., other UEs or FL parameter servers) that are close or adjacent to each other and that may participate in the ML training task, for example, including the following regarding Fig.14 or Fig.15 (2) a signaling process for cluster formation and leader selection, in which nodes participating in the ML training task can form a cluster with other discovered nodes (without network participation) and further select a cluster leader (ML model weight aggregator) for the corresponding cluster based on configured criteria, for example, including the following about Fig.16 (3) a combined signaling process for neighbor discovery and cluster formation and leader selection or election, in which nodes participating in an ML training task may join a cluster led by other nodes in response to corresponding requests from other nodes, for example, including the following about Fig.18 (4) a signaling process for cluster formation and leader selection, in which nodes participating in an ML training task are grouped together into clusters, where other nodes are led by a selected cluster leader (ML model weight aggregator) of the corresponding cluster based on configured criteria (with network participation), for example, including the following about Fig.17 and (5) a signaling process for leader election in a peer-to-peer FL setting, where nodes participating in an ML training task may elect a leader for ML model weight aggregation among other discovered nodes based on configured criteria, such as those described below. Fig.16 Signaling described. The foregoing examples of signaling procedures may apply sidelink communications between devices such as UEs (e.g., via a PC5 interface), downlink / uplink communications between UEs and network entities such as base stations or RSUs (e.g., via a Uu interface), or a combination of sidelink and downlink / uplink communications. Following these signaling procedures, network nodes with similar distributions of observed data may be grouped into clusters, thereby maintaining the homogeneity of data within the corresponding clusters. For example, UEs that train local ML models separately based on separately collected data (such as a time series of CBR measurements, achieved throughput, or other metrics) and that are affected by similar congestion levels may be grouped with other UEs into corresponding clusters led by a designated cluster leader.

[0158] After completing cluster formation and leader selection (without network participation) or leader selection (with network participation), the network node can implement side link parameter coordination in mode 2 resource allocation configuration by dynamically coordinating one or more signaling processes of clustered FL (intra-cluster or inter-cluster) or peer FL. These signaling processes may include signaling from participant nodes (e.g., UE) to leaders (e.g., UE, RSU, edge server, local FL server or other network nodes), signaling from leaders to FL parameter network entities (e.g., base stations, FL parameter servers or other network nodes), signaling from FL parameter network entities to leaders, and signaling from leaders to participant nodes.

[0159] In one example, signaling for clustered FL training may be provided, wherein a node may perform intra-FL training coordinated by a cluster leader within a corresponding cluster through messages passed between a learning node and a cluster leader. In another example, signaling for clustered FL training may be provided, wherein a node may perform inter-FL training coordinated by a FL parameter server across different clusters through message exchanges between a corresponding cluster leader and a FL parameter server. In another example, signaling for peer FL training may be provided, wherein a leader coordinates FL training between peer nodes. In another example in which none of the peer nodes operates as a leader, signaling for peer FL training may be provided, wherein a peer node synchronously coordinates FL training between other peer nodes. In another example in which none of the peer nodes operates as a leader, signaling for peer FL training may be provided, wherein a peer node asynchronously coordinates FL training between other peer nodes. The foregoing examples of signaling procedures may apply sidelink communications (via a PC5 interface) between UEs or other wireless communication devices.

[0160] Therefore, participant nodes can perform FL training within a cluster and across clusters, or within a peer-to-peer network. For example, a cluster or peer-to-peer network may be assigned a leader responsible for coordinating the FL process within its corresponding cluster or peer-to-peer network. The leader may aggregate local ML models reported by each participant node that observes similar business statistics, and the cluster leader may communicate this information to the FL parameter network entity together with the summary statistics associated with its corresponding cluster. The FL parameter network entity can dynamically control the evolution of the FL process between clusters. For example, the FL parameter network entity may aggregate ML models across clusters that observe similar data distributions about a certain metric, and the FL parameter network entity may dynamically assign participant nodes to different clusters based on the aggregated information. The FL parameter network entity may communicate inter-cluster aggregation and dynamic cluster assignments to the cluster leader, which in turn communicates this information to the participant nodes. The leader of the peer-to-peer network may similarly communicate the aggregated ML model information of the participant nodes to each participant node. The participant nodes in the cluster or peer-to-peer network may apply the aggregated ML model to their corresponding data sets to derive common SCI parameters between the participant nodes in the cluster or peer-to-peer network. Thus, sidelink communication performance may be improved between network nodes having similar congestion levels, traffic statistics, or similar attributes.

[0161] Fig.14 An example 1400 of a signaling process that allows a first node 1402 (e.g., a UE referred to herein as node 1) to discover potential neighboring nodes such as a second node 1404 (e.g., another UE referred to herein as node 2) using sidelink communication is illustrated. Referring to the previous figures, node 1 and node 2 may be, for example, UEs 104, 1104, 1202, 1302 or network nodes 906, 1002, 1052. Initially, the first node 1402 may send a neighbor discovery (ND) message 1406 or request. The ND message 1406 may be provided periodically (e.g., at a preconfigured periodicity) or in response to an event. For example, the first node 1402 may be triggered to transmit the ND message 1406 after receiving a basic safety message (BSM) indicating the presence of another node (e.g., a neighboring VUE). The ND message 1406 may be broadcast to solicit a response from any node (e.g., VUE) that is capable of decoding the ND message 1406. Alternatively, the ND message 1406 may be multicast, for example, to nodes that provide the BSM to the first node 1402. The ND message 1406 may also be transmitted using multicast option 1 to enhance the reliability of the ND message 1406 (e.g., the first node 1402 may indicate in the second stage SCI the distance at which the first node 1402 expects to receive a NACK from other nodes that failed to decode the message). In this example, the node that receives the ND message 1406 from the first node 1402 is referred to herein as node 2 (the second node 1404).

[0162] In response to receiving the ND message 1406 from the first node 1402, the second node 1404 may add the first node 1402 to a neighbor list, a communication map, or other data structure (stored at the second node 1404) indicating that the first node 1402 is an asymmetric neighbor. The second node 1404 may indicate the first node 1402 as an "asymmetric" neighbor because at this time the second node 1404 knows its communication capabilities with the first node 1402, but the first node 1402 does not yet know its communication capabilities with the second node 1404. The second node 1404 may then provide a neighbor discovery response (NDR) message 1408 confirming that the first node 1402 is an asymmetric neighbor of the second node 1404. The NDR message 1408 may be unicast provided to the first node 1402. In response to receiving the NDR message 1408, the first node 1402 may add the second node 1404 to a neighbor list, a communication map, or other data structure (stored at the first node 1402) indicating that the second node 1404 is a symmetric neighbor. The first node 1402 may indicate the second node 1404 as a "symmetric" neighbor because at this point both the first node 1402 and the second node 1404 are aware of their ability to communicate with the other node. The first node 1402 may then provide a message 1410 (e.g., via unicast) to the second node 1404 confirming that the second node 1404 is a symmetric neighbor of the first node 1402. In response to receiving the message 1410, the second node 1404 may update the state of the first node 1402 in its data structure to similarly indicate that the first node 1402 is now a symmetric neighbor. The second node 1404 may then transmit a message 1412 (e.g., via unicast) to the first node 1402 confirming that the first node 1402 has become a symmetric neighbor of the second node 1404.

[0163] After the first node 1402 and the second node 1404 have designated each other as neighbors (i.e., symmetric neighbors), the first node 1402 may send a message 1414 indicating model training related information to the second node 1404. The information may include, for example: the ML tasks in which the first node 1402 is participating or interested in participating; the available sensors at the first node 1402 and their associated data input formats; the available ML models in training, including model state, model architecture, model training parameters, and current performance (e.g., accuracy and loss); and available computing resources (e.g., whether the first node 1402 includes a CPU or GPU, information about its clock speed, available memory, etc.). For example, the message 1414 may indicate that the first node 1402 is configured to perform object detection (e.g., to identify OBBs), image classification, reinforcement learning, or other ML tasks. The message 1414 may indicate that the first node 1402 includes a camera, LIDAR, RADAR, IMU, or other sensor, and a data format that can be read by the indicated sensor. The message 1414 may indicate that the first node 1402 is configured with a neural network (e.g., Figure 6 The message 1414 may indicate the number of neurons and synapses, the logistic regression or classification model, the current values ​​of the model weights and biases, the activation function applied (e.g., ReLU), the classification accuracy, and the loss function applied to the model weight update (e.g., cross entropy loss function). The message 1414 may indicate that the CPU or GPU of the first node 1402 (e.g., the controller / processor 459 or some other processor of the UE 450) applies the indicated neural network and model to the indicated ML task, the clock speed of the CPU or GPU, and a certain amount of available memory (e.g., the memory 460 or some other memory of the UE 450) for storing information associated with the ML task, the neural network, data, etc. The message 1414 may include any combination of the foregoing items as well as other information.

[0164] In response to receiving the message 1414 including the model training related information, at operation 1415, the second node 1404 associates the information with the first node 1402 in its neighbor list, communication graph, or other data structure. If deciding whether to form a cluster with the first node 1402 to perform an ML task, the second node 1404 may later consider the information during cluster formation. For example, if the information indicates that the first node 1402 has a different ML model than the second node 1404 or has lower computing power than it, the second node 1404 may determine not to cluster with the first node 1402.

[0165] The second node 1404 may then transmit a message 1416 indicating model training related information to the first node 1402. The information may include, for example: the ML tasks that the second node 1404 is participating in or interested in participating in; the available sensors at the second node 1404 and their associated data input formats; the available ML models in training, including model status, model architecture, model training parameters, and current performance (e.g., accuracy and loss); and available computing resources (e.g., whether the second node 1404 includes a CPU or GPU, information about its clock speed, available memory, etc.). Therefore, the message 1416 may include the same type of information as the message 1414, but for the second node 1404 instead of the first node 1402. In response to receiving the message 1416 including the model training related information, at operation 1417, the first node 1402 associates the information with the second node 1404 in its corresponding data structure. If deciding whether to form a cluster with the second node 1404 to perform the ML task, the first node 1402 may similarly consider the information later during cluster formation.

[0166] Fig.15 An example 1500 of another signaling process that allows a first node 1502 (e.g., a UE similarly referred to herein as Node 1) to discover potential neighboring nodes such as a second node 1504 (e.g., another UE similarly referred to herein as Node 2) using sidelink communications is illustrated. The signaling process combines Fig.14 Neighbor discovery and model-related information exchange messages, resulting in Fig.14 Compared to the signaling process of the , fewer operations but larger message payload per operation. In addition, the signaling process allows each node to determine whether to add another node to its neighbor list or similar structure during neighbor discovery, which is different from Fig.14 This contrasts with the signaling process of , which defers this decision until the cluster formation process.

[0167] Initially, the first node 1502 may send a neighbor discovery (ND) message 1506 or request similar to Fig.14 In this example, the node receiving the ND message 1506 from the first node 1502 is referred to as the second node 1504, similar to Fig.14 However, with Fig.14 Compared to the example of , in this example, the ND message 1506 also indicates model training related information to the second node 1504. This information may include Fig.14 Similar types of information as those provided in message 1414.

[0168] In response to receiving (e.g., successfully decoding) the ND message 1506 from the first node 1502, the second node 1504 may determine whether to add the first node 1502 to its neighbor list, communication graph, or other data structure based on the model training related information in the ND message 1506. For example, if the information indicates that the first node 1502 has a different ML model than the second node 1504 or has lower computing power than it (e.g., if the first node 1502 is configured with a Markov model or a small amount of available memory for ML, while the second node 1504 is configured with a classification model or a large amount of available memory for ML), the second node 1504 may determine not to add the first node 1502 to its data structure. In this case, the second node 1504 may disregard or ignore (and therefore not respond to) the ND message 1506. Alternatively, if the information indicates, for example, that the first node 1502 has the same or similar ML model (e.g., classification model) or the same or similar computing capabilities (e.g., memory) as the second node 1504, the second node 1504 may determine to add the first node 1502 to its data structure. In this case, the second node 1504 may add the first node 1502 and its associated model training-related information to its neighbor list, communication graph, or similar data structure (stored at the second node 1504).

[0169] Subsequently, the second node 1504 may provide an NDR message 1508 to the first node 1502 (e.g., via unicast), which not only confirms the first node 1502 as a neighbor of the second node 1504, but also indicates model training related information of the second node 1504. This information may include the same type of information as the ND message 1506, but is directed to the second node 1504 instead of the first node 1502.

[0170] In response to receiving the NDR message 1508 including model training related information, the first node 1502 may similarly determine whether to add the second node 1504 to its neighbor list, communication graph, or other data structure based on the model training related information. For example, if the information indicates that the second node 1504 has a different ML model than the first node 1502 or a lower computing power than it, the first node 1502 may determine not to add the second node 1504 to its data structure. In this case, the first node 1502 may ignore or ignore (and therefore not respond to) the NDR message 1508. Alternatively, if the information indicates, for example, that the second node 1504 has the same or similar ML model or the same or similar computing power as the first node 1502, the first node 1502 may determine to add the second node 1504 to its data structure. In this case, the first node 1502 may add the second node 1504 and its associated model training related information to its neighbor list, communication graph, or similar data structure (stored at the first node 1502). Furthermore, the first node 1502 may then provide a message 1510 to the second node 1504 (eg, via unicast) confirming that the second node 1504 is a neighbor of the first node 1502 .

[0171] On the other hand, if the second node 1504 fails to receive an acknowledgment of the NDR message 1508 from the first node 1502 (or an acknowledgment that the second node 1504 has been added to the neighbor list of the first node 1502) within a preconfigured timeout interval, the second node 1504 discards the first node 1502 and its associated information from its neighbor list, communication graph, or similar data structure, such as if the first node 1502 disregards or ignores the NDR message 1508. Similarly, if the second node 1504 fails to receive an acknowledgment of the NDR message 1508 from the first node 1502 (or an acknowledgment that the second node 1504 has been added to the neighbor list of the first node 1502), the second node 1504 discards the first node 1502 and its associated information from its neighbor list, communication graph, or similar data structure. Fig.12 As illustrated in the example of , if the first node 1502 previously received an ND request 1512 from another node 1514 (third node 1514), transmitted an NDR message 1516 in response to the ND request 1512, but failed to receive an acknowledgment of the NDR message 1516 from the third node 1514 within a timeout interval 1518 (such as when the third node 1514 disregarded or ignored the NDR message 1516), then at operation 1520, the first node 1502 may discard any associated information of the third node 1514 from its neighbor list, communication graph, or similar data structure.

[0172] Fig.16 An example 1600 of a node state machine during a cluster formation or leader election process is illustrated. The state machine may indicate the state of a network node (e.g., UE) at a given time and the operations that the network node may perform to transition between states. Different network nodes may have similar state machines and may be used in different network nodes. Fig.14 or Fig.15A network node that includes at least one other node in its neighbor list, communication graph, or similar data structure may operate according to the state machine. Fig.16 The node state machine of the FL operates to form a cluster with a cluster leader or to select a leader of the FL in the peer-to-peer network. In addition, during the cluster formation or leader selection process, the signaling or communication of the node to its neighbors can be broadcast or multicast.

[0173] exist Fig.16 In an example of , a network node (e.g., UE) can potentially assume any of the following roles: leader (aggregator), follower (participant), or candidate for leader. The leader node is configured to aggregate ML model updates from follower nodes in the FL for a period of time (a timeout period or deadline as a leader). In one example, the leader node can act as an aggregator without participating in learning or model training itself, while in other examples, the leader node can participate in model training itself in addition to acting as an aggregator. The leader node can be such as Fig.11 The cluster leader in the clustered FL architecture exemplified in , or a Fig.12 A peer leader in the peer-to-peer FL architecture illustrated in Figure 1. A follower node is a participating or learning node in FL that trains the ML model and provides updates to the corresponding leader. A candidate node is a network node that competes with other candidate nodes in an election to become a leader node.

[0174] Initially, during an election cycle, a network node begins in a follower state 1602, as illustrated by transition 1604. In this state, the node may change to a candidate state 1606 in response to declaring itself a candidate for leader election, as illustrated by transition 1608. In one example of a candidacy trigger, a node may declare itself a candidate if it fails to receive a candidacy message from another node within a preconfigured timeout period that begins after the node completes neighbor discovery (e.g., message 1624 below). The timeout period is intended to provide time for the other node to complete its own neighbor discovery and determine whether to declare itself a candidate before the node makes that determination itself. In another example of a candidacy trigger, even if the node does receive a candidacy message from other nodes within a preconfigured timeout period (e.g., message 1624 below), the node may still declare itself a candidate if the node determines that the ML model information collected from other nodes during neighbor discovery is insufficient to trigger a vote for election from the node (e.g., message 1628 below). For example, if such a candidacy message originates from a node that includes a different ML model than the node or has lower computational power than the node (e.g., if one node is configured with a Markov model or a small amount of available memory for ML, while another node is configured with a classification model or a large amount of available memory for ML), the node may ignore such a candidacy message. In another example of a candidacy trigger, if after completing neighbor discovery and while monitoring the loss function of its local ML model, the node determines that its loss function indicates a low-quality ML model, the node may declare itself as a candidate. For example, if the node determines that the value of its loss function drops below a threshold, the node may be incentivized to be selected as a leader to quickly start a FL session and improve its loss function. Alternatively, the node may remain in follower state 1602 in response to taking no action, as illustrated by transition 1610.

[0175] During election, nodes in follower state 1602 may vote (e.g., based on majority rule) for nodes in candidate state 1606. If a candidate node loses election to another node but is still determined to remain a candidate for subsequent leader election, the node may remain in candidate state 1606, as illustrated by transition 1612. Alternatively, the node may switch back to follower state 1602, as illustrated by transition 1614.

[0176] On the other hand, if the candidate node wins the election (e.g., the node receives a majority of votes from other nodes during the election), the node may change to the leader state 1616, as illustrated by transition 1618. During ML model training, the leader node may perform aggregation of model updates from follower nodes in the cluster within the clustered FL architecture, such as regarding Fig.11 described, or performing aggregation of model updates from follower nodes in a peer-to-peer network with a peer-to-peer FL architecture, such as with respect to Fig.12 As described. In a clustered FL architecture, the leader node may also group follower nodes into a cluster served by the leader node, and transmit a message to the FL parameter server indicating its state as a cluster leader. In either FL architecture, the leader node may remain in leader state 1616 (as a leader) for a timeout period or duration, after which the node may change back to follower state 1602, as illustrated by transition 1620, and a new leader node may be elected. In the case of a tie where no candidate node receives a majority vote (e.g., each node receives the same number of votes), the follower node may select one of the candidate nodes as the leader node based on various tie-breaking factors. For example, a leader node may be randomly selected among the tied candidate nodes, or a leader node may be selected from any one of the tied candidate nodes that has the most available computing resources or communication quality (e.g., signal-to-noise ratio).

[0177] In one example of selection, a candidate node 1622 may multicast or broadcast a message 1624 indicating its candidacy to other nodes in its neighbor list. The candidate node 1622 may initially be in such a follower node that declares itself as a candidate for the leader to be selected in response to any of the aforementioned candidate qualification triggers. A follower node 1626 that receives the candidate qualification message and potential other candidate qualification messages from other candidate nodes may send a message 1628 indicating a vote for one of these candidate nodes. The candidate node that receives the vote from the follower node may notify the other candidate nodes of the number of votes that each candidate node has received. If the candidate node has received a majority of votes from the follower nodes in the other candidate nodes, or has received a configured or preconfigured number of votes exceeding a threshold in the case where a single candidate node is ready for selection, the node may win the selection. The leader node may become an ML model aggregator for a cluster or peer network of follower nodes that have provided voting messages to the node. The leader node may also transmit messages 1630 to follower nodes and other candidate nodes indicating its status as a leader node for a period of time or term 1632. Once the leader node's term 1632 ends, the node may transmit messages 1634 to follower nodes in its cluster or peer network indicating the end of its term.

[0178] In a clustered FL architecture, other selections involving other follower nodes and other candidate nodes may be made simultaneously, and thus multiple clusters and cluster leaders may be formed. For example, a group of follower nodes that vote for a set of candidate nodes in their neighbor lists may participate in one cluster, while another group of follower nodes that vote for another set of candidate nodes in a different neighbor list may participate in a different cluster. In a peer-to-peer FL architecture, follower nodes and candidate nodes may be within the neighbor lists or networks of other follower nodes and candidate nodes, and thus a peer group and leader may be formed from these nodes accordingly. This result is equivalent to forming a single cluster that encompasses every node in the clustered FL architecture (e.g., the peer group may effectively act as a cluster).

[0179] Fig.17 An example 1700 of a signaling process is illustrated that allows UEs 1702 (UEs 1702a, 1702b) to form a cluster (eg, Fig.11 and Fig.13 1102, 1304 in the cluster), and allow the base station 1704 to select RSU 1706 (RSU1706a, 1706b) as a cluster leader or a model weight aggregator for UE 1702 and other FL participants within the cluster. In this process, the base station 1704 may form a cluster of UEs 1702 based on the FL information (e.g., ML model-related information) of these respective UEs, and the base station may designate an RSU that is capable of communicating with these UEs as a corresponding cluster leader. For example, the base station 1704 may select one of a plurality of RSUs to act as a model weight aggregator for a group of UEs that communicate with the RSU and are indicated by the RSU as being interested in performing the same ML task in FL, having the same or similar sensors and input formats for training data in FL, having the same or similar amount of available computing resources or other parameters for FL. Therefore, the base station may cluster UEs with similar FL parameters for maximum efficiency in model training and optimization (e.g., by not grouping UEs with different ML tasks, different computing capabilities, etc.), and select an RSU for cluster leadership that is able to communicate with these UEs for coordinated model training and optimization. Fig.17 The illustrated example specifically refers to RSU 1706 as a cluster leader or model weight aggregator, but in other examples, an edge server or other network node may replace the RSU as a cluster leader or model weight aggregator. For example, the RSU may be applied to systems such as Fig. 10A In the transportation-related applications of FL shown in the example, the edge server can be applied to Fig. 10B In the manufacturing-related applications of FL shown in the example.

[0180] Initially, the UEs 1702 may each send a FL message 1708. The FL message 1708 is intended to be received by the RSU 1706, which is the intended cluster leader in this example. When sending the FL message 1708, the RSU is not aware of the UEs 1702 with which the RSU may communicate respectively (e.g., which UEs are neighbors or proximate to the respective RSU), nor the ML tasks that these UEs are interested in performing. Therefore, the FL message 1708 may serve as a discovery message, in response to which the UE and the RSU may become aware of the existence of the other.

[0181] The FL message 1708 may be provided periodically (e.g., with a preconfigured periodicity) or in response to an event. In one example of an event, the UE may be triggered to transmit the FL message 1708 upon entering the service coverage area of ​​the RSU or base station. In another example of an event, the UE may be triggered to transmit the FL message 1708 in response to the UE's decision to join a FL cluster to improve the accuracy of its ML model. For example, if the UE's classification accuracy or other evaluation metric drops below a threshold, the UE may provide the FL message 1708.

[0182] The FL message 1708 also indicates FL or model training related information of the corresponding UE. This information may include, for example: the ML tasks in which the UE is participating or interested in participating; the available sensors at the UE and their associated data input formats; the available ML models in training, including model status, model architecture, model training parameters, and current performance (e.g., accuracy and loss); and available computing resources (e.g., whether the UE includes a CPU or GPU, information about its clock speed, available memory, etc.). For example, the FL message 1708 may indicate that the UE is configured to perform object detection (e.g., to identify OBBs), image classification, reinforcement learning, or other ML tasks. The FL message 1708 may indicate that the UE includes a camera, LIDAR, RADAR, IMU, or other sensor, and the data format that can be read by the indicated sensor. The FL message 1708 may indicate that the UE is configured with a neural network (e.g., Figure 6The FL message 1708 may indicate the number of neurons and synapses, the logistic regression or classification model, the current values ​​of the model weights and biases, the activation function applied (e.g., ReLU), the classification accuracy, and the loss function applied to the model weight updates (e.g., cross entropy loss function). The FL message 1708 may indicate the UE's CPU or GPU (e.g., controller / processor 459 or some other processor of device 450) to apply the indicated neural network and model to the indicated ML task, the clock speed of the CPU or GPU, and a certain amount of available memory (e.g., memory 460 or some other memory of device 450) for storing information associated with the ML task, neural network, data, etc. The FL message 1708 may include any combination of the foregoing items as well as other information.

[0183] The FL message 1708 may be broadcast to solicit a response from any RSU that is capable of decoding the FL message. Fig.17 In the illustrated example, RSU 1 may be adjacent to (e.g., be a neighbor of) both UE 1 and UE 2 and thus successfully decode the FL messages of both UE 1 and UE 2, while RSU 2 may be adjacent to UE 2 and thus successfully decode the FL message of UE 2. Therefore, one UE may provide FL messages to multiple RSUs, and one RSU may obtain FL messages from multiple UEs, respectively.

[0184] In response to receiving and decoding the FL message 1708 from the respective UE, the RSU (e.g., RSU 1 or RSU 2) may store an identifier of the UE in a memory, store (at least temporarily) the FL information of the UE 1702 in the memory, and provide a FL message confirmation 1710 to the UE 1702 confirming receipt of its FL message. The RSU 1706 may provide the FL message confirmation 1710 to the respective UE via unicast. The FL message confirmation 1710 may also indicate the identifier of the RSU 1706 so that the UE 1702 receiving the confirmation may determine the source RSU. For example, if the UE 1702 provides the FL message 1708 to multiple RSUs 1706, the UE will obtain multiple such FL message confirmations 1710 including the respective identifiers from the respective RSUs. For example, in Fig.17In the illustrated example of , UE 2 may provide its FL message to both RSU 1 and RSU 2, and thus UE 2 may obtain a FL message confirmation from RSU 1 indicating the identifier of RSU 1 and a FL message confirmation from RSU 2 indicating the identifier of RSU 2. Similarly, UE 1 also provides its FL message to RSU 1, and thus UE 1 may similarly obtain a FL message confirmation from RSU 1 indicating the identifier of RSU 1 (but not a confirmation from RSU 2 because RSU 2 has not obtained or successfully decoded the FL message).

[0185] After obtaining the FL message confirmation from the corresponding RSU 1706, the UE can determine the RSU in the network that has decoded its FL message. Then, the UE can select one of these confirming RSUs to act as its representative for passing its FL information to the base station 1704 for cluster formation and leader selection. Similarly, other UEs can select corresponding confirming RSUs to act as their corresponding representatives for passing corresponding FL information. For example, in Fig.17 In the illustrated example of , UE 1 may select only RSU 1 (given its only acknowledgment of the FL message 1708 from the UE), while UE 2 may select either RSU 1 or RSU 2 (given the acknowledgments of the FL message 1708 from the UE by both RSUs). However, if UE 2 also happens to select RSU 1 in this example, RSU 1 will end up inefficiently transmitting the FL information of both UEs to the base station, thereby generating additional communication overhead at RSU 1 compared to the case where RSU 2 happens to be selected. Therefore, in order to minimize the communication overhead between RSU 1706 and base station 1704, UE 1702 may be configured to select a different RSU for FL message delivery. For example, in the aforementioned example, UE 2 may be configured to select RSU 2 (rather than RSU 1) in this case. Therefore, the inefficient situation where the RSU delivers FL information from multiple UEs to the base station, or the inefficient situation where multiple RSUs deliver FL information of a single UE to the base station, can be avoided.

[0186] After selecting an RSU from the confirmed RSUs, the UE 1702 may provide an RSU indication message 1712 including an identifier of the selected RSU to the confirmed RSU. Fig.17For RSU 1 and RSU 2 of UE 2 in the example of , the UE may provide an RSU indication message 1712 to multiple confirming RSUs via multicast. The message is provided via multicast to allow the confirming RSUs to respectively determine whether the RSU is indicated as a representative for FL message delivery by the UE. After receiving the RSU indication message 1712 from the corresponding UE, an RSU having an identifier matching the identifier included in the RSU indication message 1712 may determine that it is a representative for delivering the FL information of the UE, and thus maintain its storage in a memory of the FL information in the FL message 1708 from the UE. In contrast, an RSU having an identifier that does not match the identifier included in the RSU indication message 1712 may determine that it is not a representative of the UE, and thus may discard or discard the FL information of the UE from the memory, while still maintaining its storage in the memory of the UE's identifier. Regardless of the indicated RSU in the RSU indication message 1712, the RSU may continue to store identifiers of UEs that the RSU previously acknowledged in order to track its ability to communicate with these UEs and inform the base station 1704 of the list of UEs during cluster formation and leader selection.

[0187] In response to obtaining the RSU indication message 1712 from the UE 1702, the RSU 1706 may respectively provide a message 1714 to the base station 1704, the message including the UE's FL information indicating the identifier of the RSU and a list (or other data structure) of identifiers of the UEs 1702 that the RSU has previously confirmed (and therefore has communication capability). Fig.17 In the illustrated example, RSU 1 may provide FL information of UE 1 to base station 1704, RSU 2 may provide FL information of UE 2 to base station 1704, and both RSU 1 and RSU 2 may provide identifiers of UE 1 and UE 2 to base station 1704. Therefore, base station 1704 may be informed of ML tasks, sensors, model information, and computing capabilities of UE 1702 and RSU 1706 that may communicate with the corresponding UE.

[0188] In one example, one or more of the RSUs 1706 may be a learning node, rather than just a UE FL information aggregator. For example, these RSUs may have obtained their own local data set and intend to use the data set in the FL to train their own ML models. In this case, those RSUs 1706 may also provide their own FL information to the base station 1704 in a message 1714, similar to the UE's FL information. This information may include, for example: ML tasks that the RSU is participating in or interested in participating in; available ML models in training, including model status, model architecture, model training parameters, and current performance (e.g., accuracy and loss); and available computing resources (e.g., whether the RSU includes a CPU or GPU, information about its clock speed, available memory, etc.). Although in one example, the message 1714 from the RSU may include UE FL information, a UE list, and RSU FL information (if present), in other examples, any combination of the foregoing may be included in one or more such messages 1714 from the RSU.

[0189] In response to receiving the message 1714 from the respective RSUs 1706, the base station 1704 may aggregate the UE FL information (e.g., store the FL information of the respective UEs 1702 in a memory), select an RSU to act as a cluster leader based on the aggregated information, and select the UE 1702 as a member or participant of the cluster led by the respective selected RSU based on the aggregated information. For example, the base station may assign the UE 1702 to a cluster of RSUs that indicate the ability to communicate with those UEs in the message 1714 (e.g., via identifiers that include those UEs). The base station may not necessarily assign the UE to a cluster of the same RSU that identifies the UE indicated in the RSU indication message 1712. For example, in Fig.17 In the illustrated example, even if base station 1704 obtains message 1714 indicating FL information of UE 1 and UE 2 from both RSU 1 and RSU 2, respectively (in response to UE 1 and UE 2 indicating RSU 1 and RSU 2 as their representatives, respectively, in corresponding RSU indication messages 1712), the base station may assign both UE 1 and UE 2 as cluster members of the cluster led by RSU 1.

[0190] The base station 1704 may also select an RSU to serve as a cluster leader and select UEs to serve as cluster members of such an RSU based on other factors other than the communication capabilities indicated in the message 1714. For example, the base station may select any one RSU that is capable of communicating with the largest number of UEs (e.g., an RSU whose list includes the largest number of UE identifiers) as the cluster leader and group these UEs together in a cluster under that RSU. Fig.17In the illustrated example of , the base station may select RSU 1 as the cluster leader for both UE 1 and UE 2 in response to determining from the obtained UE list in message 1714 that RSU 1 can communicate with a larger number of UEs (e.g., both UE 1 and UE 2) than RSU 2 (which in this example can communicate only with UE 1), instead of selecting RSU 1 as the cluster leader for UE 1 and RSU 2 as the cluster leader for UE 2. Additionally or alternatively, the base station may select UEs 1702 with the same or similar computing capabilities to be in the same cluster, while selecting UEs with different computing capabilities to be in different clusters. For example, in Fig.17 In the illustrated example of , the base station 1704 may group UE 1 and UE 2 in the same cluster led by RSU 1 in response to determining that both UE 1 and UE 2 have high computing capabilities or both have low computing capabilities based on the FL information in the message 1714. On the contrary, if UE 1 has high computing capabilities and UE 2 has low computing capabilities, or vice versa, the base station 1704 may determine to group UE 1 and UE 2 in different clusters led by different RSUs (e.g., to avoid low-capability UEs acting as bottlenecks in model training via FL for high-capability UEs). Additionally or alternatively, in other examples, the base station may apply other criteria in its selection of cluster leaders and cluster members based on the FL information in the message 1714.

[0191] After the aggregation and selection, the base station 1704 may provide the selected RSUs with messages 1716 indicating their respective cluster information via unicast. The cluster information provided to the respective RSUs may include, for example, identifiers of UEs to be grouped into a cluster for the FL led by the RSU. Fig.17 In the illustrated example of , base station 1704 may notify RSU 1 of its status as a cluster leader or model update aggregator for a cluster including UE 1 and UE 2. For example, base station 1704 may provide RSU 1 with identifiers of UE 1 and UE 2. If the base station does not select a particular RSU as a cluster leader or aggregator, the base station may not provide such a message or cluster information to the RSU. For example, in Fig.17 In the illustrated example, the base station may not select RSU 2 as the cluster leader, and therefore the base station may not provide message 1716 to RSU 2.

[0192] After obtaining the corresponding cluster information in the message 1716 from the base station 1704, the RSU 1706 receiving such information may respectively provide a message 1718 to the UEs 1702 indicated in its corresponding cluster information. The RSU may provide its corresponding message 1718 to those UEs 1702 having the identifier assigned or indicated in the cluster information of the message 1716 via multicast. The message 1718 may inform these UEs 1702 of their status as participants or follower nodes of the FL in the cluster led by the RSU, and may instruct these UEs to provide ML model weight updates to the RSU for training and optimization during clustering FL. For example, in Fig.17 In the illustrated example of , RSU 1 may provide message 1718 to UE 1 and UE 2 via multicast, indicating that those UEs are cluster members of the cluster led by RSU 1.

[0193] Fig.18 An example 1800 of a signaling process for allowing a first node 1802 (e.g., an RSU) to form a cluster (as a cluster leader in a clustered FL architecture) with a second node 1804 (e.g., a UE) in sidelink communication is illustrated. The process may be performed by allowing the first node 1802 (e.g., a UE) to simultaneously discover and form a cluster (as a cluster leader in a clustered FL architecture) with a second node 1804 (e.g., another UE) via sidelink communication, instead of the signaling process for allowing the first node 1802 (e.g., a UE) to simultaneously discover and form a cluster (as a cluster leader in a clustered FL architecture) with a second node 1804 (e.g., another UE) via sidelink communication. Figures 14 to 17 1 and 2. The various signaling processes for neighbor discovery or cluster formation described herein. In this process, a network node (e.g., a UE or RSU) may designate itself as a cluster leader (e.g., in response to a designated trigger discussed below) rather than being selected or elected as a cluster leader by other network nodes (e.g., a UE or a base station). In addition, a network node that intends to act as a leader (e.g., in response to a designated trigger) may form a cluster with other network nodes by transmitting a message to recruit these other nodes to its cluster. The recruitment message may indicate that the network node intends to operate as a cluster leader and FL coordinator, as well as indicate ML training related information for the network node. Examples of designated triggers may be similar to those regarding Fig.16 The examples of candidacy triggers described are the same. For example, if a node fails to receive a recruitment message from another node within a preconfigured timeout period (e.g., message 1806 below), if the node determines that the ML model information collected from other nodes is insufficient to trigger a subscription from the node to its cluster (e.g., message 1808 below), or if the node determines that its loss function indicates a low-quality ML model, the node may designate itself as a leader.

[0194] Nodes that receive the recruitment message and potentially other recruitment messages from other self-designated leaders may determine whether to subscribe to any indicated cluster led by the corresponding transmitter based on the indicated ML training-related information in the corresponding message. If the network node determines to subscribe to the cluster of one of these transmitting nodes, the network node may transmit back a message requesting subscription to the cluster and further indicating the ML-related training information of the node; otherwise, the network node may ignore the recruitment message. Upon receiving the subscription message, the recruitment node may similarly determine whether to allow the subscribing node to join its cluster based on the indicated ML-related training information of the subscribing node. If a permission decision is made, the recruitment node may transmit a confirmation of cluster permission to the subscribing node. Otherwise, the recruitment node may ignore the subscription message, and the subscribing node may search to join or lead a different cluster after a specified time period.

[0195] Initially, the first node 1802 sends a message 1806 indicating that the first node 1802 is interested in recruiting other nodes to form or join a cluster led by the first node. For example, if the first node is not currently participating in an active cluster FL session, the message 1806 may be provided periodically (e.g., according to a preconfigured periodicity). Alternatively, the message 1806 may be event-triggered, for example, in response to the first node 1802 determining to train or update the ML model to improve the performance of a particular ML task. The first node 1802 may broadcast the message 1806 to any network node capable of decoding the message. Alternatively, the message 1806 may be multicast, for example, to network nodes known to the first node 1802. The message 1806 may also be transmitted using multicast option 1 to enhance the reliability of the recruitment message (e.g., the first node 1802 may indicate in the second stage SCI the distance at which the first node expects to receive NACKs from other nodes that failed to decode the message).

[0196] In addition, message 1806 may indicate model training related information of first node 1802. This information may include, for example: ML tasks that first node 1802 is participating in or interested in participating in; available sensors at first node 1802 and their associated data input formats; available ML models in training, including model status, model architecture, model training parameters, and current performance (e.g., accuracy and loss); and available computing resources (e.g., whether first node 1802 includes a CPU or GPU, information about its clock speed, available memory, etc.). Thus, message 1806 may include information related to the model training of first node 1802. Fig.14 Message 1414 or Fig.15 Similar types of information to those provided in ND message 1506.

[0197] exist Fig.18In the example of , if the nodes are in proximity to each other, the second node 1804 may receive message 1806 from the first node 1802. The second node 1804 may potentially receive other such messages from other network nodes requesting that the second node 1804 be recruited to a cluster led by the respective node. Thus, the second node 1804 may receive recruitment messages from multiple candidates for cluster leadership. In response to receiving the recruitment message, the second node may determine whether to subscribe to a cluster of one of the candidates based at least in part on the model training related information indicated in the respective message. The determination may be based on information related to the model training. Fig.15 Similar factors to those described for the neighbor list (e.g., whether the candidate node and the second node have similar ML models, computing resources, etc.).

[0198] In this example, the second node 1804 determines to form or join a cluster led by the first node 1802, and therefore the second node may provide a message 1808 (e.g., via unicast) to the first node 1802 requesting to subscribe to or join its cluster. The message 1808 may also indicate model training related information of the second node 1804. The information may include, for example: the ML tasks that the second node 1804 is participating in or interested in participating in; the available sensors at the second node and their associated data input formats; the available ML models in training, including model state, model architecture, model training parameters, and current performance (e.g., accuracy and loss); and available computing resources (e.g., whether the second node includes a CPU or GPU, information about its clock speed, available memory, etc.). Therefore, message 1808 may include the same type of information as message 1806, but for the second node 1804 instead of the first node 1802.

[0199] In response to receiving message 1808, the first node 1802 may similarly determine whether to allow the second node 1804 to join its cluster based at least in part on the model training related information indicated in message 1808. For example, if the information indicates that the second node has a different ML model than the first node or has lower computing power than it, the first node 1802 may determine not to add the second node 1804 to its cluster. In this case, the first node 1802 may disregard or ignore (and therefore not respond to) message 1808. Alternatively, if the information indicates, for example, that the nodes have the same or similar ML models or the same or similar computing power, the first node 1802 may determine to add the second node 1804 to its cluster. In this case, the first node 1802 may provide (e.g., via unicast) a message 1810 confirming that the second node 1804 is allowed to join the cluster led by the first node 1802.

[0200] On the other hand, if the first node 1802 determines not to permit the second node 1804 to join its cluster, the second node may fail to receive the message 1810 within a timeout window 1812 from the time the message 1808 was provided (because the first node 1802 disregarded or ignored the message 1808). In this case, the second node 1804 may subscribe to the cluster of another network node from which the second node previously received a recruitment request. Fig.18 In the illustrated example of , the second node 1804 may have previously received a message 1814 (similar to message 1806) from the third node 1816 requesting to recruit other nodes to its cluster, and based at least in part on the model training related information of the third node 1816 indicated in the message 1814, the second node 1804 may send a message 1814 requesting to subscribe to or join the cluster led by the third node 1816 after the timeout window 1812 expires. In response to receiving the message 1818, the third node 1816 may similarly determine whether to allow the second node 1804 to join its cluster based at least in part on the model training related information indicated in the message 1818, in which case the third node 1816 may provide (e.g., via unicast) a message 1820 confirming that the second node 1804 is allowed to join the cluster led by the third node 1816. Alternatively, instead of subscribing to the cluster of another network node, the second node 1804 may send a message 1820 (similar to messages 1806, 1814) requesting that other nodes be recruited to the cluster led by the second node 1804. For example, the second node 1804 may determine to designate itself as the cluster leader and broadcast or multicast the message 1820 to the third node 1816 after the timeout window 1812 expires. In response to receiving the message 1820, the third node 1816 may determine whether to subscribe to the cluster led by the second node 1804 based on the model training related information of the second node indicated in the message 1820.

[0201] Fig.19 An example 1900 of a signaling process is illustrated, in which a participant node 1902 (participant nodes 1902a, 1902b, 1902c) may communicate with a user in a manner such as regarding Figures 16 to 18The described selection, election, or designation of a leader node 1904 (e.g., a cluster leader or a peer leader) is followed by the execution of intra-cluster FL or peer FL. Participant nodes 1902 (e.g., UEs) may be, for example, adjacent follower or candidate nodes in a cluster Ci or a peer network. Leader node 1904 (e.g., UE, RSU, edge server, local FL server, or other network node) may be, for example, a selected, chosen, or self-designated leader of cluster Ci or a selected leader of a peer network. In one example, leader node 1904 may act as an aggregator without participating in learning or model training itself, while in other examples, leader node 1904 may participate in model training itself in addition to acting as an aggregator. Although Fig.19 The example illustrates three UEs communicating with the leader node 1904, but it should be understood that the cluster or peer network is not limited to this number of UEs and may have a different number of UEs in other examples.

[0202] Initially, at operation 1906, the leader node 1904 may initialize an ML model (eg, a cluster model W for clustering FL. Ci , or the global model W0 for peer FL G ). The initialization may include, for example, configuring the ML task (e.g., object detection), the neural network (e.g., CNN), the number of neurons and synapses, random weights and biases, the model algorithm (e.g., logistic regression), and other model parameters. After initialization, at operation 1908, the leader node 1904 may broadcast or multicast these ML model parameters to the network nodes 1902 for download.

[0203] After receiving the ML model parameters from the leader node 1904 and configuring its local ML model accordingly, at operation 1910, the participant nodes 1902 (e.g., k cluster members or peer nodes) and the potential leader node 1904 may perform local ML model training based on their respective data sets. For example, during a given FL iteration t, the respective node k (corresponding to the cluster model W according to the FL architecture) t Cik Or the global model W t k ) can train local ML models 1911 (models 1911a, 1911b, 1911c) to identify the function for minimizing the loss function F(W t Cik ) or F k (W t k ), such as the optimal model parameters for Figure 6As described. For example, a node may utilize backpropagation and stochastic gradient descent to identify optimal model weights to be applied to its corresponding neural network, which results in minimized cross entropy loss after multiple training sessions. Thereafter, at operation 1912, the participant node 1902 may upload its corresponding updated ML model parameters to the leader node 1904. To reduce transmission traffic, these parameters may include, for example, modified weights, weights that have changed beyond a threshold, or multiplicative or additive increments or percentages of these weights. For example, the participant node 1902 may send optimized ML model weights and other information to the leader node 1904. After receiving these ML model information updates from the participant nodes 1902, at operation 1914, the leader node 1904 may aggregate the updates to generate an updated ML model 1915 for the node (e.g., an updated cluster model W to be applied to the next FL iteration t+1). t+1 Ci Or the global model W t+1 G ). For example, the leader node 1904 may average the corresponding model weights indicated by the participant nodes 1902 or perform some other calculation, potentially including the corresponding model weight of the leader node 1904.

[0204] After generating the aggregated ML model information, the leader node 1904 may determine that the loss function associated with the updated ML model is no longer minimized. For example, the aggregated ML model weights may potentially result in an increased loss compared to the previous ML model weights of the individual nodes. Therefore, the leader node may transmit the aggregated ML model information to the participant nodes 1902, which may be the same nodes as before or may include additional or fewer nodes, for further local ML model training. For example, after configuring their local ML models with the updated weights, during the next FL iteration t+1, the participant nodes 1902 (and potentially the leader node 1904) may again perform local ML model training to achieve a set of optimal ML model weights for their respective models, and the participant nodes 1902 may similarly transmit their updated ML models to the leader node 1904 for further aggregation. This process may be repeated until a loss function associated with the aggregated ML model (e.g., a cluster loss function F(W) for cluster Ci) is achieved. t Ci ) or the global loss function F for peer-to-peer networks G (W t GAlternatively, the process may be repeated until a predetermined number of FL iterations in the cluster or peer network have occurred (e.g., based on the computational power of the participant nodes 1902 or the number of other ML model information), or a timeout of the leader node 1904 regarding its status as a leader has occurred (e.g., if a leader is elected).

[0205] Fig. 20 An example 2000 of a signaling process is illustrated in which a leader node 2002 (nodes 2002a, 2002b) may communicate with a FL parameter server 2004 to communicate with the FL parameter server 2004 as described above. Fig.19 The described inter-cluster FL is performed after one or more FL iterations of intra-cluster model training with the participant nodes 2005 (nodes 2005a, 2005b, 2005c, 2005d). The leader node 2002 (e.g., UE, RSU, or edge server) can be, for example, a selected, elected, or self-designated leader of the corresponding cluster 2006 (clusters 2006a, 2006b). The FL parameter server 2004 can be, for example, a base station, RSU, edge server, or other network entity that coordinates FL between the corresponding clusters 2006. In one example, the leader node 2002 can act as an aggregator without participating in learning or model training itself, while in other examples, the leader node 2002 can participate in model training itself in addition to acting as an aggregator. Although Fig. 20 The example illustrates two UEs communicating with the leader node 2002 in the corresponding cluster 2006, but it should be understood that the cluster 2006 is not limited to this number of UEs and may have a different number of UEs in other examples.

[0206] At a given moment after intra-cluster FL training, the leader node 2002 may communicate the locally updated ML model information (intra-cluster) to the FL parameter server 2004 for inter-cluster FL training. For example, the FL parameter server 2004 may periodically or aperiodically transmit a request to the cluster leader to communicate its model updates for inter-cluster FL training. Alternatively, the leader node 2002 itself may communicate its model updates for inter-cluster FL training in response to an event trigger, such as the minimization of a local loss function in the corresponding cluster, or the occurrence of a certain number of intra-cluster FL iterations. The FL parameter server 2004 may aggregate this information received from the corresponding cluster leaders to generate a global (multi-cluster) ML model update, after which the FL parameter server may provide the globally updated ML model information to the corresponding leader nodes. These leader nodes may then provide the globally updated ML model information to their corresponding participant nodes to further refine the local ML model through additional intra-cluster FL training. Intra-cluster and inter-cluster FL training may or may not be synchronized; for example, different clusters may perform intra-cluster FL training or inter-cluster FL training for the same number or different number of iterations before stopping FL training. For example, the leader node 2002 of the corresponding cluster may stop communicating local model updates (in intra-cluster training) or global model updates (in inter-cluster training) to participant nodes or FL parameter servers at the same time or at different times.

[0207] Initially, at operation 2008, the leader node 2002 of the corresponding cluster may transmit its updated ML model W of its cluster Ci to the FL parameter server 2004. t Ci , the updated ML model includes one or more previous aggregations 2007 of ML model updates from the local models 2009 (models 2009a, 2009b, 2009c, 2009d) of the follower nodes. To reduce the sent traffic, these updates may include, for example, modified weights, weights that have changed beyond a threshold, or multiplicative or additive increments or percentages of these weights. For example, the leader node 2002 may transmit its most recently generated aggregated ML model weights after a given number of intra-cluster FL iterations for further aggregation with other aggregated ML model weights generated by other leader nodes in other clusters. For example, in response to receiving a request from the FL parameter server 2004, implementing a loss function associated with the updated ML model (e.g., the cluster loss function F(W) of the cluster), t Ci), the occurrence of a predetermined number of FL iterations in a cluster, or the timeout of the leader node 2002 regarding its status as a leader, the leader node 2002 may be triggered to transmit an updated ML model for the inter-cluster FL training. Similarly, in response to similar events, the leader nodes 2002 of other corresponding clusters may respectively transmit their updated ML models (e.g., aggregated ML model weights) of their own clusters to the FL parameter server 2004 for inter-cluster FL training.

[0208] After receiving the updated ML model from the cluster leader, at operation 2010, the FL parameter server 2004 may aggregate the updates to generate an updated global ML model 2011 for the leader node (e.g., an updated global model W t+1 G ). For example, the FL parameter server 2004 may average or perform some other calculation on the corresponding aggregated model weights indicated by the leader node 2002. After generating the aggregated ML model information, the FL parameter server 2004 may determine that the global loss function associated with the updated ML model is no longer minimized. For example, the aggregated ML model weights across clusters may potentially result in an increased loss compared to the previously updated ML model weights of the individual cluster leaders. Therefore, at operation 2012, the FL parameter server may transmit the aggregated ML model information to the leader node 2002 for further training of the corresponding in-cluster ML models 2013 (models 2013a, 2013b, from the updated model W t+1 G start), such as Fig.19 After this intra-cluster FL training, the leader node 2002 may again transmit its updated or aggregated ML model information to the FL parameter server 2004 for further inter-cluster FL training, and the process may be repeated until a global loss function associated with the aggregated ML model across clusters (e.g., global loss function F G (W t G Alternatively, the process may be repeated until a predetermined number of inter-cluster FL iterations have occurred, or a timeout of the FL parameter server 2004 has occurred (eg, if a leader is elected).

[0209] Fig.21 An example 2100 of a signaling process is illustrated in which peer nodes 2102 (nodes 2102a, 2102b, 2102c, 2102d) may perform peer FL in a synchronous manner without designating or electing a leader node. Peer nodes 2102 (e.g., UEs) may be, for example, neighboring follower nodes in a peer-to-peer network. Fig.19 Unlike the example of , in this example, peer nodes 2102 can each act as their own aggregator of ML model updates received from other peer nodes. Fig.19 This approach may cause peer nodes 2102 to perform more computation and experience more communication overhead than the leader-based approach, but it may minimize or avoid aggregation bottlenecks that may otherwise occur in the event of a failure at the leader or FL parameter server (because the peer nodes themselves act as aggregators). In addition, the signaling process may be used in contexts such as Fig.14 and Fig.15 Neighbor discovery is then performed without peer nodes as described in Figures 16 to 18 Describes the formation of a cluster or the election of a leader. Therefore, Fig.21 The peer-to-peer network may include peer nodes 2102 communicating effectively within a single group or cluster with other peer nodes in their respective neighbor lists. Fig.21 The example of illustrates four UEs in a peer-to-peer network, but it should be understood that the peer-to-peer network is not limited to this number of UEs and may have a different number of UEs in other examples.

[0210] Initially, at operation 2104, a peer node 2102 may use its respective data set to initialize and train its own ML model in parallel. For example, a peer node may initialize an ML model, which may include, for example, configuring an ML task (e.g., object detection), a neural network (e.g., CNN), the number of neurons and synapses, random weights and biases, a model algorithm (e.g., logistic regression), and other model parameters. After initialization, the peer node 2102 may perform various ML model trainings based on its respective data set. For example, a respective node may use observed data to train its ML model to identify optimal model parameters for minimizing a loss function, such as with respect to Figure 6 As described. For example, a node may utilize back-propagation and stochastic gradient descent to identify optimal model weights to be applied to its respective neural network, which results in a minimized cross entropy loss after a number of training sessions or iterations. Training performed by a peer node may continue until minimization of a loss function associated with the respective ML model is achieved, or until a predetermined number of iterations or training sessions have been performed for the peer node.

[0211] After a peer node 2102 completes or stops its model training, the peer node may broadcast its updated ML model information to other peer nodes in its peer network (e.g., nodes in the neighbor list of the corresponding node). To reduce transmission traffic, this information may include, for example, modified weights, weights that have changed beyond a threshold, or multiplicative or additive increments or percentages of these weights. These broadcasts may be scheduled so that a synchronous model update exchange 2106 may be performed between each peer node 2102. After this synchronous model update exchange 2106, the peer node 2102 may have the updated ML model information (e.g., updated model weights) of other peer nodes as well as its own updated ML model information.

[0212] The peer nodes 2102 may then aggregate their updates, respectively, to generate an updated global ML model 2108 (models 2108a, 2108b, 2108c, 2108d) shared among the peer nodes. For example, the peer nodes 2102 may average or perform some other common calculation on the model weights (including their own updated ML model information) previously obtained by the corresponding peer nodes during the synchronous model update exchange 2106. Thus, the aggregated ML model information generated by the peer nodes may be the same. After the peer node 2102 generates the aggregated ML model information, the peer node may determine that the loss function associated with the updated ML model is no longer minimized, or that a predetermined number of FL iterations have not occurred. Therefore, the peer node 2102 may repeat the aforementioned process in subsequent FL iterations, i.e., perform individual ML training based on its respective data set; perform scheduled synchronous model update exchanges; and aggregate the obtained ML model updates from other peer nodes.

[0213] The peer node may continue to repeat the foregoing process in multiple FL iterations until the minimization of the loss function associated with the aggregate ML model is ultimately achieved, the various target performance metrics are met, or a predetermined number of FL iterations have occurred. For example, if a peer node determines that a classification accuracy of 75% or more has been achieved after one or more FL iterations, or if a peer node determines that it has performed a certain number of FL iterations, the peer node may stop performing model training, updating exchanges, or aggregation regardless of the state of other peer nodes. Therefore, other peer nodes seeking different target performance or different numbers of FL iterations may continue to perform the foregoing process until their corresponding conditions are met.

[0214] Fig. 22An example 2200 of a signaling process is illustrated in which peer nodes 2202 (nodes 2202a, 2202b, 2202c, 2202d) may perform peer FL in an asynchronous manner without designating or electing a leader node. Peer nodes 2202 (e.g., UEs) may be, for example, neighboring follower nodes in a peer-to-peer network. Fig.19 Unlike the example of , in this example, peer node 2202 can act as its own aggregator of ML model updates received from other peer nodes. Fig. 22 The example of exemplifies four peer nodes in a peer-to-peer network, but it should be understood that the peer-to-peer network is not limited to this number of peer nodes and may have a different number of peer nodes in other examples.

[0215] In addition, peer nodes synchronously exchange model updates with other peer nodes. Fig.21 Unlike the example of , the aggregation of model updates by one node is asynchronous with respect to other nodes. For example, in this process, one peer node (e.g., a random node) at a time may request model updates from other peer nodes and aggregate these model updates to generate a single updated ML model, rather than multiple peer nodes operating simultaneously. Furthermore, in this example, the requesting peer node may receive a model update with a performance metric that specifically meets a threshold performance level, which is different from Fig.21 , in which a peer node receives model updates from other peer nodes in its neighbor list regardless of model performance. For example, the requesting node may specifically receive model updates from peer nodes that have models associated with a classification accuracy above 75% or other threshold, or with some other relatively good performance or evaluation metric. Thus, the requesting peer node may not necessarily receive updates from every peer node in its neighbor list, and the aggregation of model updates is different from Fig.21 This method can lead to more accurate models more quickly than the previous one.

[0216] Initially, at operation 2204, a peer node 2202 may use its respective data set to initialize and train its own ML model in parallel. For example, a peer node may initialize its respective ML model, which may include, for example, configuring an ML task (e.g., object detection), a neural network (e.g., CNN), the number of neurons and synapses, random weights and biases, a model algorithm (e.g., logistic regression), and other model parameters. After initialization, the peer node 2202 may perform various ML model trainings based on its respective data set. For example, a respective node may use observed data to train its ML model to identify optimal model parameters for minimizing a loss function, such as with respect to Figure 6The training performed by the peer node may continue until a minimization of a loss function associated with the corresponding ML model is achieved, or until a predetermined number of iterations or training sessions have been performed for the peer node.

[0217] In addition, while training their respective ML models, the peer nodes 2202 may track performance metrics of their respective ML models. For example, the peer nodes may determine the classification accuracy of their respective models after a predetermined number of training sessions have occurred. Alternatively, the peer nodes may determine log loss, confusion matrix, or other evaluation metrics that indicate the current performance of their respective ML models.

[0218] After the peer node 2202 completes or stops its model training, at operation 2206, one of the peer nodes 2202 (node ​​2202a or N1 in the illustrated example) transmits a request for its ML model update of the ML model 2209 (respectively ML models 2209a, 2209b, 2209c) to other peer nodes during FL iterations. For example, the peer node transmitting the request may be a random peer node selected among the peer nodes 2202 in the network. The peer node receiving the request may determine whether to provide its corresponding ML model update based on its corresponding performance or evaluation metric. For example, if the peer node determines that its performance metric meets a preconfigured threshold (e.g., a classification accuracy of 75% or greater), then at operation 2208, the peer node may provide its updated ML model information to the requesting node. In order to reduce the transmission traffic, the information may include, for example, modified weights, weights that have changed beyond a threshold, or multiplicative or additive increments or percentages of these weights. In contrast, a peer node with a performance metric that falls below a preconfigured threshold may refrain from providing its updated ML model information to the requesting node, or otherwise ignore the request. Fig. 22 In the illustrated example, N3 (node ​​2202c) and N4 (node ​​2202d) have high performance metrics and therefore respond to the request of N1 (node ​​2202a) with their model updates, but N2 (node ​​2202b) has a low performance metric and therefore does not respond to the request.

[0219] In response to receiving updated ML model information from one or more of the peer nodes 2202 having performance metrics that meet a threshold, at operation 2210, the requesting peer node (N1 in this example) may aggregate these updates to generate an updated ML model 2211. For example, the requesting peer node may average the model weights obtained from other peer nodes in response to its request or perform some other calculation. After the requesting peer node obtains the aggregated ML model, the node may determine that the loss function associated with the aggregated ML model is no longer minimized or that a predetermined number of FL iterations have not occurred. Therefore, the requesting node may continue to perform individual ML training based on its corresponding data set until the node is triggered to perform another request in a subsequent FL iteration (e.g., in response to a random selection or other event). Alternatively, if the loss function is minimized or a certain number of FL iterations for the node have occurred, the requesting node may terminate the model training process.

[0220] At the same time, after responding to the requesting node (and regardless of its subsequent actions), another node in the peer nodes 2202 (node ​​2202b or N2 in the illustrated example) may transmit a request for its corresponding ML model update to other peer nodes (including node 2202a or N1) during subsequent FL iterations. For example, the new peer node transmitting the request may be another random peer node selected among the peer nodes 2202 in the network. The peer node 2202 receiving the request may similarly provide its model update according to its corresponding performance metric to the requesting node for ML model aggregation, and the requesting node may similarly continue ML training or terminate ML training based on the loss function, the number of FL iterations, or other factors. This process may be repeated similarly for other peer nodes, where peer nodes may transmit requests for model updates separately at different times while other peer nodes continue to monitor their model performance metrics or the number of FL iterations.

[0221] Fig.23 An example 2300 of a signaling process is illustrated in which a UE 2302 (UE 2302a, 2302b, 2302c) may optimize sidelink communication parameters in response to a clustered or peer FL. The UE 2302 may correspond to the device 410, 450, the UE 104, 702, 1104, 1202, 1302, 1702, or the node 1402, 1404, 1502, 1504, 1514, 1622, 1626, 1802, 1804, 1816, 1902, 2005, 2102, 2202, and the UE 2302 may be configured to perform as described with respect to Figure 1 , Figure 7 and Figures 11 to 22The UE may be in one or more clusters such as clusters 1102, 1304, 2006, or in a cluster such as Fig.12 , Fig.19 , Fig.21 and Fig. 22 In the illustrated peer-to-peer network.

[0222] Initially, UE 2302 collects measurements 2304 from its environment (e.g., CBR, number of received packets at the UE, number of resource reservations of other UEs, or other metrics observed instantaneously or over a configured time period). UE 2302 may store these observations in a local dataset and map the dataset to an ML model or policy to select sidelink communication parameters. For example, in Mode 2 resource allocation, UE 2302 may use reinforcement learning (or other ML tasks) and back propagation to train a neural network (e.g., Figure 6 The neural network 600) is updated to obtain updated ML model information 2306 (e.g., updated weights 616 and gradients with respect to these weights). These updates allow the model to output an optimal communication range for multicast option 1, which is associated with a maximum reward (e.g., maximum throughput, minimum latency, or other metric) based on the measurements 2304 input to the model in the case of reinforcement learning, or a minimum loss function in the case of other ML tasks such as object detection or classification. The updated ML model information 2306 may correspond to, for example, information from about Fig.11 and Figures 19 to 22 The modified weights or gradients of the described updated local model 1112 , local ML model 1911 , local ML model 2009 , updated global ML model 2108 , and updated ML model 2209 .

[0223] After training its local ML model, the UE 2302 provides a report 2310 including updated ML model information 2306 and measurements 2304 to a leader node 2308 (leader node 2308a, 2308b, e.g., a cluster leader in a clustered FL or a leader in a peer FL). For example, the UE 2302 may measure the distribution of CBRs observed over a time window, calculate an average or maximum CBR from the distribution of CBRs, and report the average or maximum CBR along with updated model weights, gradients, or other ML model information. The leader node 2308 may correspond to a device 410, 450, a UE 104, an RSU 107, 1706, a cluster leader 1108, 1306, a leader 1204, or a node 1622, 1802, 1904, 2002, and the leader node 2308 may be configured to perform as described with respect to Figure 1 , Figure 7 and Figures 11 to 22The leader node 2308 may also be in one or more clusters such as clusters 1102, 1304, 2006, or in a cluster such as Fig.12 and Fig.19 In the illustrated peer-to-peer network.

[0224] UE 2302 may obtain measurements 2304, train the ML model to obtain updated ML model information 2306, and periodically, semi-persistently, or aperiodically provide a report 2310 including these measurements and updated ML model information. Report 2310 may correspond to the operation 1912, 2208, or as described with respect to Figures 19 to 22 2004, and the FL parameter server 2312 may be configured to periodically send a report 2310 to the leader node 2308. The FL parameter server 2312 may correspond to the device 410, 450, the base station 102 / 180, 181, 500, 1308, 1704, or the FL parameter server 1106, 2004, and the FL parameter server 2312 may be configured to perform the above-described synchronization model update exchange 2106. In one example, the UE 2302 may be configured by the leader node 2308 or periodically configured by the FL parameter server 2312 to send a report 2310 to the leader node 2308. The FL parameter server 2312 may correspond to the device 410, 450, the base station 102 / 180, 181, 500, 1308, 1704, or the FL parameter server 1106, 2004, and the FL parameter server 2312 may be configured to perform the above-described synchronization model update exchange 2106. Figure 1 , Figure 7 and Figures 11 to 22 In another example, the UE 2302 may semi-persistently send the report 2310 within a configured time period, and the leader node 2308 or the FL parameter server 2312 may select the configured time period based on a period of time during which the UE 2302 intends to participate in a cluster or peer FL. In another example, the UE 2302 may send the report 2310 aperiodically in response to a request from the leader node 2308 to trigger a corresponding report instance.

[0225] The UE 2302 may determine whether to send reports 2310 periodically, semi-persistently, or aperiodically from one or more report configurations 2314 from the FL parameter server 2312 or the leader node 2308, including applicable time periods and time-frequency resources in which to send reports. In one example, the FL parameter server 2312 may send the report configuration 2314 (which may be an RRC configuration) via RRC signaling, and the UE 2302 may send the report 2310 to the leader node 2308 in the resources indicated in the report configuration 2314. In another example, the report configuration 2314 may be triggered by the leader node 2308. For example, the leader node 2308 may reserve sidelink resources in the PSSCH for sending the report configuration 2314 and receiving the report 2310, and the leader node may send the report configuration in the reserved sidelink resources indicated in the SCI. In this example, the UE may send the report 2310 to the leader node 2308 in the reserved sidelink resources. In an additional example, regardless of the source of the report configuration 2314, the UE 2302 can send the report 2310 in dedicated resources (such as preconfigured time-frequency resources) for such reports.

[0226] After receiving report 2310 from UE 2302, leader node 2308 may aggregate the report with other reports from other UEs in the same cluster or peer network, such as regarding Figures 19 to 22 For example, the leader node 2308 may receive reports 2310 from UE 1 and UE 2, including updated ML model information 2306 (e.g., weights and gradients) and measurements 2304, respectively, and the leader node 2308 may aggregate the updated ML model information of the respective UEs to form local aggregated ML model information 2316. The local aggregated ML model information 2316 may correspond to, for example, the updated ML model 1915, the intra-cluster ML model 2013, the updated global ML model 2108, and the updated ML model 2211, as described with respect to Figures 19 to 22 The leader node 2308 may then provide local aggregated ML model information 2316, including aggregated weights and gradients, to UEs 2302 in its cluster or peer network via a multicast message. Using the local aggregated ML model information 2316, the UE 2302 may continue to perform FL training, such as intra-cluster FL or peer FL, as described in relation to Fig.19 , Fig.21 and Fig. 22 as described.

[0227] In addition, the leader node 2308 may derive measurement statistics 2318 from the measurements 2304 obtained from the UE 2302. These measurement statistics may be, for example, averages, maximums, variances, frequencies, or other values ​​derived from a combination of measurements 2304 obtained by the leader node 2308 from UEs 2302 in the same cluster or peer network. For example, if the measurements 2304 from various UEs include instantaneous CBRs observed over a period of time, the leader node 2308 may calculate the average or maximum CBR on each measurement, or the variance of the observed CBRs on multiple UEs. Similarly, if the measurements 2304 from the respective UEs include average or maximum CBRs, the leader node 2308 may calculate the average of these averages, the average of the maximums, the maximum of the averages, the maximum of the maximums, or other statistics. The leader node 2308 may also or alternatively maintain a histogram of the measurements 2304 from the UE 2302, such as individually reported CBRs. For example, the leader node 2308 may derive the overall frequency of individually reported CBR values, ranges, or other measurements from the UEs 2302.

[0228] Using the measurement statistics 2318, the leader node 2308 may identify a UE 2302 that is an outlier in a cluster or peer network led or utilized by the leader node. An outlier UE may be, for example, a UE whose measurement 2304 has a significantly different value (e.g., a CBR value that differs by at least three standard deviations from the total average CBR across multiple UEs) or a different distribution of values ​​compared to the value of the measurement statistics 2318. For example, the leader node 2308 may classify the UE as an outlier in the cluster or peer network in response to determining that the average CBR of the UE differs by at least three standard deviations from the total average CBR (or nominal CBR) across multiple UEs in the cluster or network. Since the data set of the outlier UE may affect the homogeneity of the data observed within its cluster or peer network, the leader node 2308 may notify the FL parameter server 2312 or the corresponding UE of the outlier UE information 2320. Thus, the FL parameter server 2312 may reassign the outlier UE to a different cluster, the UE 2302 may participate in or lead a different cluster or peer network, or the leader node 2308 may otherwise remove the outlier UE from its cluster or peer network to maintain data homogeneity in the cluster or network.

[0229] After obtaining local aggregate ML model information 2316 (e.g., updated model weights and gradients), measurement statistics 2318 (e.g., summary statistics on reported measurements from various UEs), and outlier UE information 2320 (e.g., identifiers of outlier UEs and their measurement distributions), leader node 2308 may provide a report 2322 including the information to FL parameter server 2312. Report 2322 may correspond, for example, to information about Fig. 20 The message delivered at operation 2008 described above. The FL parameter server 2312 may similarly obtain reports 2322 including local aggregate ML model information 2316, measurement statistics 2318, and outlier UE information 2320 from other leader nodes of different clusters or networks. For example, the FL parameter server 2312 may receive reports 2322 from leader nodes 1 and 2, where leader node 1 derives its local aggregate ML model information 2316, measurement statistics 2318, and outlier UE information 2320 from updated ML model information 2306 and measurements 2304 in reports 2310 from UE 1 and UE 2, and where leader node 2 derives similar information from similar report information in reports 2310 from UE 3 and other UEs.

[0230] After receiving the report 2322 from the leader node 2308, the FL parameter server 2312 may dynamically associate clusters that experience similar conditions associated with FL training. For example, the FL parameter server 2312 may link or associate clusters led by the leader node 2308 that report similar values ​​of the measurement statistics 2318. As an example, if the report 2322 includes a CBR value (e.g., an average of the average CBRs of the individual UEs), the FL parameter server 2312 may obtain an average of the total CBR values ​​across the leader node 2308, and the FL parameter server may link the cluster associated with the report 2322 indicating a CBR value within one or two standard deviations of the average of the total CBR values ​​across the cluster.

[0231] After performing this linking or association of clusters to groups, the FL parameter server 2312 may aggregate the local aggregated ML model information 2316 obtained in reports 2322 associated with clusters of the same group, such as regarding Fig. 20 As described (e.g., inter-cluster FL). Thus, the FL parameter server 2312 may aggregate local ML model information of the leader node 2308 that reports similar measurement statistics (rather than clusters with different measurement statistics). For example, if the cluster including UEs 1 and 2 has similar measurement statistics as the cluster including UE3, the FL parameter server 2312 may aggregate the aggregated model weights of UEs 1, 2, and 3. Alternatively, if the cluster including UEs 1 and 2 has different measurement statistics than the cluster including UE 3, the FL parameter server 2312 may aggregate the aggregated model weights of UEs 1 and 2, and aggregate the aggregated model weights of UE 3 separately from the aggregated model weights of other UEs in its cluster.

[0232] In response to performing aggregation of local ML model updates for a cluster or a group of similar clusters, the FL parameter server 2312 may derive global aggregated ML model information 2324 for the cluster or a group of similar clusters. The global aggregated ML model information 2324 may correspond to, for example, Fig. 20 The FL parameter server 2312 may also reassign UEs 2302 whose identifiers are indicated in the outlier UE information 2320 to different existing clusters, or remove these UEs 2302 from their current clusters to form new clusters, such as with respect to Figures 16 to 18 The FL parameter server 2312 may maintain outlier UE reassignment information 2326 , which includes identifiers of UEs 2302 reassigned to different clusters, identifiers of leader nodes 2308 leading those different clusters, or other information based on the outlier UE information 2320 .

[0233] After obtaining the global aggregate ML model information 2324 and the outlier UE reassignment information 2326 of the corresponding cluster, the FL parameter server 2312 may separately provide a message 2328 including the information to the leader node 2308 associated with the corresponding cluster via unicast. The message 2328 may correspond to, for example, Fig. 20 The FL parameter server 2312 may similarly provide such messages with corresponding global aggregate ML model information and outlier UE reassignment information to other leader nodes from which the FL parameter server has obtained reports 2322. If the leader node guides the clusters that the FL parameter servers 2312 link together to have similar measurement statistics (in the same cluster group), the global aggregate ML model information 2324 provided to different leader nodes may include the same model updates. Otherwise, the global aggregate ML model information 2324 may include different model updates for different leader nodes or clusters.

[0234] In response to obtaining message 2328 from FL parameter server 2312, leader node 2308 may provide message 2330 including global aggregate ML model information 2324 to its corresponding UE 2302 for application in further intra-cluster FL training, such as information about Fig.19 and Fig. 20As described. For example, in response to receiving message 2328 from FL parameter server 2312, leader 1 and leader 2 may send message 2330 including global aggregate ML model information 2324 to UE 1 and 2 and UE 3, respectively. Leader node 2308 may also provide outlier UE cluster information 2332 to UEs classified as outliers of their current cluster. Outlier UE cluster information 2332 may include identifiers of other leader nodes of different clusters or other information obtained or derived from outlier UE reassignment information 2326. For example, in response to receiving message 2328 from FL parameter server 2312, if leader 1 determines that UE 2 is an outlier of leader 1's cluster and is therefore to be reassigned to leader 2's cluster based on outlier UE reassignment information 2326, leader 1 may provide outlier UE cluster information 2332 to UE 2, thereby notifying the UE of its new status as a participant node of leader 2's cluster.

[0235] In response to receiving the local aggregated ML model information 2316 (for a clustered FL or a peer FL) or the global aggregated ML model information 2324 (for a clustered FL), the UE 2302 may continue to perform as previously described with respect to Figures 19 to 22 For example, UE 1 and UE 3 may perform intra-cluster FL training in their respective clusters, followed by one or more iterations of intra-cluster FL training and one or more iterations of inter-cluster FL training until the loss functions associated with their respective ML models are minimized, such as the loss function of Figure 6 As described. Subsequently, UE 2302 may input the observed data in its respective data sets into its updated ML model to respectively output sidelink communication parameters 2334, and UE 2302 may include these sidelink communication parameters 2334 in corresponding SCI 2336 to optimize communication with other UEs 2338 (UE 2338a, 2338b or with UE 2302). For example, UE 1 and UE 3 may respectively send SCI 2336 to UE 2338, which includes the desired communication range for multicast option 1 derived from the updated ML model following the aforementioned FL training.

[0236] Therefore, re-referencing Figure 7 and Fig.13In the example 700, 1300 of, UEs 702, 1302 associated with high congestion (or other similar conditions) in street 704 may be grouped into one cluster 1304, while UEs 702, 1302 associated with low congestion (or other similar conditions) in street 706 may be grouped into different clusters 1304 with or without the assistance of base station 1308. Alternatively, UEs 702, 1302 in different intersections (or portions of the same intersection) may be participants in the same peer-to-peer network. UEs 702, 1302 may perform intra-cluster FL with cluster leader 1306 and inter-cluster FL with base station 1308, or UEs 702, 1302 may perform peer FL with other UEs in their respective zones or intersections. In addition, UEs 702, 1302 that are determined to be outliers of their current cluster may be reassigned to another cluster. After optimizing the respective ML models through the aforementioned clustering or peer FL process, the UE 702, 1302 may obtain the SCI parameters and apply them to its sidelink communications with other UEs. Thus, network performance may be improved through sidelink parameter coordination via clustering or peer FL.

[0237] Fig.24 2400 is a flow chart of a method of wireless communication. The method may be performed by a UE (e.g., UE 104, 702, 1104, 1202, 1302, 1702, 2302; device 410, 450; node 1402, 1404, 1502, 1504, 1514, 1622, 1626, 1802, 1804, 1816, 1902, 2005, 2102, 2202, apparatus 2802) or a leader node (e.g., UE 104; device 410, 450; RSU 107, 1706; cluster leader 1108, 1306; leader 1204; node 1622, 1802, 1904, 2002; leader node 2308; apparatus 2802). For example, the method may be performed by a controller / processor 459, 475 coupled to a memory 460, 476 of the device 410, 450. Optional aspects are illustrated by dashed lines. The method allows a UE or leader node to achieve sidelink parameter coordination in a mode 2 resource allocation scenario through various signaling in a cluster FL or a peer FL.

[0238] At 2402, a UE or a leader node can provide a first ML model information update and measurements associated with the first ML model information update. For example, 2402 can be performed by the FL training component 3044.

[0239] In one example, if the UE performs operation 2402, then reference Fig.23Initially, UE 2302 collects measurements 2304 from its environment (e.g., CBR, number of received packets at the UE, number of resource reservations of other UEs, or other metrics observed instantaneously or over a configured time period). UE 2302 may store these observations in a local dataset and map the dataset to an ML model or strategy to select sidelink communication parameters. For example, in Mode 2 resource allocation, UE 2302 may use reinforcement learning and back propagation to train a neural network (e.g., Figure 6 The UE 2302 trains its local ML model to obtain updated ML model information 2306 (e.g., weights 616, biases 618, subsampling or scaling factors in the subsampling layer 608, or other updated ML model information after minimization of the loss function). After training its local ML model, the UE 2302 sends a report 2310 including the updated ML model information 2306 and the measurements 2304 to a leader node 2308 (e.g., a cluster leader in a clustered FL or a leader in a peer FL).

[0240] In another example, if the leader node performs operation 2402, then reference Fig.23 , after receiving the local aggregated ML model information 2316 (e.g., updated model weights and gradients) and measurement statistics 2318 (e.g., summary statistics about reported measurements from various UEs), the leader node 2308 may send a report 2322 including the information to the FL parameter server 2312 (in this example, the first ML model information update and measurements, respectively).

[0241] At 2404, the UE or leader node can obtain aggregated ML model information updates. For example, 2404 can be performed by the FL training component 3044.

[0242] In one example, if the UE performs operation 2404, the aggregated ML model information update includes an aggregation of the first ML model information update and the second ML model information update of the first network node in the first FL cluster including the UE. Fig.23 After receiving report 2310 from UE 2302, leader node 2308 may aggregate the report with other reports from other UEs in the same cluster or peer network, such as regarding Figures 19 to 22As described (e.g., intra-cluster FL or peer FL). For example, the leader node 2308 may receive reports 2310 from UE 1 (a device in this example) and UE 2 (a first network node in a first FL cluster including UE 1 in this example), including updated ML model information 2306 (e.g., weights and gradients) and measurements 2304, respectively, and the leader node 2308 may aggregate the updated ML model information of the respective UEs to form local aggregated ML model information 2316 (aggregated ML model information update). The leader node 2308 may then send the local aggregated ML model information 2316, including aggregated weights and gradients, to the UE 2302 of its cluster or peer network via a multicast message. Thus, the UE 2302 may obtain (e.g., receive) the aggregated ML model information update.

[0243] In another example, if the leader node performs operation 2404, the aggregated ML model information update includes an aggregation of the first ML model information update and a second ML model information update of a network node in the FL cluster (e.g., a second network node in the second FL cluster). Fig.23 , the leader node 2308 of the corresponding cluster including UE 1 may receive global aggregated ML model information 2324 from the FL parameter server 2312, which is aggregated with updates of UE 3 (a second network node in this example) in a cluster different from UE 1.

[0244] Finally, at 2406, the UE or leader node can provide the network node with sidelink communication parameters in the SCI, where the sidelink communication parameters are outputs of the ML model associated with the aggregated ML model information update. For example, 2406 can be performed by the sidelink parameter coordination component 3046.

[0245] In one example, if the UE performs operation 2406, then reference Fig.23 In response to receiving the local aggregated ML model information 2316 (for clustered FL or peer FL) or the global aggregated ML model information 2324 (for clustered FL), the UE 2302 may continue to perform as previously described with respect to Figures 19 to 22 2302 may input the observed data in its respective data sets into its updated ML models to respectively output sidelink communication parameters 2334, and UE 2302 may include these sidelink communication parameters 2334 in corresponding SCI 2336 to optimize communication with other UEs 2338 (or with UE 2302). For example, UE 1 and UE 3 may respectively send SCI 2336 to UE 2338 (a network node in this example), which includes the required communication range for multicast option 1 derived from the updated ML model following the aforementioned FL training.

[0246] In another example, if the leader node performs operation 2406, then reference Fig.23 In response to receiving the local aggregate ML model information 2316 (for cluster FL or peer FL) or the global aggregate ML model information 2324 (for cluster FL), the leader node 2308 may perform the operations described previously with respect to Figures 19 to 22 The FL training described in the above description is repeated until its respective ML models are optimized. Subsequently, the leader node 2308 may input the observed data in its respective data sets into its updated ML models to respectively output side link communication parameters 2334 (e.g., the desired communication range for multicast option 1), and the leader node may include these side link communication parameters 2334 in the corresponding SCI 2336 to optimize communication with other UEs 2338 (or with UE 2302).

[0247] Fig.25 2500 is a flow chart of another method of wireless communication. The method may be performed by a leader node (e.g., UE 104; device 410, 450; RSU 107, 1706; cluster leader 1108, 1306; leader 1204; node 1622, 1802, 1904, 2002; leader node 2308; device 2802) or a FL parameter network entity (e.g., base station 102 / 180, 181, 500, 1308, 1704; device 410, 450; FL parameter server 1106, 2004, 2312; device 2902). For example, the method may be performed by a controller / processor 459, 475 coupled to a memory 460, 476 of the device 410, 450. Optional aspects are illustrated by dashed lines. This approach allows the leader node or FL parameter network entity to provide sidelink parameter coordination in Mode 2 resource allocation scenarios through various signaling in a clustered FL or peer FL.

[0248] At 2502, a leader node or FL parameter network entity can obtain a first ML model information update and measurements associated with the first ML model information update. For example, 2502 can be performed by the FL training component 3144.

[0249] In one example, if the leader node performs operation 2502, then reference Fig.23 , the leader node 2308 may receive updated ML model information 2306 (first ML model information update, e.g., updated model weights and gradients) and measurements 2304 (e.g., observed CBR, etc.) in a report 2310. The measurements may be applied to the UE's ML model to generate updated ML model information 2306.

[0250] In another example, if the FL parameter network entity performs operation 2502, then referring to Fig.23 , the FL parameter server 2312 (in this example, the FL parameter network entity) may receive a report 2322 from a leader node 2308 (the FL cluster leader of the FL cluster that includes UE 1 and UE 2), the report including local aggregated ML model information 2316 (the first ML model information update) and measurement statistics 2318 (in this example, including measurements associated with the update) derived by the leader node 2308 from measurements 2304 observed by UE 2302. These measurements may be applied to the ML models of the UEs to produce updated ML model information 2306, which the leader node 2308 aggregates into the local aggregated ML model information 2316.

[0251] At 2504, the leader node or FL parameter network entity may provide an aggregated ML model information update, wherein the aggregated ML model information update comprises an aggregation of a first ML model information update and a second ML model information update of a first network node in a first FL cluster including the device; and wherein the sidelink communication parameter in the SCI of the network node is an output of the ML model associated with the aggregated ML model information update. For example, 2504 may be performed by the FL training component 3144.

[0252] In one example, if the leader node performs operation 2504, then reference Fig.23 , the leader node 2308 may send local aggregated ML model information 2316 (aggregated ML model information update) to the UE 2302. For example, the local aggregated ML model information 2316 aggregates the updated ML model information 2306 (first ML model information update) of UE 1 with the updated ML model information 2306 (second ML model information update) of UE 2 (in this example, the first network node), where UE 1 and UE 2 are in the same corresponding cluster (in this example, devices) of the leader 1. Using the local aggregated ML model information 2316, the UE 2302, the leader node 2308, or other network nodes (e.g., UE 2338) may perform the operations described previously with respect to Figures 19 to 22 The FL training described in 2330 is repeated until its respective ML models are optimized. The network node may then input the observed data in its respective data sets into its updated ML models to respectively output side link communication parameters 2334 (e.g., the desired communication range for multicast option 1), and the network node may include these side link communication parameters 2334 in the corresponding SCI 2336 to optimize communication with other network nodes.

[0253] In another example, if the FL parameter network entity performs operation 2504, then referring to Fig.23, the FL parameter server 2312 (in this example, the FL parameter network entity) may send global aggregated ML model information 2324 (aggregated ML model information update) to the leader node 2308 (e.g., leader 1), which aggregates the local aggregated ML model information 2316 (first ML model information update) received from the leader node 2308 (e.g., leader 1) with the local aggregated ML model information (second ML model information update) of another leader node (in this example, the second network node, e.g., leader 2) in a cluster different from the leader node 2308 (e.g., leader 1). Using the local aggregated ML model information 2316, the UE 2302, the leader node 2308, or other network nodes (e.g., UE 2338) may perform the operations described previously with respect to Figures 19 to 22 The FL training described in 2330 is repeated until its respective ML models are optimized. The network node may then input the observed data in its respective data sets into its updated ML models to respectively output side link communication parameters 2334 (e.g., the desired communication range for multicast option 1), and the network node may include these side link communication parameters 2334 in the corresponding SCI 2336 to optimize communication with other network nodes.

[0254] Fig.26 2600 is a flow chart of a method of wireless communication. The method may be performed by a UE (e.g., UE 104, 702, 1104, 1202, 1302, 1702, 2302; device 410, 450; node 1402, 1404, 1502, 1504, 1514, 1622, 1626, 1802, 1804, 1816, 1902, 2005, 2102, 2202, apparatus 3002). For example, the method may be performed by a controller / processor 459, 475 coupled to a memory 460, 476 of the apparatus 410, 450. Optional aspects are illustrated by dashed lines. The method allows the UE to implement sidelink parameter coordination in a mode 2 resource allocation scenario through various signaling in a clustered FL or a peer FL.

[0255] At 2602, the flowchart determines whether to apply such as Fig.17 and Fig.18 described network-assisted cluster formation, or whether to apply techniques such as Fig.16 and Fig.18 2604 and 2606. Otherwise, if network assistance is applied in cluster formation, the UE may perform operations 2608, 2610, and 2612. The flowchart then converges at operation 2614.

[0256] At 2604, the UE may provide a first message including first FL information of the device to the first network node. For example, 2604 may be performed by neighbor discovery component 3040. In one example, the first FL information includes at least one of: a machine learning task of the device; information associated with an available sensor coupled to the device for the machine learning task; an available ML model associated with the machine learning task; or available computing resources of the device for the machine learning task. In one example, the machine learning task may include at least one of: a reinforcement learning task; an object detection task; or a classification task. For example, referring to Fig.14 , after the first node 1402 (node ​​1) and the second node 1404 (node ​​2) have designated each other as neighbors (i.e., symmetric neighbors), node 1 may send a message 1414 (a first message in this example) to node 2 indicating information related to its model training. This information may include, for example: ML tasks that node 1 is participating in or interested in participating in (e.g., reinforcement learning, object detection, or classification); available sensors at node 1 and their associated data input formats; available ML models in training, including model state, model architecture, model training parameters, and current performance (e.g., accuracy and loss); or available computing resources (e.g., whether node 1 includes a CPU or GPU, information about its clock speed, available memory, etc.). Message 1414 may include any combination of the foregoing, as well as other information.

[0257] At 2606, the UE may obtain a second message including second FL information of the first network node. For example, 2606 may be performed by neighbor discovery component 3040. For example, referring to Fig.14 In response to receiving message 1414 including information related to model training, the second node 1404 (node ​​2) associates the information with the first node 1402 (node ​​1) in its neighbor list, communication graph, or other data structure. Node 2 may then send a message 1416 indicating information related to model training to node 1. Thus, the first node 1402 may obtain message 1416 (a second message in this example) from the second node 1404. The information may include, for example: the ML tasks that node 2 is participating in or interested in participating in; available sensors at node 2 and their associated data input formats; available ML models in training, including model state, model architecture, model training parameters, and current performance (e.g., accuracy and loss); and available computing resources (e.g., whether node 2 includes a CPU or GPU, information about its clock speed, available memory, etc.). Thus, message 1416 may include the same type of information as message 1414, but for node 2 instead of node 1.

[0258] Based on the first FL information and the second FL information, the first network node may be in the first cluster. Fig.11 , Fig.13 , Fig.14 and Fig. 20 , the first node 1402 and the second node 1404 may be grouped into clusters 1102, 1304, 2006 based on the first FL information previously provided in the first message at 2604 and the second FL information previously obtained in the second message at 2606. For example, in response to Fig.14 and Fig.15 The neighbor discovery process or Fig.18 After a simultaneous neighbor discovery and cluster formation process where the first node 1402 is in the second node's neighbor list, communication graph, or other data structure (or vice versa), the first node 1402 and the second node 1404 may be assigned to the same cluster.

[0259] At 2608, the UE may obtain a message (first message) indicating a FL cluster leader of a first FL cluster including the device. For example, 2608 may be performed by cluster formation component 3042. For example, at 2610, the UE may provide a second message including FL information, and at 2612, the UE may provide a third message indicating a second network node, wherein the message (first message) indicates one of the second network node or the third network node as the FL cluster leader of the first FL cluster based on the FL information. For example, 2610 and 2612 may be performed by cluster formation component 3042. In one example, reference Fig.17 , the UE 1702 may send a FL message 1708 to the RSU 1706 that is the intended cluster leader. The FL message 1708 may indicate FL or model training related information of the UE 1702. In addition, the UE 1702 may send an RSU indication message 1712 to the RSU 1706, which includes an identifier of the selected RSU (second network node) to act as a representative of the UE to pass its FL information to the base station 1704 for cluster formation and leader selection. In response, the UE 1702 may receive a message 1718 (first message) from the RSU 1706 (second network node) that identified the UE in its RSU indication message 1712 or from a different RSU (third network node), which notifies the UE that it is a member of a cluster led by the corresponding RSU (second network node or third network node) (e.g., Fig.11 , Fig.13 and Fig. 20 The RSU selected as the cluster leader and the UEs selected as members or participants of the cluster led by the RSU may be based on the aggregated FL information from UE 1702 and other UEs.

[0260] At 2614, the UE may provide a first ML model information update and a measurement associated with the first ML model information update. For example, 2614 may be performed by the FL training component 3044. In one example, the measurement may include at least one of the following: a CBR, a number of received packets at the device, or a number of resource reservations of the first network node. In one example, the first ML model information update may include at least one of the following: a weight; a gradient; or a scaling factor. The scaling factor may be, for example, a multiplicand (e.g., the update may be a multiplicative increment) or a summand (e.g., the update may be an additive increment). For example, referring to Fig.23 Initially, UE 2302 collects measurements 2304 from its environment (e.g., CBR, number of received packets at the UE, number of resource reservations of other UEs, or other metrics observed instantaneously or over a configured time period). UE 2302 may store these observations in a local dataset and map the dataset to an ML model or strategy to select sidelink communication parameters. For example, in Mode 2 resource allocation, UE 2302 may use reinforcement learning and back propagation to train a neural network (e.g., Figure 6 The UE 2302 trains its local ML model to obtain updated ML model information 2306 (e.g., weights 616, biases 618, subsampling or scaling factors in the subsampling layer 608, or other updated ML model information after minimization of the loss function). After training its local ML model, the UE 2302 sends a report 2310 including the updated ML model information 2306 and the measurements 2304 to a leader node 2308 (e.g., a cluster leader in a clustered FL or a leader in a peer FL). To reduce transmission traffic, the updated ML model information may include, for example, modified weights, weights that have changed beyond a threshold, or multiplicative or additive increments or percentages of these weights.

[0261] In one example, the measurements may also include statistics associated with the CBR within a configured time window. Fig.23 , UE 2302 may measure the distribution of CBRs observed over a time window, calculate an average or maximum CBR from the distribution of CBRs, and send the average or maximum CBR in a report along with updated model weights, gradients, or other ML model information.

[0262] In one example, the UE may provide first ML model information updates and measurements at a periodicity configured by one of the FL cluster leader or the FL parameter network entity. Fig.23, UE 2302 may send report 2310 to leader node 2308 periodically as configured by leader node 2308 or configured by FL parameter server 2312. In another example, UE may provide first ML model information updates and measurements semi-persistently within a configured time window. For example, referring to Fig.23 , UE 2302 may semi-persistently send reports 2310 for a configured time period, and the leader node 2308 or FL parameter server 2312 may select the configured time period based on a period of time during which UE 2302 intends to participate in a cluster or peer FL. In another example, the UE may provide first ML model information updates and measurements aperiodically in response to a request from the FL cluster leader. For example, referring to Fig.23 , UE 2302 may send reports 2310 non-periodically in response to a request from leader node 2308 to trigger a corresponding reporting instance.

[0263] At 2616, the UE may obtain an RRC configuration from the FL parameter network entity, the RRC configuration indicating resources for reporting the first ML model information update and measurement. For example, 2616 may be performed by the FL training component 3044. The UE may provide the first ML model information update and measurement in the resource. For example, reference Fig.23 , the FL parameter server 2312 may send a report configuration 2314 (which may be an RRC configuration) to the UE 2302 via RRC signaling, and after receiving the report configuration 2314, the UE 2302 may send a report 2310 to the leader node 2308 in the resources indicated in the report configuration 2314.

[0264] At 2618, the UE may obtain a message from the FL cluster leader triggering the sending of a report of the first ML model information update and measurement. For example, 2618 may be performed by the FL training component 3044. The UE may provide the first ML model information update and measurement in the reserved side link resources in response to the message. For example, referring to Fig.23 , report configuration 2314 may be triggered by leader node 2308. For example, leader node 2308 may reserve side link resources in PSSCH for sending report configuration 2314 and receiving report 2310, and the leader node may send a message including the report configuration in the reserved side link resources indicated in SCI. In response to obtaining report configuration 2314 (message), the UE may send report 2310 to leader node 2308 in the reserved side link resources.

[0265] In one example, the UE may provide the first ML model information update and measurement in a dedicated resource for reporting the first ML model information update and measurement. Fig.23Regardless of the source of the report configuration 2314, the UE 2302 may send the report 2310 in dedicated resources for such reporting, such as preconfigured time-frequency resources.

[0266] At 2620, the UE may obtain an aggregated ML model information update, wherein the aggregated ML model information update includes an aggregation of a first ML model information update and a second ML model information update of a first network node in a first FL cluster including the device. For example, 2620 may be performed by the FL training component 3044. The first FL cluster may be a group of network nodes in a clustered FL architecture (e.g., where UEs may be grouped into different clusters) or a peer FL architecture (e.g., where UEs are in a single cluster or peer node group). For example, referring to Fig.23 After receiving report 2310 from UE 2302, leader node 2308 may aggregate the report with other reports from other UEs in the same cluster or peer network, such as regarding Figures 19 to 22 As described (e.g., intra-cluster FL or peer FL). For example, the leader node 2308 may receive reports 2310 from UE 1 (a device in this example) and UE 2 (a first network node in a first FL cluster including UE 1 in this example), including updated ML model information 2306 (e.g., weights and gradients) and measurements 2304, respectively, and the leader node 2308 may aggregate the updated ML model information of the respective UEs to form local aggregated ML model information 2316 (aggregated ML model information update). The leader node 2308 may then send the local aggregated ML model information 2316, including aggregated weights and gradients, to the UE 2302 of its cluster or peer network via a multicast message. Thus, the UE 2302 may obtain (e.g., receive) the aggregated ML model information update.

[0267] In one example, the UE may provide the first ML model information update and measurement to the FL cluster leader of the first FL cluster. Fig.23 , UE 2302 sends a report 2310 including updated ML model information 2306 and measurements 2304 to a leader node 2308 , where in this example, leader node 2308 is a cluster leader of clusters 1102 , 1304 , 2006 including UE 2302 .

[0268] In one example, the aggregated ML model information update from the FL cluster leader may also include additional ML model information updates for the second network nodes in the second FL cluster. Fig.23In response to obtaining message 2328 from FL parameter server 2312, leader node 2308 may send message 2330 including global aggregate ML model information 2324 to its corresponding UE 2302, respectively, for application to further intra-cluster FL training, such as regarding Fig.19 and Fig. 20 For example, leader 1 may send a message 2330 to UE 1 including global aggregated ML model information 2324 (in this example, an aggregated ML model information update) aggregated with updates of UE 3 (in this example, a second network node) in a cluster different from UE 1.

[0269] At 2622, the UE may obtain a message indicating that the device is a cluster member of a third FL cluster based on the measurement being an outlier of a statistic associated with the second measurement of the first network node. For example, 2622 may be performed by the FL training component 3044. For example, referring to Fig.23 , the leader node 2308 may send outlier UE cluster information 2332 in a message to the UE indicated as an outlier of its current cluster. An outlier UE may be, for example, a UE whose measurement 2304 has a significantly different value (e.g., CBR value) or a different value distribution than the value of the measurement statistic 2318 (in this example, a statistic) derived from the measurement 2304 (including the second measurement in this example) obtained from other UEs (including the first network node). In response to receiving the message 2328 from the FL parameter server 2312, the leader 1 may send the outlier UE cluster information 2332 to UE 2, thereby notifying the UE of its new status as a participant node of the leader 2's cluster (in this example, the third FL cluster).

[0270] At 2624, the UE may provide the network node with a sidelink communication parameter in the SCI, wherein the sidelink communication parameter is an output of the ML model associated with the aggregated ML model information update. For example, 2624 may be performed by the sidelink parameter coordination component 3046. For example, referring to Fig.23 In response to receiving the local aggregated ML model information 2316 (for clustered FL or peer FL) or the global aggregated ML model information 2324 (for clustered FL), the UE 2302 may continue to perform as previously described with respect to Figures 19 to 222302 may input the observed data in its respective data sets into its updated ML models to respectively output sidelink communication parameters 2334, and UE 2302 may include these sidelink communication parameters 2334 in corresponding SCI 2336 to optimize communication with other UEs 2338 (or with UE 2302). For example, UE 1 and UE 3 may respectively send SCI 2336 to UE 2338 (a network node in this example), which includes the required communication range for multicast option 1 derived from the updated ML model following the aforementioned FL training.

[0271] In one example, the first network node and the second network node may be the same network node or different network nodes. In one example, the first FL cluster and the second FL cluster may be the same FL cluster or different FL clusters.

[0272] In one example, the first network node may be a first UE, and the second network node may be a second UE or an RSU. In this example, the device may be a third UE and the network node may be a second RSU. Alternatively, the device and the network node may be UEs.

[0273] Fig. 27 2700 is a flow chart of a method of wireless communication. The method may be performed by a leader node (e.g., UE 104; device 410, 450; RSU 107, 1706; cluster leader 1108, 1306; leader 1204; node 1622, 1802, 1904, 2002; leader node 2308; device 3002). For example, the method may be performed by a controller / processor 459, 475 coupled to a memory 460, 476 of the device 410, 450. Optional aspects are illustrated by dashed lines. The method allows the leader node to achieve side link parameter coordination in a mode 2 resource allocation scenario through various signaling in a cluster FL or a peer FL.

[0274] At 2702, the flowchart determines whether to apply such as Fig.17 and Fig.18 described network-assisted cluster formation, or whether to apply techniques such as Fig.16 and Fig.18 2704 and 2706. Otherwise, if network assistance is applied in cluster formation, the leader node may perform operations 2708 and 2710. The flowchart then converges at operation 2712.

[0275] At 2704, the leader node may provide a first message including first FL information of the device to the first network node. For example, 2704 may be performed by neighbor discovery component 3040. In one example, the first FL information may include at least one of: a machine learning task for the device; information associated with an available sensor coupled to the device for the machine learning task; an available ML model associated with the machine learning task; or available computing resources of the device for the machine learning task. In one example, the machine learning task may include at least one of: a reinforcement learning task; an object detection task; or a classification task. For example, the leader node may perform a task associated with the UE at 2604 regarding the first FL information of the device. Fig.26 Perform similar operations.

[0276] At 2706, the leader node can obtain a second message including second FL information of the first network node. For example, 2706 can be performed by neighbor discovery component 3040. Based on the first FL information and the second FL information, the first network node can be in a first FL cluster. For example, the leader node can perform communication with the UE at 2606 regarding Fig.26 Perform similar operations.

[0277] At 2708, the leader node may provide a message indicating that the device is a FL cluster leader of the first FL cluster. For example, 2708 may be performed by cluster formation component 3042. For example, at 2710, the leader node may obtain a second message including FL information of the first network node; wherein the message is responsive to the FL information. For example, 2710 may be performed by cluster formation component 3042. For example, the leader node may perform communication with the UE at 2608 and 2610 regarding Fig.26 Perform operations similar to those performed by

[0278] At 2712, the leader node may provide a first ML model information update and a measurement associated with the first ML model information update. For example, 2712 may be performed by the FL training component 3044. In one example, the first ML model information update may include at least one of: a weight; a gradient; or a scaling factor. For example, referring to Fig.23 , after receiving the local aggregated ML model information 2316 (e.g., updated model weights and gradients) and measurement statistics 2318 (e.g., summary statistics about reported measurements from various UEs), the leader node 2308 may send a report 2322 including the information to the FL parameter server 2312 (in this example, the first ML model information update and measurements, respectively).

[0279] At 2714, the leader node can obtain an aggregated ML model information update, wherein the aggregated ML model information update includes an aggregation of the first ML model information update and a second ML model information update of a network node in the FL cluster (e.g., a second network node in a second FL cluster, e.g., after operation 2708). For example, 2720 can be performed by the FL training component 3044. The second FL cluster can be a group of network nodes in a clustered FL architecture (e.g., where UEs can be grouped into different clusters) or a peer FL architecture (e.g., where UEs are in a single cluster or peer node group). For example, referring to Fig.23 , the leader node 2308 of the corresponding cluster including UE 1 may receive global aggregated ML model information 2324 from the FL parameter server 2312, which is aggregated with updates of UE 3 (a second network node in this example) in a cluster different from UE 1.

[0280] In one example, the leader node may provide the first ML model information updates and measurements to the FL parameter network entity. Fig.23 , the leader node 2308 can send a report 2322 to the FL parameter server 2312.

[0281] At 2716, the leader node may obtain a report including a third ML model information update from a first network node in the first FL cluster, wherein the first ML model information update includes a second aggregation of the third ML model information update and additional ML model information updates from additional network nodes in the first FL cluster. For example, 2714 may be performed by the FL training component 3044. For example, referring to Fig.23 , a leader node 2308 (cluster leader) may receive a report 2310 including updated ML model information 2306 from a UE 2302 in its corresponding cluster (in this example, a first network node). The local aggregated ML model information 2316 sent by the leader node 2308 to the FL parameter server 2312 may include an aggregation (a second aggregation) of updated ML model information 2306 from multiple UEs 2302 (e.g., UE 1 and UE 2, where in this example, UE 2 is an additional network node in the same cluster as UE 1).

[0282] At 2718, the leader node can provide the second aggregation to the first network node in the first FL cluster. For example, 2716 can be performed by the FL training component 3044. For example, referring to Fig.23 , the leader node 2308 (in this example, the cluster leader) may send local aggregated ML model information 2316 including an aggregation (second aggregation) of the updated ML model information 2306 of UE 1 and UE 2 to UE 1 (in this example, the first network node) in its corresponding cluster.

[0283] In one example, the report may include a second measurement associated with the third ML model information update, and the measurement may include statistics associated with the second measurement. Fig.23 , the report 2310 obtained from the UE 2302 may include measurements 2304 (e.g., in this example, second measurements, including CBR, CR, etc.), which the UE observes in its environment and maps these measurements to its local ML model (e.g., using reinforcement learning and back propagation) to obtain updated ML model information 2306 (third ML model information update). The leader node 2308 may derive its own measurements or measurement statistics 2318 (e.g., average or maximum CBR, variance of CBR, etc.) from the measurements 2304 obtained from the UE, and the measurements (including statistics associated with the second measurements or measurements 2304) are reported to the FL parameter server 2312 in the report 2322.

[0284] At 2720, the leader node may provide the identifier of the first network node and the second measurement to the FL parameter network entity. For example, 2718 may be performed by the FL training component 3044. The second measurement may be a statistical outlier. For example, reference Fig.23 , the leader node 2308 may send outlier UE information 2320 to the FL parameter server 2312, which includes measurements 2304 observed by the leader node 2308 as outliers of the measurement statistics 2318, and an identifier of the UE 2302 that observed such outlier measurements. For example, if the observed CBR of UE 2 differs from the average CBR of other UEs in the same cluster by more than three standard deviations, the leader node may notify the FL parameter server of the identifier of UE 2 (in this example, the first network node).

[0285] At 2722, the leader node may provide aggregated ML model information updates to the first network node in the first FL cluster. For example, 2722 may be performed by the FL training component 3044. For example, referring to Fig.23 , after the leader node 2308 receives the global aggregate ML model information 2324 from the FL parameter server 2312, the leader node 2308 may send the global aggregate ML model information 2324 (aggregate ML model information update) to UE 1 in its corresponding cluster (the first network node in this example).

[0286] At 2724, the leader node can provide the network node with a sidelink communication parameter in the SCI, wherein the sidelink communication parameter is an output of the ML model associated with the aggregated ML model information update. For example, 2724 can be performed by the sidelink parameter coordination component 3046. For example, referring to Fig.23In response to receiving the local aggregate ML model information 2316 (for cluster FL or peer FL) or the global aggregate ML model information 2324 (for cluster FL), the leader node 2308 may perform the operations described previously with respect to Figures 19 to 22 The FL training described in the above description is repeated until its respective ML models are optimized. Subsequently, the leader node 2308 may input the observed data in its respective data sets into its updated ML models to respectively output side link communication parameters 2334 (e.g., the desired communication range for multicast option 1), and the leader node may include these side link communication parameters 2334 in the corresponding SCI 2336 to optimize communication with other UEs 2338 (or with UE 2302).

[0287] In one example, the first network node and the second network node may be the same network node or different network nodes. In one example, the first FL cluster and the second FL cluster may be the same FL cluster or different FL clusters.

[0288] In one example, the first network node may be a first UE, and the second network node may be a second UE or an RSU. In this example, the device may be a third UE and the network node may be a second RSU. Alternatively, the device may be a second RSU and the network node may be a third UE. Alternatively, the device and the network node may be UEs. Alternatively, the device and the network node may be RSUs.

[0289] Fig.28 2800 is a flowchart of another method of wireless communication. The method may be performed by a leader node (e.g., UE 104; device 410, 450; RSU 107, 1706; cluster leader 1108, 1306; leader 1204; node 1622, 1802, 1904, 2002; leader node 2308; device 3002). For example, the method may be performed by a controller / processor 459, 475 coupled to a memory 460, 476 of the device 410, 450. Optional aspects are illustrated by dashed lines. The method allows the leader node to provide sidelink parameter coordination in a mode 2 resource allocation scenario through various signaling in a cluster FL or a peer FL.

[0290] At 2802, the flowchart determines whether to apply such as Fig.17 and Fig.18 described network-assisted cluster formation, or whether to apply techniques such as Fig.16 and Fig.18The process of describing the cluster formation without network assistance diverges. If network assistance is not applied in the cluster formation, the leader node may perform operations 2804 and 2806. Otherwise, if network assistance is applied in the cluster formation, the leader node may perform operations 2808 and 2810. The flowchart then converges at operation 2812.

[0291] At 2804, the leader node may provide a first message including first FL information of the device to the first network node. For example, 2804 may be performed by neighbor discovery component 3140. In one example, the first FL information may include at least one of: a machine learning task of the device; an available sensor coupled to the device for the machine learning task; an available ML model associated with the machine learning task; or an available computing resource of the device for the machine learning task. In one example, the machine learning task may include at least one of: a reinforcement learning task; an object detection task; or a classification task. For example, the leader node may perform a task associated with the UE at 2604 regarding the first FL information of the device. Fig.26 Perform similar operations.

[0292] At 2806, the leader node can obtain a second message including second FL information of the first network node. For example, 2806 can be performed by neighbor discovery component 3140. Based on the first FL information and the second FL information, the first network node can be in a first FL cluster. For example, the leader node can perform communication with the UE at 2606 regarding Fig.26 Perform similar operations.

[0293] At 2808, the leader node may obtain a message indicating a FL cluster leader of a first FL cluster including the device. For example, 2808 may be performed by cluster formation component 3142. For example, referring to Fig.17 , RSU 1 (the FL cluster leader in this example) may receive a message 1716 including cluster information from base station 1704 , which indicates that the RSU leads a cluster including itself (device) and UE 1702 .

[0294] At 2810, the leader node may provide a message indicating that the device is the FL cluster leader of the first FL cluster. For example, 2810 may be performed by the cluster formation component 3142. For example, referring to Fig.18 , the first node 1802 (FL cluster leader) may send a message 1810 that permits the second node 1804 to join its cluster and thereby indicates that the first node 1802 (device) leads the cluster including the second node 1804.

[0295] At 2812, the leader node may obtain a first ML model information update and a measurement associated with the first ML model information update. For example, 2812 may be performed by the FL training component 3144. The first ML model information update may include at least one of the following: a weight; a gradient; or a scaling factor. For example, reference Fig.23 , the leader node 2308 may receive updated ML model information 2306 (first ML model information update, e.g., updated model weights and gradients) and measurements 2304 (e.g., observed CBR, etc.) in a report 2310. The measurements may be applied to the UE's ML model to generate updated ML model information 2306.

[0296] At 2814, the leader node may provide an aggregated ML model information update, wherein the aggregated ML model information update comprises an aggregation of a first ML model information update and a second ML model information update of a first network node in a first FL cluster including the device; and wherein a sidelink communication parameter in the SCI of the network node is an output of the ML model associated with the aggregated ML model information update. For example, 2814 may be performed by the FL training component 3144. The first FL cluster may be a group of network nodes in a clustered FL architecture (e.g., where UEs may be grouped into different clusters) or a peer FL architecture (e.g., where UEs are in a single cluster or peer node group). For example, referring to Fig.23 , the leader node 2308 may send local aggregated ML model information 2316 (aggregated ML model information update) to the UE 2302. For example, the local aggregated ML model information 2316 aggregates the updated ML model information 2306 (first ML model information update) of UE 1 with the updated ML model information 2306 (second ML model information update) of UE 2 (in this example, the first network node), where UE 1 and UE 2 are in the same corresponding cluster (in this example, devices) of the leader 1. Using the local aggregated ML model information 2316, the UE 2302, the leader node 2308, or other network nodes (e.g., UE 2338) may perform the operations described previously with respect to Figures 19 to 22 The FL training described in 2330 is repeated until its respective ML models are optimized. The network node may then input the observed data in its respective data sets into its updated ML models to respectively output side link communication parameters 2334 (e.g., the desired communication range for multicast option 1), and the network node may include these side link communication parameters 2334 in the corresponding SCI 2336 to optimize communication with other network nodes.

[0297] In one example, the leader node may obtain first ML model information updates and measurements from a third network node in the first FL cluster. Fig.23, the leader node 2308 may receive updated ML model information 2306 (first ML model information update) and measurements 2304 in a report 2310 from UE 1 or UE 2 (the third network node in this example) in the same cluster.

[0298] In one example, the leader node can provide aggregated ML model information updates to the FL parameter network entity. Fig.23 , the leader node 2308 may send local aggregated ML model information 2316 (aggregate ML model information update) in a report 2322 to the FL parameter server 2312 (in this example, the FL parameter network entity).

[0299] At 2816, the leader node can provide statistics associated with the measurements to the FL parameter network entity. For example, 2816 can be performed by the FL training component 3144. For example, referring to Fig.23 , the leader node 2308 may send measurement statistics 2318 (including statistics derived from measurements 2304 ) in a report 2322 to a FL parameter server 2312 (in this example, a FL parameter network entity).

[0300] At 2818, the leader node may provide an identifier of the first network node and an individual measurement of the first network node to the FL parameter network entity, wherein the individual measurement of the first network node is a statistical outlier. For example, 2818 may be performed by the FL training component 3144. For example, referring to Fig.23 , the leader node 2308 may send outlier UE information 2320 to the FL parameter server 2312, which includes measurements 2304 observed by the leader node 2308 as outliers of the measurement statistics 2318, and an identifier of the UE 2302 that observed such outlier measurements. For example, if the observed CBR of UE 2 differs from the average CBR of other UEs in the same cluster by more than three standard deviations, the leader node may notify the FL parameter server of the identifier of UE 2 (in this example, the first network node).

[0301] At 2820, the leader node may obtain from the FL parameter network entity a second aggregation of the first ML model information update and additional ML model information updates for a second network node in a second FL cluster. For example, 2820 may be performed by the FL training component 3144. For example, referring to Fig.23, the leader node 2308 (e.g., Leader 1) may receive global aggregated ML model information 2324 from the FL parameter server 2312 (in this example, the FL parameter network entity), which aggregates the local aggregated ML model information 2316 (first ML model information update) of the leader node 2308 (e.g., Leader 1) with the local aggregated ML model information (additional ML model information update) of another leader node (in this example, the second network node, e.g., Leader 2) in a cluster different from the leader node 2308 (e.g., Leader 1) (corresponding to a second aggregation).

[0302] At 2822, the leader node may provide the second aggregation to the first network node in the first FL cluster. For example, 2822 may be performed by the FL training component 3144. For example, referring to Fig.23 , the leader node 2308 may send global aggregated ML model information 2324 (including the second aggregation mentioned at 2820) to UE 1 or UE2 in its cluster (in this example, the first network node).

[0303] In one example, the first network node and the second network node may be the same network node or different network nodes. In one example, the first FL cluster and the second FL cluster may be the same FL cluster or different FL clusters.

[0304] In one example, the first network node may be a first UE, the second network node may be a second UE or an RSU, and the FL parameter network entity may be a base station or a server. In one example, the device may be a second RSU and the network node may be a third UE. In another example, the device and the network node may be RSUs.

[0305] Fig.29 2900 is a flow chart of a method of wireless communication. The method may be performed by a FL parameter network entity (e.g., base station 102 / 180, 181, 500, 1308, 1704; device 410, 450; FL parameter server 1106, 2004, 2312; apparatus 3102). Optional aspects are illustrated by dashed lines. The method allows the FL parameter network entity to provide sidelink parameter coordination in a mode 2 resource allocation scenario by clustering various signaling in the FL.

[0306] At 2902, the FL parameter network entity may obtain a first ML model information update and a measurement associated with the first ML model information update. For example, 2902 may be performed by the FL training component 3144. In one example, the FL parameter network entity may obtain the first ML model information update and the measurement from the FL cluster leader of the first FL cluster. For example, referring to Fig.23, the FL parameter server 2312 (in this example, the FL parameter network entity) may receive a report 2322 from a leader node 2308 (the FL cluster leader of the FL cluster that includes UE 1 and UE 2), the report including local aggregated ML model information 2316 (the first ML model information update) and measurement statistics 2318 (in this example, including measurements associated with the update) derived by the leader node 2308 from measurements 2304 observed by UE 2302. These measurements may be applied to the ML models of the UEs to produce updated ML model information 2306, which the leader node 2308 aggregates into the local aggregated ML model information 2316.

[0307] At 2904, the FL parameter network entity may obtain an identifier of a first network node in a first FL cluster and an individual measurement of the first network node from the FL cluster leader, wherein the individual measurement of the first network node is an outlier of the measurement from the FL cluster leader. For example, 2904 may be performed by the FL training component 3144. For example, referring to Fig.23 , FL parameter server 2312 (in this example, a FL parameter network entity) may receive outlier UE information 2320 from leader node 2308 (in this example, a FL cluster leader), including measurements 2304 observed by leader node 2308 as outliers of measurement statistics 2318, and an identifier of UE 2302 that observed such outlier measurements. For example, if the observed CBR of UE 2 differs by more than three standard deviations from the average CBR of other UEs in the same cluster, the leader node may notify the FL parameter server of the identifier of UE 2 (in this example, the first network node).

[0308] At 2906, the FL parameter network entity may provide an aggregated ML model information update, wherein the aggregated ML model information update comprises an aggregation of a first ML model information update and a second ML model information update of a network node in the FL cluster (e.g., a second network node in a second FL cluster, e.g., after operation 2904); and wherein the sidelink communication parameter in the SCI of the network node is an output of the ML model associated with the aggregated ML model information update. For example, 2906 may be performed by the FL training component 3144. The second FL cluster may be a group of network nodes in a clustered FL architecture (e.g., where UEs may be grouped into different clusters) or a peer FL architecture (e.g., where UEs are in a single cluster or peer node group). For example, referring to Fig.23, the FL parameter server 2312 (in this example, the FL parameter network entity) may send global aggregated ML model information 2324 (aggregated ML model information update) to the leader node 2308 (e.g., leader 1), which aggregates the local aggregated ML model information 2316 (first ML model information update) received from the leader node 2308 (e.g., leader 1) with the local aggregated ML model information (second ML model information update) of another leader node (in this example, the second network node, e.g., leader 2) in a cluster different from the leader node 2308 (e.g., leader 1). Using the local aggregated ML model information 2316, the UE 2302, the leader node 2308, or other network nodes (e.g., UE 2338) may perform the operations described previously with respect to Figures 19 to 22 The FL training described in 2330 is repeated until its respective ML models are optimized. The network node may then input the observed data in its respective data sets into its updated ML models to respectively output side link communication parameters 2334 (e.g., the desired communication range for multicast option 1), and the network node may include these side link communication parameters 2334 in the corresponding SCI 2336 to optimize communication with other network nodes.

[0309] In one example, the first ML model information update may include a second aggregation of additional ML model information updates of the first network nodes in the first FL cluster. Fig.23 , the local aggregated ML model information 2316 (first ML model information update) may aggregate (second aggregation) the updated ML model information 2306 (additional ML model information update) of the UE 2302 (including the first network node) in the cluster (first FL cluster) led by the leader node 2308 .

[0310] At 2908, the FL parameter network entity may provide a message indicating that the first network node is a cluster member of a third FL cluster based on the individual measurement being an outlier. For example, 2908 may be performed by the FL training component 3144. For example, referring to Fig.23, the FL parameter server 2312 (in this example, the FL parameter network entity) may send a message 2328 including the outlier UE reassignment information 2326 to the leader node 2308. The leader node 2308 may then send outlier UE cluster information 2332 in the message to the UE indicated as the outlier of its current cluster in the outlier UE reassignment information 2326. An outlier UE may be, for example, a UE whose measurement 2304 has a significantly different value (e.g., CBR value) or a different value distribution than the value of the measurement statistic 2318 (in this example, the statistic) derived from the measurement 2304 (in this example, including the second measurement) obtained from other UEs (including the first network node). As an example, the outlier UE cluster information 2332 may indicate that UE 2 (in this example, the first network node) will be reassigned as a participant node (cluster member) in the cluster (FL cluster) led by leader 1 to a different cluster (in this example, the third FL cluster) led by leader 2.

[0311] In one example, the first network node may be a first UE, and the second network node may be a second UE or an RSU. In one example, the device may be a FL parameter network entity, and the network node may be a second RSU. In another example, the device may be a FL parameter network entity, and the network node may be a third UE. The FL parameter network entity may be a base station or a server.

[0312] Fig.303000 is a diagram illustrating an example of a hardware implementation for an apparatus 3002. The apparatus 3002 is a network node such as a UE or RSU, and includes a cellular baseband processor 3004 (also referred to as a modem) coupled to a cellular RF transceiver 3022 and one or more subscriber identity modules (SIM) cards 3020, an application processor 3006 coupled to a secure digital (SD) card 3008 and a screen 3010, a Bluetooth module 3012, a wireless local area network (WLAN) module 3014, a global positioning system (GPS) module 3016, and a power source 3018. The cellular baseband processor 3004 communicates with the UE 104, the RSU 107, the BS 102 / 180, and / or other network nodes through the cellular RF transceiver 3022. The cellular baseband processor 3004 may include a computer-readable medium / memory. The computer-readable medium / memory may be non-transitory. The cellular baseband processor 3004 is responsible for general processing, including executing software stored on the computer-readable medium / memory. The software, when executed by the cellular baseband processor 3004, causes the cellular baseband processor 3004 to perform the various functions described above. The computer-readable medium / memory may also be used to store data manipulated by the cellular baseband processor 3004 when executing the software. The cellular baseband processor 3004 also includes a receiving component 3030, a communication manager 3032, and a sending component 3034. The communication manager 3032 includes one or more of the illustrated components. The components within the communication manager 3032 may be stored in a computer-readable medium / memory and / or configured as hardware within the cellular baseband processor 3004. The cellular baseband processor 3004 may be a component of the device 410, 450, and may include a memory 460, 476 and / or at least one of the following: a TX processor 428, 468, an RX processor 456, 470, and a controller / processor 459, 475. In one configuration, the apparatus 3002 may be a modem chip and include only the baseband processor 3004, and in another configuration, the apparatus 3002 may be an entire UE, RSU, or other network node (e.g., see Figure 4 devices 410, 450) and include additional modules of the apparatus 3002.

[0313] Communications manager 3032 includes a neighbor discovery component 3040 configured to provide a message including FL information to a network node, e.g., as described in conjunction with 2604 or 2704. Neighbor discovery component 3040 is also configured to obtain a message including FL information of a network node, e.g., as described in conjunction with 2606 or 2706.

[0314] The communication manager 3032 also includes a cluster forming component 3042 configured to obtain a message indicating a FL cluster leader of the FL cluster, e.g., as described in conjunction with 2608. The cluster forming component 3042 is further configured to provide a message including FL information, e.g., as described in conjunction with 2610. The cluster forming component 3042 is further configured to provide a message indicating the network node, wherein the message indicates one of the network node or another network node as a FL cluster leader of the FL cluster based on the FL information, e.g., as described in conjunction with 2612. The cluster forming component 3042 is further configured to provide a message indicating the FL cluster leader of the FL cluster, e.g., as described in conjunction with 2708. The cluster forming component 3042 is further configured to obtain a message including FL information of the network node; wherein the message is responsive to the FL information, e.g., as described in conjunction with 2710.

[0315] The communication manager 3032 also includes a FL training component 3044 configured to provide a first ML model information update and measurements associated with the first ML model information update, for example, as described in conjunction with 2402, 2614, or 2712. The FL training component 3044 is further configured to obtain an RRC configuration from a FL parameter network entity, the RRC configuration indicating resources for reporting the first ML model information update and the measurements, for example, as described in conjunction with 2616. The FL training component 3044 is further configured to obtain a message from a FL cluster leader triggering the sending of a report of the first ML model information update and the measurements, for example, as described in conjunction with 2618. The FL training component 3044 is further configured to obtain an aggregated ML model information update, wherein the aggregated ML model information update includes an aggregation of the first ML model information update and second ML model information updates of network nodes in the FL cluster, for example, as described in conjunction with 2404, 2620, or 2714. The FL training component 3044 is further configured to obtain a message indicating cluster membership in another FL cluster based on the measurement being an outlier of a statistic associated with another measurement of the network node, e.g., as described in conjunction with 2622. The FL training component 3044 is further configured to obtain a report including an ML model information update from a cluster member in the FL cluster, wherein the ML model information update includes an aggregation of the ML model information update with additional ML model information updates from additional network nodes in the FL cluster, e.g., as described in conjunction with 2716. The FL training component 3044 is further configured to provide the aggregation to the network node in the FL cluster, e.g., as described in conjunction with 2718. The FL training component 3044 is further configured to provide an identifier of the network node and the other measurement to the FL parameter network entity, e.g., as described in conjunction with 2720. The FL training component is further configured to provide the aggregated ML model information update to another network node in another FL cluster, e.g., as described in conjunction with 2722.

[0316] The communication manager 3032 also includes a sidelink parameter coordination component 3046, which is configured to provide a sidelink communication parameter in the SCI to the network node, wherein the sidelink communication parameter is an output of the ML model associated with the aggregated ML model information update, for example, as described in conjunction with 2406, 2624 or 2724.

[0317] The device may include executing Fig.24 , Fig.26 and Fig. 27 The additional components of each box of the algorithm in the preceding flowchart. Therefore, Fig.24 , Fig.26 and Fig. 27Each block in the aforementioned flow chart of can be performed by a component, and the device may include one or more of these components. These components can be one or more hardware components, which are specifically configured to perform the process / algorithm, implemented by a processor configured to perform the process / algorithm, stored in a computer-readable medium for implementation by a processor, or some combination thereof.

[0318] In one configuration, the device 3002 (and specifically, the cellular baseband processor 3004) includes: a component for providing a first ML model information update and a measurement associated with the first ML model information update; a component for obtaining an aggregated ML model information update, wherein the aggregated ML model information update includes an aggregation of the first ML model information update and a second ML model information update of one of the following items: a first network node in a first FL cluster including the device, or a second network node in a second FL cluster; and wherein the component for providing is also configured to provide a sidelink communication parameter in an SCI to the network node, wherein the sidelink communication parameter is an output of the ML model associated with the aggregated ML model information update.

[0319] In one configuration, the means for providing is configured to provide the first ML model information update and the measurement to a FL cluster leader of the first FL cluster; and wherein the aggregation is performed with the second ML model information update of the first network node in the first FL cluster.

[0320] In one configuration, the measurement includes at least one of: a channel busy rate (CBR), a number of received packets at the device, or a number of resource reservations of the first network node.

[0321] In one configuration, the measurements also include statistics associated with the CBR within a configured time window.

[0322] In one configuration, the means for providing is configured to provide the first ML model information updates and measurements at a periodicity configured by one of the FL cluster leader or the FL parameter network entity.

[0323] In one configuration, the means for providing is configured to provide the first ML model information updates and measurements semi-persistently within a configured time window.

[0324] In one configuration, the means for providing is configured to aperiodically provide the first ML model information updates and measurements in response to a request from the FL cluster leader.

[0325] In one configuration, the component for obtaining is further configured to: obtain an RRC configuration from the FL parameter network entity, the RRC configuration indicating resources for reporting the first ML model information update and measurement; wherein the component for providing is configured to provide the first ML model information update and measurement in the resource.

[0326] In one configuration, the component for obtaining is further configured to: obtain a message from the FL cluster leader triggering the sending of a report of the first ML model information update and measurement; wherein the component for providing is configured to provide the first ML model information update and measurement in the reserved side link resources in response to the message.

[0327] In one configuration, the means for providing is configured to provide the first ML model information updates and measurements in a dedicated resource for reporting the first ML model information updates and measurements.

[0328] In one configuration, the aggregated ML model information updates from the FL cluster leader also include additional ML model information updates for a second network node in a second FL cluster.

[0329] In one configuration, the means for obtaining is further configured to obtain a message indicating that the apparatus is a cluster member of a third FL cluster based on the measurement being an outlier of a statistic associated with a second measurement of the first network node.

[0330] In one configuration, the means for providing is configured to provide the first ML model information update and the measurement to the FL parameter network entity; and wherein the aggregation is performed together with a second ML model information update of a second network node in a second FL cluster.

[0331] In one configuration, the component for obtaining is further configured to: obtain a report including a third ML model information update from a first network node in the first FL cluster; wherein the first ML model information update includes a second aggregation of the third ML model information update and additional ML model information updates from additional network nodes in the first FL cluster.

[0332] In one configuration, the means for providing is further configured to provide the second aggregation to the first network node in the first FL cluster.

[0333] In one configuration, the report includes a second measurement associated with the third ML model information update, and the measurement includes statistics associated with the second measurement.

[0334] In one configuration, the means for providing is further configured to provide the FL parameters network entity with an identifier o...

Claims

1. A device for wireless communication, the device comprising: processor; and a memory coupled to the processor, wherein the processor is configured to: providing a first machine learning (ML) model information update and a measurement associated with the first ML model information update; Obtaining an aggregated ML model information update, wherein the aggregated ML model information update comprises an aggregation of the first ML model information update and a second ML model information update of one of: a first network node in a first federated learning (FL) cluster comprising the apparatus, or a second network node in a second FL cluster; as well as A sidelink communication parameter in sidelink control information (SCI) is provided to a network node, wherein the sidelink communication parameter is an output of the ML model associated with the aggregated ML model information update.

2. The apparatus of claim 1 , wherein the processor is further configured to provide the first ML model information update and the measurement to a FL cluster leader of the first FL cluster, and The aggregation is performed together with updating the second ML model information of the first network node in the first FL cluster.

3. The apparatus of claim 2, wherein the measurement comprises at least one of: a channel busy rate (CBR), a number of received packets at the apparatus, or a number of resource reservations of the first network node.

4. The apparatus of claim 3, wherein the measurements further comprise statistics associated with the CBR within a configured time window.

5. The device according to claim 2, wherein the processor is configured to provide the first ML model information update and the measurement with a periodicity configured by one of the FL cluster leader or a FL parameter network entity; wherein the processor is configured to provide the first ML model information update and the measurement semi-persistently within a configured time window; or Wherein the processor is configured to aperiodically provide the first ML model information update and the measurement in response to a request from the FL cluster leader.

6. The apparatus of claim 2, wherein the processor is further configured to: obtaining a radio resource control (RRC) configuration from a FL parameters network entity, the RRC configuration indicating resources for reporting the first ML model information update and the measurement; Wherein the processor is configured to provide the first ML model information update and the measurement in the resource.

7. The apparatus of claim 2, wherein the processor is further configured to: obtaining, from the FL cluster leader, a message triggering the sending of the first ML model information update and the measured report; Wherein the processor is configured to provide the first ML model information update and the measurement in a reserved side link resource in response to the message.

8. The apparatus of claim 2, wherein the processor is configured to provide the first ML model information update and the measurement in a dedicated resource for reporting the first ML model information update and the measurement.

9. The apparatus of claim 2, wherein the aggregated ML model information updates from the FL cluster leader also include additional ML model information updates for the second network nodes in the second FL cluster.

10. The apparatus of claim 2, wherein the processor is further configured to: A message indicating that the apparatus is a cluster member of a third FL cluster is obtained based on the measurement being an outlier of a statistic associated with a second measurement of the first network node.

11. The device according to claim 1, wherein the processor is configured to provide the first ML model information update and the measurement to a FL parameter network entity; and The aggregation is performed together with updating the second ML model information of the second network node in the second FL cluster.

12. The apparatus of claim 11, wherein the processor is further configured to: obtaining, from the first network node in the first FL cluster, a report including a third ML model information update; The first ML model information update includes a second aggregation of the third ML model information update and additional ML model information updates from additional network nodes in the first FL cluster.

13. The apparatus of claim 12, wherein the processor is further configured to: The second aggregation is provided to the first network node in the first FL cluster.

14. The apparatus of claim 12, wherein the report comprises a second measurement associated with the third ML model information update, and the measurement comprises statistics associated with the second measurement.

15. The apparatus of claim 14, wherein the processor is further configured to: An identifier of the first network node and the second measurement are provided to the FL parameters network entity, wherein the second measurement is an outlier of the statistic.

16. The apparatus of claim 11, wherein the processor is further configured to: The aggregated ML model information update is provided to the first network node in the first FL cluster.

17. The apparatus of claim 1, wherein the processor is further configured to: providing a first message including first FL information of the device to the first network node; and obtaining a second message including second FL information of the first network node; Based on the first FL information and the second FL information, the first network node is in the first FL cluster.

18. The device according to claim 17, The first FL information includes at least one of the following: a machine learning task for the device; information associated with available sensors coupled to the device for the machine learning task; An available ML model associated with the machine learning task; or the available computing resources of the device for the machine learning task; and The machine learning task includes at least one of the following: Reinforcement learning tasks; Object detection tasks; or Classification task.

19. The apparatus of claim 1, wherein the processor is further configured to: obtaining a message indicating a FL cluster leader of the first FL cluster including the device; providing a second message including FL information; and providing a third message indicating the second network node; Wherein the message indicates one of the second network node or a third network node as the FL cluster leader of the first FL cluster based on the FL information.

20. The apparatus of claim 1, wherein the processor is further configured to: providing a message indicating that the device is a FL cluster leader of the first FL cluster; and obtaining a second message including FL information of the first network node; The message is in response to the FL information.

21. The device according to claim 1, wherein the first network node and the second network node are the same network node or different network nodes; and The first FL cluster and the second FL cluster are the same FL cluster or different FL clusters.

22. The apparatus of claim 1, wherein the first ML model information update comprises at least one of the following: Weight; gradient; or A scaling factor, which is either a multiplicand or an augend.

23. The apparatus according to claim 1, further comprising: a second processor configured to send the sidelink communications parameters to the network node using one or more antennas; wherein the first network node is a first user equipment (UE); wherein the second network node is a second UE or a road side unit (RSU); and in: The device is a third UE, and the network node is a second RSU; The device is the second RSU, and the network node is the third UE; The apparatus and the network node are UE; or The device and the network node are RSUs.

24. An apparatus for wireless communication, the apparatus comprising: processor; and a memory coupled to the processor, wherein the processor is configured to: obtaining a first machine learning (ML) model information update and a measurement associated with the first ML model information update; and Providing an aggregated ML model information update, wherein the aggregated ML model information update comprises an aggregation of the first ML model information update and a second ML model information update of one of: a first network node in a first federated learning (FL) cluster comprising the apparatus, or a second network node in a second FL cluster; And wherein the sidelink communication parameters in the sidelink control information (SCI) of the network node are outputs of the ML model associated with the aggregated ML model information update.

25. The device according to claim 24, wherein the processor is further configured to obtain the first ML model information update and the measurement from a third network node in the first FL cluster; and The aggregation is performed together with updating the second ML model information of the first network node in the first FL cluster.

26. The apparatus of claim 25, wherein the processor is further configured to provide the aggregated ML model information update to a FL parameter network entity.

27. The apparatus of claim 25, wherein the processor is further configured to: Statistics associated with the measurements are provided to the FL parameters network entity.

28. The apparatus of claim 27, wherein the processor is further configured to: An identifier of the first network node and an individual measurement of the first network node are provided to the FL parameters network entity, wherein the individual measurement of the first network node is an outlier of the statistic.

29. The apparatus of claim 25, wherein the processor is further configured to: obtaining, from a FL parameter network entity, a second aggregation of the first ML model information update and additional ML model information updates for the second network nodes in the second FL cluster; and The second aggregation is provided to the first network node in the first FL cluster.

30. The apparatus of claim 24, wherein the processor is further configured to obtain the first ML model information update and the measurement from a FL cluster leader of the first FL cluster; and The aggregation is performed together with updating the second ML model information of the second network node in the second FL cluster.

31. The apparatus of claim 30, wherein the first ML model information update comprises an aggregation of additional ML model information updates for the first network nodes in the first FL cluster.

32. The apparatus of claim 30, wherein the processor is further configured to: obtaining, from the FL cluster leader, an identifier of the first network node in the first FL cluster and an individual measurement of the first network node, wherein the individual measurement of the first network node is an outlier of the measurement from the FL cluster leader; and A message is provided indicating that the first network node is a cluster member of a third FL cluster based on the individual measurement being the outlier.

33. The apparatus of claim 24, further comprising: a second processor configured to send the sidelink communications parameters to the network node using one or more antennas; wherein the first network node is a first user equipment (UE); wherein the second network node is a second UE or a road side unit (RSU); in: The device is a second RSU, and the network node is a third UE; The device and the network node are RSUs; The apparatus is a FL parameter network entity, and the network node is the second RSU; or The apparatus is the FL parameters network entity, and the network node is the third UE; and The FL parameter network entity is a base station or a server.

34. A method for wireless communication at a network entity, the method comprising: providing a first machine learning (ML) model information update and a measurement associated with the first ML model information update; Obtaining an aggregated ML model information update, wherein the aggregated ML model information update comprises an aggregation of the first ML model information update and a second ML model information update of one of: a first network node in a first federated learning (FL) cluster comprising the network entity, or a second network node in a second FL cluster; as well as A sidelink communication parameter in sidelink control information (SCI) is provided to a network node, wherein the sidelink communication parameter is an output of the ML model associated with the aggregated ML model information update.

35. A method for wireless communication at a network entity, the method comprising: obtaining a first machine learning (ML) model information update and a measurement associated with the first ML model information update; Providing an aggregated ML model information update, wherein the aggregated ML model information update comprises an aggregation of the first ML model information update and a second ML model information update of one of: a first network node in a first federated learning (FL) cluster comprising the network entity, or a second network node in a second FL cluster; and The sidelink communication parameters in the sidelink control information (SCI) of the network node are outputs of the ML model associated with the aggregated ML model information update.

Images