Systems and methods for federated learning

Abstract

At least one aspect relates to a system, method, apparatus, or computer-readable medium. In some embodiments, a wireless communication node may send a first message to a wireless communication device including a configuration for performing federated learning (FL). In some embodiments, the wireless communication device estimates a time interval for performing FL. In some embodiments, a wireless communication device in the RRC_IDLE state may start FL before t1-t2, where t1 represents the local model update deadline and t2 represents the estimated time interval for performing FL.

Description

Technical Field

[0001] This disclosure generally relates to wireless communications, including but not limited to systems and methods for federated learning. Background Technology

[0002] The standards organization Third Generation Partnership Project (3GPP) is currently developing a new radio interface known as 5G New Radio (5G NR) and a Next Generation Packet Core Network (NG-CN or NGC). 5G NR will have three main components: the 5G Access Network (5G-AN), the 5G Core Network (5GC), and User Equipment (UE). To facilitate the implementation of different data services and needs, the elements of the 5GC (also known as network functions) have been simplified, with some based on software and others on hardware, allowing for customization as needed. Summary of the Invention

[0003] The exemplary embodiments disclosed herein are intended to address problems related to one or more issues raised in the prior art, and to provide additional features that will become apparent when taken in conjunction with the accompanying drawings and the following detailed description. Exemplary systems, methods, apparatuses, and computer program products are disclosed herein according to various embodiments. However, it should be understood that these embodiments are given by way of example and not by way of limitation, and that various modifications can be made to the disclosed embodiments by those skilled in the art who read this disclosure, while still remaining within the scope of this disclosure.

[0004] At least one aspect relates to a system, method, apparatus, or computer-readable medium. In some embodiments, a wireless communication node may send a first message to a wireless communication device including a configuration for performing Federated Learning (FL). In some embodiments, the wireless communication device estimates a time interval for performing FL. In some embodiments, a wireless communication device in an RRC_IDLE state may start FL before t1-t2, where t1 represents the local model update deadline and t2 represents the estimated time interval for performing FL. The configuration includes at least one of the following: FL task ID; FL task index; FL task; FL model; FL validity period; AI / ML / FL region; connected_allowed indicator; idle allowed indicator; inactive allowed indicator; priority; preemptible indicator; power threshold; CPU threshold; expected completion time or deadline for local model update; or periodicity.

[0005] In some embodiments, FL validity time indicates the duration of the configured FL task. In some embodiments, FL model indicates the address of the FL model. In some embodiments, the AI / ML / FL area includes at least one of the following fields: one or more Tracking Area Codes (TACs); a list of cell IDs; or an AI / ML / FL area code. In some embodiments, connected_allowed indicates that the wireless communication device is allowed to execute FL in the RRC_CONNECTED state. In some embodiments, idle allowed indicates that the wireless communication device is allowed to execute FL in the RRC_IDLE state, and inactive allowed indicates that the wireless communication device is allowed to execute FL in the RRC_INACTIVE state. In some embodiments, power threshold indicates a threshold of the remaining power percentage of the wireless communication device or a threshold of the CPU usage percentage of the wireless communication device. In some embodiments, priority indicates the priority level of the associated FL task. Preemptibility indicates whether the associated FL task can be preempted when a higher-priority FL task is scheduled / executed.

[0006] In some embodiments, the wireless communication device may receive a first message from the wireless communication node. In some embodiments, the wireless communication node may operatively act as a centralized node in the FL network. In some embodiments, the first message may be sent via system information or via broadcast / multicast signaling. In some embodiments, the wireless communication node may send an RRC reconfiguration or RRC release message including the configuration to the wireless communication device. In some embodiments, the wireless communication device in the RRC_IDLE or RRC_INACTIVE state receives the FL model directly from the system information.

[0007] In some embodiments, the wireless communication device may send a second message requesting an FL model to the wireless communication node, wherein the second message includes at least one of the following fields: FL task ID; FL task index; FL task; FL interest indication; or model request indication, and receive an FL model from the wireless communication node. In some embodiments, the FL model indicates the address of the FL model. If the FL model is not yet available, the wireless communication device receives the FL model via system information when the FL model becomes available; otherwise, the wireless communication device needs to enter the RRC_Connected state to retrieve the latest FL model from the wireless communication node or via a URL link.

[0008] In some embodiments, the wireless communication device is configured to transition from an RRC_IDLE or RRC_INACTIVE state to an RRC_Connected state in order to receive FL models from a wireless communication node or URL link. In some embodiments, this configuration includes at least one of the following: FL task ID; FL task index; FL task; FL model; FL validity period; AI / ML / FL region; connected_allowed indicator; idle allowed indicator; inactive allowed indicator; priority; preemptible indicator; power threshold; or CPU threshold.

[0009] In some embodiments, the wireless communication device may send an RRC setup message to the wireless communication node. In the RRCSetup or RRC message, the wireless communication device sets the setup reason value to mo-Data or ai-Data. In some embodiments, the wireless communication device may send a local model and / or performance data used to train the local model to the wireless communication node. In some embodiments, the performance data includes at least one of the following fields: start time; end time; RRC status; elapsed training time; consumed computing resources; consumed storage resources; or consumed power. In some embodiments, the wireless communication device immediately transitions from the RRC_IDLE or RRC_INACTIVE state to the RRC_Connected state after completing FL (Fluid Filtering). In some embodiments, the wireless communication device in the RRC_IDLE or RRC_INACTIVE state waits until the deadline for updating the local model is approaching before transitioning from the RRC_IDLE or RRC_INACTIVE state to the RRC_Connected state. In some embodiments, the wireless communication device may send an RRC setup message to the wireless communication node. In the RRCSetupRequest or RRCResumeRequest message, the wireless communication device sets the setup reason value to mo-Data or ai-Data.

[0010] In some embodiments, a wireless communication device may receive a paging message from a wireless communication node indicating an AI / ML / FL model update. In some embodiments, the paging message includes at least one of the following fields: AI / ML / FL model ID; AI / ML / FL model function; or AI / ML / FL model update indication. Upon receiving the paging message, the device may also retrieve the updated FL model from the wireless communication node. If the wireless communication device is in the RRC_IDLE or RRC_INACTIVE state, it needs to enter the RRC_Connected state before retrieving the updated FL model. In some embodiments, the wireless communication device sends an FL model request to the wireless communication node. In some embodiments, the FL model request includes at least one of the following fields: FL task ID; FL task index; FL task; FL interest indication; or model request indication.

[0011] At least one aspect relates to a system, method, apparatus, or computer-readable medium that includes the following: A wireless communication node can send to a neighboring wireless communication node a FL task, AI / ML / FL model, AI / ML / FL function, and / or AI / ML / FL model ID supported by the wireless communication node. In some embodiments, the wireless communication node can send a new global AI / ML / FL model and / or paging indication to the neighboring wireless communication node. In some embodiments, the wireless communication node and the neighboring wireless communication node belong to the same AI / ML / FL region. Upon receiving a new global AI / ML / FL model and / or paging indication, the neighboring wireless communication node can initiate a paging procedure to send a paging message to the wireless communication device. In some embodiments, the paging message includes at least one of the following fields: AI / ML / FL model ID; AI / ML / FL function; or AI / ML / FL model update indication.

[0012] In some embodiments, a wireless communication device may receive an updated FL model from system information of neighboring wireless communication nodes. In some embodiments, a wireless communication device may transition from an RRC_IDLE state or an RRC_INACTIVE state to an RRC_Connected state to receive an updated FL model from neighboring wireless communication nodes. In some embodiments, a wireless communication device may determine whether it is within an FL region upon receiving FL information via system information. In some embodiments, a wireless communication device may determine that the configuration is valid for performing FL and reporting local model updates, and may remove the configuration if it is no longer valid. Attached Figure Description

[0013] Various exemplary embodiments of the present invention are described in detail below with reference to the accompanying drawings or illustrations. These drawings are provided for illustrative purposes only and depict only exemplary embodiments of the present invention to facilitate the reader's understanding of the present invention. Therefore, these drawings should not be considered as limitations on the breadth, scope, or applicability of the present invention. It should be noted that these drawings are not necessarily drawn to scale for clarity and ease of explanation.

[0014] Figure 1 An exemplary cellular communication network in which the technology 104 disclosed herein can be implemented, according to embodiments of the present disclosure, is shown; Figure 2 Block diagrams of exemplary base stations and user equipment according to some embodiments of the present disclosure are shown; Figures 3A-3C An exemplary base station coordinated DL / FL scenario according to embodiments of the present disclosure is illustrated; Figure 4 The process of sending FL configuration to UE1 via dedicated signaling according to an embodiment of the present disclosure is illustrated; Figure 5 An exemplary process for federated training of a UE according to embodiments of this disclosure is shown; Figure 6 An exemplary procedure is shown for a potential signaling procedure of a FL for a UE in the RRC_IDLE state according to an embodiment of this disclosure; Figure 7 An exemplary process is shown in which a UE, according to an embodiment of the present disclosure, enters the RRC_IDLE state with pending FL tasks for federated learning; Figure 8 An exemplary procedure for joining an RRC_INACTIVE UE to an FL according to an embodiment of this disclosure is shown; Figure 9 An exemplary procedure for a signaling process for paging a UE to perform an FL model update, according to an embodiment of the present disclosure, is shown; Figure 10 An exemplary process of interaction between one or more BSs for paging AI / ML model updates is illustrated according to embodiments of this disclosure; Figure 11 An exemplary process for FL region inspection according to embodiments of the present disclosure is shown; Figure 12 An exemplary protocol stack 1200 of an AI sublayer located above the PDCP layer according to an embodiment of the present disclosure is shown. Detailed Implementation

[0015] A. Mobile communication technology and environment Figure 1 An exemplary wireless communication network and / or system 100 according to embodiments of this disclosure, in which the techniques disclosed herein may be implemented, is illustrated. In the following discussion, the wireless communication network 100 may be any wireless network, such as a cellular network or a narrowband Internet of Things (NB-IoT) network, referred to herein as "network 100". Such an exemplary network 100 includes a base station 102 (hereinafter referred to as "BS 102"; also called a wireless communication node) and a user equipment 104 (hereinafter referred to as "UE 104"; also called a wireless communication device), which may communicate with each other via a communication link 110 (e.g., a wireless communication channel). Network 100 also includes cell clusters 126, 130, 132, 134, 136, 138, and 140 covering a geographic area 101. Figure 1In this context, BS 102 and UE 104 are contained within the corresponding geographical boundaries of cell 126. Each of the other cells 130, 132, 134, 136, 138, and 140 may include at least one base station that operates with its allocated bandwidth to provide sufficient wireless coverage to its target users.

[0016] For example, BS 102 can operate with the allocated channel transmission bandwidth to provide sufficient coverage to UE 104. BS 102 and UE 104 can communicate via downlink radio frame 118 and uplink radio frame 124, respectively. Each radio frame 118 / 124 can be further divided into subframes 120 / 127, and the subframes may include data symbols 122 / 128. In this disclosure, BS 102 and UE 104 are described herein as non-limiting examples of "communication nodes" in a general sense, which can implement the methods disclosed herein. According to various embodiments of this scheme, such communication nodes may be able to perform wireless and / or wired communication.

[0017] Figure 2 A block diagram of an exemplary wireless communication system 200 for transmitting and receiving wireless communication signals (e.g., OFDM / OFDMA signals) according to some embodiments of this solution is shown. System 200 may include components and elements configured to support known or conventional operating characteristics, which do not need to be described in detail herein. In one illustrative embodiment, system 200 may be used to communicate (e.g., transmit and receive) data symbols in a wireless communication environment, such as those described above. Figure 1 Wireless communication environment 100.

[0018] System 200 generally includes a base station 202 (hereinafter referred to as "BS 202") and a user equipment 204 (hereinafter referred to as "UE 204"). BS 202 includes a BS (base station) transceiver module 210, a BS antenna 212, a BS processor module 214, a BS memory module 216, and a network communication module 218. These modules are coupled and interconnected with each other via a data communication bus 220 as needed. UE 204 includes a UE (user equipment) transceiver module 230, a UE antenna 232, a UE memory module 234, and a UE processor module 236. These modules are coupled and interconnected with each other via a data communication bus 240 as needed. BS 202 communicates with UE 204 via a communication channel 250, which can be any wireless channel or other medium suitable for data transmission as described herein.

[0019] As those skilled in the art will understand, system 200 may also include, in addition to Figure 2Any number of modules other than those shown herein. Those skilled in the art will understand that the various illustrative blocks, modules, circuits, and processing logic described in connection with the embodiments disclosed herein can be implemented in hardware, computer-readable software, firmware, or any practical combination thereof. To clearly illustrate this interchangeability and compatibility of hardware, firmware, and software, various illustrative components, blocks, modules, circuits, and steps have been generally described above in terms of their functionality. Whether such functionality is implemented as hardware, firmware, or software can depend on the specific application and the design constraints imposed on the system as a whole. Those skilled in the art can implement such functionality appropriately for each specific application, but such implementation decisions should not be construed as limiting the scope of this disclosure.

[0020] According to some embodiments, UE transceiver 230 may be referred to herein as "uplink" transceiver 230, which includes a radio frequency (RF) transmitter and an RF receiver, each including circuitry coupled to antenna 232. A duplex switch (not shown) may alternatively couple the uplink transmitter or receiver to the uplink antenna in a time-duplex manner. Similarly, according to some embodiments, BS transceiver 210 may be referred to herein as "downlink" transceiver 210, which includes an RF transmitter and an RF receiver, each including circuitry coupled to antenna 212. A downlink duplex switch may alternatively couple the downlink transmitter or receiver to downlink antenna 212 in a time-duplex manner. The operation of the two transceiver modules 210 and 230 may be time-coordinated such that while the downlink transmitter is coupled to downlink antenna 212, the uplink receiver circuitry is coupled to uplink antenna 232 to receive transmissions via wireless transmission link 250. Conversely, the operation of the two transceivers 210 and 230 can be coordinated in time, such that while the uplink transmitter is coupled to the uplink antenna 232, the downlink receiver is coupled to the downlink antenna 212 to receive transmissions via the wireless transmission link 250. In some embodiments, there is tight time synchronization between duplex direction changes and minimal guard time.

[0021] UE transceiver 230 and base transceiver 210 are configured to communicate via wireless data communication link 250 and cooperate with appropriately configured RF antenna devices 212 / 232, which may support specific wireless communication protocols and modulation schemes. In some illustrative embodiments, UE transceiver 210 and base transceiver 210 are configured to support industry standards, such as Long Term Evolution (LTE) and emerging 5G standards. However, it should be understood that this disclosure is not necessarily limited to application to specific standards and associated protocols. Rather, UE transceiver 230 and base transceiver 210 may be configured to support alternative or additional wireless data communication protocols, including future standards or variations thereof.

[0022] According to various embodiments, BS 202 may be, for example, an evolved Node B (eNB), a serving eNB, a target eNB, a femtocell, or a picocell. In some embodiments, UE 204 may be embodied in various types of user equipment, such as mobile phones, smartphones, personal digital assistants (PDAs), tablets, laptops, wearable computing devices, etc. Processor modules 214 and 236 may be implemented or carried out using general-purpose processors, content-addressable memory, digital signal processors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), any suitable programmable logic devices, discrete gate or transistor logic, discrete hardware components, or any combination thereof, designed to perform the functions described herein. In this way, the processor may be implemented as a microprocessor, a controller, a microcontroller, a state machine, etc. The processor may also be implemented as a combination of computing devices, such as a combination of a digital signal processor and a microprocessor, multiple microprocessors, one or more microprocessors combined with a digital signal processor core, or any other such configuration.

[0023] Furthermore, the steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be directly embodied in hardware, firmware, software modules executed by processor modules 214 and 236 respectively, or any actual combination thereof. Memory modules 216 and 234 can be implemented as RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art. In this regard, memory modules 216 and 234 can be coupled to processor modules 210 and 230 respectively, such that processor modules 210 and 230 can read information from and write information to memory modules 216 and 234 respectively. Memory modules 216 and 234 can also be integrated into their respective processor modules 210 and 230. In some embodiments, memory modules 216 and 234 may each include a cache memory for storing temporary variables or other intermediate information during the execution of instructions executed by processor modules 210 and 230 respectively. Memory modules 216 and 234 may each include non-volatile memory for storing instructions to be executed by processor modules 210 and 230, respectively.

[0024] Network communication module 218 generally refers to the hardware, software, firmware, processing logic, and / or other components of base station 202 that enable bidirectional communication between base station transceiver 210 and other network components and communication nodes configured to communicate with base station 202. For example, network communication module 218 may be configured to support Internet or WiMAX services. In a typical deployment (but not limited to), network communication module 218 provides an 802.3 Ethernet interface, enabling base station transceiver 210 to communicate with conventional Ethernet-based computer networks. In this way, network communication module 218 may include a physical interface for connecting to a computer network (e.g., a Mobile Switching Center (MSC)). The terms “configured for,” “configured to,” and variations thereof, used herein with respect to a specified operation or function, refer to devices, components, circuits, structures, machines, signals, etc., that are physically constructed, programmed, formatted, and / or arranged to perform that specified operation or function.

[0025] The Open Systems Interconnection (OSI) model (referred to herein as the "OSI model") is a conceptual and logical layout that defines network communications used by systems (e.g., wireless communication devices, wireless communication nodes) that are open to interconnection and communication with other systems. The model is divided into seven sub-components or layers, each representing a conceptual set of services provided to its upper and lower layers. The OSI model also defines logical networks and effectively describes computer packet transmissions using different layer protocols. The OSI model may also be referred to as the seven-layer OSI model or the seven-layer model. In some embodiments, the first layer may be the physical layer. In some embodiments, the second layer may be the Medium Access Control (MAC) layer. In some embodiments, the third layer may be the Radio Link Control (RLC) layer. In some embodiments, the fourth layer may be the Packet Data Convergence Protocol (PDCP) layer. In some embodiments, the fifth layer may be the Radio Resource Control (RRC) layer. In some embodiments, the sixth layer may be the Non-Access Stratum (NAS) layer or the Internet Protocol (IP) layer, while the seventh layer may be other layers.

[0026] Various exemplary embodiments of this solution are described below with reference to the accompanying drawings to enable those skilled in the art to make and use this solution. As will be apparent to those skilled in the art upon reading this disclosure, various changes or modifications can be made to the examples described herein without departing from the scope of this solution. Therefore, this solution is not limited to the exemplary embodiments and applications described and illustrated herein. Furthermore, the specific order or hierarchy of steps in the methods disclosed herein is merely exemplary. Based on design preferences, the specific order or hierarchy of steps in the disclosed methods or processes may be rearranged while still remaining within the scope of this solution. Therefore, those skilled in the art will understand that the methods and techniques disclosed herein present various steps or actions in an exemplary order, and unless otherwise expressly stated, this solution is not limited to the specific order or hierarchy presented.

[0027] B. Systems and methods for federated learning This disclosure proposes a federated learning design for UEs in RRC_IDLE or RRC_INACTIVE states in wireless networks, aiming to support federated or distributed learning in different RRC states. Based on this invention, not only RRC_Connected UEs but also RRC_IDLE / INACTIVE UEs can participate in federated / distributed learning, which expands the applicable scenarios of federated / distributed learning. Furthermore, this disclosure also proposes a user plane design for AI / ML model management in wireless networks.

[0028] In a rapidly changing wireless environment, AI / ML models running on devices or networks can continuously adapt to new environments to maintain desired performance. Advances in mobile edge computing and caching technologies enable BS 102 to store and analyze the behavior of UE 104 in wireless networks.

[0029] Generally, UE 104 can send raw data to a centralized network entity, which can then train an ML model based on that raw data. Training a Deep Neural Network (DNN) model can take anywhere from hours to days. After training, the centralized network entity can use the DNN model for inference, or it can send the model back to the UE for inference. However, training based on a centralized network entity means that a large amount of training data must be transmitted from UE 104 to the centralized network entity, resulting in unacceptable communication overhead and data privacy pressure on the network side.

[0030] On the other hand, distributed learning or federated learning (FL) can be introduced into wireless networks. For distributed / federated learning, each compute node can train its own local model using local data. Network nodes can communicate with each other to exchange local model updates and build a global model. By integrating DL / FL into the wireless network, local data can be stored on UE 104 and trained locally by UE 104. Local model updates can then be aggregated and updated by a centralized network entity. This approach addresses the pressing need for data privacy on the UE while effectively coordinating a large number of UE 104 nodes.

[0031] Regarding wireless networks, centralized network entities can take the following forms: application servers, Operation Administration and Maintenance (OAM) servers, network functions for AI in the core network, BS 102 (e.g., Centralized Unit (CU) and / or Distributed Unit (DU)), AI controllers, Mobile Edge Computing (MEC), UE devices, etc. Centralized network entities for DL / FL can reside at each possible network entity. In this way, the centralized network entity is responsible for selecting participating UEs, configuring them, delivering global models to participating UEs, collecting local model updates, and performing model aggregation for forming global models / inference.

[0032] In some cases, if BS 102 is used as a centralized network entity for AI, then the DL / FL scenario coordinated by the base station is as follows: Figures 3A-3C As shown. Figure 3A As shown, this involves a BS 102, which coordinates the DL / FL execution by UE 104 served by that BS 102. As... Figure 3B and Figure 3C As shown, multiple BS 102s participate in the federated learning / inference process. In this way, one BS 102 can coordinate with other BS 102s to jointly perform federated learning / inference.

[0033] Regarding UE 104, if UE 104 is interested in DL / FL, both RRC_Connected UEs and RRC_IDLE / INACTIVE UEs can join DL / FL. A given DL / FL region can be configured for UE 104. If UE 104 moves out of this DL / FL region, UE 104 can discard its previous DL / FL configuration. Otherwise, UE 104 can continue to perform DL / FL regardless of its state. After completing one round of training, RRC_IDLE / INACTIVE UE 104 can enter the RRC_Connected state and report local model updates to BS 102.

[0034] In this disclosure, federated learning (FL) can be distributed learning (DL). Furthermore, FL configuration can be viewed as DL configuration. FL configuration can originate from other network entities, such as application servers, OAM servers, network functions used for AI in the core network, BS 102 (e.g., CU and / or DU), AI controllers, MEC, UE 104 devices, etc.

[0035] Dedicated signaling Figure 4 The procedure 400, in which FL configuration is sent to UE1 via dedicated signaling, is illustrated. At 402, the centralized network entity sends the FL configuration to UE1 via dedicated signaling. The FL configuration may include at least one of the following fields: FL task ID, FL task index, FL task, FL model, FL validity period, AI / ML / FL region, connected_allowed, idle allowed, inactive allowed, priority, preemptibility indication, power threshold, CPU threshold, expected completion time or deadline for local model updates, period, etc.

[0036] The validity period indicates the duration of a configured FL task. In some embodiments, if a validity period is configured for an FL task, UE1 can start an FL validity timer. If the validity timer expires, UE1 can discard the corresponding FL configuration. While the validity timer is running, UE1 can perform federated learning, even if it enters the RRC_IDLE or RRC_INACTIVE state. Alternatively, an FL validity timer can be configured for each FL task. Therefore, each FL task can be associated with a different validity period.

[0037] The AI / ML / FL region may include at least one of the following fields: one or more TACs, a list of cell IDs, or an AI / ML / FL region code. This AI / ML / FL region indicates the valid region for the configured FL task or AI / ML / FL model.

[0038] The "connected allowed" indication means that UE1 is allowed to perform federated learning in the RRC_CONNECTED state. The "idle allowed" indication means that UE1 is allowed to perform federated learning in the RRC_IDLE state. The "inactive allowed" indication means that UE1 is allowed to perform federated learning in the RRC_INACTIVE state.

[0039] The battery threshold indicates a threshold for the remaining battery percentage of UE 104. For example, if BS1 configures the battery threshold to 50%, federated training can be performed when the remaining battery level of UE 104 is greater than 50%. Furthermore, the battery threshold can be set to infinity. Therefore, federated training can be performed when UE 104 is plugged in. On the other hand, the CPU threshold indicates a threshold for the CPU utilization percentage. For example, if BS1 configures the CPU utilization percentage to 30%, federated training can be performed when the CPU utilization percentage is below 30%. Furthermore, the CPU threshold can be a CPU utilization level. The CPU utilization level can be high, medium, or low. For example, a CPU utilization percentage of 100% to 70% can be considered high, 70% to 40% can be considered medium, and below 40% can be considered low. If the FL configuration indicates a low or medium CPU threshold, federated training can be performed when the UE's CPU utilization level is low or medium.

[0040] Priority indicates the priority level of the associated FL tasks. In this case, if multiple FL tasks are configured for UE 104, UE 104 can prioritize FL tasks with higher priority. For example, when the UE's CPU utilization percentage is low, UE 104 can prioritize executing FL task 1. Later, the UE's CPU utilization percentage increases and approaches the CPU threshold. In this case, FL task 2, which has a lower priority, can not be executed until FL task 1 is completed.

[0041] FL configuration via system information A centralized network entity (e.g., BS1) can send FL configuration to UE 104 via system information. For example, BS1 can send at least one of the following fields via system information: FL task ID, FL task index, FL task, FL model, FL validity period, FL region, connected_allowed, idle allowed, inactive allowed, priority, preemptibility indication, power threshold, CPU threshold, deadline for local model update submission, or period.

[0042] For UE 104 with FL capability, UE 104 can receive FL task information via system information. If UE 104 is interested in one or more FL tasks, UE 104 can check its eligibility to participate in a particular FL task based on information such as FL region, connected_allowed, idleallowed, inactive allowed, priority, preemptibility indication, power threshold, CPU threshold, or the deadline for local update submission. Detailed checks of each field can follow the description in the previous examples.

[0043] In some cases, if a FL-capable UE 104 in the RRC_IDLE state receives an idle_allowed instruction and an infinite power threshold from BS1 for a given FL task, UE 104 checks that the FL task allows the RRC_IDLE state, and since UE 104 is connected to power, the infinite power threshold can be met. In this case, UE 104 can check if the FL model is available. If UE 104 retrieves the FL model, it can begin federated training based on local data. However, if the FL model is not yet available, UE 104 can receive the FL model via system information, where applicable. Otherwise, UE 104 may need to enter the RRC_Connected state and retrieve the latest FL model from BS 102 or via a URL link. Based on this FL model, UE 104 can perform the FL task based on local data. After completing the federated training for this iteration, UE 104 can report the local model update to the BS.

[0044] FL configuration via MBS In addition to dedicated signaling and system information, broadcast / multicast based on Multicast Broadcast Service (MBS) can also be considered for FL configuration and FL model delivery. Specifically, a Group Radio Network Temporary Identifier (G-RNTI) can be assigned to the FL task, and UE 104 interested in the FL task can monitor the G-RNTI and receive the corresponding configuration.

[0045] Example 1 When the inactivity timer expires, UE 104 can enter the RRC IDLE / INACTIVE state. Figure 5 An exemplary procedure 500 for federated training for UE 104 is shown. For UE 104 previously configured with an FL task, UE 104 can continue the ongoing federated training based on local data at 502 and enter the RRC_IDLE state at 504. Once one iteration of federated training is completed at 506, UE 104 can enter the RRC_Connected state at 508 to send local model updates to BS 102. Alternatively, as Figure 5 As shown, UE 104 can enter the RRC_Connected state just before the deadline for submitting local model updates.

[0046] Figure 6An exemplary procedure 600 is shown for a potential signaling procedure for a FL (Federated Learning) for a UE 104 in the RRC_IDLE state. At 602, UE 1 may send AI / ML-related capability and / or UE 104 preference information to BS1. Based on the UE 104 capability report or UE 104 preference information, BS1 may select a given number of UE 104s at 604 to join federated learning for a given FL task. Assuming UE 1 is one of the selected UEs, BS1 sends FL configuration request information to UE 1 at 606. The FL configuration request information may include at least one of the following fields: FL task ID, FL task index, FL task, FL model, FL effective time, FL region, connected_allowed, idle allowed, inactive allowed, priority, preemptibility indication, power threshold, CPU threshold, and expected completion time or deadline for model update.

[0047] Upon receiving an FL configuration request, UE1 sends an FL configuration response or FL configuration confirmation message to BS1 at point 608. After BS1 completes negotiation with UE104 and receives acceptances from the required number of UE104 devices, BS1 sends the model for federated learning to UE1 at point 610. UE1 can then begin model training based on local data. During the federated training process, UE1 can enter the RRC_IDLE state at point 612. In this case, the RRC release message may include FL configuration for either the RRC_IDLE or RRC_INACTIVE state. This FL configuration may include at least one of the following fields: FL task ID, FL task index, FL task, FL model, FL effective time, FL region, connected_allowed, idle allowed, inactiveallowed, priority, preemptible indication, power threshold, and CPU threshold.

[0048] After UE 104 enters the RRC_IDLE state, it can continue federated training at point 614. After UE 104 completes federated training, it can re-enter the RRC_Connected state at point 616. During the RRC connection establishment process, UE 104 can set the establishment reason value to mo-Data or ai-Data in the RRCSetupRequest or RRCResumeRequest request message. After UE 104 establishes an RRC connection with BS1, UE1 can send a local model update to BS1 at point 618. Furthermore, UE1 can send performance data from this local model training to BS1. The performance data may include at least one of the following fields: start time, end time, RRC state, elapsed time of local training, consumed computing resources, consumed storage resources, or consumed power.

[0049] After receiving local model updates from a significant number of UEs 104, BS1 performs FL aggregation based on these local model updates at 620. At 622, BS1 sends the aggregated model to the participating UEs 104 for the next round of federated training. Once a predefined accuracy is reached, BS1 can send the model for inference to allow UEs 104 to participate. It should be noted that UEs 104 can enter the RRC_Connected state immediately after completing federated training. Alternatively, UEs 104 can wait until the local model update deadline is approaching. UEs 104 then enter the RRC_Connected state and send the local model update to BS1 just before the local model update deadline.

[0050] Example 2 Figure 7An exemplary procedure 700 is shown for a UE to enter the RRC_IDLE state with pending FL tasks. When UE1 is in the RRC_Connected state, UE1 receives the FL configuration and the global model for FL from the centralized network entity at 702. The FL configuration received by UE1 indicates idle_allowed. Subsequently, due to the inactivity timer timeout, the UE enters the RRC_IDLE state at 704. Although UE1 has entered the RRC_IDLE state, UE1 can still continue federated training according to the FL configuration at 706. UE1 can estimate the duration of the FL training task. Based on the local model update deadline (t1) of one iteration and the estimated time interval (t2) for executing the FL training task, the RRC_IDLE UE1 can start federated learning at least before time (t1-t2). After UE1 completes federated training, UE1 can re-enter the RRC_Connected state at 708. During the RRC connection establishment process, UE 104 can set the establishment reason value to mo-Data or ai-Data in the RRCSetupRequest or RRCResumeRequest request message. After UE1 establishes the RRC connection, UE1 can send local model updates to the centralized network entity at 710. Furthermore, UE1 can send performance data of local model training to the centralized network entity. Performance data may include at least one of the following fields: start time, end time, elapsed time of local training, consumed computing resources, consumed storage resources, consumed power, etc.

[0051] After receiving local model updates from a significant number of UEs 104, the centralized network entity performs FL aggregation based on these local model updates. The centralized network entity then sends the aggregated model to the participating UEs 104 for the next round of federated training. Once a predefined accuracy is reached, the centralized network entity can send the model for inference to the involved UEs 104.

[0052] Figure 8An exemplary procedure 800 for an RRC_INACTIVE UE to join an FL is shown. On the other hand, for the RRC_INACTIVE UE at 802, it can receive FL configuration broadcast by the serving cell at 804. For example, the serving cell can send at least one of the following fields via system information: FL task ID, FL task index, FL task, FL model, FL validity period, FL region, connected_allowed, idle allowed, inactive allowed, priority, preemptibility indication, power threshold, CPU threshold, deadline for local model update submission, period, etc. If the RRC_INACTIVE UE 104 is interested in one or more FL tasks, UE 104 can check its eligibility to participate in the FL task based on the FL configuration information (e.g., inactive allowed, priority, preemptibility indication, power threshold, CPU threshold, or deadline for local model update submission).

[0053] In some cases, if a UE 104 with FL capability in the RRC_INACTIVE state determines that it is eligible to participate in the FL task, UE 104 can check if the FL model is available. If UE 104 has not yet received the FL model, UE 104 checks the FL model information in the FL configuration. If the FL model is broadcast via system information, the RRC_INACTIVE UE 104 can receive the FL model directly from the system information. On the other hand, if the FL model indicates that the latest FL model is in BS 102 or a URL link is provided, the RRC_INACTIVE UE 104 needs to enter the RRC_Connected state at 806 to retrieve the latest FL model from BS 102 or via the URL link at 808. Specifically, assuming the FL model indicates that the model is in the BS, UE 104 can send an FL model request to the BS. The FL model request may include at least one of the following fields: FL task ID, FL task index, FL task, FL interest indication, and model request indication. Based on the FL model request, the BS sends the corresponding model to the UE.

[0054] After receiving the FL model, UE 104 can perform the FL task based on local data at 810. During the federated training process, UE 104 can re-enter the RRC_INACTIVE state at 812. However, federated training is continuously performed by the RRC_INACTIVE UE.

[0055] After completing the federated training for this iteration, UE 104 needs to report the local model update to the centralized network entity. In this case, UE 104 can re-enter the RRC_Connected state at 814 so that it can report the local model update to the centralized network entity at 816, just before the local model update submission deadline.

[0056] Example 3 Figure 9 An exemplary procedure 900 for a signaling process used to page a UE for an FL model update is shown. At 902, BS1 sends an FL configuration to UE1 in the RRC_Connected state via dedicated signaling. The FL configuration may include at least one of the following fields: FL task ID, FL task index, FL task, FL model, FL validity period, FL region, connected_allowed, idle allowed, inactive allowed, priority, preemptibility indication, power threshold, CPU threshold, deadline for local model update submission, or period.

[0057] If the FL configuration received by UE1 includes a validity period, UE104 can start the FL validity timer. Subsequently, due to the inactivity timer expiring, UE104 enters the RRC_INACTIVE state at 904. Although UE1 has entered the RRC_IDLE state, since the FL validity timer has not yet expired, UE1 can still continue federated training according to the FL configuration. If BS1 performs FL aggregation based on these local model updates and derives an updated global FL model, BS1 needs to send the aggregated model to the participating UE104 for the next round of federated training. Considering that some participating UE104 has entered the INACTIVE state, BS1 can initiate a paging procedure. At 906, BS1 sends a paging message to UE104 indicating an AI / ML model update. Specifically, the paging message may include at least one of the following fields: AI / ML / FL model ID, AI / ML / FL model function, and AI / ML / FL model update indication.

[0058] When UE1 receives a paging message indicating an FL model update, and UE1 is one of the participating UEs 104 in the FL task, UE1 can enter the RRC_Connected state at 908 to retrieve the latest FL model from BS1. Specifically, if the FL model is in BS 102, UE104 can send an FL model request to BS 102. The FL model request may include at least one of the following fields: FL task ID, FL task index, FL task, FL interest indication, and model request indication. Based on the FL model request, BS sends the corresponding model to UE104 at 910. Upon receiving the updated global model, UE104 performs an FL training task based on local data at 912. After UE1 completes federated training, UE1 sends a local model update to BS1 at 914.

[0059] Example 4 Figure 10 An exemplary process 1000 for interaction between one or more BSs 102 for paging AI / ML model updates is illustrated. To support FLs within a given FL region, BS1 can send its supported FL tasks, AI / ML / FL models, AI / ML / FL functions, and AI / ML / FL model IDs to a neighboring BS (e.g., BS2) at 1002. Similarly, BS1 can send its supported FL tasks, AI / ML / FL models, AI / ML / FL functions, and AI / ML / FL model IDs to neighboring BSs. In this way, the BSs can obtain information about the supported AI / ML / FL regions and AI / ML / FL models.

[0060] BS1 initiates an FL task and selects several UEs 104 to join the federated training. For this FL task, UEs 104 with RRC_Connected, RRC_IDLE, and RRC_INACTIVE settings are allowed to join the federated training. If BS1 receives local model updates from a significant number of UEs 104, BS1 performs FL aggregation at 1004 based on these local model updates. BS1 sends the aggregated model to the participating UEs 104 for the next round of federated training. Alternatively, BS1 can send a paging message indicating an FL model update as shown in procedure 900. Furthermore, BS1 can send a new global AI / ML / FL model and / or paging indication to neighboring BSs belonging to the same AI / ML / FL region at 1006. Upon receiving such information, BS2 can also initiate a paging procedure. BS2 sends a paging message to UE1 at 1008. The paging message may include at least one of the following fields: AI / ML / FL model ID, AI / ML / FL model function, and AI / ML / FL model update indication.

[0061] When UE1, in the RRC_IDLE or RRC_INACTIVE state, receives a paging message from BS2 indicating an FL model update, UE1 can either receive the updated FL model from system information if available, or enter the RRC_Connected state and retrieve the latest FL model from BS2. Furthermore, a similar approach can be used for AI / ML model updates. For example, if the AI / ML model is updated, the BS can initiate a paging message indicating the AI / ML model update. Upon receiving such a paging message, the RRC_IDLE / INACTIVE UE104 can either receive the updated AI / ML model from system information if available, or enter the RRC_Connected state and retrieve the latest AI / ML model from the BS.

[0062] Example 5 Figure 11 An exemplary procedure 1100 for FL area checking is shown. At 1102, the RRC_IDLE UE moves from one cell, BS 102, or Transmission Reception Point (TRP) to another cell, BS 102, or TRP. Each cell, BS 102, or TRP can broadcast FL area information via system information at 1104. UE 104 can then... Figure 10 The procedure involves checking whether the broadcast FL region information matches its configured FL region. If at least one FL region matches, UE 104 determines that it is within the corresponding FL region. When UE 104 moves within the configured FL region, the FL configuration is valid, and UE 104 needs to perform federated training and report local model updates. If UE 104 moves out of the configured FL region at 1106, the FL configuration is no longer valid, and the associated FL configuration can be removed at 1108.

[0063] Figure 12 A flowchart of method 1200 for federated learning is shown. Method 1200 can be combined with the methods described in this paper. Figures 1-11 The method 1200 may be performed by any one or more of the detailed components and devices. Generally, in some embodiments, method 1200 may be performed by a wireless communication node (e.g., base station (BS) 102). According to embodiments, additional, fewer, or different operations may be performed in method 1200. At least one aspect of these operations relates to a system, method, apparatus, or computer-readable medium.

[0064] At step 1205, the wireless communication node may send a first message to the wireless communication device including a configuration for performing federated learning (FL). The wireless communication device estimates the time interval for performing FL. The wireless communication device in the RRC_IDLE state may start FL before t1-t2, where t1 represents the local model update deadline and t2 represents the estimated time interval for performing FL. The configuration includes at least one of the following: FL task ID; FL task index; FL task; FL model; FL validity period; AI / ML / FL region; connected_allowed indicator; idle allowed indicator; inactiveallowed indicator; priority; preemptible indicator; power threshold; CPU threshold; expected completion time or deadline for local model update; or period.

[0065] FL Valid Time indicates the duration of the configured FL task. FL Model indicates the address of the FL model. AI / ML / FL Region includes at least one of the following fields: one or more TACs; a list of cell IDs; or an AI / ML / FL region code. `connected_allowed` indicates that the wireless communication device is allowed to execute FL in the RRC_CONNECTED state. `idleallowed` indicates that the wireless communication device is allowed to execute FL in the RRC_IDLE state, and `inactiveallowed` indicates that the wireless communication device is allowed to execute FL in the RRC_INACTIVE state. Battery Threshold indicates a threshold for the remaining battery percentage of the wireless communication device or a threshold for the CPU usage percentage of the wireless communication device. Priority indicates the priority level of the associated FL task. Preemptibility indicates whether the associated FL task can be preempted when a higher-priority FL task is scheduled / executed.

[0066] At step 1210, the wireless communication device can receive a first message from the wireless communication node. The wireless communication node can operatively act as a centralized node in the FL network. The first message can be sent via system information or via broadcast / multicast signaling. The wireless communication node can send an RRC reconfiguration or RRC release message including the configuration to the wireless communication device. The wireless communication device in the RRC_IDLE or RRC_INACTIVE state receives the FL model directly from the system information.

[0067] The wireless communication device can send a second message requesting an FL model to the wireless communication node, wherein the second message includes at least one of the following fields: FL task ID; FL task index; FL task; FL interest indication; or model request indication, and receive an FL model from the wireless communication node. The FL model indicates the address of the FL model. If the FL model is not yet available, the wireless communication device receives the FL model via system information when the FL model becomes available; otherwise, the wireless communication device needs to enter the RRC_Connected state to retrieve the latest FL model from the wireless communication node or via a URL link.

[0068] The wireless communication device is configured to transition from the RRC_IDLE or RRC_INACTIVE state to the RRC_Connected state in order to receive FL models from a wireless communication node or URL link. This configuration includes at least one of the following: FL task ID; FL task index; FL task; FL model; FL validity period; AI / ML / FL region; connected_allowed indicator; idleallowed indicator; inactive allowed indicator; priority; preemptible indicator; power threshold; or CPU threshold.

[0069] The wireless communication device can send an RRC setup message to the wireless communication node. In the RRCSetupRequest or RRCResumeRequest message, the wireless communication device sets the setup reason value to mo-Data or ai-Data. The wireless communication device can send the local model and / or performance data used to train the local model to the wireless communication node. The performance data includes at least one of the following fields: start time; end time; RRC status; elapsed training time; computational resources consumed; storage resources consumed; or power consumption. The wireless communication device immediately transitions from the RRC_IDLE or RRC_INACTIVE state to the RRC_Connected state after completing the FL (Fluid Flow). A wireless communication device in the RRC_IDLE or RRC_INACTIVE state waits until the deadline for updating the local model is approaching before transitioning from the RRC_IDLE or RRC_INACTIVE state to the RRC_Connected state. The wireless communication device can send an RRC setup message to the wireless communication node. In the RRCSetupRequest or RRCResumeRequest message, the wireless communication device sets the setup reason value to mo-Data or ai-Data.

[0070] The wireless communication device can receive a paging message from the wireless communication node indicating an AI / ML / FL model update. The paging message includes at least one of the following fields: AI / ML / FL model ID; AI / ML / FL model function; or AI / ML / FL model update indication. Upon receiving the paging message, the device also retrieves the updated FL model from the wireless communication node. If the wireless communication device is in the RRC_IDLE or RRC_INACTIVE state, it must enter the RRC_Connected state before retrieving the updated FL model. The wireless communication device sends an FL model request to the wireless communication node. The FL model request includes at least one of the following fields: FL task ID; FL task index; FL task; FL interest indication; or model request indication.

[0071] A wireless communication node can send the FL tasks, AI / ML / FL models, AI / ML / FL functions, and / or AI / ML / FL model IDs it supports to neighboring wireless communication nodes. A wireless communication node can also send new global AI / ML / FL models and / or paging indications to neighboring wireless communication nodes. The wireless communication node and its neighboring wireless communication nodes belong to the same AI / ML / FL region. Upon receiving a new global AI / ML / FL model and / or paging indication, a neighboring wireless communication node can initiate a paging process to send a paging message to the wireless communication device. The paging message includes at least one of the following fields: AI / ML / FL model ID; AI / ML / FL function; or AI / ML / FL model update indication. The wireless communication device can receive updated FL models from the system information of neighboring wireless communication nodes. The wireless communication device can enter the RRC_Connected state from the RRC_IDLE or RRC_INACTIVE state to receive updated FL models from neighboring wireless communication nodes. The wireless communication device can determine whether it is within the FL region when it receives system information about the FL region. The wireless communication device can determine whether the configuration is valid for performing FL and reporting local model updates, and can remove the configuration if it is no longer valid.

[0072] While various embodiments of this solution have been described above, it should be understood that they are given by way of example only and not by way of limitation. Similarly, various illustrations may depict exemplary architectures or configurations provided to enable those skilled in the art to understand the exemplary features and functionality of this solution. However, those skilled in the art will understand that this solution is not limited to the exemplary architectures or configurations shown, but can be implemented using various alternative architectures and configurations. Furthermore, as those skilled in the art will understand, one or more features of one embodiment may be combined with one or more features of another embodiment described herein. Therefore, the breadth and scope of this disclosure should not be limited by any of the illustrative embodiments described above.

[0073] It should also be understood that any reference to elements in this document, such as “first” or “second”, generally does not limit the number or order of these elements. Rather, these references may be used herein as a convenient way to distinguish between two or more elements or instances of a single element. Therefore, references to a first element and a second element do not imply that only two elements can be used, or that the first element must precede the second element in some way.

[0074] Furthermore, those skilled in the art will understand that information and signals can be represented using any of a variety of different techniques and skills. For example, data, instructions, commands, information, signals, bits, and symbols that may be mentioned in the above description can be represented by voltage, current, electromagnetic waves, magnetic fields or particles, light fields or particles, or any combination thereof.

[0075] Those skilled in the art will further understand that any of the various illustrative logic blocks, modules, processors, devices, circuits, methods, and functions described in conjunction with the aspects disclosed herein can be implemented by electronic hardware (e.g., digital implementation, analog implementation, or a combination of both), firmware, various forms of program or design code containing instructions (which may be referred to herein as "software" or "software module" for convenience), or any combination of these technologies. To clearly illustrate this interchangeability of hardware, firmware, and software, various illustrative components, blocks, modules, circuits, and steps have been generally described above in terms of their functionality. Whether such functionality is implemented as hardware, firmware, or software, or a combination of these technologies, depends on the specific application and the design constraints imposed on the system as a whole. Those skilled in the art can implement the described functionality in various ways for each specific application, but such implementation decisions will not depart from the scope of this disclosure.

[0076] Furthermore, those skilled in the art will understand that the various illustrative logic blocks, modules, devices, components, and circuits described herein can be implemented within or executed by an integrated circuit (IC), which may include a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, or any combination thereof. Logic blocks, modules, and circuits may also include antennas and / or transceivers for communicating with various components within a network or device. A general-purpose processor may be a microprocessor, but alternatively, the processor may be any conventional processor, controller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors combined with a DSP core, or any other suitable configuration configured to perform the functions described herein.

[0077] If implemented in software, these functions can be stored as one or more instructions or code on a computer-readable medium. Therefore, the steps of the methods or algorithms disclosed herein can be implemented as software stored on a computer-readable medium. Computer-readable media include computer storage media and communication media, including any medium capable of transferring computer programs or code from one place to another. Storage media can be any available medium accessible by a computer. By way of example and not limitation, such computer-readable media can include RAM, ROM, EEPROM, CD-ROM or other optical disc storage, disk storage or other magnetic storage devices, or any other medium capable of storing desired program code in the form of instructions or data structures and accessible by a computer.

[0078] In this document, the term "module" as used herein refers to software, firmware, hardware, and any combination of these elements used to perform the associated functions described herein. Furthermore, for ease of discussion, individual modules are described as discrete modules; however, as will be apparent to those skilled in the art, two or more modules may be combined to form a single module that performs the associated functions according to embodiments of this solution.

[0079] Furthermore, in embodiments of this solution, memory or other storage devices and communication components may be employed. It is understood that, for clarity, the above description has referenced various functional units and processors in describing embodiments of this solution. However, it will be apparent that any suitable allocation of functionality among different functional units, processing logic elements, or domains can be used without diminishing the effectiveness of this solution. For example, a function shown to be performed by a separate processing logic element or controller may be performed by the same processing logic element or controller. Therefore, references to specific functional units are merely references to suitable means for providing said functionality and do not indicate a strict logical or physical structure or organization.

[0080] Various modifications to the embodiments described in this disclosure will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other embodiments without departing from the scope of this disclosure. Therefore, this disclosure is not intended to be limited to the embodiments shown herein, but should be accorded the widest scope consistent with the novel features and principles disclosed herein, as set forth in the following claims.

Claims

1. A wireless communication method, comprising: The wireless communication device receives a first message from the wireless communication node, the first message including configuration for performing federated learning (FL).

2. The wireless communication method according to claim 1, wherein the wireless communication node is operatively configured to act as a centralized node in an FL network.

3. The wireless communication method according to claim 2, wherein the FL network can be at least one of an application server, an OAM server, a network function for AI in the core network, a base station (BS), a CU, a DU, an AI controller, an MEC, or a user equipment (UE).

4. The wireless communication method according to claim 1, wherein the configuration includes at least one of the following: FL task ID; FL task index; FL task; FL model; FL validity period; AI / ML / FL region; connected_allowed indicator; idleallowed indicator; inactive allowed indicator; priority; preemptible indicator; power threshold; CPU threshold; expected completion time or deadline for local model update; or period.

5. The wireless communication method according to claim 4, wherein the FL effective time indicates the duration of the configured FL task.

6. The wireless communication method according to claim 4, wherein the FL model indicates the address of the FL model.

7. The wireless communication method according to claim 4, wherein the AI / ML / FL region includes at least one of the following fields: one or more TACs; a list of cell IDs; or an AI / ML / FL region code.

8. The wireless communication method of claim 4, wherein the connected_allowed indication specifies that the wireless communication device is allowed to execute the FL in the RRC_CONNECTED state, wherein the idle allowed indication specifies that the wireless communication device is allowed to execute the FL in the RRC_IDLE state, and wherein the inactive allowed indication specifies that the wireless communication device is allowed to execute the FL in the RRC_INACTIVE state.

9. The wireless communication method according to claim 4, wherein the power threshold indicates a threshold of the remaining power percentage of the wireless communication device, and wherein the CPU threshold indicates a threshold of the CPU occupancy percentage of the wireless communication device.

10. The wireless communication method of claim 4, wherein the priority indicates the priority level of the associated FL task.

11. The wireless communication method of claim 4, wherein the preemptibility indication specifies whether the associated FL task can be preempted when a higher-priority FL task is scheduled / executed.

12. The wireless communication method according to claim 1, wherein the first message can be transmitted via system information or via broadcast / multicast signaling.

13. The wireless communication method according to claim 1, further comprising: The wireless communication node sends an RRC reconfiguration or RRC release message to the wireless communication device, the RRC reconfiguration or RRC release message including the configuration for performing the FL.

14. The wireless communication method of claim 13, wherein the configuration includes at least one of the following: FL task ID; FL task index; FL task; FL model; FL validity period; AI / ML / FL region; connected_allowed indicator; idle allowed indicator; inactive allowed indicator; priority; preemptible indicator; power threshold; CPU threshold; expected completion time or deadline for local model update; or period.

15. The wireless communication method according to claim 1, further comprising: The wireless communication device sends an RRCSetupRequest or RRCResumeRequest message to the wireless communication node. In the RRCSetupRequest message, the wireless communication device sets the establishment reason value to mo-Data or ai-Data.

16. The wireless communication method according to claim 1, further comprising: The wireless communication device sends local model updates and / or performance data used to train the local model to the wireless communication node.

17. The wireless communication method according to claim 16, wherein the performance data includes at least one of the following fields: start time; end time; RRC status; training elapsed time; computing resources consumed; storage resources consumed; or power consumption.

18. The wireless communication method according to claim 16, wherein the wireless communication device immediately enters the RRC_Connected state from the RRC_IDLE or RRC_INACTIVE state after completing the FL.

19. The wireless communication method according to claim 16, wherein the wireless communication device in the RRC_IDLE or RRC_INACTIVE state waits until the expected completion time or deadline of the local model update is approaching before entering the RRC_Connected state from the RRC_IDLE or RRC_INACTIVE state.

20. The wireless communication method of claim 1, wherein the wireless communication device estimates the time interval for performing the FL.

21. The wireless communication method of claim 20, wherein the wireless communication device in the RRC_IDLE state is able to start the FL before t1-t2, wherein t1 represents the expected completion time or deadline of the local model update, and t2 represents the estimated time interval for performing the FL.

22. The wireless communication method according to any one of claims 1, 18, or 19, further comprising: The wireless communication device sends an RRCSetupRequest or RRCResumeRequest message to the wireless communication node. In the RRCSetupRequest message, the wireless communication device sets the establishment reason value to mo-Data or ai-Data.

23. The wireless communication method according to claim 1, 6, or 12, wherein if the FL model is included in system information, the wireless communication device receives the FL model via the system information; otherwise, the wireless communication device in the RRC_IDLE or RRC_INACTIVE state needs to enter the RRC_Connected state to retrieve the latest FL model from the wireless communication node or via a URL link.

24. The wireless communication method according to claim 1 or 23, further comprising: The wireless communication device sends a second message to the wireless communication node requesting the FL model, wherein the second message includes at least one of the following fields: FL task ID; FL task index; FL task; FL interest indication; or model request indication.

25. The wireless communication method of claim 24, wherein the wireless communication device receives the FL model from the wireless communication node based on a request from the wireless communication device.

26. The wireless communication method according to claim 1, further comprising: The wireless communication device receives a paging message from the wireless communication node indicating an update to the AI / ML / FL model; The paging message includes at least one of the following fields: AI / ML / FL model ID; AI / ML / FL model function; or AI / ML / FL model update indication.

27. The wireless communication method according to claim 26, wherein, Upon receiving the paging message, the method further includes: the wireless communication device retrieving an updated FL model from the wireless communication node.

28. The wireless communication method of claim 27, wherein if the wireless communication device is in the RRC_IDLE or RRC_INACTIVE state, the wireless communication device enters the RRC_Connected state before retrieving the updated FL model.

29. The wireless communication method of claim 28, wherein the wireless communication device sends an FL model request to the wireless communication node, and wherein the FL model request includes at least one of the following fields: FL task ID; FL task index; FL task; FL interest indication; or model request indication.

30. The wireless communication method according to claim 1, further comprising: When a wireless communication device receives information about a FL area via system information, it determines whether the wireless communication device is within the FL area.

31. The wireless communication method according to claim 30, wherein determining that the wireless communication device is within the FL area further includes: The wireless communication device determines that the configuration is valid for executing the FL.

32. The wireless communication method of claim 30, wherein determining that the wireless communication device is outside the FL region further comprises: The wireless communication device determines that the configuration is no longer valid and removes the configuration.

33. A wireless communication method, comprising: The wireless communication node sends at least one of the following fields to the neighboring wireless communication node: the FL task supported by the wireless communication node, the AI / ML / FL model, the paging indication, the AI / ML / FL function and / or the AI / ML / FL model ID.

34. The wireless communication method according to claim 33, wherein the wireless communication node and the adjacent wireless communication node belong to the same AI / ML / FL region.

35. The wireless communication method according to claim 33, wherein, Upon receiving a new global AI / ML / FL model and / or the paging indication, the adjacent wireless communication node can initiate a paging process to send a paging message to the wireless communication device; The paging message includes at least one of the following fields: the AI / ML / FL model ID; the AI / ML / FL function; or the AI / ML / FL model update indication.

36. The wireless communication method according to claim 35, further comprising: The wireless communication device receives an updated FL model from the system information of the adjacent wireless communication node; or The wireless communication device enters the RRC_Connected state from the RRC_IDLE state or the RRC_INACTIVE state to receive the updated FL model from the adjacent wireless communication node.

37. A wireless communication method, comprising: The wireless communication node sends a first message to the wireless communication device, the first message including configuration for performing federated learning (FL).

38. A wireless communication device comprising a processor and a memory, wherein the processor is configured to read code from the memory and implement the method according to any one of claims 1 to 36.

39. A computer program product comprising a processor and a memory, wherein the processor is configured to read code from the memory and implement the method according to any one of claims 1 to 36.

Images