Updating a machine learning model

Abstract

A method is disclosed, the method comprising: transmitting, to at least one user equipment, a machine learning model for performing at least one prediction of at least one communication parameter in at least one cell; transmitting, to the at least one user equipment, at least one of: an indication to update the machine learning model, or a configuration indicating at least one condition for determining whether the machine learning model is to be updated; and receiving, from the at least one user equipment, an updated version of the machine learning model.

Description

Technical Field

[0001] The following example embodiments relate to wireless communication and machine learning. Background Technology

[0002] Machine learning models can be used in a variety of use cases in wireless communication. However, effectively updating machine learning models presents challenges. Summary of the Invention

[0003] The scope of protection sought by the various example embodiments is defined by the independent claims. Example embodiments and features (if any) described in this specification that are not within the scope of the independent claims should be interpreted as examples helpful in understanding the various embodiments.

[0004] According to one aspect, an apparatus is provided, the apparatus comprising at least one processor and at least one memory, the at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to at least: receive a machine learning model from a network node, the machine learning model being used to perform at least one prediction of at least one communication parameter in at least one cell; receive from the network node at least one of: an instruction for updating the machine learning model, or a configuration indicating at least one condition for determining whether to update the machine learning model; use the machine learning model to perform at least one prediction of at least one communication parameter in at least one cell; determine whether to update the machine learning model based on the instruction or at least one condition; based on the determination that the machine learning model needs to be updated, obtain an updated version of the machine learning model by updating the machine learning model based on at least one prediction; and send the updated version of the machine learning model to the network node.

[0005] According to another aspect, an apparatus is provided, comprising: components for receiving a machine learning model from a network node, the machine learning model being used to perform at least one prediction of at least one communication parameter in at least one cell; components for receiving from the network node at least one of: an instruction for updating the machine learning model, or a configuration indicating at least one condition for determining whether to update the machine learning model; components for using the machine learning model to perform at least one prediction of at least one communication parameter in at least one cell; components for determining whether to update the machine learning model based on the instruction or at least one condition; components for obtaining an updated version of the machine learning model by updating the machine learning model based on at least one prediction based on the determination that the machine learning model should be updated; and components for sending the updated version of the machine learning model to the network node.

[0006] According to another aspect, a method is provided, the method comprising: receiving a machine learning model from a network node, the machine learning model being used to perform at least one prediction of at least one communication parameter in at least one cell; receiving from the network node at least one of the following: an instruction for updating the machine learning model, or a configuration indicating at least one condition for determining whether to update the machine learning model; using the machine learning model to perform at least one prediction of at least one communication parameter in at least one cell; determining whether to update the machine learning model based on the instruction or at least one condition; obtaining an updated version of the machine learning model by updating the machine learning model based on at least one prediction based on the determination that the machine learning model needs to be updated; and sending the updated version of the machine learning model to the network node.

[0007] According to another aspect, a computer program is provided, the computer program including instructions that, when executed by a device, cause the device to perform at least the following: receiving a machine learning model from a network node, the machine learning model being used to perform at least one prediction of at least one communication parameter in at least one cell; receiving from the network node at least one of the following: an instruction for updating the machine learning model, or a configuration of at least one condition for determining whether to update the machine learning model; using the machine learning model to perform at least one prediction of at least one communication parameter in at least one cell; determining whether to update the machine learning model based on the instruction or at least one condition; obtaining an updated version of the machine learning model by updating the machine learning model based on at least one prediction based on the determination that the machine learning model needs to be updated; and sending the updated version of the machine learning model to the network node.

[0008] According to another aspect, a computer-readable medium is provided, comprising program instructions that, when executed by an apparatus, cause the apparatus to perform at least the following: receiving a machine learning model from a network node, the machine learning model being used to perform at least one prediction of at least one communication parameter in at least one cell; receiving from the network node at least one of the following: an instruction for updating the machine learning model, or a configuration indicating at least one condition for determining whether to update the machine learning model; using the machine learning model to perform at least one prediction of at least one communication parameter in at least one cell; determining whether to update the machine learning model based on the instruction or at least one condition; obtaining an updated version of the machine learning model by updating the machine learning model based on at least one prediction, based on the determination that the machine learning model needs to be updated; and sending the updated version of the machine learning model to the network node.

[0009] According to another aspect, a non-transitory computer-readable medium is provided, the medium comprising program instructions that, when executed by a device, cause the device to perform at least the following: receiving a machine learning model from a network node, the machine learning model being used to perform at least one prediction of at least one communication parameter in at least one cell; receiving from the network node at least one of the following: an instruction for updating the machine learning model, or a configuration indicating at least one condition for determining whether to update the machine learning model; using the machine learning model to perform at least one prediction of at least one communication parameter in at least one cell; determining whether to update the machine learning model based on the instruction or at least one condition; obtaining an updated version of the machine learning model by updating the machine learning model based on at least one prediction based on the determination that the machine learning model needs to be updated; and sending the updated version of the machine learning model to the network node.

[0010] According to another aspect, an apparatus is provided, comprising at least one processor and at least one memory, the at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to at least: send a machine learning model to at least one user equipment, the machine learning model being used to perform at least one prediction of at least one communication parameter in at least one cell; send to the at least one user equipment at least one of the following: an instruction to update the machine learning model, or an instruction to configure at least one condition for determining whether to update the machine learning model; and receive an updated version of the machine learning model from the at least one user equipment.

[0011] According to another aspect, an apparatus is provided, comprising: components for transmitting a machine learning model to at least one user equipment, the machine learning model being used to perform at least one prediction of at least one communication parameter in at least one cell; components for transmitting to at least one of the following to the at least one user equipment: an instruction for updating the machine learning model, or an instruction for configuring at least one condition for determining whether to update the machine learning model; and components for receiving an updated version of the machine learning model from the at least one user equipment.

[0012] According to another aspect, a method is provided, the method comprising: sending a machine learning model to at least one user equipment for performing at least one prediction of at least one communication parameter in at least one cell; sending at least one of the following to at least one user equipment: an instruction to update the machine learning model, or a configuration indicating at least one condition for determining whether to update the machine learning model; and receiving an updated version of the machine learning model from at least one user equipment.

[0013] According to another aspect, a computer program is provided, the computer program including instructions that, when executed by a device, cause the device to perform at least the following: sending a machine learning model to at least one user equipment, the machine learning model being used to perform at least one prediction of at least one communication parameter in at least one cell; sending at least one of the following to at least one user equipment: an instruction to update the machine learning model, or an instruction to configure at least one condition for determining whether to update the machine learning model; and receiving an updated version of the machine learning model from at least one user equipment.

[0014] According to another aspect, a computer-readable medium is provided, comprising program instructions that, when executed by a device, cause the device to perform at least the following: sending a machine learning model to at least one user equipment for performing at least one prediction of at least one communication parameter in at least one cell; sending at least one of the following to the at least one user equipment: an instruction to update the machine learning model, or an instruction to configure at least one condition for determining whether to update the machine learning model; and receiving an updated version of the machine learning model from the at least one user equipment.

[0015] According to another aspect, a non-transitory computer-readable medium is provided, the medium including program instructions that, when executed by a device, cause the device to perform at least the following: sending a machine learning model to at least one user equipment for performing at least one prediction of at least one communication parameter in at least one cell; sending at least one of the following to at least one user equipment: an instruction to update the machine learning model, or an instruction to configure at least one condition for determining whether to update the machine learning model; and receiving an updated version of the machine learning model from at least one user equipment. Attached Figure Description

[0016] In the following description, various exemplary embodiments will be described in more detail with reference to the accompanying drawings, in which...

[0017] Figure 1A An example of a wireless communication network is illustrated;

[0018] Figure 1B An example of a communication system is illustrated;

[0019] Figure 2 An example of a synchronization signal block is illustrated;

[0020] Figure 3 The illustration shows an example of beamforming for a user equipment with a multi-panel configuration;

[0021] Figure 4A An example of a communication system supported by artificial intelligence or machine learning is illustrated;

[0022] Figure 4B An example of a communication system supported by artificial intelligence or machine learning is illustrated;

[0023] Figure 5 The diagram illustrates the signal flow graph;

[0024] Figure 6 The diagram illustrates the signal flow graph;

[0025] Figure 7 The flowchart is shown.

[0026] Figure 8 The flowchart is shown.

[0027] Figure 9 The flowchart is shown.

[0028] Figure 10 The flowchart is shown.

[0029] Figure 11 The flowchart is shown.

[0030] Figure 12 The illustration shows an example of spatial domain prediction based on a convolutional neural network model;

[0031] Figure 13 The illustration shows an example of time-domain prediction based on a long short-term memory recurrent neural network model;

[0032] Figure 14 The illustration shows an example of a prediction based on a deep reinforcement learning model;

[0033] Figure 15 An example of the device is illustrated; and

[0034] Figure 16 An example of the device is illustrated. Detailed Implementation

[0035] The following embodiments are exemplary. Although the specification may refer to "a," "an," or "some" embodiments in several places in the text, this does not necessarily mean that each reference to the same embodiment(s) or a particular feature applies only to a single embodiment. Individual features of different embodiments may also be combined to provide other embodiments.

[0036] Some of the example embodiments described herein can be implemented in a wireless communication network, including a radio access network based on one or more of the following radio access technologies: Global System for Mobile Communications (GSM) or any other second-generation radio access technology, Universal Mobile Telecommunications System (UMTS, 3G) based on Basic Wideband Code Division Multiple Access (W-CDMA), High-Speed ​​Packet Access (HSPA), Long Term Evolution (LTE), LTE Advanced, Fourth Generation (4G), Fifth Generation (5G), 5G New Radio (NR), Advanced 5G (i.e., 3GPP NR Rel-18 and later), or Sixth Generation (6G). Some examples of radio access networks include Universal Mobile Telecommunications System (UMTS) Radio Access Network (UTRAN), Evolved Universal Terrestrial Radio Access Network (E-UTRA), or Next Generation Radio Access Network (NG-RAN). The wireless communication network may also include a core network, and some example embodiments can also be applied to the network functions of the core network.

[0037] It should be noted that the embodiments are not limited to the wireless communication network given as an example, but those skilled in the art can also apply the solution to other wireless communication networks or systems with the necessary characteristics. For example, some example embodiments can also be applied to communication systems based on the IEEE 802.11 standard or communication systems based on the IEEE 802.15 standard.

[0038] Figure 1A An example of a simplified wireless communication network is depicted, showing some physical and logical entities. Figure 1A The connections shown can be physical or logical connections. Those skilled in the art will understand that wireless communication networks may also include, in addition to... Figure 1A Other physical and logical entities besides the physical and logical entities shown.

[0039] However, the exemplary embodiments described herein are not limited to the wireless communication networks given as examples, but those skilled in the art can apply the embodiments described herein to other wireless communication networks with the necessary characteristics.

[0040] Figure 1A The example wireless communication network shown includes an access network such as a radio access network (RAN) and a core network 110.

[0041] Figure 1AUser equipment (UEs) 100 and 102 are illustrated, configured to wirelessly connect to an access node (AN) 104 of an access network on one or more communication channels in a radio cell. AN 104 may be an evolved Node B (eNB or eNodeB) or a next-generation Node B (gNB or gNodeB) providing the radio cell. A wireless connection from the UE to the access node 104 (e.g., a radio link) may be referred to as an uplink (UL) or reverse link, while a wireless connection from the access node to the UE (e.g., a radio link) may be referred to as a downlink (DL) or forward link. UE 100 may also communicate directly with UE 102 via a wireless connection commonly referred to as a sidelink (SL), and vice versa. It should be understood that the access node 104 or its functionality may be implemented using any node, host, server, or access point, etc., suitable for providing such functionality.

[0042] An access network may include more than one access node, in which case the access nodes may also be configured to communicate with each other via wired or wireless links. These links between access nodes may be used to send and receive control plane signaling, or to route data from one access node to another.

[0043] An access node may include a computing device configured to control the radio resources of the access node. An access node may also be referred to as a base station, base transceiver station (BTS), access point, cell site, radio access node, or any other type of node capable of wirelessly connecting to a UE (e.g., UE 100, 102). An access node may include or be coupled to a transceiver. From the transceiver of the access node, a connection may be provided to an antenna element that establishes a bidirectional radio link to UE 100, 102. The antenna element may include an antenna or antenna element, or multiple antennas or antenna elements.

[0044] Access node 104 can also connect to core network (CN) 110. Core network 110 may include an evolved packet core (EPC) network and / or a fifth-generation core network (5GC). EPC may include network entities such as a serving gateway (S-GW for routing and forwarding data packets), a packet data network gateway (P-GW) for providing UE connectivity to external packet data networks, and a mobility management entity (MME). 5GC may include network functions such as user plane functions (UPF), access and mobility management functions (AMF), and location management functions (LMF).

[0045] The core network 110 can also communicate with or utilize services provided by one or more external networks 113, such as the public switched telephone network or the Internet. For example, in a 5G wireless communication network, the UPF of the core network 110 can be configured to communicate with an external data network via the N6 interface. In an LTE wireless communication network, the P-GW of the core network 110 can be configured to communicate with an external data network.

[0046] The UEs 100 and 102 shown are a type of device that can be allocated and assigned resources on an air interface. UEs 100 and 102 may also be referred to as wireless communication devices, subscriber units, mobile stations, remote terminals, access terminals, user terminals, terminal equipment, or user equipment, to name just a few. A UE can be a computing device operating with or without a Subscriber Identity Module (SIM), including but not limited to the following types of computing devices: mobile phones, smartphones, personal digital assistants (PDAs), handheld devices, computing devices including wireless modems (e.g., alarm or measuring devices), laptop computers, desktop computers, tablet computers, game consoles, laptops, multimedia devices, capability reduction (RedCap) devices, wearable devices with radio components (e.g., watches, headphones, or glasses), sensors including wireless modems, or any computing device including a wireless modem integrated into a vehicle.

[0047] It should be understood that a UE can also be a nearly exclusive uplink-only device, an example of which could be a camera or camcorder that loads images or video clips onto a network. A UE can also be a device capable of operating in an Internet of Things (IoT) network, a scenario where objects can be provided with the ability to transmit data over a network without requiring human-to-human or human-to-computer interaction. A UE can also leverage the cloud. In some applications, computation can be performed in the cloud or within another UE.

[0048] Wireless communication networks can also support the use of cloud services. For example, at least a portion of the core network operation can be performed as a cloud service (this is in...). Figure 1A (The "cloud" in the middle is 114). Wireless communication networks may also include a central control entity, etc., to provide facilities for different operators' wireless communication networks to cooperate, for example, in spectrum sharing.

[0049] 5G enables the use of multiple-input multiple-output (MIMO) antennas in access nodes 104 and / or UEs 100, 102, far more base stations or access nodes than LTE networks (the so-called small cell concept), including macro sites operating in cooperation with smaller sites, and the use of various radio technologies depending on service requirements, use cases, and / or available spectrum. 5G wireless communication networks can support a wide range of use cases and related applications, including video streaming, augmented reality, different data sharing methods, and various forms of machine-type applications such as (massive) machine-type communication (mMTC), including vehicle safety, various sensors, and real-time control.

[0050] In 5G wireless communication networks, access nodes and / or UEs can have multiple radio interfaces, namely sub-6GHz, cmWave, and mmWave, and can also be integrated with existing legacy radio access technologies such as LTE. For example, integration with LTE can be implemented as a system where LTE provides macro coverage, and 5G radio interface access can be aggregated to LTE from small cells. In other words, 5G wireless communication networks can support inter-RAT operability (such as LTE-5G) and inter-RI operability (inter-radio interface operability, such as sub-6GHz, cmWave, and mmWave). One concept considered for 5G wireless communication networks could be network slicing, where multiple independent and dedicated virtual subnets (network instances) can be created within substantially the same infrastructure to run services with different requirements for latency, reliability, throughput, and mobility.

[0051] 5G enables analytics and knowledge generation at the data source. This approach can involve leveraging resources that may not have continuous network connectivity, such as laptops, smartphones, tablets, and sensors. Multi-access edge computing (MEC) can provide a distributed computing environment for application and service hosting. It can also store and process content near cellular subscribers for faster response times. Edge computing can encompass a wide range of technologies, such as wireless sensor networks, mobile data acquisition, mobile signature analytics, collaborative distributed peer-to-peer self-organizing networks and processing (which can also be categorized as local cloud / fog computing and grid / mesh computing), dew computing, mobile edge computing, cloudlets, distributed data storage and retrieval, autonomous self-healing networks, remote cloud services, augmented and virtual reality, data caching, the Internet of Things (massive connectivity and / or time-critical), and critical communications (autonomous vehicles, traffic safety, real-time analytics, time-critical control, healthcare applications).

[0052] In some example embodiments, an access node (e.g., access node 104) may include: a radio unit (RU) comprising radio transceivers (TRXs), i.e., transmitters (Tx) and receivers (Rx); one or more distributed units (DUs) 105, which may be used for so-called Layer 1 (L1) processing and real-time Layer 2 (L2) processing; and a central unit (CU) 108 (also called a centralized unit), which may be used for non-real-time Layer 2 and Layer 3 (L3) processing. The CU 108 may be connected to one or more DUs 105, for example, via an F1 interface. Such embodiments of the access node enable the centralization of the CU relative to the cell site and the DU, while the DU can be more distributed and may even be retained at the cell site. The CU and DU together may also be referred to as a baseband or baseband unit (BBU). The CU and DU may also be included in a radio access point (RAP).

[0053] CU 108 can be a logical node that hosts the Radio Resource Control (RRC), Service Data Adaptation Protocol (SDAP), and / or Packet Data Convergence Protocol (PDCP) of the access node's NR protocol stack. DU 105 can be a logical node that hosts the Radio Link Control (RLC), Media Access Control (MAC), and / or Physical (PHY) layers of the access node's NR protocol stack. The operation of the DU can be at least partially controlled by the CU. It should also be understood that the functional distribution between DU 105 and CU 108 can vary depending on the implementation. The CU may include a control plane (CU-CP), which can be a logical node that hosts the control plane portions of the RRC and PDCP protocols of the access node's NR protocol stack. The CU may also include a user plane (CU-UP), which can be a logical node that hosts the user plane portions of the PDCP protocol and the SDAP protocol of the access node's CU.

[0054] Cloud computing systems can also be used to provide CU 108 and / or DU 105. CUs provided by cloud computing systems can be referred to as virtualized CUs (vCUs). In addition to vCUs, virtualized DUs (vDUs) provided by cloud computing systems can also exist. Furthermore, there can be a combination where DUs can be implemented on so-called bare-metal solutions, such as application-specific integrated circuits (ASICs) or customer-specific standard product (CSSP) system-on-chips (SoCs).

[0055] By leveraging Network Functions Virtualization (NFV) and Software-Defined Networking (SDN), edge cloud can be introduced into the access network (e.g., RAN). Using edge cloud can represent access node operations that should be performed, at least partially, in a computing system operatively coupled to a Remote Radio Header (RRH) or Radio Unit (RU) at the access node. Alternatively, access node operations can be performed on a distributed computing system or cloud computing system located at the access node. The application of cloud RAN architecture enables real-time RAN functions to be performed at the access network (e.g., in DU 105) and non-real-time functions to be performed in a centralized manner (e.g., in CU 108).

[0056] It should also be understood that, compared to LTE or 5G, the functional allocation between core network operations and access node operations may differ, or even cease to exist, in future wireless communication networks. Other technological advancements that may be available include big data and all-IP, which could potentially transform how wireless communication networks are built and managed. 5G (or New Radio, NR) wireless communication networks can support multiple hierarchical structures, where multiple access edge computing (MEC) servers can be placed between the core network 110 and access node 104. It should be understood that MEC can also be applied to LTE wireless communication networks.

[0057] 5G wireless communication networks (“5G networks”) may also include non-terrestrial communication networks, such as satellite communication networks, to enhance or supplement the coverage of 5G radio access networks. For example, satellite communications can support data transmission between the 5G radio access network and the core network, thereby enabling broader network coverage. Possible use cases could be providing service continuity for machine-to-machine (M2M) or Internet of Things (IoT) devices or for passengers on board vehicles, or ensuring the service availability of critical communications and future rail / maritime / aviation communications. Satellite communications can utilize geostationary orbit (GEO) satellite systems or low Earth orbit (LEO) satellite systems, particularly mega-constellations (systems deploying hundreds of (nano) satellites). A given satellite 106 in a mega-constellation can cover several satellite-enabled network entities that create terrestrial cells. Terrestrial cells can be created by ground relay access nodes or by access nodes 104 located on the ground or in satellites.

[0058] Those skilled in the art will understand that Figure 1AThe access node 104 shown is merely an example of a portion of an access network (e.g., a radio access network), and in practice, an access network may include multiple access nodes, UEs 100 and 102 may access multiple radio cells, and the access network may also include other equipment, such as physical layer relay access nodes or other entities. At least one of the access nodes may be a home eNodeB or a home gNodeB. A home gNodeB or a home eNodeB is an access node that can be used to provide indoor coverage in a home, office, or other indoor environment.

[0059] Furthermore, within the geographical area of ​​an access network (e.g., a radio access network), multiple different types of radio cells and multiple radio cells can be provided. Radio cells can be macrocells (or umbrella cells), which can be large areas with diameters of up to tens of kilometers, or smaller cells such as microcells, femtocells, or picocells. Figure 1A Multiple access nodes can provide any type of these cells. A cellular radio network can be implemented as a multi-layered access network comprising several types of radio cells. In a multi-layered access network, an access node can provide one or more types of radio cells, and therefore providing such a multi-layered access network may require multiple access nodes.

[0060] To meet the need for improved access network performance, the concept of "plug-and-play" access nodes can be introduced. Besides the home eNodeB or home gNodeB, access networks capable of using "plug-and-play" access nodes can also include home node B gateways or HNB-GWs ( Figure 1A (Not shown in the image). An HNB-GW, which can be installed in an operator's access network, can aggregate services from a large number of home eNodeBs or home gNodeBs back to the operator's core network.

[0061] 5G NR access links can operate in millimeter wave (mmWave), Asia Pacific Hertz (THz) bands, and higher frequency ranges, which are inherently more susceptible to higher path loss and penetration loss. In the higher frequency range, both the gNB and UE can employ front-end circuitry with, for example, multiple beam patterns to overcome the limitations of the propagation channel. 5G NR access links can also operate at lower frequencies, where the gNB can use beamforming, but the UE can operate using isotropic or omnidirectional beam patterns (i.e., the UE can operate without beamforming).

[0062] A beam refers to the directional transmission or reception of radio signals. A beam can also be represented as a spatial filter, spatial direction, or angle. Beams can be formed using advanced antenna techniques known as beamforming. For example, beamforming can be beneficial in 5G NR because it enables support for higher frequency bands (e.g., millimeter-wave frequencies) and their massive MIMO capabilities. Beams can be categorized into downlink beams and uplink beams.

[0063] Downlink beamforming can be performed by a gNB to send signals to the UE in a specific direction. By focusing transmission energy in the direction of the intended UE, downlink beamforming can improve signal strength and overall communication quality, while minimizing interference to other UEs and reducing power consumption.

[0064] Uplink beamforming can be performed by the UE to transmit signals to the gNB in ​​a specific direction. By focusing transmission energy in the direction of the gNB, uplink beamforming can enhance the communication link between the UE and the gNB, which can improve received signal strength, reduce interference, and extend the UE's range.

[0065] Beamforming can involve using large antenna arrays at both the gNB and the UE, allowing multiple beams to be formed simultaneously. This enables features such as spatial multiplexing and multi-user MIMO (MU-MIMO), which can further improve the capacity and efficiency of 5G networks.

[0066] Network (e.g., Figure 1A The wireless communication network shown can also support the use of multiple Transmitter Receiver Points (TRPs). This can be called multiple Transmitter Receiver Point (multiple TRP) operation. Multiple TRP operation can support, for example, two or more TRPs. Therefore, for example, UE 100, 102 can receive data via multiple TRPs. Different TRPs can be controlled, for example, by access node 104 (such as gNB).

[0067] TRP is a term used to refer to a physical point in network infrastructure where both signal transmission and reception can occur. In other words, a TRP is a point that can handle uplink and downlink communication between the UE and the network.

[0068] Figure 1B An example of the system is illustrated. Figure 1B It can be understood as depicting Figure 1A It is part of the wireless communication network, but has higher accuracy for multi-TRP scenarios.

[0069] The cell area can use one or more TRPs 104A, 104B (e.g., gNB) from access node 104 (e.g., gNB). Figure 1BThe access node 104 provides one or more beams 121, 122, 123, 124, 125, 126 (TRP#1...#X) for coverage. It should be noted that the access node 104 can provide one or more cells. Here, the term "cell" refers to a radio cell, which represents the coverage area served by the access node 104.

[0070] Given beams 121, 122, 123, 124, 125, and 126, they can carry identifiers to enable UE 100 to identify the beams and perform measurements (e.g., received power, reference signal received power), and associate the measurements with that specific identifier. For example, a cell can be covered by multiple synchronization signal blocks (SSBs) represented as SSB#0…#L. A given SSB can be identified based on an identifier (SSB time location index) carried by the SSB. A given SSB can also carry the identifier of the cell it is associated with.

[0071] It should be noted that the number of TRPs and the number of beams can also be related to... Figure 1B The differences are shown.

[0072] Figure 2 The diagram illustrates an example of the time-frequency structure of an SSB. This type of SSB can be used, for example, in 5G NR. The SSB includes a Primary Synchronization Signal (PSS) 201, a Secondary Synchronization Signal (SSS) 202, and a Physical Broadcast Channel (PBCH) 203. Figure 2 In the example, both PSS 201 and SSS 202 occupy one orthogonal frequency division multiplexing (OFDM) symbol and 127 subcarriers. Figure 2 In the example, PBCH spans three OFDM symbols (OFDM symbols #1, #2 and #3) and 240 subcarriers, but leaves an unused portion in the middle for SSS 202 on one symbol (OFDM symbol #2).

[0073] The potential temporal location of an SSB within a half-frame can be determined by the subcarrier spacing, where the network (e.g., gNB) can configure the period of the half-frame for transmitting the SSB. During the half-frame, different SSBs can be transmitted in different spatial directions (i.e., using different beams, spanning the cell's coverage area). Multiple SSBs can be transmitted within the carrier's frequency range.

[0074] To enable a UE to locate a cell upon entering the communication system and to locate new cells while moving within the system, the UE can use the PSS, SSS, and PBCH to obtain the information needed to access the target cell. The PSS and SSS can be periodically transmitted from the access node along with the PBCH on the downlink. After the UE successfully detects the PSS and / or SSS, it acquires knowledge about the target cell's synchronization and physical cell identity (PCI), and then the UE prepares to decode the PBCH. The PBCH carries information needed for further system access, such as obtaining the target cell's System Information Block Type 1 (SIB1). The PSS, SSS, and PBCH can be collectively referred to as the SSB. The SSB can also be called the synchronization and PBCH block or the SS / PBCH block. To cover the entire cell space, the access node can transmit multiple SSBs in different directions (beams) within a so-called SSB burst.

[0075] Furthermore, for downlink measurements, a signal for beam management can be configured, called the Non-Zero Power Channel State Information Reference Signal (NZP-CSI-RS). Therefore, from a downlink perspective, beam management can be performed using both the SSB and CSI-RS signals. The SSB signal can be always-on, and its period is fixed, while the CSI-RS can be configured specifically for a given UE (e.g., with different periods and bandwidths). For example, the CSI-RS can be used to train a narrower beam (a higher-gain beam) by associating it with the SSB beam. For example, the UE can be configured to report... N The highest quality SSB beam, and the network (e.g., gNB) can further configure the UE to report associated with a specific SSB (e.g., SSB#0). M The highest quality CSI-RS (#0...#K) beams. This association can be configured by a network (e.g., gNB). For example, the association can be a spatial association (e.g., quasi-co-location), where the reception of a first signal (e.g., SSB#0) can be used to determine the potential reception of at least one of the second signals (e.g., CSI-RS#0...#K).

[0076] 5G NR also supports UE beamforming. At lower frequencies (e.g., below 6 GHz), the UE may not use beamforming and can operate using omnidirectional beamforming (e.g., equal gain for transmission and / or reception in all directions). However, at higher frequencies (e.g., above 6 GHz), the UE may have one or more antenna panels (or distributed antenna elements) forming one or more beams, such as... Figure 3 As shown. Individual beams that can be formed by the antenna panels used by the UE and / or the individual beams that the UE is able to form can be marked or indexed.

[0077] Figure 3 An example of UE beamforming with a multi-panel configuration is illustrated. Figure 3 In the example, UE 300 includes multiple antenna panels 311, 312, 313, and 314 capable of forming multiple beams 321, 322, 323, 324, 325, and 326 in different directions. It should be noted that the number of antenna panels and the number of beams can also be related to... Figure 3 The differences are as shown. Given beams 321, 322, 323, 324, 325, 326 may be associated with beam identifiers or indices (e.g., 0…K). Alternatively or additionally, given antenna panels 311, 312, 313, 314 may be associated with identifiers or indices. Given antenna panels 311, 312, 313, 314 may include one or more antenna elements.

[0078] For communication purposes, the UE may be configured with at least one beam (identified by a reference signal) as a reference for receiving and / or transmitting data and control channels. For example, the UE may be configured with one or more Physical Downlink Control Channels (PDCCHs) and / or one or more Physical Downlink Shared Channels (PDSCHs), which can be received on one or more downlink beams. Furthermore, the UE may be configured with one or more Physical Uplink Control Channels (PUCCHs) and / or one or more Physical Uplink Shared Channels (PUSCHs), which can be transmitted using the downlink beam as a reference. Additionally, the UE may be able to perform beamforming (i.e., it may be able to form UL and / or DL ​​beams for transmission and reception respectively), or the UE may use omnidirectional transmission and reception.

[0079] Artificial intelligence (AI) and machine learning (ML) for the 5G NR air interface are being explored for various use cases, such as performance, complexity, and potential specification impacts. One example of such use cases is beam management (e.g., beam prediction in the time and / or spatial domains to reduce overhead and latency, and improve beam selection accuracy).

[0080] When a UE is configured to apply an AI or ML model, the UE can request a different model pre-configured by the network, or the network can pre-configure a different model for the UE (e.g., based on the UE's capabilities). However, this presents several challenges, such as: 1) how the network can efficiently update the model on the UE side (e.g., when the UE is configured to update or train an ML model), and 2) what the conditions are for the UE to update the model configured or pre-configured by the network.

[0081] Some example implementations can address the aforementioned challenges by providing methods for collaboratively training ML models in wireless communication networks (e.g., in a beam management context).

[0082] Some example implementations introduce asynchronous ML model updates on the UE side, where the UE can be configured with one or more conditions (e.g., threshold conditions) or criteria to indicate when the UE should update the ML model. For example, the UE can apply reinforcement learning schemes to apply rewards or penalties to update the ML model. For example, the ML model can be used for both uplink and downlink communication parameter prediction (e.g., DL and UL beam prediction).

[0083] However, the following uses the principles and terminology of 5G radio access technology to describe some example embodiments, without limiting the example embodiments to 5G radio access technology.

[0084] Figure 4A The illustration shows an example of a communication system with AI / ML support to which some sample implementations can be applied. Figure 4A The system includes UE 400, RAN node 404, and AI / ML function 407. RAN node 404 can correspond to... Figure 1A and Figure 1B Access node 104. UE 400 can correspond to Figure 1A and Figure 1B UE 100.

[0085] AI / ML function 407 may include an ML model, which consists of an input space, an output space, and training and inference functions. For example, the AI / ML function may include a first K beam prediction function, i.e., prediction of the first 1, first 4, or first K beams in uplink and / or downlink transmissions. The AI / ML function may also include a legacy mode, a non-ML function that allows RAN nodes to switch to legacy mode when AI / ML prediction fails.

[0086] exist Figure 4A In the example, AI / ML function 407 is included in a network entity separate from RAN node 404.

[0087] Interface 411 can be used to exchange information between AI / ML function 407 (which may include the AI / ML model itself) and RAN node 404 (e.g., a base station including communication protocols such as RRC, MAC and / or PHY).

[0088] Interface 412 can be used to exchange information between UE 400 and RAN node 404 and / or between UE 400 and AI / ML function 407 via RAN communication protocol.

[0089] Interface 413 can be used to exchange information between UE 400 and AI / ML function 407 (e.g., using RAN communication protocol as a container, i.e. via interface 412).

[0090] Figure 4B The illustration shows an example of a communication system with AI / ML support to which some sample implementations can be applied. Figure 4B The system includes UE 400, RAN node 404, and AI / ML function 407. RAN node 404 can correspond to... Figure 1A and Figure 1B Access node 104. UE 400 can correspond to Figure 1A and Figure 1B UE 100.

[0091] exist Figure 4B In the example, AI / ML function 407 is embedded in RAN node 404 (e.g., in gNB).

[0092] However, it should be noted that some example embodiments do not depend on any particular type of network deployment architecture. Therefore, some example embodiments are not limited to... Figure 4A and Figure 4B The example shown.

[0093] In one example embodiment, the method for asynchronous ML model updates is performed by a network node (e.g., RAN node 404 or a separate network entity 407) based on updates received from one or more UEs 400.

[0094] A network node can provide a ML model to the UE, which the UE uses to make at least one prediction of at least one communication parameter. The network node can also configure the UE to evaluate the predictive performance of the ML model and train it by calculating rewards or penalties. The network node can also configure the UE to run the ML model. Furthermore, the network node can configure the UE to run or train the ML model by providing instructions to the UE in the ML model configuration or pre-configuration. Finally, the network node can configure the UE to run inference by providing another instruction to the UE in the ML inference configuration.

[0095] Network nodes can select an ML model using predefined criteria (e.g., UE capabilities in antenna configuration and other information that can be used to obtain input parameters (such as measurements) for the ML model), which is then provided to the UE. Furthermore, network nodes can verify updated versions of the ML model received from the UE to determine if the updated version provides performance improvements. If the updated version does not improve performance, the network node can discard the updated version of the ML model.

[0096] Figure 5 A signal flow diagram according to an example embodiment is illustrated. Although Figure 5The diagram shows two UEs, but it should be noted that the number of UEs can be more than two. In other words, one or more UEs can exist. Furthermore, Figure 5 The signaling procedures shown can be expanded and applied according to the actual number of UEs.

[0097] refer to Figure 5 At point 501, the network node sends an indication of support for one or more pre-configured machine learning models used to perform at least one prediction of at least one communication parameter in at least one cell. In other words, the network node may, for example, broadcast or announce support for the pre-configured and used machine learning models in system information. Figures 12-14 The diagram illustrates some examples of ML models.

[0098] This instruction is received by one or more UEs 100, 102, 400, such as the first UE (UE1) 100 and the second UE (UE2) 102. Here, the terms "first UE" and "second UE" are used to distinguish UEs, and they do not necessarily indicate a specific order or specific identifier of the UEs.

[0099] Network nodes can be radio access network nodes such as gNBs (e.g., 104, 404, etc.). Figure 4B (as shown), or a network node can be a separate network entity that includes AI / ML functionality 407 (e.g., Figure 4A (As shown).

[0100] For example, at least one communication parameter may include at least one of the following: beam index or identifier (e.g., SSB or CSI-RS index), beam index quality (e.g., reference signal received power and / or signal-to-interference-plus-noise ratio), or the duration of the quality of the beam index or identifier (e.g., the period during which a particular reference signal is detectable or its quality is above a quality threshold). For example, at least one communication parameter may be a downlink beam or an uplink beam (e.g., a corresponding downlink reference signal of a DL or UL beam that can be used for communication).

[0101] One or more machine learning models may include a set of machine learning model types (e.g., model type A, model type B, ..., model type Y) or IDs for predicting communication parameters based on the UE.

[0102] Support indications can provide information or references to one or more machine learning models, which can be referenced using index values ​​or type values ​​(e.g., model type A). For example, one or more machine learning models may have a predefined structure associated with index values ​​or type values, where indications specifying support for one or more pre-configured machine learning models can include the index value or type value for each of the one or more machine learning models. The index value or type value can indicate at least one of the following: the number of input vectors of the machine learning model (e.g., an artificial neural network), the number of output vectors of the machine learning model, the number of nodes of the machine learning model, the number of layers of the machine learning model, or the type of neural network of the machine learning model.

[0103] In other words, type values ​​can correspond to a specific type of ML model, such as an artificial neural network with X input parameters and Y output variables (i.e., observation space vector X, action space Y), where X and Y can be known based on references to indexes or type values.

[0104] For example, model type A may have specific input parameters, which may include the use of UE location, multiple antenna panels, and measurement capabilities (e.g., accuracy).

[0105] As another example, model type B can have specific input parameters that assume no UE location is used for prediction, the UE does not have multiple antenna panels, measurement capabilities (e.g., accuracy) are low, and so on.

[0106] Network nodes can support multiple model types, up to model type Y, and different models can involve different combinations of UE capabilities.

[0107] At point 502, the first UE sends an indication to the network node specifying the first UE's ability to perform at least one of the following: predict at least one communication parameter, or train at least one of one or more machine learning models. The network node receives the capability indication.

[0108] For example, the first UE may indicate during the connection establishment phase, or upon entering at least one cell, or during the connected state that it is able to predict at least one communication parameter of one or more ML model types announced by the network node, and / or indicate that the first UE supports (or does not support) online training of a given ML model type announced by the network node.

[0109] The first UE may additionally indicate to the network node at least one parameter indicating the communication capabilities of the first UE, wherein the at least one parameter may include at least one of the following: the number of antenna panels(s) of the first UE, the relative positions of the antenna panels to each other, the number of beams of each antenna panel (e.g., the (maximum) number of beams that the first UE can form), the beam coverage number of each antenna panel, a beamforming codebook defining how the first UE forms a beam mesh, etc.

[0110] The first UE may send a capability indication in response to receiving an indication specifying support for one or more pre-configured machine learning models. Alternatively, the first UE may send the capability indication to the network node before receiving the indication specifying support for one or more pre-configured machine learning models (i.e., 502 may be executed before 501). In the latter case, in response to receiving the capability indication from the first UE, the network node may send an indication specifying support for one or more pre-configured machine learning models to the first UE via dedicated signaling.

[0111] At point 503, the second UE sends an instruction to the network node instructing the second UE to perform at least one of the following: predict at least one communication parameter, or train at least one of one or more machine learning models online. The network node receives the capability instruction.

[0112] For example, the second UE may indicate during the connection establishment phase, or upon entering at least one cell, or during the connected state that it is able to predict at least one communication parameter of one or more ML model types announced by the network node, and / or indicate that the second UE supports (or does not support) online training of a given ML model type announced by the network node.

[0113] The second UE may additionally indicate to the network node at least one parameter indicating the communication capabilities of the second UE, wherein the at least one parameter may include at least one of the following: the number of antenna panels(s) of the second UE, the relative positions of the antenna panels to each other, the number of beams of each antenna panel (e.g., the maximum number of beams that the second UE can form), the beam coverage number of each antenna panel, a beamforming codebook defining how the second UE forms a beam mesh, etc.

[0114] The second UE may send a capability indication in response to receiving an indication specifying support for one or more pre-configured machine learning models. Alternatively, the second UE may send the capability indication to the network node before receiving the indication specifying support for one or more pre-configured machine learning models (i.e., 503 may be executed before 501). In the latter case, in response to receiving the capability indication from the second UE, the network node may send an indication specifying support for one or more pre-configured machine learning models to the second UE via dedicated signaling.

[0115] At point 504, the network node determines the machine learning model to be sent to one or more machine learning models in the first UE based on the capability indication of the first UE. Furthermore, the network node determines the machine learning model to be sent to one or more machine learning models in the second UE based on the capability indication of the second UE.

[0116] In this example, the network node can determine that the first UE and the second UE have the same or similar capabilities (regarding the ML model to be applied or pre-configured), which allows the network node to provide the same type of ML model (e.g., model type A or ID A) to both UEs. Here, the machine learning model provided to the first UE can be referred to as the first machine learning model, and the machine learning model provided to the second UE can be referred to as the second machine learning model (although the first and second machine learning models can be of the same type). The first and second machine learning models can be represented as NN_a1 (i.e., state 1 of ML model type A).

[0117] At point 505, the network node sends a first machine learning model to the first UE, the first machine learning model being used to perform at least one prediction of at least one communication parameter in at least one cell. The first UE receives the first machine learning model. The first machine learning model may be associated with at least one of the following: a timestamp or identifier (e.g., date, time reference, time, label, identifier, or index of the ML model state), the identifier of the first UE, the version number of the first machine learning model, or any identifier or time parameter that can be used to distinguish different versions of the first machine learning model.

[0118] At point 506, the network node sends a second machine learning model to the second UE, which is used to perform at least one prediction of at least one communication parameter in at least one cell. The second UE receives the second machine learning model. The second machine learning model may be associated with at least one of the following: a timestamp or identifier (e.g., date, time reference, time, label, identifier, or index of the ML model state), the identifier of the second UE, the version number of the second machine learning model, or any identifier or time parameter that can be used to distinguish different versions of the second machine learning model.

[0119] At point 507, the network node sends a configuration to the first UE, which indicates at least one condition for determining whether to update the first machine learning model. The first UE receives the configuration.

[0120] At point 508, the network node sends a configuration to the second UE, which indicates at least one condition for determining whether to update the second machine learning model. The second UE receives the configuration.

[0121] For example, at least one condition may include at least one of the following: the UE performs at least one prediction of at least one communication parameter, receives configuration of at least one communication parameter to be used for communication with network nodes (e.g., for one or more channels such as PDSCH, PDCCH, PUSCH or PUCCH), monitors the prediction quality or utilization of at least one communication parameter for a predefined duration, or the amount of correlation between at least one prediction and at least one observation of at least one communication parameter.

[0122] At point 509, the first UE uses a first machine learning model to perform at least one prediction of at least one communication parameter in at least one cell. The at least one prediction may include at least one predicted value for at least one communication parameter.

[0123] For example, the first machine learning model can be configured for beam prediction in both the uplink and downlink. In this case, at least one prediction can be performed by providing input information to the first machine learning model, which includes at least one of the following: an identifier of at least one downlink beam, an identifier of at least one uplink beam, a reference signal received power value measured on at least one downlink beam, a reference signal received power value measured on at least one uplink beam, a threshold for the reference signal received power, a threshold for the signal-to-interference-plus-noise ratio (SINR), or one or more antenna panel indices.

[0124] At least one prediction may include at least one of the following: a prediction identifier for at least one downlink beam, a predicted reference signal received power value for at least one downlink beam, a predicted signal-to-interference-plus-noise ratio (SIR) for at least one downlink beam, a prediction identifier for at least one uplink beam, a predicted reference signal received power value for at least one uplink beam, a predicted SIR for at least one uplink beam, a quality threshold for the reference signal received power, or a quality threshold for the SIR.

[0125] At point 510, the first UE determines whether to update the first machine learning model based on at least one condition.

[0126] For example, if the first UE performs at least one prediction of at least one communication parameter (e.g., downlink beam and / or uplink beam), the first UE can determine that it needs to update the first machine learning model.

[0127] As another example, if a network node configures a predicted value (e.g., a predicted output) for at least one communication parameter for communication between the first UE and the network node (e.g., for beaming for at least one of the following: downlink control information reception, downlink data reception, uplink control information reception, or uplink data reception), the first UE may determine whether to update the first machine learning model, or alternatively begin monitoring the prediction quality of the first machine learning model. In other words, if the network node applies prediction (e.g., predicted beaming) in its actual configuration, the first UE may update the first machine learning model or initiate monitoring of the prediction quality.

[0128] As another example, if the first UE monitors the prediction quality of at least one communication parameter or the utilization rate of the predicted value of at least one communication parameter (e.g., within a predefined duration or when that duration is observed to be within the range limits of the predicted duration), the first UE can determine whether to update the first machine learning model. In other words, the first UE can perform communication and compare it with the prediction to determine whether to update the first machine learning model.

[0129] As another example, if the predicted value of at least one communication parameter (e.g., communication quality) is related to the actual observed (measured) value of at least one communication parameter, the first UE can determine to update the first machine learning model by rewarding the first machine learning model.

[0130] As another example, if the predicted value of at least one communication parameter (e.g., communication quality) is not related to the actual observed (measured) value of at least one communication parameter, the first UE may determine to update the first machine learning model by penalty.

[0131] At point 511, based on the determination that the first machine learning model needs updating (i.e., if at least one condition is met), the first UE obtains an updated version of the first machine learning model by updating the first machine learning model based on at least one prediction performed at the first UE. Updating can refer to training the first machine learning model. The first UE can perform prediction and updating once or multiple times (i.e., perform one or more training iterations). The updated version of the first machine learning model can be represented as NN_2a-UE1 (i.e., state 2 of ML model type A associated with the first UE). The network node can configure how many training iterations and / or updates the first UE should perform.

[0132] For example, the first UE can determine whether at least one prediction performed at the first UE is related to at least one observation of at least one communication parameter (i.e., an actual measurement or an observed or determined communication quality). If at least one prediction is related to at least one observation, the first UE can update the first machine learning model by rewarding it. If at least one prediction is not related to at least one observation, the first UE can update the first machine learning model by penalizing it.

[0133] If the first UE cannot determine whether at least one prediction is related to at least one observation (i.e., related or unrelated) (e.g., due to a link outage or failure or handover), the first UE may not perform any updates to the first machine learning model, or the first UE may avoid updating the first machine learning model until it is able to determine whether at least one prediction is related to at least one observation (i.e., related or unrelated).

[0134] At point 512, the first UE sends an updated version of the first machine learning model to the network node. The network node receives the updated version of the first machine learning model.

[0135] For example, an updated version of the first machine learning model may be sent to the network node based on the completion of a predefined number of training iterations for updating the first machine learning model, or when or after the connection between the first UE and the network node is terminated. The first UE may also send an indication of the number of training iterations performed to update the first machine learning model along with the updated version of the first machine learning model to the network node.

[0136] The first UE can also indicate to the network node whether the first machine learning model has been updated.

[0137] At point 513, the second UE uses a second machine learning model to perform at least one prediction of at least one communication parameter in at least one cell. The at least one prediction may include at least one predicted value for the at least one communication parameter.

[0138] At point 514, the second UE determines whether to update the second machine learning model based on at least one condition.

[0139] At point 515, based on the determination that the second machine learning model needs updating (i.e., if at least one condition is met), the second UE obtains an updated version of the second machine learning model by updating the second machine learning model based on at least one prediction performed at the second UE. Updating can refer to training the second machine learning model. The second UE can perform prediction and updating once or multiple times (i.e., perform one or more training iterations). The updated version of the second machine learning model can be represented as NN_2a-UE2 (i.e., state 2 of ML model type A associated with the second UE). The network node can configure how many training iterations and / or updates the second UE should perform.

[0140] For example, the second UE can determine whether at least one prediction performed at the second UE is related to at least one observation (i.e., actual measurement) of at least one communication parameter. If at least one prediction is related to at least one observation, the second UE can update the second machine learning model by rewarding it. If at least one prediction is not related to at least one observation, the second UE can update the second machine learning model by penalizing it.

[0141] If the second UE cannot determine whether at least one prediction is related to at least one observation (i.e., related or unrelated) (e.g., due to a link outage or failure or switchover), the second UE may not perform any updates to the second machine learning model, or the second UE may avoid updating the second machine learning model until it is able to determine whether at least one prediction is related to at least one observation (i.e., related or unrelated).

[0142] At point 516, the second UE sends an updated version of the second machine learning model to the network node. The network node receives the updated version of the second machine learning model.

[0143] For example, an updated version of the second machine learning model may be sent to the network node based on the completion of a predefined number of training iterations for updating the second machine learning model, or when or after the connection between the second UE and the network node terminates (e.g., when transitioning from a connected state, and / or when leaving at least one cell or tracking area (cell set)). The second UE may also send an indication of the number of training iterations performed for updating the second machine learning model to the network node along with the updated version of the second machine learning model.

[0144] The second UE can also indicate to the network node whether the second machine learning model has been updated.

[0145] At point 517, a network node can obtain a merged machine learning model by merging (combining) an updated version of the first machine learning model (NN_a2-UE1) with an updated version of the second machine learning model (NN_a2-UE2). The merged machine learning model can be represented as NN_a2 (i.e., state 2 of ML model type A). Alternatively, a network node can apply the updates of NN_a2-UE2 to NN_a2-UE1 to form NN_a2.

[0146] In one example, network nodes can merge ML models by performing an ensemble approach, where network nodes can combine features and functionalities of ML models (e.g., artificial neural network models).

[0147] In another example, network nodes can merge ML models by concatenating ML models trained by different UEs. In this case, network nodes can take different inputs (from ML models provided by the UE) and concatenate them into the same ML model. However, the result of concatenating datasets can have more dimensions than the original dataset.

[0148] In another example, network nodes can merge ML models by averaging ML models. For example, network nodes can average ML models and use the average as a new model. In this example, network nodes can take a simple average or a weighted average of the ML models. In the case of a weighted average, network nodes can assign different weights to different ML models based on their performance. For example, ML model averaging could refer to per-node (neuron) averaging, where the weights of the input vectors applied to the corresponding nodes of one or more ML models are averaged as described herein.

[0149] At point 518, a network node can verify at least one of the following: the merged machine learning model, an updated version of the first machine learning model, or an updated version of the second machine learning model.

[0150] For example, network nodes can validate a merged machine learning model by determining whether the merged machine learning model (NN_a2) provides a performance improvement compared to the previous version of the machine learning model (NN_a1).

[0151] Alternatively, network nodes can validate their updated versions before merging the first and second machine learning models. In this case, network nodes can determine whether the updated version of the first machine learning model (NN_a2-UE1) provides a performance improvement compared to the previous version of the first machine learning model (NN_a1), and whether the updated version of the second machine learning model (NN_a2-UE2) provides a performance improvement compared to the previous version of the second machine learning model (NN_a1).

[0152] At point 519, if the merged machine learning model (or an updated version of the first and second machine learning models) provides performance improvements, the network node may send the merged machine learning model to the first and second UEs to perform at least one subsequent prediction of at least one communication parameter in at least one cell. Alternatively or additionally, the merged machine learning model may be sent to one or more other UEs besides the first and second UEs.

[0153] Alternatively, if the merged machine learning model does not provide performance improvements, network nodes can discard the merged machine learning model and revert to the previous version.

[0154] In some examples, not all UEs are capable of prediction and online training. Functions of UE1 and UE2 (e.g., 509-512 and 513-516) can be executed simultaneously or in combination. Figure 5 The different execution orders shown.

[0155] Figure 6 The diagram illustrates a signal flow graph according to an example embodiment. In this example embodiment, the network node predicts UE based on communication quality monitoring and instructs the UE to update the machine learning model used (e.g., referring to a specific prediction instance).

[0156] although Figure 6 The example shows one UE, but it should be noted that the number of UEs can be more than one. In other words, one or more UEs can exist. Furthermore, Figure 6 The signaling procedures shown can be expanded and applied according to the actual number of UEs.

[0157] refer to Figure 6 At point 601, the network node sends an indication of support for one or more pre-configured machine learning models used to perform at least one prediction of at least one communication parameter in at least one cell. In other words, the network node may, for example, broadcast or announce support for the pre-configured and used one or more machine learning models in system information. Figures 12-14 The diagram illustrates some examples of ML models. This instruction is received by UE 100 and 400.

[0158] Network nodes can be radio access network nodes such as gNBs (e.g., 104, 404, etc.). Figure 4B (as shown), or a network node can be a separate network entity that includes AI / ML functionality 407 (e.g., Figure 4A (As shown).

[0159] For example, at least one communication parameter may include at least one of the following: beam index or identifier (e.g., SSB or CSI-RS index), beam index quality (e.g., reference signal received power and / or signal-to-interference-plus-noise ratio), or the duration of the quality of the beam index or identifier (e.g., the period during which a particular reference signal is detectable or its quality is above a quality threshold). For example, at least one communication parameter may be a downlink beam or an uplink beam (e.g., a corresponding downlink reference signal of a DL or UL beam that can be used for communication).

[0160] One or more machine learning models may include a set of machine learning model types (e.g., model type A, model type B, ..., model type Y) for predicting communication parameters based on the UE.

[0161] Support indications can provide information or references to one or more machine learning models, which can be referenced using index values ​​or type values ​​(e.g., model type A). For example, one or more machine learning models may have a predefined structure associated with index values, identifier values, or type values, where indications specifying support for one or more pre-configured machine learning models can include the index value or type value for each of the one or more machine learning models. The index value or type value can indicate at least one of the following: the number of input vectors of the machine learning model (e.g., an artificial neural network), the number of output vectors of the machine learning model, the number of nodes of the machine learning model, the number of layers of the machine learning model, or the type of neural network of the machine learning model.

[0162] In other words, type values ​​can correspond to a specific type of ML model, such as an artificial neural network with X input parameters and Y output variables (i.e., observation space vector X, action space Y), where X and Y can be known based on references to indexes or type values.

[0163] For example, model type A may have specific input parameters, which may include the use of UE location, multiple antenna panels, and measurement capabilities (e.g., accuracy).

[0164] As another example, model type B can have specific input parameters that assume no UE location is used for prediction, the UE does not have multiple antenna panels, measurement capabilities (e.g., accuracy) are low, and so on.

[0165] Network nodes can support multiple model types, up to model type Y, and different models can involve different combinations of UE capabilities.

[0166] At point 602, the UE sends an indication to the network node specifying the UE's ability to perform at least one of the following: predict at least one communication parameter, or train at least one of one or more machine learning models. The network node receives the capability indication.

[0167] For example, the UE may indicate during the connection establishment phase, or upon entering at least one cell, or during the connected state that it is able to predict at least one communication parameter of one or more ML model types advertised by the network node, and / or indicate that the UE supports (or does not support) online training of a given ML model type advertised by the network node.

[0168] The UE may additionally indicate to the network node at least one parameter that specifies the UE’s communication capabilities, wherein at least one parameter may include at least one of the following: the number of antenna panels(s) of the UE, the relative positions of the antenna panels to each other, the number of beams of each antenna panel (e.g., the maximum number of beams that the UE can form), the beam coverage number of each antenna panel, a beamforming codebook that defines how the UE forms a beam grid, etc.

[0169] The UE may send a capability indication in response to receiving an indication specifying support for one or more pre-configured machine learning models. Alternatively, the UE may send the capability indication to the network node before receiving the indication specifying support for one or more pre-configured machine learning models (i.e., step 602 may be performed before step 601). In the latter case, in response to receiving the capability indication from the UE, the network node may send an indication specifying support for one or more pre-configured machine learning models to the UE via dedicated signaling.

[0170] At 603, the network node determines the machine learning model to be sent to the UE from one or more machine learning models based on the UE's capabilities.

[0171] At position 604, the network node sends a machine learning model to the UE, which is used to perform at least one prediction of at least one communication parameter in at least one cell. The UE receives the machine learning model.

[0172] A machine learning model can be associated with at least one of the following: a timestamp or identifier (e.g., a date, time reference, time, label, identifier, or index of the ML model state), an identifier of the UE, a version number of the machine learning model, or any identifier or time parameter that can be used to distinguish different versions of the machine learning model.

[0173] At point 605, the network node sends a configuration to the UE for predicting at least one communication parameter. The UE receives this configuration.

[0174] For example, the configuration may indicate at least one of the following: one or more thresholds, one or more functions, and / or one or more parameters for predicting at least one communication parameter and / or estimating prediction quality. In one example, these may include beam quality thresholds for determining the correlation and / or decorrelation of the prediction (with respect to observations), input parameters for the prediction (e.g., L1-RSRP measurements), and / or output parameters, such as the number K of the first K beams predicted (in terms of quality). RSRP is an abbreviation for Reference Signal Received Power.

[0175] At 606, based on this configuration, the UE uses a machine learning model to perform at least one prediction of at least one communication parameter in at least one cell. The at least one prediction may include at least one predicted value for at least one communication parameter.

[0176] The UE can tag or index at least one prediction of at least one communication parameter, and store at least one prediction along with the tag or index in its internal memory. The stored at least one prediction can be referenced by a network node using an update command, and / or used by the UE to update a machine learning model.

[0177] For example, a machine learning model can be configured for beam prediction in both uplink and downlink. In this case, at least one prediction can be performed by providing input information to the machine learning model, which includes at least one of the following: an identifier of at least one downlink beam, an identifier of at least one uplink beam, a reference signal received power value measured on at least one downlink beam, a reference signal received power value measured on at least one uplink beam, a threshold for the reference signal received power, a threshold for the signal-to-interference-plus-noise ratio (SINR), or one or more antenna panel indices.

[0178] At least one prediction may include at least one of the following: a prediction identifier for at least one downlink beam, a predicted reference signal received power value for at least one downlink beam, a predicted signal-to-interference-plus-noise ratio (SIR) for at least one downlink beam, a prediction identifier for at least one uplink beam, a predicted reference signal received power value for at least one uplink beam, a predicted SIR for at least one uplink beam, a quality threshold for the reference signal received power, or a quality threshold for the SIR.

[0179] At point 607, the UE sends information indicating at least one prediction performed at the UE, along with a tag or index, to the network node. In other words, the UE reports the prediction output to the network node according to the configuration provided by the network node. The network node receives this information.

[0180] At point 608, the network node monitors at least one prediction of the UE based on the communication quality between the network node and the UE. In other words, the network node can monitor communication based on at least one prediction reported from the UE, and therefore the network node can observe the actual performance related to at least one communication parameter and compare it with at least one prediction.

[0181] For example, a network node can determine whether at least one prediction is related to at least one observed (measured) value of at least one communication parameter.

[0182] At point 609, based on monitoring, the network node sends an instruction to the UE to update the machine learning model (e.g., referring to a specific prediction instance). The UE receives this instruction.

[0183] For example, if at least one prediction (e.g., communication quality) is related to at least one observation (such as that observed by a network node), the indication can instruct the machine learning model to be updated by rewarding the machine learning model.

[0184] As another example, if at least one prediction (e.g., communication quality) is not correlated with at least one observation (such as that observed by a network node), the indication can instruct the machine learning model to be updated by penalizing it.

[0185] If a network node cannot determine whether at least one prediction (e.g., communication quality) is related to at least one observation (i.e., related or unrelated) (e.g., due to a link drop or failure or handover), the network node may avoid sending an instruction to update the machine learning model until it is able to determine whether at least one prediction is related to at least one observation, or the network node may send an explicit instruction to the UE not to update the machine learning model.

[0186] At 610, the UE determines to update the machine learning model based on the instruction received at 609.

[0187] At point 611, based on the determination that the machine learning model needs updating, the UE obtains an updated version of the machine learning model by updating the model based on at least one prediction stored at the UE. Updating can refer to training the machine learning model. The UE can perform prediction and updating once or multiple times (i.e., perform one or more training iterations). The network node can configure how many training iterations and / or updates the UE should perform.

[0188] At point 612, the UE sends an updated version of the machine learning model to the network node. The network node receives the updated version of the machine learning model.

[0189] For example, an updated version of the machine learning model can be sent to the network node based on the number of training iterations used to update the machine learning model reaching a predefined number, or when or after the connection between the UE and the network node is terminated (e.g., when transitioning out of a connected state, and / or when leaving at least one cell or tracking area (cell set)). The UE can also send an indication of the number of training iterations performed to update the machine learning model along with the machine learning model to the network node.

[0190] The UE can also indicate to network nodes whether the machine learning model has been updated.

[0191] At point 613, network nodes can verify updated versions of the machine learning model. In this case, the network node can determine whether the updated version of the machine learning model provides a performance improvement compared to the previous version.

[0192] If an updated version of the machine learning model does not provide performance improvements, network nodes can discard the updated version and revert to the previous version.

[0193] At point 614, if an updated version of the machine learning model provides performance improvements, the network node can obtain a merged machine learning model by merging the updated version of the machine learning model with one or more other machine learning models received from one or more other UEs.

[0194] As an alternative to 613, network nodes can validate merged machine learning models instead of updated versions of machine learning models provided by the UE.

[0195] At point 615, if an updated version of the machine learning model (or a merged machine learning model) provides performance improvements, the network node may send the merged machine learning model or the updated version of the machine learning model to the UE to perform at least one subsequent prediction of at least one communication parameter in at least one cell. Alternatively, the merged machine learning model or the updated version of the machine learning model may be sent to one or more other UEs.

[0196] Figure 7 The illustration shows a flowchart of a method performed by device 1500 according to an example embodiment. For example, device 1500 may be, or include, user equipment 100, 102, 400, or be included in, user equipment 100, 102, 400.

[0197] refer to Figure 7In box 701, the device receives a machine learning model from a network node, which is used to perform at least one prediction of at least one communication parameter in at least one cell. Some examples of ML models are shown in... Figures 12-14 The illustrations are shown in the middle. However, ML models are not limited to these examples.

[0198] A machine learning model can be associated with at least one of the following: a timestamp (e.g., the current date), an identifier of the UE, a version number of the machine learning model, or any identifier or time parameter that can be used to distinguish different versions of the machine learning model.

[0199] Network nodes can be radio access network nodes such as gNBs (e.g., 104, 404, etc.). Figure 4B (as shown), or a network node can be a separate network entity that includes AI / ML functionality 407 (e.g., Figure 4A (As shown).

[0200] In box 702, the device receives from a network node at least one of the following: an instruction to update the machine learning model, or a configuration indicating at least one condition for determining whether to update the machine learning model.

[0201] In box 703, the device uses a machine learning model to perform at least one prediction of at least one communication parameter in at least one cell.

[0202] In box 704, the device determines whether to update the machine learning model based on the instruction or at least one condition;

[0203] In box 705, the device, based on determining that the machine learning model needs updating, obtains an updated version of the machine learning model by updating the machine learning model based on at least one prediction; and

[0204] In box 706, the device sends an updated version of the machine learning model to the network nodes.

[0205] For example, at least one communication parameter may include at least one of the following: beam index, beam index quality, or beam index quality duration.

[0206] For example, at least one condition may include at least one of the following: performing at least one prediction of at least one communication parameter, receiving configuration of at least one communication parameter to be used for communication with network nodes (e.g., for one or more channels such as PDSCH, PDCCH, PUSCH, or PUCCH), monitoring the prediction quality or utilization of at least one communication parameter for a predefined duration, or the amount of correlation between at least one prediction and at least one observation of at least one communication parameter.

[0207] The device can: receive from a network node an instruction indicating support for one or more pre-configured machine learning models for performing at least one prediction of at least one communication parameter in at least one cell; and send to the network node an instruction indicating the ability to: predict at least one communication parameter, or train at least one of the one or more machine learning models.

[0208] One or more machine learning models have a predefined structure associated with index values ​​or type values, wherein the indications for support for the pre-configured one or more machine learning models may include the index value or type value of each of the one or more machine learning models.

[0209] In one example, the device can: determine whether at least one prediction is related to at least one observation of at least one communication parameter; and update the machine learning model by rewarding the machine learning model based on the determination that at least one prediction is related to at least one observation.

[0210] In another example, the device can: determine whether at least one prediction is related to at least one observation of at least one communication parameter; and update the machine learning model by penalizing the machine learning model based on the determination that at least one prediction is not related to at least one observation.

[0211] In another example, the device may: attempt to determine whether at least one prediction is related to at least one observation of at least one communication parameter; and avoid updating the machine learning model until it is possible to determine whether at least one prediction is related to at least one observation, based on the inability to determine whether at least one prediction is related to at least one observation.

[0212] The device can: receive from a network node a configuration for predicting at least one communication parameter, wherein at least one prediction is performed based on the configuration; and send to the network node information indicating at least one prediction.

[0213] The device can: tag or index at least one prediction of at least one communication parameter; and store at least one prediction together with a label or index, wherein a machine learning model can be updated based on the stored at least one prediction.

[0214] An updated version of the machine learning model can be sent to the network node when the training iterations used to update the machine learning model reach a predefined number, or when or after the connection between the device and the network node is terminated.

[0215] The device can send instructions to network nodes specifying the number of training iterations to be performed to update the machine learning model.

[0216] The machine learning model can be configured for beam prediction in uplink and downlink, wherein at least one prediction may include at least one of the following: a prediction identifier for at least one downlink beam, a predicted reference signal received power value for at least one downlink beam, a predicted signal-to-interference-plus-noise ratio (SINNR) for at least one downlink beam, a prediction identifier for at least one uplink beam, a predicted reference signal received power value for at least one uplink beam, a predicted SINNR for at least one uplink beam, a quality threshold for the reference signal received power, or a quality threshold for the SINNR.

[0217] In one example, at least one prediction may include multiple predictions in a time series.

[0218] At least one prediction can be performed by providing input information to a machine learning model, which includes at least one of the following: a threshold for the received power of a reference signal or a threshold for the signal-to-interference-plus-noise ratio.

[0219] Figure 8 The illustration shows a flowchart of a method performed by device 1500 according to an example embodiment. For example, device 1500 may be, or include, user equipment 100, 102, 400, or be included in, user equipment 100, 102, 400.

[0220] refer to Figure 8 In box 801, the device receives a machine learning model from a network node, which is used to perform at least one prediction of at least one communication parameter in at least one cell. Some examples of ML models are shown in... Figures 12-14 The illustrations are shown in the middle. However, ML models are not limited to these examples.

[0221] Network nodes can be radio access network nodes such as gNBs (e.g., 104, 404, etc.). Figure 4B (as shown), or a network node can be a separate network entity that includes AI / ML functionality 407 (e.g., Figure 4A (As shown).

[0222] In box 802, the device receives from a network node a configuration indicating at least one condition for determining whether to update the machine learning model.

[0223] For example, at least one condition may include the amount of correlation between at least one prediction and at least one observation of at least one communication parameter. If at least one prediction is correlated with at least one observation of at least one communication parameter, the device may update the machine learning model by rewarding the machine learning model. If at least one prediction is not correlated with at least one observation of at least one communication parameter, the device may update the machine learning model by penalizing the machine learning model.

[0224] In box 803, the device uses a machine learning model to perform at least one prediction of at least one communication parameter in at least one cell.

[0225] In block 804, the device attempts to determine whether at least one prediction is related to at least one observation of at least one communication parameter.

[0226] In box 805, based on the inability (e.g., due to link downtime or failure or switching) to determine whether at least one prediction is related to at least one observation (box 804: No), the device avoids updating the machine learning model until it is possible to determine whether at least one prediction is related to at least one observation.

[0227] Alternatively, in block 806, based on the ability to determine whether at least one prediction is related to at least one observation (block 804: yes), the device determines whether at least one prediction is related to at least one observation of at least one communication parameter.

[0228] In one example, the correlation between at least one prediction and at least one observation of at least one communication parameter can refer to the quality of the reference signal and / or the difference between at least one prediction (predicted value) and at least one observation (measured value). This difference can be determined based on a threshold between the absolute values ​​of the predicted value and the observed value (measured value).

[0229] For example, if the absolute value of the difference between the predicted value and the observed value (measured value) is less than or equal to a threshold, this indicates that the predicted value is correlated with the observed value. For example, if the observed value is within ±X of the predicted value, then the predicted value can be determined to be correlated. The value X (i.e., the threshold) can be configured by the network nodes.

[0230] As another example, if the absolute value of the difference between the predicted value and the observed value (measured value) is higher than a threshold, this can indicate that the predicted value and the observed value are uncorrelated, or that there is insufficient correlation between them. For example, if the observed value is not within ±X of the predicted value, it can be determined that the predicted value is uncorrelated. The value X (i.e., the threshold) can be configured by the network nodes.

[0231] In box 807, based on determining that at least one prediction is related to at least one observation (box 806: yes), the device updates the machine learning model by rewarding the machine learning model.

[0232] Alternatively, in box 808, based on determining that at least one prediction is unrelated to at least one observation (box 806: No), the device updates the machine learning model by penalizing the machine learning model.

[0233] In box 809, following box 807 or box 808, the device sends an updated version of the machine learning model to the network nodes.

[0234] Figure 9 A flowchart illustrating a method performed by apparatus 1600 according to an example embodiment is shown. For example, apparatus 1600 may be a network node, or include a network node, or be included in a network node. The network node may be a radio access network node 104, 404, such as a gNB (e.g., as...). Figure 4B (as shown), or a network node can be a separate network entity that includes AI / ML functionality 407 (e.g., Figure 4A (As shown).

[0235] refer to Figure 9 In box 901, the device sends a machine learning model to at least one user equipment 100, 102, 400, the machine learning model being used to perform at least one prediction of at least one communication parameter in at least one cell.

[0236] A machine learning model can be associated with at least one of the following: a timestamp (e.g., the current date), an identifier of the UE, a version number of the machine learning model, or any identifier or time parameter that can be used to distinguish different versions of the machine learning model.

[0237] In block 902, the device sends at least one of the following to at least one user equipment: an instruction to update the machine learning model, or a configuration indicating at least one condition for determining whether to update the machine learning model.

[0238] In box 903, the device receives an updated version of the machine learning model from at least one user device.

[0239] For example, at least one communication parameter may include at least one of the following: beam index, beam index quality, or beam index quality duration.

[0240] At least one condition may include at least one of the following: performing at least one prediction of at least one communication parameter, receiving configuration of at least one communication parameter to be used for communication with network nodes, monitoring the prediction quality or utilization of at least one communication parameter for a predefined duration, or the correlation between at least one prediction and at least one observation of at least one communication parameter.

[0241] The device can: send an instruction indicating support for one or more pre-configured machine learning models for performing at least one prediction of at least one communication parameter in at least one cell; receive from at least one user equipment an instruction indicating the capability to: predict at least one communication parameter, or train at least one of one or more machine learning models; and determine, based on the capability, the machine learning model to be sent to the at least one user equipment.

[0242] The device can: send a configuration for predicting at least one communication parameter to at least one user equipment; receive information from at least one user equipment indicating at least one prediction of at least one communication parameter; monitor at least one prediction of at least one user equipment based on the communication quality between the device and at least one user equipment; and send an instruction to at least one user equipment to update a machine learning model based on the monitoring.

[0243] In one example, the device may: determine whether at least one prediction is related to at least one observation of at least one communication parameter; and based on the determination that at least one prediction is related to at least one observation, send an instruction to at least one user equipment to update the machine learning model by rewarding the machine learning model.

[0244] In another example, the device may: determine whether at least one prediction is related to at least one observation of at least one communication parameter; and based on the determination that at least one prediction is not related to at least one observation, send an instruction to at least one user equipment to update the machine learning model by penalizing the machine learning model.

[0245] In another example, the device may: attempt to determine whether at least one prediction is related to at least one observation of at least one communication parameter; and based on the inability to determine whether at least one prediction is related to at least one observation, avoid sending an instruction to at least one user device to update the machine learning model until it is possible to determine whether at least one prediction is related to at least one observation.

[0246] In another example, the device may: attempt to determine whether at least one prediction is related to at least one observation of at least one communication parameter; and, based on the inability to determine whether at least one prediction is related to at least one observation, send an instruction to at least one user device not to update the machine learning model.

[0247] In one example, the device may: determine whether an updated version of the machine learning model provides a performance improvement compared to a previous version of the machine learning model; and based on the determination that the updated version provides a performance improvement, send the updated version of the machine learning model to one or more other user equipment to perform at least one subsequent prediction of at least one communication parameter in at least one cell.

[0248] In another example, the device can: determine whether an updated version of the machine learning model provides a performance improvement compared to a previous version of the machine learning model; and discard the updated version of the machine learning model based on the determination that the updated version does not provide a performance improvement.

[0249] The device can obtain a merged machine learning model by merging an updated version of the machine learning model with one or more other machine learning models received from one or more other user devices.

[0250] Figure 10 A flowchart illustrating a method performed by apparatus 1600 according to an example embodiment is shown. For example, apparatus 1600 may be a network node, or include a network node, or be included in a network node. The network node may be a radio access network node 104, 404, such as a gNB (e.g., as...). Figure 4B (as shown), or a network node can be a separate network entity that includes AI / ML functionality 407 (e.g., Figure 4A (As shown).

[0251] refer to Figure 10 In box 1001, the device sends a machine learning model to at least one user equipment, the machine learning model being used to perform at least one prediction of at least one communication parameter in at least one cell. Some examples of ML models are shown in... Figures 12-14 The illustrations are shown in the middle. However, ML models are not limited to these examples.

[0252] In block 1002, the device sends a configuration for predicting at least one communication parameter to at least one user equipment.

[0253] In block 1003, the device receives from at least one user equipment information indicating at least one prediction of at least one communication parameter to be performed at at least one user equipment.

[0254] In box 1004, the device attempts to determine whether at least one prediction is related to at least one observation of at least one communication parameter.

[0255] In block 1005, based on the inability (e.g., due to link downtime or failure or switching) to determine whether at least one prediction is related to at least one observation (block 1004: No), the device avoids sending an instruction to at least one user equipment to update the machine learning model until it can be determined whether at least one prediction is related to at least one observation. Alternatively, the device may send an instruction to at least one user equipment not to update the machine learning model.

[0256] In block 1006, based on the ability to determine whether at least one prediction is related to at least one observation (block 1004: yes), the device determines whether at least one prediction is related to at least one observation of at least one communication parameter.

[0257] In one example, the correlation between at least one prediction and at least one observation of at least one communication parameter can refer to the quality of the reference signal and / or the difference between at least one prediction (predicted value) and at least one observation (measured value). This difference can be determined based on a threshold between the absolute values ​​of the predicted value and the observed value (measured value).

[0258] For example, if the difference between the absolute values ​​of the predicted and observed values ​​(measured values) is below a threshold, this can indicate that the predicted values ​​are correlated with the observed values.

[0259] As another example, if the difference between the absolute values ​​of the predicted and observed values ​​(measured values) is higher than a threshold, this may indicate that the predicted values ​​are not correlated with the observed values, or that there is not enough correlation between the predicted and observed values.

[0260] In box 1007, based on determining that at least one prediction is related to at least one observation (box 1006: yes), the device sends an instruction to at least one user equipment to update the machine learning model by rewarding the machine learning model.

[0261] Alternatively, in box 1008, based on determining that at least one prediction is unrelated to at least one observation (box 1006: No), the device sends an instruction to at least one user equipment to update the machine learning model by penalizing the machine learning model.

[0262] In box 1009, following box 1007 or box 1008, the device receives an updated version of the machine learning model from at least one user device.

[0263] Figure 11 A flowchart illustrating a method performed by apparatus 1600 according to an example embodiment is shown. For example, apparatus 1600 may be a network node, or include a network node, or be included in a network node. The network node may be a radio access network node 104, 404, such as a gNB (e.g., as...). Figure 4B(as shown), or a network node can be a separate network entity that includes AI / ML functionality 407 (e.g., Figure 4A (As shown).

[0264] refer to Figure 11 In box 1101, the device sends a machine learning model to at least one user equipment 100, 102, 400, which is used to perform at least one prediction of at least one communication parameter in at least one cell. Some examples of ML models are shown in... Figures 12-14 The illustrations are shown in the middle. However, ML models are not limited to these examples.

[0265] In block 1102, the device sends at least one of the following to at least one user equipment: an instruction to update the machine learning model, or a configuration indicating at least one condition for determining whether to update the machine learning model.

[0266] In box 1103, the device receives an updated version of the machine learning model from at least one user device.

[0267] In box 1104, the device determines whether the updated version of the machine learning model provides a performance improvement compared to the previous version of the machine learning model.

[0268] For example, the device can determine the errors or error rate of the updated version and compare it with a performance threshold.

[0269] In box 1105, based on the determination that the updated version does not provide performance improvements (box 1104: No), the device discards the updated version of the machine learning model.

[0270] Alternatively, in box 1106, based on determining an updated version to provide performance improvements (box 1104: Yes), the device sends an updated version of the machine learning model to one or more other user equipment to perform at least one subsequent prediction of at least one communication parameter in at least one cell.

[0271] In another option, the device can obtain a merged machine learning model by merging an updated version of the machine learning model with one or more other machine learning models received from one or more other user devices. In this case, the device can send the merged machine learning model instead of the updated version of the machine learning model provided by the user device.

[0272] The above is made with the help of Figures 5-11The described boxes, related functions, and information exchanges (messages) are not in an absolute chronological order, and some of them may occur simultaneously or in a different order than described. Other functions may also be executed between or within them, and other information may be sent, and / or other rules may be applied. Some boxes or portions of boxes, or one or more messages, may also be omitted or replaced with the corresponding boxes or portions of boxes, or one or more messages.

[0273] As used herein, “at least one of the following: ” and “at least one of ” and similar wording (where a list of two or more elements is connected by “and” or “or”) means at least any one of these elements, or at least any two or more of these elements, or at least all of these elements.

[0274] Figure 12 An example of spatial prediction based on a convolutional neural network (CNN) model 1200 is illustrated. The CNN model 1200 can be used on the UE side for beam prediction in both uplink and downlink aspects.

[0275] The UE can deploy a CNN model 1200 to perform training in spatial prediction, where the CNN model 1200 can be updated based on a quality threshold in the spatial domain (at time t). A network node can trigger at least one condition (e.g., one or more coefficients or indicators) to configure the UE for training the CNN model 1200.

[0276] For downlink beam management, the input information 1201 of the CNN model 1200 may include at least one of the following: one or more DL beam IDs, reference signal received power (RSRP) of one or more DL beam IDs, signal-to-interference-plus-noise ratio (SINR) of one or more DL beam IDs, a threshold for RSRP, a threshold for SINR, one or more antenna panel indices, and / or one or more beam indices received at the UE.

[0277] The RSRP and SINR thresholds in the input information can prevent low-quality input information from being fed into the CNN model 1200, as low-quality input information may lead to errors in prediction or unwanted results.

[0278] For downlink beam management, the output information 1202 of the CNN model 1200 may include at least one of the following: one or more predicted DL beam IDs, predicted RSRPs of one or more DL beam IDs, predicted SINRs of one or more DL beam IDs, a quality threshold for RSRPs, and / or a quality threshold for SINRs in the spatial domain (at a specific time).

[0279] Multiple quality thresholds can be used to prevent network nodes from indicating beam IDs with low quality (based on prediction).

[0280] The UE can also be triggered by a network node to configure the CNN model 1200 for uplink transmission. For uplink beam management, the input information 1201 of the CNN model 1200 may include at least one of the following: one or more UL beam IDs, RSRP of one or more UL beam IDs, SINR of one or more UL beam IDs, threshold of RSRP, threshold of SINR and / or one or more antenna panel indices.

[0281] For uplink beam management, the output information 1202 of the CNN model 1200 may include at least one of the following: one or more predicted UL beam IDs, predicted RSRPs of one or more UL beam IDs, predicted SINRs of one or more UL beam IDs, a quality threshold for RSRPs, and / or a quality threshold for SINRs in the spatial domain (at a specific time).

[0282] Figure 13 The illustration shows an example of temporal prediction based on a Long Short-Term Memory (LSTM) Recurrent Neural Network (RNN) model 1300. The LSTM RNN model 1300 can be used on the UE side for beam prediction in both uplink and downlink aspects. With this model, the UE can predict output and quality thresholds in a time-series manner.

[0283] A network node can trigger at least one condition (e.g., one or more coefficients or indicators) to configure the UE for training an LSTM RNN model 1300.

[0284] For downlink beam management, the input information 1301 of the LSTM RNN model 1300 may include at least one of the following: one or more DL beam IDs in a time series, the reference signal received power (RSRP) of one or more DL beam IDs in a time series, the signal-to-interference-plus-noise ratio (SINR) of one or more DL beam IDs in a time series, a threshold for RSRP in a time series, a threshold for SINR in a time series, one or more antenna panel indices, and / or one or more beam indices received at the UE. In other words, the input may be a time series (e.g., t-T1, ..., t-2, t).

[0285] For downlink beam management, the output information 1302 of the LSTM RNN model 1300 may include at least one of the following: one or more predicted DL beam IDs in the time series, the predicted RSRP of one or more DL beam IDs in the time series, the predicted SINR of one or more DL beam IDs in the time series, a quality threshold for RSRP over a duration, and / or a quality threshold for SINR over a duration. In other words, the output may be a time series following the input time series (e.g., t+2, ..., t+t).

[0286] The UE can also be triggered by a network node to configure the LSTM RNN model 1300 for uplink transmission. For uplink beam management, the input information 1301 of the LSTM RNN model 1300 may include at least one of the following: one or more UL beam IDs in a time series, RSRP of one or more UL beam IDs in a time series, SINR of one or more UL beam IDs in a time series, a threshold for RSRP in a time series, a threshold for SINR in a time series, and / or one or more antenna panel indices. In other words, the input can be a time series (e.g., t-T1, ..., t-2, t).

[0287] For uplink beam management, the output information 1302 of the LSTM RNN model 1300 may include at least one of the following: one or more predicted UL beam IDs in the time series, the predicted RSRP of one or more UL beam IDs in the time series, the predicted SINR of one or more UL beam IDs in the time series, a quality threshold for RSRP over a duration, and / or a quality threshold for SINR over a duration. In other words, the output may be a time series following the input time series (e.g., t+2, ..., t+t).

[0288] Figure 14 The illustration shows an example of a prediction based on a deep reinforcement learning (DRL) model 1400. The DRL model 1400 can be used on the UE side for beam prediction in both the uplink and downlink aspects.

[0289] Reinforcement learning is a type of machine learning in which an agent (i.e., the UE in this case) learns to make decisions by interacting with its environment. The learning process involves receiving feedback, in the form of rewards or penalties, based on the agent's behavior. The agent's goal is to learn a policy for mapping states to actions to maximize cumulative rewards over time.

[0290] In reinforcement learning, rewards are the positive feedback an agent receives when it takes an ideal action in a given state, while penalties (often called negative rewards) are the negative feedback received when it takes a less than ideal action. Agents learn to make better decisions by trying to maximize rewards and minimize penalties over time.

[0291] The observation space 1401 represents the set of all possible states or observations that the agent may encounter when interacting with the environment 1405. A given state in the observation space is a description of the current state of the environment 1405 and is used by the agent to make decisions. The observation space can be discrete, where states are represented by a finite set of distinct values; or it can be continuous, where states are represented by continuous variables.

[0292] In deep reinforcement learning, the observation space can be processed by a neural network (such as a deep neural network or a convolutional neural network), which is used as a function approximator to learn the optimal policy or value function.

[0293] Action space 1402 represents the set of all possible actions an agent can take in a given state. A given action in the action space corresponds to a decision or control input that the agent can apply to influence the environment 1405 and transition to a new state. Like the observation space, the action space can be discrete, containing a finite set of different actions; or it can be continuous, where actions are represented by continuous variables.

[0294] In deep reinforcement learning, neural networks learn to map observations in the observation space to appropriate actions in the action space to maximize cumulative rewards over time. The learning process involves updating the parameters of the neural network based on observed rewards and penalties, using techniques such as Q-learning, policy gradients, or actor critique methods.

[0295] In this example, the UE can deploy a DRL model 1400 to perform training, which takes into account penalty and reward conditions. The DRL model 1400 can be updated based on the penalty and reward conditions.

[0296] For example, for downlink beam prediction, a network node can trigger the UE to configure DRL model 1400 to predict at least one of the following: one or more DL beam IDs, RSRP of one or more DL beam IDs, SINR of one or more DL beam IDs, a quality threshold for RSRP, and / or a quality threshold for SINR. In other words, for downlink beam prediction, action space 1402 can include one or more of these outputs.

[0297] For downlink beam prediction, the observation space 1401 may include at least one of the following: one or more DL beam IDs, the reference signal received power (RSRP) of one or more DL beam IDs, the signal-to-interference-plus-noise ratio (SINR) of one or more DL beam IDs, a threshold for RSRP, a threshold for SINR, one or more antenna panel indices, and / or one or more beam indices received at the UE. These inputs may come from previous sequences.

[0298] The network node can also trigger the UE to configure DRL model 1400 for uplink transmission. For uplink beam prediction, DRL model 1400 can predict at least one of the following: one or more UL beam IDs, RSRP of one or more UL beam IDs, SINR of one or more UL beam IDs, a quality threshold for RSRP, and / or a quality threshold for SINR. In other words, for uplink beam prediction, action space 1402 can include one or more of these outputs. The quality thresholds (multiple) can be considered in a similar manner to downlink transmission, but are related to uplink beams.

[0299] For uplink beam prediction, the observation space 1401 may include at least one of the following: one or more UL beam IDs, RSRP of one or more UL beam IDs, SINR of one or more UL beam IDs, threshold of RSRP, threshold of SINR, and / or one or more antenna panel indices. These inputs may come from previous sequences.

[0300] In action space 1402, the UE can determine the output (action space) that takes into account a quality threshold. If the quality threshold is higher than a predefined threshold, then the quality threshold can be satisfied.

[0301] Action 1403 may include, for example, configuring a (new) service beam or maintaining the current beam.

[0302] The reward and penalty assessment 1404 may include at least one of the following: mapping the predicted RSRP value of a given beam ID to an observed channel quality indicator (CQI) value, mapping the predicted SINR value of a given beam ID to an observed CQI value, or mapping the predicted RSRP value and the predicted SINR value to a throughput value.

[0303] RSRP can be L1-RSRP or L3-RSRP. The UE can continue to use the current action space to calculate the reward. If the quality threshold and / or the L1-RSRP threshold or the L3-RSRP threshold is higher than a predefined threshold, the UE can update the action space 1402.

[0304] However, if the quality threshold and / or L1-RSRP threshold or L3-RSRP threshold is less than a predefined threshold, the UE may not update the action space 1402, but the UE may find a new action space from the observation space 1401 without calculating the reward.

[0305] Figure 15 An example of an apparatus 1500 is illustrated, which includes components for performing one or more of the example embodiments described above. For example, apparatus 1500 may be, or include, a device such as user equipment (UE) 100, 102, 400, etc., or may be included in or incorporated into such a device.

[0306] User equipment can also be referred to as wireless communication equipment, subscriber unit, mobile station, remote terminal, access terminal, user terminal, terminal equipment, or user equipment.

[0307] Apparatus 1500 may include a circuit system or chipset suitable for implementing one or more of the example embodiments described above. For example, apparatus 1500 may include at least one processor 1510. At least one processor 1510 interprets instructions (e.g., computer program instructions) and processes data. At least one processor 1510 may include one or more programmable processors. At least one processor 1510 may include programmable hardware with embedded firmware, and alternatively or additionally may include one or more application-specific integrated circuits (ASICs).

[0308] At least one processor 1510 is coupled to at least one memory 1520. The at least one processor is configured to write data to and read data from the at least one memory 1520. The at least one memory 1520 may include one or more memory cells. Memory cells may be volatile or non-volatile. It should be noted that one or more non-volatile memory cells and one or more volatile memory cells may be present, or alternatively, one or more non-volatile memory cells may be present, or alternatively, one or more volatile memory cells may be present. Volatile memory may be, for example, random access memory (RAM), dynamic random access memory (DRAM), or synchronous dynamic random access memory (SDRAM). Non-volatile memory may be, for example, read-only memory (ROM), programmable read-only memory (PROM), electrically erasable programmable read-only memory (EEPROM), flash memory, optical storage devices, or magnetic storage devices. Generally, memory may be referred to as a non-transitory computer-readable medium. The term "non-transitory" as used herein refers to a limitation on the medium itself (i.e., tangible, not tactile), rather than a limitation on the persistence of data storage (e.g., RAM and ROM). At least one memory 1520 stores computer-readable instructions that are executed by at least one processor 1510 to perform one or more of the example embodiments described above. For example, non-volatile memory stores computer-readable instructions, and at least one processor 1510 uses volatile memory for temporary storage of data and / or instructions to execute instructions. Computer-readable instructions may refer to computer program code.

[0309] Computer-readable instructions may have been pre-stored in at least one memory 1520, or alternatively or additionally, they may be received by the device via an electromagnetic carrier signal, and / or copied from a physical entity such as a computer program product. Execution of the computer-readable instructions by at least one processor 1510 causes the device 1500 to perform one or more of the above-described example embodiments. That is, at least one processor storing the instructions and at least one memory may provide components for providing or causing execution of any of the above methods and / or blocks.

[0310] In the context of this document, "memory" or "a computer-readable medium" or "a plurality of computer-readable media" can be any one or more nontransitory media or components that can contain, store, transmit, propagate, or transfer instructions for use by or in connection with an instruction execution system, apparatus, or device, such as a computer. The term "nontransitory" as used herein refers to a limitation on the medium itself (i.e., tangible, not tactile), not to a limitation on the persistence of data storage (e.g., RAM and ROM).

[0311] The device 1500 may also include or be connected to the input unit 1530. The input unit 1530 may include one or more interfaces for receiving input. The one or more interfaces may include, for example, one or more temperature, motion, and / or orientation sensors, one or more cameras, one or more accelerometers, one or more microphones, one or more buttons, and / or one or more touch detection units. In addition, the input unit 1530 may include interfaces to which external devices can be connected.

[0312] The device 1500 may also include an output unit 1540. The output unit may include or be connected to one or more displays capable of rendering visual content, such as a light-emitting diode (LED) display, a liquid crystal display (LCD), and / or a liquid crystal on silicon (LCoS) display. The output unit 1540 may also include one or more audio outputs. The one or more audio outputs may be, for example, speakers.

[0313] Device 1500 also includes a connection unit 1550. Connection unit 1550 enables wireless connectivity to one or more external devices. Connection unit 1550 includes at least one transmitter and at least one receiver, which may be integrated into device 1500 or connected to the transmitter and receiver. The at least one transmitter includes at least one transmitting antenna, and the at least one receiver includes at least one receiving antenna. Connection unit 1550 may include an integrated circuit or a set of integrated circuits providing wireless communication capabilities to device 1500. Alternatively, the wireless connection may be a hardwired application-specific integrated circuit (ASIC). Connection unit 1550 may also provide components for performing at least some of the blocks or functions of one or more of the example embodiments described above. Connection unit 1550 may include one or more components controlled by a corresponding control unit, such as a power amplifier, digital front end (DFE), analog-to-digital converter (ADC), digital-to-analog converter (DAC), frequency converter, modulator (demodulator), and / or encoder / decoder circuitry.

[0314] It should be noted that device 1500 may also include Figure 15 Various components are not shown. These various components can be hardware components and / or software components.

[0315] Figure 16 An example of an apparatus 1600 including components for performing one or more of the example embodiments described above is illustrated. For example, apparatus 1600 may be an apparatus such as a network node, or may include such an apparatus, or be included in such an apparatus. The network node may be a radio access network node 104, 404, such as a gNB (e.g., as...). Figure 4B (as shown), or a network node can be a separate network entity that includes AI / ML functionality 407 (e.g., Figure 4A (As shown).

[0316] For example, a network node can also be referred to as a network element, a next-generation radio access network (NG-RAN) node, a NodeB, an eNB, a gNB, a base transceiver station (BTS), a base station, an NR base station, a 5G base station, an access node, an access point (AP), a cell site, a relay node, a repeater, an integrated access and backhaul (IAB) node, an IAB donor node, a distributed unit (DU), a central unit (CU), a baseband unit (BBU), a radio unit (RU), a radio head, a remote radio head end (RRH), or a transmit and receive point (TRP).

[0317] Apparatus 1600 may include, for example, a circuit system or chipset suitable for implementing one or more of the example embodiments described above. Apparatus 1600 may be an electronic device including one or more electronic circuit systems. Apparatus 1600 may include a communication control circuit system 1610 (such as at least one processor) and at least one memory 1620 storing instructions 1622, which, when executed by at least one processor, cause apparatus 1600 to perform one or more of the example embodiments described above. For example, such instructions 1622 may include computer program code (software). At least one processor and at least one memory storing instructions may provide components for providing or causing execution of any of the methods and / or blocks described above.

[0318] The processor is coupled to memory 1620. The processor is configured to read data from and write data to memory 1620. Memory 1620 may include one or more memory cells. Memory cells may be volatile or non-volatile. It should be noted that one or more non-volatile memory cells and one or more volatile memory cells may be present, or alternatively, one or more non-volatile memory cells may be present, or alternatively, one or more volatile memory cells may be present. Volatile memory may be, for example, random access memory (RAM), dynamic random access memory (DRAM), or synchronous dynamic random access memory (SDRAM). Non-volatile memory may be, for example, read-only memory (ROM), programmable read-only memory (PROM), electrically erasable programmable read-only memory (EEPROM), flash memory, optical storage device, or magnetic storage device. Generally, memory may be referred to as a non-transitory computer-readable medium. The term "non-transitory" as used herein is a limitation on the medium itself (i.e., tangible, not tactile), rather than a limitation on the persistence of data storage (e.g., RAM and ROM). Memory 1620 stores computer-readable instructions that are executed by the processor. For example, non-volatile memory stores computer-readable instructions, while the processor uses volatile memory for temporary storage of data and / or instructions to execute instructions.

[0319] The computer-readable instructions may have been pre-stored in memory 1620, or alternatively or additionally, they may be received by the device via an electromagnetic carrier signal, and / or copied from a physical entity such as a computer program product. Execution of the computer-readable instructions causes device 1600 to perform one or more of the functions described above.

[0320] The memory 1620 can be implemented using any suitable data storage technology, such as semiconductor-based memory devices, flash memory, magnetic memory devices and systems, optical memory devices and systems, fixed memory, and / or removable memory. The memory may include a configuration database for storing configuration data, such as a current list of neighboring cells, and, in some example embodiments, the structure of frames used in detected neighboring cells.

[0321] Device 1600 may also include or be connected to communication interface 1630, such as a radio unit, which includes hardware and / or software for establishing a communication connection with one or more wireless communication devices according to one or more communication protocols. Communication interface 1630 includes at least one transmitter (Tx) and at least one receiver (Rx), which may be integrated into device 1600 or connected to it. Communication interface 1630 may provide components for performing some blocks of the above-described example embodiments. Communication interface 1630 may include one or more components, such as a power amplifier, digital front end (DFE), analog-to-digital converter (ADC), digital-to-analog converter (DAC), frequency converter, modulator (demodulator), and / or encoder / decoder circuitry.

[0322] Communication interface 1630 provides the device with radio communication capabilities for communication within a wireless communication network. For example, the communication interface may provide a radio interface to one or more wireless communication devices. Device 1600 may also include or be connected to another interface toward the core network, such as a network coordinator device or AMF, and / or include or be connected to an access node of the wireless communication network.

[0323] The apparatus 1600 may also include a scheduler 1640 configured to allocate radio resources. The scheduler 1640 may be configured together with the communication control circuitry system 1610 or separately.

[0324] It should be noted that device 1600 may also include Figure 16 Various components are not shown. These various components can be hardware components and / or software components.

[0325] As used in this application, the term "circuit system" may refer to one or more or all of the following: a) a hardware circuit implementation only (such as an implementation only in analog and / or digital circuit systems); and b) a combination of hardware circuits and software, such as (if applicable): i) a combination of (multiple) analog and / or digital hardware circuits with software / firmware, and (multiple) hardware processors having software (including (multiple) digital signal processors, software, and any part of (multiple) memories, which work together to cause a device (such as a mobile phone) to perform various functions); and c) (multiple) hardware circuits and / or (multiple) processors, such as (multiple) microprocessors or a portion thereof, which require software (e.g., firmware) to operate, but may be absent when the software is not required to operate.

[0326] This definition of circuit system applies to all uses of the term in this application, including in any claim. As another example, as used in this application, the term circuit system also covers only the implementation of hardware circuitry or a processor (or processors) or a portion of hardware circuitry or processing and its accompanying software and / or firmware. For example, if applicable to a particular claim element, the term circuit system also covers baseband integrated circuits or processor integrated circuits for mobile devices, or similar integrated circuits in servers, cellular network devices, or other computing or network devices.

[0327] The techniques and methods described herein can be implemented in various ways. For example, these techniques can be implemented in hardware (one or more devices), firmware (one or more devices), software (one or more modules), or a combination thereof. For hardware implementation, the apparatus(s) of the example embodiments can be implemented within one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), graphics processing units (GPUs), processors, controllers, microcontrollers, microprocessors, other electronic units designed to perform the functions described herein, or a combination thereof. For firmware or software, implementation can be achieved by modules (e.g., processes, functions, etc.) of at least one chipset that perform the functions described herein. Software code can be stored in memory cells and executed by a processor. Memory cells can be implemented within the processor or external to the processor. In the latter case, as is known in the art, memory cells can be communicatively coupled to the processor in various ways. Furthermore, those skilled in the art will understand that the components of the systems described herein can be rearranged and / or supplemented by additional components to facilitate the implementation of the various aspects described, etc., and they are not limited to the precise configurations illustrated in the given figures.

[0328] It will be apparent to those skilled in the art that the inventive concept can be implemented in various ways as technology advances. The embodiments are not limited to the exemplary embodiments described above, but may vary within the scope of the claims. Therefore, all words and expressions should be interpreted broadly, and they are intended to illustrate rather than limit the embodiments.

Claims

1. An apparatus comprising at least one processor and at least one memory, the at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to at least: Receive a machine learning model from a network node, the machine learning model being used to perform at least one prediction of at least one communication parameter in at least one cell; Receive at least one of the following from the network node: an instruction to update the machine learning model, or a configuration indicating at least one condition for determining whether to update the machine learning model; The machine learning model is used to perform at least one prediction of at least one communication parameter in the at least one cell; Whether to update the machine learning model is determined based on the instruction or at least one of the conditions; Based on the determination that the machine learning model needs to be updated, an updated version of the machine learning model is obtained by updating the machine learning model based on the at least one prediction; as well as The updated version of the machine learning model is sent to the network node.

2. The apparatus according to claim 1, further comprising: Receive from the network node an instruction indicating support for one or more pre-configured machine learning models, the one or more machine learning models being used to perform the at least one prediction of the at least one communication parameter in the at least one cell; and Send an instruction to the network node indicating the ability to predict at least one of the following: predict at least one communication parameter, or train at least one of the one or more machine learning models.

3. The apparatus according to any one of the preceding claims, wherein the at least one condition comprises at least one of the following: Perform the at least one prediction of the at least one communication parameter. Receive configuration of at least one communication parameter to be used for communication with the network node. Monitor the predicted quality or utilization of at least one communication parameter within a predefined duration, or The amount of correlation between the at least one prediction and at least one observation of the at least one communication parameter.

4. The apparatus according to any one of the preceding claims is further configured such that: Determine whether the at least one prediction is related to at least one observation of the at least one communication parameter; and Based on the determination that the at least one prediction is related to the at least one observation. The machine learning model is updated by rewarding it.

5. The apparatus according to any one of claims 1 to 3, further comprising: Determine whether the at least one prediction is related to at least one observation of the at least one communication parameter; and Based on the determination that the at least one prediction is not correlated with the at least one observation. The machine learning model is updated by penalizing it.

6. The apparatus according to any one of the preceding claims is further configured such that: An attempt is made to determine whether the at least one prediction is related to at least one observation of the at least one communication parameter; and Since it cannot be determined whether the at least one prediction is related to the at least one observation. Avoid updating the machine learning model until it can be determined whether the at least one prediction is related to the at least one observation.

7. An apparatus comprising at least one processor and at least one memory, the at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to at least: Send a machine learning model to at least one user equipment, the machine learning model being used to perform at least one prediction of at least one communication parameter in at least one cell; Send to the at least one user equipment at least one of the following: an instruction to update the machine learning model, or a configuration indicating at least one condition for determining whether to update the machine learning model; as well as Receive an updated version of the machine learning model from the at least one user device.

8. The apparatus according to claim 7, further comprising: Send an instruction indicating support for one or more pre-configured machine learning models, which are used to perform at least one prediction of at least one communication parameter in the at least one cell; Receive an instruction from the at least one user equipment indicating the ability to: predict the at least one communication parameter, or train at least one of the one or more machine learning models; as well as The machine learning model to be sent to the at least one user device is determined based on the said capability.

9. The apparatus according to any one of claims 7 to 8, further comprising: Send a configuration for predicting the at least one communication parameter to the at least one user equipment; Receive information indicating the at least one communication parameter from the at least one user equipment; The at least one prediction of the at least one user equipment is monitored based on the communication quality between the device and the at least one user equipment. as well as The instruction to update the machine learning model is sent to the at least one user device based on the monitoring.

10. The apparatus according to any one of claims 7 to 9, further comprising: Determine whether the at least one prediction is related to at least one observation of the at least one communication parameter; and Based on the determination that the at least one prediction is related to the at least one observation. Send an instruction to the at least one user device to update the machine learning model by rewarding the machine learning model.

11. The apparatus according to any one of claims 7 to 9, further comprising: Determine whether the at least one prediction is related to at least one observation of the at least one communication parameter; and Based on the determination that the at least one prediction is not correlated with the at least one observation. Send an instruction to the at least one user device to update the machine learning model by penalizing it.

12. The apparatus according to any one of claims 7 to 11, further comprising: Determine whether the updated version of the machine learning model provides a performance improvement compared to the previous version of the machine learning model; and Based on the determination that the updated version provides the performance improvement, the updated version of the machine learning model is sent to one or more other user equipment to perform at least one subsequent prediction of the at least one communication parameter in the at least one cell.

13. The apparatus according to any one of claims 7 to 11, further comprising: Determine whether the updated version of the machine learning model provides a performance improvement compared to the previous version of the machine learning model; and Based on the determination that the updated version does not provide the performance improvement, the updated version of the machine learning model is discarded.

14. A method comprising: Receive a machine learning model from a network node, the machine learning model being used to perform at least one prediction of at least one communication parameter in at least one cell; Receive at least one of the following from the network node: an instruction to update the machine learning model, or a configuration indicating at least one condition for determining whether to update the machine learning model; The machine learning model is used to perform at least one prediction of at least one communication parameter in the at least one cell; Whether to update the machine learning model is determined based on the instruction or at least one of the conditions; Based on the determination that the machine learning model needs to be updated, an updated version of the machine learning model is obtained by updating the machine learning model based on the at least one prediction; as well as The updated version of the machine learning model is sent to the network node.

15. A method comprising: Send a machine learning model to at least one user equipment, the machine learning model being used to perform at least one prediction of at least one communication parameter in at least one cell; Send to the at least one user equipment at least one of the following: an instruction to update the machine learning model, or a configuration indicating at least one condition for determining whether to update the machine learning model; as well as Receive an updated version of the machine learning model from the at least one user device.

Images