WO2024223044A1

WO2024223044A1 - Updating machine learning model

Info

Publication number: WO2024223044A1
Application number: PCT/EP2023/061121
Authority: WO
Inventors: Timo Koskela; Tachporn SANGUANPUAK
Original assignee: Nokia Technologies Oy
Priority date: 2023-04-27
Filing date: 2023-04-27
Publication date: 2024-10-31

Abstract

Disclosed is a 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 to update the machine learning model; and receiving, from the at least one user equipment, an updated version of the machine learning model.

Description

UPDATING MACHINE LEARNING MODEL

FIELD

The following example embodiments relate to wireless communication and to machine learning.

BACKGROUND

Machine learning models may be used for various use cases in wireless communication. However, there is a challenge in how to efficiently update the machine learning models.

BRIEF DESCRIPTION

The scope of protection sought for various example embodiments is set out by the independent claims. The example embodiments and features, if any, described in this specification that do not fall under the scope of the independent claims are to be interpreted as examples useful for understanding various embodiments.

According to an aspect, there is provided an apparatus comprising at least one processor, and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to: receive, from a network node, a machine learning model for performing 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 indication to update the machine learning model, or a configuration indicating at least one condition for determining whether to update the machine learning model; perform the at least one prediction of the at least one communication parameter in the at least one cell using the machine learning model; determine, based on the indication or the at least one condition, whether to update the machine learning model; obtain, based on determining to update the machine learning model, an updated version of the machine learning model by updating the machine learning model based on the at least one prediction; and transmit, to the network node, the updated version of the machine learning model. According to another aspect, there is provided an apparatus comprising: means for receiving, from a network node, a machine learning model for performing at least one prediction of at least one communication parameter in at least one cell; means for receiving, from the network node, at least one of: an indication to update the machine learning model, or a configuration indicating at least one condition for determining whether to update the machine learning model; means for performing the at least one prediction of the at least one communication parameter in the at least one cell using the machine learning model; means for determining, based on the indication or the at least one condition, whether to update the machine learning model; means for obtaining, based on determining to update the machine learning model, an updated version of the machine learning model by updating the machine learning model based on the at least one prediction; and means for transmitting, to the network node, the updated version of the machine learning model.

According to another aspect, there is provided a method comprising: receiving, from a network node, a machine learning model for performing at least one prediction of at least one communication parameter in at least one cell; receiving, from the network node, at least one of: an indication to update the machine learning model, or a configuration indicating at least one condition for determining whether to update the machine learning model; performing the at least one prediction of the at least one communication parameter in the at least one cell using the machine learning model; determining, based on the indication or the at least one condition, whether to update the machine learning model; obtaining, based on determining to update the machine learning model, an updated version of the machine learning model by updating the machine learning model based on the at least one prediction; and transmitting, to the network node, the updated version of the machine learning model.

According to another aspect, there is provided a computer program comprising instructions which, when executed by an apparatus, cause the apparatus to perform at least the following: receiving, from a network node, a machine learning model for performing at least one prediction of at least one communication parameter in at least one cell; receiving, from the network node, at least one of: an indication to update the machine learning model, or a configuration indicating at least one condition for determining whether to update the machine learning model; performing the at least one prediction of the at least one communication parameter in the at least one cell using the machine learning model; determining, based on the indication or the at least one condition, whether to update the machine learning model; obtaining, based on determining to update the machine learning model, an updated version of the machine learning model by updating the machine learning model based on the at least one prediction; and transmitting, to the network node, the updated version of the machine learning model.

According to another aspect, there is provided a computer readable medium comprising program instructions which, when executed by an apparatus, cause the apparatus to perform at least the following: receiving, from a network node, a machine learning model for performing at least one prediction of at least one communication parameter in at least one cell; receiving, from the network node, at least one of: an indication to update the machine learning model, or a configuration indicating at least one condition for determining whether to update the machine learning model; performing the at least one prediction of the at least one communication parameter in the at least one cell using the machine learning model; determining, based on the indication or the at least one condition, whether to update the machine learning model; obtaining, based on determining to update the machine learning model, an updated version of the machine learning model by updating the machine learning model based on the at least one prediction; and transmitting, to the network node, the updated version of the machine learning model.

According to another aspect, there is provided a non-transitory computer readable medium comprising program instructions which, when executed by an apparatus, cause the apparatus to perform at least the following: receiving, from a network node, a machine learning model for performing at least one prediction of at least one communication parameter in at least one cell; receiving, from the network node, at least one of: an indication to update the machine learning model, or a configuration indicating at least one condition for determining whether to update the machine learning model; performing the at least one prediction of the at least one communication parameter in the at least one cell using the machine learning model; determining, based on the indication or the at least one condition, whether to update the machine learning model; obtaining, based on determining to update the machine learning model, an updated version of the machine learning model by updating the machine learning model based on the at least one prediction; and transmitting, to the network node, the updated version of the machine learning model.

According to another aspect, there is provided an apparatus comprising at least one processor, and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to: transmit, 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; transmit, 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 to update the machine learning model; and receive, from the at least one user equipment, an updated version of the machine learning model.

According to another aspect, there is provided an apparatus comprising: means for 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; means for 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 to update the machine learning model; and means for receiving, from the at least one user equipment, an updated version of the machine learning model.

According to another aspect, there is provided a 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 to update the machine learning model; and receiving, from the at least one user equipment, an updated version of the machine learning model.

According to another aspect, there is provided a computer program comprising instructions which, when executed by an apparatus, cause the apparatus to perform at least the following: 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 to update the machine learning model; and receiving, from the at least one user equipment, an updated version of the machine learning model.

According to another aspect, there is provided a computer readable medium comprising program instructions which, when executed by an apparatus, cause the apparatus to perform at least the following: 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 to update the machine learning model; and receiving, from the at least one user equipment, an updated version of the machine learning model.

According to another aspect, there is provided a non-transitory computer readable medium comprising program instructions which, when executed by an apparatus, cause the apparatus to perform at least the following: 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 to update the machine learning model; and receiving, from the at least one user equipment, an updated version of the machine learning model.

LIST OF DRAWINGS

In the following, various example embodiments will be described in greater detail with reference to the accompanying drawings, in which

FIG. 1A illustrates an example of a wireless communication network;

FIG. IB illustrates an example of a communication system;

FIG. 2 illustrates an example of a synchronization signal block;

FIG. 3 illustrates an example of user equipment beamforming with a multi-panel configuration;

FIG. 4A illustrates an example of a communication system with artificial intelligence or machine learning support;

FIG. 4B illustrates an example of a communication system with artificial intelligence or machine learning support;

FIG. 5 illustrates a signal flow diagram;

FIG. 6 illustrates a signal flow diagram;

FIG. 7 illustrates a flow chart;

FIG. 8 illustrates a flow chart;

FIG. 9 illustrates a flow chart;

FIG. 10 illustrates a flow chart;

FIG. 11 illustrates a flow chart;

FIG. 12 illustrates an example of spatial domain prediction based on a convolutional neural network model;

FIG. 13 illustrates an example of time domain prediction based on a long-short-term-memory recurrent neural network model;

FIG. 14 illustrates an example of prediction based on a deep reinforcement learning model;

FIG. 15 illustrates an example of an apparatus; and

FIG. 16 illustrates an example of an apparatus.

DETAILED DESCRIPTION The following embodiments are exemplifying. Although the specification may refer to “an”, “one”, or “some” embodiment(s) in several locations of the text, this does not necessarily mean that each reference is made to the same embodiment(s), or that a particular feature only applies to a single embodiment. Single features of different embodiments may also be combined to provide other embodiments.

Some example embodiments described herein may be implemented in a wireless communication network comprising 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 Telecommunication 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), 5G-Advanced (i.e., 3GPP NR Rel-18 and beyond), or sixth generation (6G). Some examples of radio access networks include the universal mobile telecommunications system (UMTS) radio access network (UTRAN), the Evolved Universal Terrestrial Radio Access network (E-UTRA), or the next generation radio access network (NG-RAN). The wireless communication network may further comprise a core network, and some example embodiments may also be applied to network functions of the core network.

It should be noted that the embodiments are not restricted to the wireless communication network given as an example, but a person skilled in the art may also apply the solution to other wireless communication networks or systems provided with necessary properties. For example, some example embodiments may also be applied to a communication system based on IEEE 802.11 specifications, or a communication system based on IEEE 802.15 specifications.

FIG. 1A depicts an example of a simplified wireless communication network showing some physical and logical entities. The connections shown in FIG. 1A may be physical connections or logical connections. It is apparent to a person skilled in the art that the wireless communication network may also comprise other physical and logical entities than those shown in FIG. 1A.

The example embodiments described herein are not, however, restricted to the wireless communication network given as an example but a person skilled in the art may apply the embodiments described herein to other wireless communication networks provided with necessary properties.

The example wireless communication network shown in FIG. 1A includes an access network, such as a radio access network (RAN), and a core network 110.

FIG. 1A shows user equipment (UE) 100, 102 configured to be in a wireless connection on one or more communication channels in a radio cell with an access node (AN) 104 of an access network. The AN 104 may be an evolved Node B (abbreviated as eNB or eNodeB) or a next generation Node B (abbreviated as gNB or gNodeB), providing the radio cell. The wireless connection (e.g., radio link) from a UE to the access node 104 may be called uplink (UL) or reverse link, and the wireless connection (e.g., radio link) from the access node to the UE may be called downlink (DL) or forward link. UE 100 may also communicate directly with UE 102, and vice versa, via a wireless connection generally referred to as a sidelink (SL). It should be appreciated that the access node 104 or its functionalities may be implemented by using any node, host, server or access point etc. entity suitable for providing such functionalities.

The access network may comprise more than one access node, in which case the access nodes may also be configured to communicate with one another over links, wired or wireless. These links between access nodes may be used for sending and receiving control plane signaling and also for routing data from one access node to another access node.

The access node may comprise a computing device configured to control the radio resources of the access node. The access node may also be referred to as a base station, a base transceiver station (BTS), an access point, a cell site, a radio access node or any other type of node capable of being in a wireless connection with a UE (e.g., UEs 100, 102). The access node may include or be coupled to transceivers. From the transceivers of the access node, a connection may be provided to an antenna unit that establishes bi-directional radio links to UEs 100, 102. The antenna unit may comprise an antenna or antenna element, or a plurality of antennas or antenna elements.

The access node 104 may further be connected to a core network (CN) 110. The core network 110 may comprise an evolved packet core (EPC) network and/or a 5^th generation core network (5GC). The EPC may comprise network entities, such as a serving gateway (S-GW for routing and forwarding data packets), a packet data network gateway (P-GW) for providing connectivity of UEs to external packet data networks, and a mobility management entity (MME). The 5GC may comprise network functions, such as a user plane function (UPF), an access and mobility management function (AMF), and a location management function (LMF).

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

The illustrated UE 100, 102 is one type of an apparatus to which resources on the air interface maybe allocated and assigned. The UE 100, 102 may also be called a wireless communication device, a subscriber unit, a mobile station, a remote terminal, an access terminal, a user terminal, a terminal device, or a user device just to mention but a few names. The UE may be a computing device operating with or without a subscriber identification module (SIM), including, but not limited to, the following types of computing devices: a mobile phone, a smartphone, a personal digital assistant (PDA), a handset, a computing device comprising a wireless modem (e.g., an alarm or measurement device, etc.), a laptop computer, a desktop computer, a tablet, a game console, a notebook, a multimedia device, a reduced capability (RedCap) device, a wearable device (e.g., a watch, earphones or eyeglasses) with radio parts, a sensor comprising a wireless modem, or any computing device comprising a wireless modem integrated in a vehicle.

It should be appreciated that a UE may also be a nearly exclusive uplink- only device, of which an example may be a camera or video camera loading images or video clips to a network. A UE may also be a device having capability to operate in an Internet of Things (loT) network, which is a scenario in which objects maybe provided with the ability to transfer data over a network without requiring human- to-human or human-to-computer interaction. The UE may also utilize cloud. In some applications, the computation may be carried out in the cloud or in another UE.

The wireless communication network may also be able to support the usage of cloud services, for example at least part of core network operations may be carried out as a cloud service (this is depicted in FIG. 1A by “cloud” 114). The wireless communication network may also comprise a central control entity, or the like, providing facilities for wireless communication networks of different operators to cooperate for example in spectrum sharing.

5G enables using multiple input - multiple output (M1M0) antennas in the access node 104 and/or the UE 100, 102, many more base stations or access nodes than an LTE network (a so-called small cell concept), including macro sites operating in co-operation with smaller stations and employing a variety of radio technologies depending on service needs, use cases and/or spectrum available. 5G wireless communication networks may support a wide range of use cases and related applications including video streaming, augmented reality, different ways of data sharing and various forms of machine type applications, such as (massive) machine-type communications (mMTC), including vehicular safety, different sensors and real-time control.

In 5G wireless communication networks, access nodes and/or UEs may have multiple radio interfaces, namely below 6GHz, cmWave and mmWave, and also being integrable with existing legacy radio access technologies, such as the LTE. Integration with the LTE may be implemented, for example, as a system, where macro coverage may be provided by the LTE, and 5G radio interface access may come from small cells by aggregation to the LTE. In other words, a 5G wireless communication network may support both inter-RAT operability (such as LTE-5G) and inter-Rl operability (inter-radio interface operability, such as below 6GHz - cmWave - mmWave). One of the concepts considered to be used in 5G wireless communication networks may be network slicing, in which multiple independent and dedicated virtual sub-networks (network instances) may be created within the substantially same infrastructure to run services that have different requirements on latency, reliability, throughput and mobility.

5G may enable analytics and knowledge generation to occur at the source of the data. This approach may involve leveraging resources that may not be continuously connected to a network, such as laptops, smartphones, tablets and sensors. Multi-access edge computing (MEC) may provide a distributed computing environment for application and service hosting. It may also have the ability to store and process content in close proximity to cellular subscribers for faster response time. Edge computing may cover a wide range of technologies, such as wireless sensor networks, mobile data acquisition, mobile signature analysis, cooperative distributed peer-to-peer ad hoc networking and processing also classifiable as local cloud/fog computing and grid/mesh computing, dew computing, mobile edge computing, cloudlet, distributed data storage and retrieval, autonomic self-healing networks, remote cloud services, augmented and virtual reality, data caching, Internet of Things (massive connectivity and/or latency critical), critical communications (autonomous vehicles, traffic safety, realtime analytics, time-critical control, healthcare applications).

In some example embodiments, an access node (e.g., access node 104) may comprise: a radio unit (RU) comprising a radio transceiver (TRX), i.e., a transmitter (Tx) and a receiver (Rx); one or more distributed units (DUs) 105 that may be used for the so-called Layer 1 (LI) processing and real-time Layer 2 (L2) processing; and a central unit (CU) 108 (also known as a centralized unit) that may be used for non-real-time L2 and Layer 3 (L3) processing. The CU 108 may be connected to the one or more DUs 105 for example via an Fl interface. Such an embodiment of the access node may enable the centralization of CUs relative to the cell sites and DUs, whereas DUs may be more distributed and may even remain at cell sites. The CU and DU together may also be referred to as baseband or a baseband unit (BBU). The CU and DU may also be comprised in a radio access point (RAP).

The CU 108 may be a logical node hosting radio resource control (RRC), service data adaptation protocol (SDAP) and/or packet data convergence protocol (PDCP), of the NR protocol stack for an access node. The DU 105 may be a logical node hosting radio link control (RLC), medium access control (MAC) and/or physical (PHY) layers of the NR protocol stack for the access node. The operations of the DU may be at least partly controlled by the CU. It should also be understood that the distribution of functions between DU 105 and CU 108 may vary depending on implementation. The CU may comprise a control plane (CU-CP), which may be a logical node hosting the RRC and the control plane part of the PDCP protocol of the NR protocol stack for the access node. The CU may further comprise a user plane (CU-UP), which may be a logical node hosting the user plane part of the PDCP protocol and the SDAP protocol of the CU for the access node.

Cloud computing systems may also be used to provide the CU 108 and/or DU 105. A CU provided by a cloud computing system may be referred to as a virtualized CU (vCU). In addition to the vCU, there may also be a virtualized DU (vDU) provided by a cloud computing system. Furthermore, there may also be a combination, where the DU maybe implemented on so-called bare metal solutions, for example application-specific integrated circuit (ASIC) or customer-specific standard product (CSSP) system-on-a-chip (SoC).

Edge cloud may be brought into the access network (e.g., RAN) by utilizing network function virtualization (NFV) and software defined networking (SDN). Using edge cloud may mean access node operations to be carried out, at least partly, in a computing system operationally coupled to a remote radio head (RRH) or a radio unit (RU) of an access node. It is also possible that access node operations may be performed on a distributed computing system or a cloud computing system located at the access node. Application of cloud RAN architecture enables RAN real-time functions being carried out at the access network (e.g., in a DU 105) and non-real-time functions being carried out in a centralized manner (e.g., in a CU 108).

It should also be understood that the distribution of functions between core network operations and access node operations may differ in future wireless communication networks compared to that of the LTE or 5G, or even be nonexistent. Some other technology advancements that may be used include big data and all-lP, which may change the way wireless communication networks are being constructed and managed. 5G (or new radio, NR) wireless communication networks may support multiple hierarchies, where multi-access edge computing (MEC) servers may be placed between the core network 110 and the access node 104. It should be appreciated that MEC may be applied in LTE wireless communication networks as well.

A 5G wireless communication network (“5G network”) may also comprise a non-terrestrial communication network, such as a satellite communication network, to enhance or complement the coverage of the 5G radio access network. For example, satellite communication may support the transfer of data between the 5G radio access network and the core network, enabling more extensive network coverage. Possible use cases may be providing service continuity for machine-to-machine (M2M) or Internet of Things (loT) devices or for passengers on board of vehicles, or ensuring service availability for critical communications, and future railway/maritime/aeronautical communications. Satellite communication may utilize geostationary earth orbit (GEO) satellite systems, but also low earth orbit (LEO) satellite systems, in particular megaconstellations (systems in which hundreds of (nano)satellites are deployed). A given satellite 106 in the mega-constellation may cover several satellite-enabled network entities that create on-ground cells. The on-ground cells may be created through an on-ground relay access node or by an access node 104 located on- ground or in a satellite.

It is obvious for a person skilled in the art that the access node 104 depicted in FIG. 1A is just an example of a part of an access network (e.g., a radio access network) and in practice, the access network may comprise a plurality of access nodes, the UEs 100, 102 may have access to a plurality of radio cells, and the access network may also comprise other apparatuses, 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 a type of access node that may be used to provide indoor coverage inside a home, office, or other indoor environment.

Additionally, in a geographical area of an access network (e.g., a radio access network), a plurality of different kinds of radio cells as well as a plurality of radio cells may be provided. Radio cells may be macro cells (or umbrella cells) which may be large cells having a diameter of up to tens of kilometers, or smaller cells such as micro-, femto- or picocells. The access node(s) of FIG. 1A may provide any kind of these cells. A cellular radio network may be implemented as a multilayer access networks including several kinds of radio cells. In multilayer access networks, one access node may provide one kind of a radio cell or radio cells, and thus a plurality of access nodes may be needed to provide such a multilayer access network.

For fulfilling the need for improving performance of access networks, the concept of “plug-and-play” access nodes may be introduced. An access network which may be able to use “plug-and-play” access nodes, may include, in addition to Home eNodeBs or Home gNodeBs, a Home Node B gateway, or HNB-GW (not shown in FIG. 1A). An HNB-GW, which may be installed within an operator’s access network, may aggregate traffic from a large number of Home eNodeBs or Home gNodeBs back to a core network of the operator.

The 5G NR access link may operate in the millimeter wave (mmWave), sub-terahertz (THz) bands, and higher frequency ranges, which are intrinsically more susceptible to higher path loss and penetration loss. In the higher frequency ranges, both the gNB and the UE may employ front-end circuits with, for example, multiple beam patterns to combat the drawbacks of the propagation channel. The 5G NR access link may also operate in the lower frequencies, where the gNB may use beamforming, but the UE may operate with isotropic or omni-directional beam pattern (i.e., the UE may not use beamforming). A beam refers to directional transmission or reception of a radio signal. A beam may also be represented as a spatial filter, spatial direction, or angle. Beams may be formed using an advanced antenna technology called beamforming. Beamforming may be beneficial for example in 5G NR because of its ability to support higher frequency bands (e.g., mmWave frequencies) and its massive multiple-input, multiple-output (M1M0) capabilities. Beams may be classified into downlink beams and uplink beams.

Downlink beams may be formed by the gNB to transmit signals towards the UE in a specific direction. By focusing the transmission energy in the direction of the intended UE, downlink beamforming can improve the signal strength and overall communication quality, while also minimizing interference with other UEs and reducing power consumption.

Uplink beams may be formed by the UE to transmit signals towards the gNB in a specific direction. Uplink beamforming may enhance the communication link between the UE and the gNB by focusing the transmission energy in the direction of the gNB, which can improve the received signal strength, reduce interference, and extend the range of the UE.

Beamforming may involve the use of large antenna arrays at both the gNB and the UE, allowing for the formation of multiple beams simultaneously. This enables features such as spatial multiplexing and multi-user M1M0 (MU-M1M0), which can further increase the capacity and efficiency of the 5G network.

The network (e.g., the wireless communication network illustrated in FIG. 1A) may further support utilizing multiple transmission and reception points (TRPs). This may be referred to as multiple transmission and reception point (multi-TRP) operation. Multi-TRP operation may support, for example, two or more TRPs. Thus, for example, the UE 100, 102 may receive data via a plurality of TRPs. The different TRPs may be controlled, for example, by the access node 104, such as a gNB.

A TRP is a term used to represent a physical point in the network infrastructure, where both transmission and reception of signals may occur. In other words, a TRP is a point that may handle both uplink and downlink communication between the UE and the network.

FIG. IB illustrates an example of a system. FIG. IB may be understood to depict a part of the wireless communication network of FIG. 1A, but with greater accuracy with respect to the multi-TRP scenario.

The cell area may be covered using one or more beams 121, 122, 123, 124, 125, 126 provided by one or more TRPs 104A, 104B (e.g., TRP#1...#X in FIG. IB) of the access node 104 (e.g., gNB). It should be noted that the access node 104 may provide one or more cells. Herein the term “cell” refers to a radio cell, which represents a coverage area served by the access node 104.

A given beam 121, 122, 123, 124, 125, 126 may carry an identifier enabling the UE 100 to identify the beam and perform measurements (e.g., received power, reference signal received power) and associate the measurements with that specific identifier. As an example, the cell may be covered by a number of synchronization signal blocks (SSBs) denoted as SSB#O...#L. A given SSB may be identified based on the identifier carried by the SSB (SSB time location index). A given SSB may also carry the identifier for the cell that it is associated with.

It should be noted that the number of TRPs and the number of beams may also be different from what is shown in FIG. IB.

FIG. 2 illustrates an example of the time-frequency structure of an SSB. This kind of SSB may be used, for example, in 5G NR. The SSB comprises a primary synchronization signal (PSS) 201, a secondary synchronization signal (SSS) 202, and a physical broadcast channel (PBCH) 203. In the example of FIG. 2, the PSS 201 and the SSS 202 both occupy 1 orthogonal frequency-division multiplexing (OFDM) symbol and 127 subcarriers. In the example of FIG. 2, the PBCH spans across three OFDM symbols (OFDM symbols #1, #2, and #3) and 240 subcarriers, but on one symbol (OFDM symbol #2) leaving an unused part in the middle for the SSS 202.

The potential time locations of SSBs within a half-frame maybe dictated by subcarrier spacing, wherein the network (e.g., gNB) may configure the periodicity of half-frames in which SSBs are transmitted. During a half-frame, different SSBs may be transmitted in different spatial directions (i.e., using different beams, spanning the coverage area of a cell). Within the frequency span of a carrier, multiple SSBs can be transmitted.

To enable a UE to find a cell when entering a communication system, as well as to find new cells when moving within the communication system, the UE may use the PSS, SSS and PBCH to derive the information needed to access the target cell. The PSS and SSS may be transmitted from the access node periodically on the downlink along with the PBCH. Once the UE successfully detects the PSS and/or SSS, it obtains knowledge about the synchronization and physical cell identity (PCI) of the target cell, and the UE is then ready to decode the PBCH. The PBCH carries information needed for further system access, for example to acquire the system information block type 1 (S1B1) of the target cell. The PSS and SSS along with the PBCH can be jointly referred to as the SSB. The SSB may also be referred to as a synchronization and PBCH block or as an SS/PBCH block. Aiming to cover the whole cell space, the access node may transmit multiple SSBs in different directions (beams) in a so-called SSB burst.

Furthermore, for downlink measurement, signals for beam management, called non-zero power channel state information reference signals (NZP-CS1-RS) may be configured. Thus, the beam management from downlink perspective may be performed using SSB and CS1-RS signals. The SSB signals may be always-on signals and the periodicity of SSB is fixed, whereas CS1-RS may be configured in dedicated manner for a given UE (e.g., with different periodicities and bandwidth). As an example, the CS1-RS may be used to train narrower beams (higher gain beams) through the association with an SSB beam. For example, the UE may be configured to report N highest quality SSB beams, and the network (e.g., gNB) may further configure the UE to report M highest quality CS1-RS (#0...#K) beams associated with the specific SSB (e.g., SSB#O). This association may be configured by the network (e.g., gNB). The association may be, for example, spatial association (e.g., quasi co-location), wherein reception of a first signal (e.g., SSB#O) may be used to determine potential reception of at least one of the second signals (e.g., CS1-RS #0...#K).

5G NR also supports UE beamforming. In lower frequencies (e.g., below 6 GHz), the UE may not use beamforming and may operate with an omnidirectional beam (e.g., equal gain on all directions for transmission and/or reception). However, in higher frequencies (e.g., above 6 GHz) the UE may have one or more antenna panels (or distributed antenna elements) that form one or more beams as illustrated in FIG. 3. It may be possible to label or index the antenna panels used by the UE, and/or the individual beams that the UE is capable of forming.

FIG. 3 illustrates an example of UE beamforming with a multi-panel configuration. In the example of FIG. 3, the UE 300 comprises a plurality of antenna panels 311, 312, 313, 314 thatmaybe capable of forming multiple beams 321, 322, 323, 324, 325, 326 in different directions. It should be noted that the number of antenna panels and the number of beams may also be different from what is shown in FIG. 3. A given beam 321, 322, 323, 324, 325, 326 maybe associated with a beam identifier or index (e.g., 0...K). Alternatively, or additionally, a given antenna panel 311, 312, 313, 314 may be associated with an identifier or index. A given antenna panel 311, 312, 313, 314 may comprise one or more antenna elements.

For communication purposes, the UE may be configured with at least one beam (identified by a reference signal) used as reference for receiving and/or transmitting data and control channels. As an 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) that may be received on one or more downlink beams. Additionally, the UE may be configured with one or more physical uplink control channels (PUCCHs) and/or one or more physical uplink shared channels (PUSCHs) that may be transmitted by using the downlink beams as reference. Furthermore, the UE may be capable of beamforming (i.e., it may be capable of forming UL and/or DL beams for transmission and reception, respectively), or the UE may use omnidirectional transmission and reception.

Artificial intelligence (Al) and machine learning (ML) for the 5G NR air interface are being studied for different use cases regarding aspects such as performance, complexity and potential specification impact. One example of such a use case is beam management (e.g., beam prediction in time and/or spatial domain for overhead and latency reduction, as well as beam selection accuracy improvement). When the UE is configured to apply an Al or ML model, the UE may request different models provisioned by the network, or the network may provision different models to the UE (e.g., based on the UE capability). However, this involves some challenges, such as: 1) how the network can efficiently update the model (e.g., in case the UE is configured to update or train the ML model) at the UE side, and 2) what are the conditions for the UE to update the model configured or provisioned by the network.

Some example embodiments may address the above challenges by providing a method for collaborative training of an ML model in a wireless communication network (e.g., in a beam management context).

Some example embodiments introduce an asynchronous ML model update at the UE side, where the UE may be configured with one or more conditions (e.g., threshold conditions) or criteria indicating when the UE should update the ML model. For example, the UE may apply a reinforcement learning scheme to apply a reward or a penalty for updating the ML model. The ML model may be used, for example, for both uplink and downlink communication parameter prediction (e.g., DL and UL beam prediction).

Some example embodiments are described below using principles and terminology of 5G radio access technology without limiting the example embodiments to 5G radio access technology, however.

FIG. 4A illustrates an example of a communication system with Al/ML support, to which some example embodiments may be applied. The system of FIG. 4A comprises a UE 400, a RAN node 404, and Al/ML functions 407. The RAN node 404 may correspond to the access node 104 of FIG. 1A and FIG. IB. The UE 400 may correspond to UE 100 of FIG. 1A and FIG. IB.

The Al/ML functions 407 may include an ML model, which consists of input, output spaces, training, and inference functions. For example, the Al/ML functions may include Top-K beam prediction functionalities, i.e., Top-1, Top-4, or Top-K beams prediction in uplink and/or downlink transmission. The Al/ML functions may also include a legacy mode, which is a non-ML function, such that the RAN node can switch to the legacy mode, when the Al/ML prediction failed. In the example of FIG. 4A, the AI/ML functions 407 are comprised in a separate network entity from the RAN node 404.

The interface 411 may be used to exchange information between the AI/ML functions 407 (which may include the AI/ML model itself) and the RAN node 404 (e.g., base station comprising communication protocols such as RRC, MAC, and/or PHY].

The interface 412 may be used to exchange information between the UE 400 and the RAN node 404, and/or between the UE 400 and the AI/ML functions 407 via the RAN communication protocol.

The interface 413 may be used to exchange information between the UE 400 and the AI/ML functions 407 (e.g., using RAN communication protocol as a container, i.e., via the interface 412).

FIG. 4B illustrates an example of a communication system with AI/ML support, to which some example embodiments may be applied. The system of FIG. 4B comprises a UE 400, a RAN node 404, and AI/ML functions 407. The RAN node 404 may correspond to the access node 104 of FIG. 1A and FIG. IB. The UE 400 may correspond to UE 100 of FIG. 1A and FIG. IB.

In the example of FIG. 4B, the AI/ML functions 407 are embedded in the RAN node 404 (e.g., in a gNB).

However, it should be noted that some example embodiments are not dependent on any specific type of network deployment architecture. Thus, some example embodiments are not limited to the examples shown in FIG. 4A and FIG. 4B.

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

The network node may provide a UE with an ML model that the UE utilizes for at least one prediction of at least one communication parameter. The network node may further configure the UE to evaluate the prediction performance of the ML model and train the ML model by calculating a reward or penalty. The network node may further configure the UE to run the ML model. The network node may further configure the UE to run or train the ML model by providing an indication in the ML model configuration or provision to the UE. The network node may further configure the UE to run inference by providing another indication in ML inference configuration to the UE.

The network node may use a pre-defined criteria (e.g., UE capability in terms of antenna configuration and other information that may be used to obtain input parameters for the ML model such as measurements) for selecting an ML model, which is then provided to the UE. Furthermore, the network node may validate the updated version of the ML model received from the UE to determine whether the updated version provides a performance improvement. The network node may discard the updated version of the ML model, if the updated version does not improve the performance.

FIG. 5 illustrates a signal flow diagram according to an example embodiment. Although two UEs are shown in FIG. 5, it should be noted that the number of UEs may also be different than two. In other words, there may be one or more UEs. In addition, the signaling procedure illustrated in FIG. 5 may be extended and applied according to the actual number of UEs.

Referring to FIG. 5, at 501, a network node transmits an indication indicating support for provisioning one or more machine learning models for performing at least one prediction of at least one communication parameter in at least one cell. In other words, the network node may broadcast or advertise, for example in system information, the support for provisioning and using the one or more machine learning models. Some examples of the ML models are illustrated in FIGS. 12-14.

The indication is received by one or more UEs 100, 102, 400, such as a first UE (UE1) 100 and a second UE (UE2) 102. Herein the terms 'first UE’ and 'second UE’ are used to distinguish the UEs, and they do not necessarily mean a specific order or specific identifiers of the UEs.

The network node may be a radio access network node 104, 404 such as a gNB (e.g., as shown in FIG. 4B), or the network node may be a separate network entity comprising Al/ML functions 407 (e.g., as shown in FIG. 4A). The at least one communication parameter may comprise, for example, at least one of: a beam index or identifier (e.g., SSB or CS1-RS index), quality of the beam index (e.g., reference signal received power and/or signal-to-interference- plus-noise ratio), or a duration of the quality of the beam index or identifier (e.g., time period that a specific reference signal is detectable or has quality above a quality threshold value). For example, the at least one communication parameter may be a downlink beam or an uplink beam (e.g., the corresponding downlink reference signal of the DL or UL beam that may be used for communication).

The one or more machine learning models may comprise a set of machine learning model types (e.g., Model type A, Model type B, ... Model type Y) or IDs for UE-based communication parameter prediction.

The support indication may provide information or reference to the one or more machine learning models that may be referred to with an index value or type value (e.g., Model type A) or similar. For example, the one or more machine learning models may have a pre-defined structure associated with an index value or a type value, wherein the indication indicating the support for provisioning the one or more machine learning models may comprise the index value or the type value per machine learning model of the one or more machine learning models. The index value or type value may indicate, for example, at least one of: number of input vectors of the machine learning model (e.g., artificial neural network), number of output vectors of the machine learning model, number of nodes of the machine learning model, number of layers of the machine learning model, or neural network type of the machine learning model.

In other words, the type value may 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), wherein the X and Y may be known based on the reference to the index or type value.

As an example, a Model type A may have specific input parameters that may comprise of using the UE location, multiple antenna panels, and measurement capability (e.g., accuracy). As another example, a Model type B may have specific input parameters that assume that no UE location is used for prediction, UE does not have multiple antenna panels, and measurement capability (e.g., accuracy) is lower, etc.

A number of model types up to Model type Y may be supported by the network node, and different models may involve a different mix of UE capabilities.

At 502, the first UE transmits, to the network node, an indication indicating a capability of the first UE for at least one of: predicting the at least one communication parameter, or training of at least one of the one or more machine learning models. The network node receives the capability indication.

For example, the first UE may indicate during the connection setup phase, or when entering the at least one cell, or during the connected state, that it is capable of predicting the at least one communication parameter for one or more ML model types advertised by the network node, and/or that the first UE supports (or does not support) online training of a given ML model type advertised by the network node.

The first UE may additionally indicate, to the network node, at least one parameter indicative of communication capabilities of the first UE, wherein the at least one parameter may comprise at least one of: number of antenna panel(s) of the first UE, relative location of antenna panels with respect to one another, number of beams per antenna panel (e.g., (maximum) number of beams that the first UE can form), coverage of the beams in degrees per antenna panel, the beamforming codebook defining how the first UE forms the grid of beams, etc.

The first UE may transmit the capability indication in response to receiving the indication indicating support for provisioning the one or more machine learning models. In another option, the first UE may transmit the capability indication to the network node prior to receiving the indication indicating support for provisioning the one or more machine learning models (i.e., 502 may be performed before 501). In this latter case, the network node may transmit the indication indicating support for provisioning the one or more machine learning models to the first UE via dedicated signaling as a response for receiving the capability indication from the first UE. At 503, the second UE transmits, to the network node, an indication indicating a capability of the second UE for at least one of: predicting the at least one communication parameter, or online training of at least one of the one or more machine learning models. The network node receives the capability indication.

For example, the second UE may indicate during the connection setup phase, or when entering the at least one cell, or during the connected state, that it is capable of predicting the at least one communication parameter for one or more ML model types advertised by the network node, and/or that the second UE supports (or does not support) online training of a given ML model type advertised by the network node.

The second UE may additionally indicate, to the network node, at least one parameter indicative of communication capabilities of the second UE, wherein the at least one parameter may comprise at least one of: number of antenna panel(s) of the second UE, relative location of antenna panels with respect to one another, number of beams per antenna panel (e.g., maximum number of beams that the second UE can form), coverage of the beams in degrees per antenna panel, the beamforming codebook defining how the second UE forms the grid of beams, etc.

The second UE may transmit the capability indication in response to receiving the indication indicating support for provisioning the one or more machine learning models. In another option, the second UE may transmit the capability indication to the network node prior to receiving the indication indicating support for provisioning the one or more machine learning models (i.e., 503 may be performed before 501). In this latter case, the network node may transmit the indication indicating support for provisioning the one or more machine learning models to the second UE via dedicated signaling as a response for receiving the capability indication from the second UE.

At 504, the network node determines, based on the capability indication of the first UE, a machine learning model of the one or more machine learning models to be transmitted to the first UE. Furthermore, the network node determines, based on the capability indication of the second UE, a machine learning model of the one or more machine learning models to be transmitted to the second UE.

In this example, the network node may determine that the first UE and the second UE have the same or similar capabilities (with respect to the ML model to be applied or provisioned), which may cause the network node to provide both UEs with the same type of ML model (e.g., Model type A or ID A). Herein the machine learning model provided to the first UE may be referred to as a first machine learning model, and the machine learning model provided to the second UE may be referred to as a second machine learning model (although the first machine learning model and the second machine learning model may be of the same type). The first machine learning model and the second machine learning model may be denoted as NN_al (i.e., state 1 of ML model type A).

At 505, the network node transmits, to the first UE, the first machine learning model for performing the at least one prediction of the at least one communication parameter in the 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: a time stamp or identifier (e.g., the date, the time reference, time, label, identifier or index for the ML model state), an identifier of the first UE, a version number of the first machine learning model, or any identifier or time parameter that can be used to distinguish between different versions of the first machine learning model.

At 506, the network node transmits, to the second UE, the second machine learning model for performing the at least one prediction of the at least one communication parameter in the 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: a time stamp or identifier (e.g., the date, the time reference, time, label, identifier or index for the ML model state), an identifier of the second UE, a version number of the second machine learning model, or any identifier or time parameter that can be used to distinguish between different versions of the second machine learning model.

At 507, the network node transmits, to the first UE, a configuration indicating at least one condition for determining whether to update the first machine learning model. The first UE receives the configuration.

At 508, the network node transmits, to the second UE, a configuration indicating at least one condition for determining whether to update the second machine learning model. The second UE receives the configuration.

For example, the at least one condition may comprise at least one of: UE performing the at least one prediction of the at least one communication parameter, receiving a configuration for the at least one communication parameter to be used for communication with the network node (e.g., for one or more channels such as: PDSCH, PDCCH, PUSCH, or PUCCH), monitoring predicted quality or utilization of the at least one communication parameter for a pre-defined duration, or an amount of correlation between the at least one prediction and at least one observed value of the at least one communication parameter.

At 509, the first UE performs the at least one prediction of the at least one communication parameter in the at least one cell using the first machine learning model. The at least one prediction may comprise at least one predicted value of the at least one communication parameter.

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

The at least one prediction may comprise at least one of: a predicted identifier of at least one downlink beam, a predicted reference signal received power value of the at least one downlink beam, a predicted signal-to-interference- plus-noise ratio of the at least one downlink beam, a predicted identifier of at least one uplink beam, a predicted reference signal received power value of the at least one uplink beam, a predicted signal-to-interference-plus-noise ratio of the at least 1 one uplink beam, a quality threshold for reference signal received power, or a quality threshold for signal-to-interference-plus-noise ratio.

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

For example, the first UE may determine to update the first machine learning model, if the first UE performs the at least one prediction of the at least one communication parameter (e.g., a downlink beam and/or an uplink beam).

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

As another example, the first UE may determine to update the first machine learning model, if the first UE monitors the predicted quality of the at least one communication parameter or the utilization of the predicted value of the at least one communication parameter (e.g., for a pre-defined duration or the duration is observed to be within a range limit with the prediction duration). In other words, the first UE may perform a communication and compare it with the prediction to determine whether to update the first machine learning model.

As another example, the first UE may determine to update the first machine learning model by rewarding it, if the predicted value of the at least one communication parameter (e.g., communication quality) correlates with the actual observed (measured) value of the at least one communication parameter.

As another example, the first UE may determine to update the first machine learning model by penalizing it, if the predicted value of the at least one communication parameter (e.g., communication quality) does not correlate with the actual observed (measured) value of the at least one communication parameter.

At 511, based on determining to update the first machine learning model (i.e., if the at least one condition is fulfilled), the first UE obtains an updated version of the first machine learning model by updating the first machine learning model based on the at least one prediction performed at the first UE. The updating may refer to training the first machine learning model. The first UE may perform the prediction and the updating one or more times (i.e., perform one or more training iterations). The updated version of the first machine learning model may be denoted as NN_2a-UEl (i.e., state 2 of ML model type A associated with the first UE). The network node may configure how many training iterations and/or updates the first UE should perform.

For example, the first UE may determine whether the at least one prediction performed at the first UE correlates with at least one observed value (i.e., actual measured value or observed or determined communication quality) of the at least one communication parameter. If the at least one prediction correlates with the at least one observed value, then the first UE may update the first machine learning model by rewarding the first machine learning model. If the at least one prediction does not correlate with the at least one observed value, then the first UE may update the first machine learning model by penalizing the first machine learning model.

If the first UE is not able to determine whether the at least one prediction correlates (i.e., correlates or does not correlate) with the at least one observed value (e.g., due to link drop or failure or handover), then the first UE may not perform any update for the first machine learning model, or the first UE may refrain from updating the first machine learning model until it is able to determine whether the at least one prediction correlates (i.e., correlates or does not correlate) with the at least one observed value.

At 512, the first UE transmits, to the network node, the updated version of the first machine learning model. The network node receives the updated version of the first machine learning model. For example, the updated version of the first machine learning model may be transmitted to the network node based on a pre-defined number of training iterations being reached for updating the first machine learning model, or upon or after termination of a connection between the first UE and the network node. The first UE may also transmit, to the network node, together with the updated version of the first machine learning model, an indication indicating a number of training iterations performed for updating the first machine learning model.

The first UE may further indicate, to the network node, whether the first machine learning model was updated or not.

At 513, the second UE performs the at least one prediction of the at least one communication parameter in the at least one cell using the second machine learning model. The at least one prediction may comprise at least one predicted value of the at least one communication parameter.

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

At 515, based on determining to update the second machine learning model (i.e., if the at least one condition is fulfilled), the second UE obtains an updated version of the second machine learning model by updating the second machine learning model based on the at least one prediction performed at the second UE. The updating may refer to training the second machine learning model. The second UE may perform the prediction and the updating one or more times (i.e., perform one or more training iterations). The updated version of the second machine learning model may be denoted as NN_2a-UE2 (i.e., state 2 of ML model type A associated with the second UE). The network node may configure how many training iterations and/or updates the second UE should perform.

For example, the second UE may determine whether the at least one prediction performed at the second UE correlates with at least one observed value (i.e., actual measured value) of the at least one communication parameter. If the at least one prediction correlates with the at least one observed value, then the second UE may update the second machine learning model by rewarding the second machine learning model. If the at least one prediction does not correlate with the at least one observed value, then the second UE may update the second machine learning model by penalizing the second machine learning model.

If the second UE is not able to determine whether the at least one prediction correlates (i.e., correlates or does not correlate) with the at least one observed value (e.g., due to link drop or failure or handover), then the second UE may not perform any update for the second machine learning model, or the second UE may refrain from updating the second machine learning model until it is able to determine whether the at least one prediction correlates (i.e., correlates or does not correlate) with the at least one observed value.

At 516, the second UE transmits, to the network node, the updated version of the second machine learning model. The network node receives the updated version of the second machine learning model.

For example, the updated version of the second machine learning model may be transmitted to the network node based on a pre-defined number of training iterations being reached for updating the second machine learning model, or upon or after termination of a connection between the second UE and the network node (e.g., when transitioning out of the connected state, and/or upon leaving the at least one cell or tracking area (set of cells)). The second UE may also transmit, to the network node, together with the updated version of the second machine learning model, an indication indicating a number of training iterations performed for updating the second machine learning model.

The second UE may further indicate, to the network node, whether the second machine learning model was updated or not.

At 517, the network node may obtain a merged machine learning model by merging (combining) the updated version of the first machine learning model (NN_a2-UEl) with the updated version of the second machine learning model (NN_a2-UE2). The merged machine learning model may be denoted as NN_a2 (i.e., state 2 of ML model type A). In another option, the network node may apply the updates of NN_a2-UE2 on NN_a2-UEl to form NN_a2.

In one example, the network node may merge the ML models by performing an ensemble method, wherein the network node may combine the features and functionalities of the ML models (e.g., artificial neural network models).

In another example, the network node may merge the ML models by concatenating the ML models trained by different UEs. In this case, the network node may take different inputs (from the ML models provided by the UEs) and concatenate them into the same ML model. However, the results of the concatenated dataset may have more dimensions than the original ones.

In another example, the network node may merge the ML models by averaging the ML models. For example, the network node may average the ML models and use the average as a new model. In this example, the network node may take a simple average or a weighted average of the ML models. In the case of the weighted average, the network node may give different weights to different ML models based on the performance of the ML models. As an example, the ML model averaging may refer to per node (neuron) averaging, wherein the weights applied for the input vector of the corresponding nodes of one or more ML models are averaged as described herein.

At 518, the network node may validate at least one of: the merged machine learning model, the updated version of the first machine learning model, or the updated version of the second machine learning model.

For example, the network node may validate the merged machine learning model by determining whether the merged machine learning model (NN_a2) provides a performance improvement compared to a previous version of the machine learning model (NN_al).

Alternatively, the network node may validate the updated version of the first machine learning model and the updated version of the second machine learning model prior to merging them. In this case, the network node may determine whether the updated version of the first machine learning model (NN_a2-UEl) provides a performance improvement compared to a previous version of the first machine learning model (NN_al), and whether the updated version of the second machine learning model (NN_a2-UE2) provides a performance improvement compared to a previous version of the second machine learning model (NN_al).

At 519, if the merged machine learning model (or the updated version of the first machine learning model and the second machine learning model) provides the performance improvement, the network node may transmit the merged machine learning model to the first UE and the second UE for performing at least one subsequent prediction of the at least one communication parameter in the at least one cell using the merged machine learning model. Alternatively, or additionally, the merged machine learning model may be transmitted to one or more other UEs than the first UE and the second UE.

Alternatively, if the merged machine learning model does not provide the performance improvement, then the network node may discard the merged machine learning model and revert back to the previous version.

In some examples, not all the UEs may be capable of doing both the prediction and the online training. The functions of UE1 and UE2 (e.g., 509-512 and 513-516) may be performed simultaneously or in a different order than shown in FIG. 5.

FIG. 6 illustrates a signal flow diagram according to an example embodiment. In this example embodiment, the network node monitors the UE prediction based on the communication quality and indicates the UE to update the used machine learning model (e.g., with a reference to a specific prediction instance).

Although one UE is shown in FIG. 6, it should be noted that the number of UEs may also be more than one. In other words, there may be one or more UEs. In addition, the signaling procedure illustrated in FIG. 6 may be extended and applied according to the actual number of UEs.

Referring to FIG. 6, at 601, a network node transmits an indication indicating support for provisioning one or more machine learning models for performing at least one prediction of at least one communication parameter in at least one cell. In other words, the network node may broadcast or advertise, for example in system information, the support for provisioning and using the one or more machine learning models. Some examples of the ML models are illustrated in FIGS. 12-14. The indication is received by a UE 100, 400.

The network node may be a radio access network node 104, 404 such as a gNB (e.g., as shown in FIG. 4B), or the network node may be a separate network entity comprising Al/ML functions 407 (e.g., as shown in FIG. 4A).

The at least one communication parameter may comprise, for example, at least one of: a beam index or identifier (e.g., SSB or CS1-RS index), quality of the beam index (e.g., reference signal received power and/or signal-to-interference- plus-noise ratio), or a duration of the quality of the beam index or identifier (e.g., time period that a specific reference signal is detectable or has quality above a quality threshold value). For example, the at least one communication parameter may be a downlink beam or an uplink beam (e.g., the corresponding downlink reference signal of the DL or UL beam that may be used for communication).

The one or more machine learning models may comprise a set of machine learning model types (e.g., Model type A, Model type B, ... Model type Y) for UE-based communication parameter prediction.

The support indication may provide information or reference to the one or more machine learning models that may be referred to with an index value or type value (e.g., Model type A) or similar. For example, the one or more machine learning models may have a pre-defined structure associated with an index or identifier value or a type value, wherein the indication indicating the support for provisioning the one or more machine learning models may comprise the index value or the type value per machine learning model of the one or more machine learning models. The index value or type value may indicate, for example, at least one of: number of input vectors of the machine learning model (e.g., artificial neural network), number of output vectors of the machine learning model, number of nodes of the machine learning model, number of layers of the machine learning model, or neural network type of the machine learning model.

As an example, a Model type A may have specific input parameters that may comprise of using the UE location, multiple antenna panels, measurement capability (e.g., accuracy).

As another example, a Model type B may have specific input parameters that assume that no UE location is used for prediction, UE does not have multiple antenna panels, measurement capability (e.g., accuracy) is lower, etc.

At 602, the UE transmits, to the network node, an indication indicating a capability of the UE for at least one of: predicting the at least one communication parameter, or training at least one of the one or more machine learning models. The network node receives the capability indication.

For example, the UE may indicate during the connection setup phase, or when entering the at least one cell, or during the connected state, that it is capable of predicting the at least one communication parameter for one or more ML model types advertised by the network node, and/or that the UE supports (or does not support) online training of a given ML model type advertised by the network node.

The UE may additionally indicate, to the network node, at least one parameter indicative of communication capabilities of the UE, wherein the at least one parameter may comprise at least one of: number of antenna panel(s) of the UE, relative location of antenna panels with respect to one another, number of beams per antenna panel (e.g., maximum number of beams that the UE can form), coverage of the beams in degrees per antenna panel, the beamforming codebook defining how the UE forms the grid of beams, etc.

The UE may transmit the capability indication in response to receiving the indication indicating support for provisioning the one or more machine learning models. In another option, the UE may transmit the capability indication to the network node prior to receiving the indication indicating support for provisioning the one or more machine learning models (i.e., 602 maybe performed before 601). In this latter case, the network node may transmit the indication indicating support for provisioning the one or more machine learning models to the UE via dedicated signaling as a response for receiving the capability indication from the UE.

At 603, the network node determines, based on the capability of the UE, a machine learning model of the one or more machine learning models to be transmitted to the UE.

At 604, the network node transmits, to the UE, the machine learning model for performing the at least one prediction of the at least one communication parameter in the at least one cell. The UE receives the machine learning model.

The machine learning model may be associated with at least one of: a time stamp or identifier (e.g., the date, the time reference, time, label, identifier or index for 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 between different versions of the machine learning model.

At 605, the network node transmits, to the UE, a configuration for predicting the at least one communication parameter. The UE receives the configuration.

For example, the configuration may indicate at least one of: one or more of thresholds, one or more functions, and / or one or more parameters for predicting the at least one communication parameter and/or estimating the quality of the prediction. In one example, these may comprise a beam quality threshold for determining the correlation and/or decorrelation of the prediction (with observed value), input parameters used for prediction (e.g. Ll-RSRP measurements), and/or output parameters such as number K of top-K beams (in terms of quality) that are predicted. RSRP is an abbreviation for reference signal received power.

At 606, based on the configuration, the UE performs the at least one prediction of the at least one communication parameter in the at least one cell using the machine learning model. The at least one prediction may comprise at least one predicted value of the at least one communication parameter. The UE may label or index the at least one prediction of the at least one communication parameter, and store the at least one prediction with the label or index for example in its internal memory. The stored at least one prediction may be referred by the network node with an update command, and/or used by the UE to update the machine learning model.

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

The at least one prediction may comprise at least one of: a predicted identifier of at least one downlink beam, a predicted reference signal received power value of the at least one downlink beam, a predicted signal-to-interference- plus-noise ratio of the at least one downlink beam, a predicted identifier of at least one uplink beam, a predicted reference signal received power value of the at least one uplink beam, a predicted signal-to-interference-plus-noise ratio of the at least one uplink beam, a quality threshold for reference signal received power, or a quality threshold for signal-to-interference-plus-noise ratio.

At 607, the UE transmits, to the network node, information indicating the at least one prediction performed at the UE together with the label or index. 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 the information.

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

For example, the network node may determine whether the at least one prediction correlates with at least one observed (measured) value of the at least one communication parameter.

At 609, based on the monitoring, the network node transmits, to the UE, an indication to update the machine learning model (e.g., with a reference to a specific prediction instance). The UE receives the indication.

For example, if the at least one prediction (e.g., communication quality) correlates with the at least one observed value (as observed by the network node), then the indication may indicate to update the machine learning model by rewarding the machine learning model.

As another example, if the at least one prediction (e.g., communication quality) does not correlate with the at least one observed value (as observed by the network node), then the indication may indicate to update the machine learning model by penalizing the machine learning model.

If the network node is not able to determine whether the at least one prediction (e.g., communication quality) correlates (i.e., correlates or does not correlate) with the at least one observed value (e.g., due to link drop or failure or handover), then the network node may refrain from transmitting the indication to update the machine learning model until it is able to determine whether the at least one prediction correlates with the at least one observed value, or the network node may transmit an explicit indication to the UE to not update the machine learning model.

At 610, the UE determines, based on the indication received at 609, to update the machine learning model.

At 611, based on determining to update the machine learning model, the UE obtains an updated version of the machine learning model by updating the machine learning model based on the at least one prediction stored at the UE. The updating may refer to training the machine learning model. The UE may perform the prediction and the updating one or more times (i.e., perform one or more training iterations). The network node may configure how many training iterations and/or updates the UE should perform.

At 612, the UE transmits, to the network node, the updated version of the machine learning model. The network node receives the updated version of the machine learning model.

For example, the updated version of the machine learning model may be transmitted to the network node based on a pre-defined number of training iterations being reached for updating the machine learning model, or upon or after termination of a connection between the UE and the network node (e.g., when transitioning out of the connected state, and/or upon leaving the at least one cell or tracking area (set of cells)). The UE may also transmit, to the network node, together with the machine learning model, an indication indicating a number of training iterations performed for updating the machine learning model.

The UE may further indicate, to the network node, whether the machine learning model was updated or not.

At 613, the network node may validate the updated version of the machine learning model. In this case, the network node may determine whether the updated version of the machine learning model provides a performance improvement compared to a previous version of the machine learning model.

If the updated version of the machine learning model does not provide the performance improvement, then the network node may discard the updated version of the machine learning model and revert back to the previous version of the machine learning model.

At 614, if the updated version of the machine learning model provides the performance improvement, the network node may 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.

As an alternative to 613, the network node may validate the merged machine learning model instead of the updated version of the machine learning model provided by the UE. At 615, if the updated version of the machine learning model (or the merged machine learning model) provides the performance improvement, the network node may transmit the merged machine learning model, or the updated version of the machine learning model, to the UE for performing at least one subsequent prediction of the at least one communication parameter in the at least one cell. Alternatively, the merged machine learning model, or the updated version of the machine learning model, may be transmitted to one or more other UEs.

FIG. 7 illustrates a flow chart according to an example embodiment of a method performed by an apparatus 1500. For example, the apparatus 1500 may be, or comprise, or be comprised in, a user equipment 100, 102, 400.

Referring to FIG. 7, in block 701, the apparatus receives, from a network node, a machine learning model for performing at least one prediction of at least one communication parameter in at least one cell. Some examples of the ML model are illustrated in FIGS. 12-14. However, the ML model is not limited to these examples.

The machine learning model may be associated with at least one of: a time stamp (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 between different versions of the machine learning model.

In block 702, the apparatus receives, from the network node, at least one of: an indication to update the machine learning model, or a configuration indicating at least one condition for determining whether to update the machine learning model.

In block 703, the apparatus performs the at least one prediction of the at least one communication parameter in the at least one cell using the machine learning model.

In block 704, the apparatus determines, based on the indication or the at least one condition, whether to update the machine learning model; In block 705, the apparatus obtains, based on determining to update the machine learning model, an updated version of the machine learning model by updating the machine learning model based on the at least one prediction; and

In block 706, the apparatus transmits, to the network node, the updated version of the machine learning model.

For example, the at least one communication parameter may comprise at least one of: a beam index, quality of the beam index, or a duration of the quality of the beam index.

For example, the at least one condition may comprise at least one of: performing the at least one prediction of the at least one communication parameter, receiving a configuration for the at least one communication parameter to be used for communication with the network node (e.g., for one or more channels such as: PDSCH, PDCCH, PUSCH, or PUCCH), monitoring predicted quality or utilization of the at least one communication parameter for a pre-defined duration, or an amount of correlation between the at least one prediction and at least one observed value of the at least one communication parameter.

The apparatus may receive, from the network node, an indication indicating support for provisioning one or more machine learning models for performing the at least one prediction of the at least one communication parameter in the at least one cell; and transmit, to the network node, an indication indicating a capability of for at least one of: predicting the at least one communication parameter, or training at least one of the one or more machine learning models.

The one or more machine learning models have a pre-defined structure associated with an index value or a type value, wherein the indication indicating the support for provisioning the one or more machine learning models may comprise the index value or the type value per machine learning model of the one or more machine learning models.

In one example, the apparatus may determine whether the at least one prediction correlates with at least one observed value of the at least one communication parameter; and, based on determining that the at least one prediction correlates with the at least one observed value, update the machine learning model by rewarding the machine learning model.

In another example, the apparatus may determine whether the at least one prediction correlates with at least one observed value of the at least one communication parameter; and based on determining that the at least one prediction does not correlate with the at least one observed value, update the machine learning model by penalizing the machine learning model.

In another example, the apparats may attempt to determine whether the at least one prediction correlates with at least one observed value of the at least one communication parameter; and based on not being able to determine whether the at least one prediction correlates with the at least one observed value, refrain from updating the machine learning model until being able to determine whether the at least one prediction correlates with the at least one observed value.

The apparatus may receive, from the network node, a configuration for predicting the at least one communication parameter, wherein the at least one prediction is performed based on the configuration; and transmit, to the network node, information indicating the at least one prediction.

The apparatus may label or index the at least one prediction of the at least one communication parameter; and store the at least one prediction with the label or index, wherein the machine learning model may be updated based on the stored at least one prediction.

The updated version of the machine learning model may be transmitted to the network node based on a pre-defined number of training iterations being reached for updating the machine learning model, or upon or after termination of a connection between the apparatus and the network node.

The apparatus may transmit, to the network node, an indication indicating a number of training iterations performed for updating the machine learning model.

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

In one example, the at least one prediction may comprise multiple predictions in a time sequence.

The at least one prediction may be performed by providing, to the machine learning model, input information comprising at least one of: a threshold for reference signal received power, or a threshold for signal-to-interference-plus- noise ratio.

FIG. 8 illustrates a flow chart according to an example embodiment of a method performed by an apparatus 1500. For example, the apparatus 1500 may be, or comprise, or be comprised in, a user equipment 100, 102, 400.

Referring to FIG. 8, in block 801, the apparatus receives, from a network node, a machine learning model for performing at least one prediction of at least one communication parameter in at least one cell. Some examples of the ML model are illustrated in FIGS. 12-14. However, the ML model is not limited to these examples.

In block 802, the apparatus receives, from the network node, a configuration indicating at least one condition for determining whether to update the machine learning model.

For example, the at least one condition may comprise an amount of correlation between the at least one prediction and at least one observed value of the at least one communication parameter. If the at least one prediction correlates with the at least one observed value of the at least one communication parameter, then the apparatus may update the machine learning model by rewarding it. If the at least one prediction does not correlate with the at least one observed value of the at least one communication parameter, then the apparatus may update the machine learning model by penalizing it.

In block 803, the apparatus performs the at least one prediction of the at least one communication parameter in the at least one cell using the machine learning model.

In block 804, the apparatus attempts to determine whether the at least one prediction correlates with at least one observed value of the at least one communication parameter.

In block 805, based on not being able (e.g., due to link drop or failure or handover) to determine whether the at least one prediction correlates with the at least one observed value (block 804: no), the apparatus refrains from updating the machine learning model until being able to determine whether the at least one prediction correlates with the at least one observed value.

Alternatively, in block 806, based on being able to determine whether the at least one prediction correlates with the at least one observed value (block 804: yes), the apparatus determines whether the at least one prediction correlates with at least one observed value of the at least one communication parameter.

In one example, the correlation between the at least one prediction and the at least one observed value of the at least one communication parameter may refer to, for example, reference signal quality and/or the difference between the at least one prediction (predicted value) and the at least one observed value (measured value). The difference may be determined based on a threshold value between the absolute values of the predicted value and the observed value (measured value).

For example, if the absolute value of the difference between the predicted value and the observed value (measured value) is below or equal to the threshold value, then this may indicate that the predicted value correlates with the observed value. As an example, if the observed value is within the range of ±X of the predicted value, the predicted value may be determined to be correlating. The value X (i.e., the threshold value) may be configured by the network node. As another example, if the absolute value of the difference between the predicted value and the observed value (measured value) is above the threshold, then this may indicate that the predicted value does not correlate with the observed value, or that there is not sufficient correlation between the predicted value and the observed value. As an example, if the observed value is not within the range of ±X of the predicted value, the predicted value may be determined not to be correlating. The value X (i.e., the threshold value) may be configured by the network node.

In block 807, based on determining that the at least one prediction correlates with the at least one observed value (block 806: yes), the apparatus updates the machine learning model by rewarding the machine learning model.

Alternatively, in block 808, based on determining that the at least one prediction does not correlate with the at least one observed value (block 806: no), the apparatus updates the machine learning model by penalizing the machine learning model.

In block 809, following block 807 or block 808, the apparatus transmits the updated version of the machine learning model to the network node.

FIG. 9 illustrates a flow chart according to an example embodiment of a method performed by an apparatus 1600. For example, the apparatus 1600 may be, or comprise, or be comprised in, a network node. The network node may be a radio access network node 104, 404 such as a gNB (e.g., as shown in FIG. 4B), or the network node may be a separate network entity comprising Al/ML functions 407 (e.g., as shown in FIG. 4A).

Referring to FIG. 9, in block 901, the apparatus transmits, to at least one user equipment 100, 102, 400, a machine learning model for performing at least one prediction of at least one communication parameter in at least one cell.

The machine learning model may be associated with at least one of: a time stamp (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 between different versions of the machine learning model. In block 902, the apparatus transmits, 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 to update the machine learning model.

In block 903, the apparatus receives, from the at least one user equipment, an updated version of the machine learning model.

The at least one condition may comprise at least one of: performing the at least one prediction of the at least one communication parameter, receiving a configuration for the at least one communication parameter to be used for communication with the network node, monitoring predicted quality or utilization of the at least one communication parameter for a pre-defined duration, or an amount of correlation between the at least one prediction and at least one observed value of the at least one communication parameter.

The apparatus may transmit an indication indicating support for provisioning one or more machine learning models for performing the at least one prediction of the at least one communication parameter in the at least one cell; receive, from the at least one user equipment, an indication indicating a capability for at least one of: predicting the at least one communication parameter, or training at least one of the one or more machine learning models; and determine, based on the capability, the machine learning model to be transmitted to the at least one user equipment.

The apparatus may transmit, to the at least one user equipment, a configuration for predicting the at least one communication parameter; receive, from the at least one user equipment, information indicating the at least one prediction of the at least one communication parameter; monitor the at least one prediction of the at least one user equipment based on communication quality between the apparatus and the at least one user equipment; and transmit, based on the monitoring, to the at least one user equipment, the indication to update the machine learning model.

In one example, the apparatus may determine whether the at least one prediction correlates with at least one observed value of the at least one communication parameter; and based on determining that the at least one prediction correlates with the at least one observed value, transmit, to the at least one user equipment, the indication to update the machine learning model by rewarding the machine learning model.

In another example, the apparatus may determine whether the at least one prediction correlates with at least one observed value of the at least one communication parameter; and based on determining that the at least one prediction does not correlate with the at least one observed value, transmit, to the at least one user equipment, the indication to update the machine learning model by penalizing the machine learning model.

In another example, the apparatus may attempt to determine whether the at least one prediction correlates with at least one observed value of the at least one communication parameter; and based on not being able to determine whether the at least one prediction correlates with the at least one observed value, refrain from transmitting, to the at least one user equipment, the indication to update the machine learning model until being able to determine whether the at least one prediction correlates with the at least one observed value.

In another example, the apparatus may attempt to determine whether the at least one prediction correlates with at least one observed value of the at least one communication parameter; and based on not being able to determine whether the at least one prediction correlates with the at least one observed value, transmit, to the at least one user equipment, an indication to not update the machine learning model.

In one example, the apparatus may determine whether the updated version of the machine learning model provides a performance improvement compared to a previous version of the machine learning model; and based on determining that the updated version provides the performance improvement, transmit the updated version of the machine learning model to one or more other user equipments for performing at least one subsequent prediction of the at least one communication parameter in the at least one cell.

In another example, the apparatus may determine whether the updated version of the machine learning model provides a performance improvement compared to a previous version of the machine learning model; and based on determining that the updated version does not provide the performance improvement, discard the updated version of the machine learning model.

The apparatus may 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 user equipments.

FIG. 10 illustrates a flow chart according to an example embodiment of a method performed by an apparatus 1600. For example, the apparatus 1600 may be, or comprise, or be comprised in, a network node. The network node may be a radio access network node 104, 404 such as a gNB (e.g., as shown in FIG. 4B), or the network node may be a separate network entity comprising Al/ML functions 407 (e.g., as shown in FIG. 4A).

Referring to FIG. 10, in block 1001, the apparatus transmits, 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. Some examples of the ML model are illustrated in FIGS. 12-14. However, the ML model is not limited to these examples.

In block 1002, the apparatus transmits, to the at least one user equipment, a configuration for predicting the at least one communication parameter.

In block 1003, the apparatus receives, from the at least one user equipment, information indicating the at least one prediction of the at least one communication parameter performed at the at least one user equipment.

In block 1004, the apparatus attempts to determine whether the at least one prediction correlates with at least one observed value of the at least one communication parameter. In block 1005, based on not being able (e.g., due to link drop or failure or handover) to determine whether the at least one prediction correlates with the at least one observed value (block 1004: no), the apparatus refrains from transmitting, to the at least one user equipment, an indication to update the machine learning model until being able to determine whether the at least one prediction correlates with the at least one observed value. Alternatively, the apparatus may transmit, to the at least one user equipment, an indication to not update the machine learning model. in block 1006, based on being able to determine whether the at least one prediction correlates with the at least one observed value (block 1004: yes), the apparatus determines whether the at least one prediction correlates with at least one observed value of the at least one communication parameter.

For example, if the difference between the absolute values of the predicted value and the observed value (measured value) is below the threshold value, then this may indicate that the predicted value correlates with the observed value.

As another example, if the difference between the absolute values of the predicted value and the observed value (measured value) is above the threshold, then this may indicate that the predicted value does not correlate with the observed value, or that there is not sufficient correlation between the predicted value and the observed value.

In block 1007, based on determining that the at least one prediction correlates with the at least one observed value (block 1006: yes), the apparatus transmits, to the at least one user equipment, an indication to update the machine learning model by rewarding the machine learning model.

Alternatively, in block 1008, based on determining that the at least one prediction does not correlate with the at least one observed value (block 1006: no), the apparatus transmits, to the at least one user equipment, an indication to update the machine learning model by penalizing the machine learning model.

In block 1009, following block 1007 or block 1008, the apparatus receives, from the at least one user equipment, an updated version of the machine learning model.

FIG. 11 illustrates a flow chart according to an example embodiment of a method performed by an apparatus 1600. For example, the apparatus 1600 may be, or comprise, or be comprised in, a network node. The network node may be a radio access network node 104, 404 such as a gNB (e.g., as shown in FIG. 4B), or the network node may be a separate network entity comprising Al/ML functions 407 (e.g., as shown in FIG. 4A).

Referring to FIG. 11, in block 1101, the apparatus transmits, to at least one user equipment 100, 102, 400, a machine learning model for performing at least one prediction of at least one communication parameter in at least one cell. Some examples of the ML model are illustrated in FIGS. 12-14. However, the ML model is not limited to these examples.

In block 1102, the apparatus transmits, 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 to update the machine learning model.

In block 1103, the apparatus receives, from the at least one user equipment, an updated version of the machine learning model.

In block 1104, the apparatus determines whether the updated version of the machine learning model provides a performance improvement compared to a previous version of the machine learning model.

For example, the apparatus may determine an error or error rate of the updated version and compare it with a performance threshold. In block 1105, based on determining that the updated version does not provide the performance improvement (block 1104: no), the apparatus discards the updated version of the machine learning model.

Alternatively, in block 1106, based on determining that the updated version provides the performance improvement (block 1104: yes), the apparatus transmits the updated version of the machine learning model to one or more other user equipments for performing at least one subsequent prediction of the at least one communication parameter in the at least one cell.

In another option, the apparatus may 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 user equipments, in which case the apparatus may transmit the merged machine learning model instead of the updated version of the machine learning model provided from the user equipment.

The blocks, related functions, and information exchanges (messages) described above by means of FIGS. 5-11 are in no absolute chronological order, and some of them may be performed simultaneously or in an order differing from the described one. Other functions can also be executed between them or within them, and other information may be sent, and/or other rules applied. Some of the blocks or part of the blocks or one or more pieces of information can also be left out or replaced by a corresponding block or part of the block or one or more pieces of information.

As used herein, “at least one of the following: <a list of two or more elements>” and “at least one of <a list of two or more elements>” and similar wording, where the list of two or more elements are joined by “and” or “or”, mean at least any one of the elements, or at least any two or more of the elements, or at least all the elements.

FIG. 12 illustrates an example of spatial domain prediction based on a convolutional neural network (CNN) model 1200. The CNN model 1200 may be used at the UE side for beam prediction both in uplink and downlink aspects.

The UE may deploy the CNN model 1200 to perform training in spatial domain prediction, wherein the CNN model 1200 may be the updated based on a quality threshold in spatial domain (at time instant). The network node may trigger at least one condition (e.g., one or more coefficients or indicators) to configure the UE to train the CNN model 1200.

For downlink beam management, the input information 1201 of the CNN model 1200 may comprise, for example, at least one of: one or more DL beam IDs, reference signal received power (RSRP) of the one or more DL beam IDs, signal-to-interference-plus-noise ratio (S1NR) of the one or more DL beam IDs, a threshold for RSRP, a threshold for S1NR, one or more antenna panel indices, and/or one or more beam indices received at the UE.

The RSRP threshold and the S1NR threshold in the input information may prevent low-quality input information from being fed to the CNN model 1200, since the low-quality input information could cause an error or unwanted results in the prediction.

For downlink beam management, the output information 1202 of the CNN model 1200 may comprise at least one of: one or more predicted DL beam IDs, predicted RSRP of the one or more DL beam IDs, predicted S1NR of the one or more DL beam IDs, a quality threshold for RSRP, and/or a quality threshold for S1NR in spatial domain (at a particular time).

The quality threshold(s) may be used to prevent beam IDs that have low quality (based on the prediction) from being indicated to the network node.

The UE may be triggered by the network node to configure the CNN model 1200 for uplink transmission as well. For uplink beam management, the input information 1201 of the CNN model 1200 may comprise at least one of: one or more UL beam IDs, RSRP of the one or more UL beam IDs, S1NR of the one or more UL beam IDs, a threshold for RSRP, a threshold for S1NR, and/or one or more antenna panel indices.

For uplink beam management, the output information 1202 of the CNN model 1200 may comprise at least one of: one or more predicted UL beam IDs, predicted RSRP of the one or more UL beam IDs, predicted S1NR of the one or more UL beam IDs, a quality threshold for RSRP, and/or a quality threshold for S1NR in spatial domain (at a particular time).

FIG. 13 illustrates an example of time domain prediction based on a long-short-term-memory (LSTM) recurrent neural network (RNN) model 1300. The LSTM RNN model 1300 may be used at the UE side for beam prediction both in uplink and downlink aspects. With this model, the UE can predict outputs and the quality thresholds in a time sequence.

The network node may trigger at least one condition (e.g., one or more coefficients or indicators) to configure the UE to train the LSTM RNN model 1300.

For downlink beam management, the input information 1301 of the LSTM RNN model 1300 may comprise, for example, at least one of: one or more DL beam IDs in time sequence, reference signal received power (RSRP) of the one or more DL beam IDs in time sequence, signal-to-interference-plus-noise ratio (S1NR) of the one or more DL beam IDs in time sequence, a threshold for RSRP in time sequence, a threshold for S1NR in time sequence, one or more antenna panel indices, and/or one or more beam indices received at the UE. In other words, the inputs may be in terms of a time sequence (e.g., t-Tl,...,t-2, t).

For downlink beam management, the output information 1302 of the LSTM RNN model 1300 may comprise at least one of: one or more predicted DL beam IDs in time sequence, predicted RSRP of the one or more DL beam IDs in time sequence, predicted S1NR of the one or more DL beam IDs in time sequence, a quality threshold for RSRP in time duration, and/or a quality threshold for S1NR in time duration. In other words, the outputs may be in terms of a time sequence (e.g., t+2,...,t+T) after the time sequence of the inputs.

The UE may be triggered by the network node to configure the LSTM RNN model 1300 for uplink transmission as well. For uplink beam management, the input information 1301 of the LSTM RNN model 1300 may comprise at least one of: one or more UL beam IDs in time sequence, RSRP of the one or more UL beam IDs in time sequence, S1NR of the one or more UL beam IDs in time sequence, a threshold for RSRP in time sequence, a threshold for S1NR in time sequence, and/or one or more antenna panel indices. In other words, the inputs may be in terms of a time sequence (e.g., t-Tl,...,t-2, t). For uplink beam management, the output information 1302 ofthe LSTM RNN model 1300 may comprise at least one of: one or more predicted UL beam IDs in time sequence, predicted RSRP of the one or more UL beam IDs in time sequence, predicted S1NR of the one or more UL beam IDs in time sequence, a quality threshold for RSRP in time duration, and/or a quality threshold for S1NR in time duration. In other words, the outputs may be in terms of a time sequence (e.g., t+2,...,t+T) after the time sequence of the inputs.

FIG. 14 illustrates an example of prediction based on a deep reinforcement learning (DRL) model 1400. The DRL model 1400 may be used at the UE side for beam prediction both in uplink and downlink aspects.

Reinforcement learning is a type of machine learning where an agent (i.e., the UE in this case) learns to make decisions by interacting with an environment 1405. The learning process involves receiving feedback in the form of rewards or penalties based on the agent's actions. The goal of the agent is to learn a policy that maps states to actions in order to maximize the cumulative reward over time.

In reinforcement learning, a reward is a positive feedback received by the agent for taking a desirable action in a given state, while a penalty, often referred to as a negative reward, is a negative feedback received for taking an undesirable action. The agent learns to make better decisions by trying to maximize the rewards and minimize the penalties over time.

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

In deep reinforcement learning, the observation space may be processed by a neural network (such as a deep neural network or a convolutional neural network) that serves as a function approximator to learn the optimal policy or value function.

The action space 1402 represents the set of all possible actions that the 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 may be discrete, with a finite set of distinct actions, or continuous, with actions represented by continuous variables.

In deep reinforcement learning, the neural network learns to map the observations from the observation space to appropriate actions in the action space to maximize the cumulative reward over time. The learning process involves updating the neural network's parameters based on the observed rewards and penalties, using techniques such as Q-learning, policy gradients, or actor-critic methods.

In this example, the UE may deploy the DRL model 1400 to perform training, where the penalty and reward conditions are considered. The DRL model 1400 may be updated depending on the penalty and reward conditions.

For example, for downlink beam prediction, the network node may trigger the UE to configure the DRL model 1400 to predict at least one of: one or more DL beam IDs, RSRP of the one or more DL beam IDs, S1NR of the one or more DL beam IDs, a quality threshold for RSRP, and/or a quality threshold for S1NR. In other words, for downlink beam prediction, the action space 1402 may comprise one or more of these outputs.

For downlink beam prediction, the observation space 1401 may comprise at least one of: one or more DL beam IDs, reference signal received power (RSRP) of the one or more DL beam IDs, signal-to-interference-plus-noise ratio (S1NR) of the one or more DL beam IDs, a threshold for RSRP, a threshold for S1NR, one or more antenna panel indices, and/or one or more beam indices received at the UE. These inputs may be from a previous sequence.

The network node may also trigger the UE to configure the DRL model 1400 for uplink transmission as well. For uplink beam prediction, the DRL model 1400 may predict at least one of: one or more UL beam IDs, RSRP of the one or more UL beam IDs, SINR of the 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, the action space 1402 may comprise one or more of these outputs. The quality threshold(s) may be considered in a similar manner as for downlink transmission, but with respect to uplink beams.

For uplink beam prediction, the observation space 1401 may comprise at least one of: one or more UL beam IDs, RSRP of the one or more UL beam IDs, SINR of the one or more UL beam IDs, a threshold for RSRP, a threshold for SINR, and/or one or more antenna panel indices. These inputs may be from a previous sequence.

In the action space 1402, the UE may determine the output (action space) where the quality threshold is considered. The quality threshold may be satisfied, if it is above a pre-defined threshold.

The action 1403 may comprise, for example, configuring a (new) serving beam or maintaining the current beam.

The reward and penalty evaluation 1404 may comprise, for example, at least one of: mapping a predicted RSRP value of a given beam ID to an observed channel quality indicator (CQI) value, mapping a 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.

The RSRP may be Ll-RSRP or L3-RSRP. The UE may keep using the current action space to calculate the reward. If the quality threshold and/or Ll- RSRP threshold or L3-RSRP threshold is above the pre-defined threshold, the UE may update the action space 1402.

However, if the quality threshold and/or Ll-RSRP threshold or L3-RSRP threshold are less than the pre-defined 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 a reward.

FIG. 15 illustrates an example of an apparatus 1500 comprising means for performing one or more of the example embodiments described above. For example, the apparatus 1500 may be an apparatus such as, or comprising, or comprised in, a user equipment (UE) 100, 102, 400.

The user equipment may also be called a wireless communication device, a subscriber unit, a mobile station, a remote terminal, an access terminal, a user terminal, a terminal device, or a user device.

The apparatus 1500 may comprise a circuitry or a chipset applicable for realizing one or more of the example embodiments described above. For example, the apparatus 1500 may comprise at least one processor 1510. The at least one processor 1510 interprets instructions (e.g., computer program instructions) and processes data. The at least one processor 1510 may comprise one or more programmable processors. The at least one processor 1510 may comprise programmable hardware with embedded firmware and may, alternatively or additionally, comprise one or more application-specific integrated circuits (ASICs).

The at least one processor 1510 is coupled to at least one memory 1520. The at least one processor is configured to read and write data to and from the at least one memory 1520. The at least one memory 1520 may comprise one or more memory units. The memory units may be volatile or non-volatile. It is to be noted that there may be one or more units of non-volatile memory and one or more units of volatile memory or, alternatively, one or more units of non-volatile memory, or, alternatively, one or more units of volatile memory. 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), electronically erasable programmable read-only memory (EEPROM), flash memory, optical storage or magnetic storage. In general, memories may be referred to as non-transitory computer readable media. The term “non-transitory,” as used herein, is a limitation of the medium itself (i.e., tangible, not a signal) as opposed to a limitation on data storage persistency (e.g., RAM vs. ROM). The at least one memory 1520 stores computer readable instructions that are executed by the at least one processor 1510 to perform one or more of the example embodiments described above. For example, non-volatile memory stores the computer readable instructions, and the at least one processor 1510 executes the instructions using volatile memory for temporary storage of data and/or instructions. The computer readable instructions may refer to computer program code.

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

In the context of this document, a “memory” or “computer-readable media” or “computer-readable medium” may be any non-transitory media or medium or means that can contain, store, communicate, propagate or transport the instructions for use by or in connection with an instruction execution system, apparatus, or device, such as a computer. The term “non-transitory,” as used herein, is a limitation of the medium itself (i.e., tangible, not a signal) as opposed to a limitation on data storage persistency (e.g., RAM vs. ROM).

The apparatus 1500 may further comprise, or be connected to, an input unit 1530. The input unit 1530 may comprise one or more interfaces for receiving input. The one or more interfaces may comprise 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. Further, the input unit 1530 may comprise an interface to which external devices may connect to.

The apparatus 1500 may also comprise an output unit 1540. The output unit may comprise 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 further comprise one or more audio outputs. The one or more audio outputs may be for example loudspeakers.

The apparatus 1500 further comprises a connectivity unit 1550. The connectivity unit 1550 enables wireless connectivity to one or more external devices. The connectivity unit 1550 comprises at least one transmitter and at least one receiver that may be integrated to the apparatus 1500 or that the apparatus 1500 may be connected to. The at least one transmitter comprises at least one transmission antenna, and the at least one receiver comprises at least one receiving antenna. The connectivity unit 1550 may comprise an integrated circuit or a set of integrated circuits that provide the wireless communication capability for the apparatus 1500. Alternatively, the wireless connectivity may be a hardwired application-specific integrated circuit (ASIC). The connectivity unit 1550 may also provide means for performing at least some of the blocks or functions of one or more example embodiments described above. The connectivity unit 1550 may comprise one or more components, such as: power amplifier, digital front end (DFE), analog-to-digital converter (ADC), digital-to-analog converter (DAC), frequency converter, (de)modulator, and/or encoder/decoder circuitries, controlled by the corresponding controlling units.

It is to be noted that the apparatus 1500 may further comprise various components not illustrated in FIG. 15. The various components may be hardware components and/or software components.

FIG. 16 illustrates an example of an apparatus 1600 comprising means for performing one or more of the example embodiments described above. For example, the apparatus 1600 may be an apparatus such as, or comprising, or comprised in, a network node. The network node may be a radio access network node 104, 404 such as a gNB (e.g., as shown in FIG. 4B), or the network node may be a separate network entity comprising Al/ML functions 407 (e.g., as shown in F1G. 4A).

The network node may also be referred to, for example, 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 (1AB) 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 (RRH), or a transmission and reception point (TRP).

The apparatus 1600 may comprise, for example, a circuitry or a chipset applicable for realizing one or more of the example embodiments described above. The apparatus 1600 may be an electronic device comprising one or more electronic circuitries. The apparatus 1600 may comprise a communication control circuitry 1610 such as at least one processor, and at least one memory 1620 storing instructions 1622 which, when executed by the at least one processor, cause the apparatus 1600 to carry out one or more of the example embodiments described above. Such instructions 1622 may, for example, include computer program code (software). The at least one processor and the at least one memory storing the instructions may provide the means for providing or causing the performance of any of the methods and/or blocks described above.

The processor is coupled to the memory 1620. The processor is configured to read and write data to and from the memory 1620. The memory 1620 may comprise one or more memory units. The memory units may be volatile or non-volatile. It is to be noted that there may be one or more units of non-volatile memory and one or more units of volatile memory or, alternatively, one or more units of non-volatile memory, or, alternatively, one or more units of volatile memory. 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), electronically erasable programmable read-only memory (EEPROM), flash memory, optical storage or magnetic storage. In general, memories may be referred to as non-transitory computer readable media. The term “non-transitory,” as used herein, is a limitation of the medium itself (i.e., tangible, not a signal) as opposed to a limitation on data storage persistency (e.g., RAM vs. ROM). The memory 1620 stores computer readable instructions that are executed by the processor. For example, non-volatile memory stores the computer readable instructions, and the processor executes the instructions using volatile memory for temporary storage of data and/or instructions.

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

The memory 1620 may 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 comprise a configuration database for storing configuration data, such as a current neighbour cell list, and, in some example embodiments, structures of frames used in the detected neighbour cells.

The apparatus 1600 may further comprise or be connected to a communication interface 1630, such as a radio unit, comprising hardware and/or software for realizing communication connectivity with one or more wireless communication devices according to one or more communication protocols. The communication interface 1630 comprises at least one transmitter (Tx) and at least one receiver (Rx) that may be integrated to the apparatus 1600 or that the apparatus 1600 may be connected to. The communication interface 1630 may provide means for performing some of the blocks for one or more example embodiments described above. The communication interface 1630 may comprise one or more components, such as: power amplifier, digital front end (DFE), analog- to-digital converter (ADC), digital-to-analog converter (DAC), frequency converter, (de) modulator, and/or encoder/decoder circuitries, controlled by the corresponding controlling units.

The communication interface 1630 provides the apparatus with radio communication capabilities to communicate in the wireless communication network. The communication interface may, for example, provide a radio interface to one or more wireless communication devices. The apparatus 1600 may further comprise or be connected to another interface towards a core network such as the network coordinator apparatus or AMF, and/or to the access nodes of the wireless communication network.

The apparatus 1600 may further comprise a scheduler 1640 that is configured to allocate radio resources. The scheduler 1640 may be configured along with the communication control circuitry 1610 or it may be separately configured.

It is to be noted that the apparatus 1600 may further comprise various components not illustrated in FIG. 16. The various components may be hardware components and/or software components.

As used in this application, the term “circuitry” may refer to one or more or all of the following: a) hardware-only circuit implementations (such as implementations in only analog and/or digital circuitry); and b) combinations of hardware circuits and software, such as (as applicable): i) a combination of analog and/or digital hardware circuit(s) with software/firmware and ii) any portions of hardware processor(s) with software (including digital signal processor(s), software, and memory(ies) that work together to cause an apparatus, such as a mobile phone, to perform various functions); and c) hardware circuit(s) and/or processor(s), such as a microprocessor(s) or a portion of a microprocessor(s), that requires software (for example firmware) for operation, but the software may not be present when it is not needed for operation.

This definition of circuitry applies to all uses of this term in this application, including in any claims. As a further example, as used in this application, the term circuitry also covers an implementation of merely a hardware circuit or processor (or multiple processors) or portion of a hardware circuit or processor and its (or their) accompanying software and/or firmware. The term circuitry also covers, for example and if applicable to the particular claim element, a baseband integrated circuit or processor integrated circuit for a mobile device or a similar integrated circuit in server, a cellular network device, or other computing or network device. The techniques and methods described herein may be implemented by various means. For example, these techniques may be implemented in hardware (one or more devices), firmware (one or more devices), software (one or more modules), or combinations thereof. For a hardware implementation, the apparatus(es) of example embodiments may 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, micro-controllers, microprocessors, other electronic units designed to perform the functions described herein, or a combination thereof. For firmware or software, the implementation can be carried out through modules of at least one chipset (for example procedures, functions, and so on) that perform the functions described herein. The software codes may be stored in a memory unit and executed by processors. The memory unit may be implemented within the processor or externally to the processor. In the latter case, it can be communicatively coupled to the processor via various means, as is known in the art. Additionally, the components of the systems described herein may be rearranged and/or complemented by additional components in order to facilitate the achievements of the various aspects, etc., described with regard thereto, and they are not limited to the precise configurations set forth in the given figures, as will be appreciated by one skilled in the art.

It will be obvious to a person skilled in the art that, as technology advances, the inventive concept may be implemented in various ways. The embodiments are not limited to the example 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, not to restrict, the embodiments.

Claims

1. An apparatus comprising at least one processor, and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to: receive, from a network node, a machine learning model for performing 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 indication to update the machine learning model, or a configuration indicating at least one condition for determining whether to update the machine learning model; perform the at least one prediction of the at least one communication parameter in the at least one cell using the machine learning model; determine, based on the indication or the at least one condition, whether to update the machine learning model; obtain, based on determining to update the machine learning model, an updated version of the machine learning model by updating the machine learning model based on the at least one prediction; and transmit, to the network node, the updated version of the machine learning model.

2. The apparatus according to claim 1, further being caused to: receive, from the network node, an indication indicating support for provisioning one or more machine learning models for performing the at least one prediction of the at least one communication parameter in the at least one cell; and transmit, to the network node, an indication indicating a capability of for at least one of: predicting the at least one communication parameter, or training at least one of the one or more machine learning models.

3. The apparatus according to any preceding claim, wherein the at least one condition comprises at least one of: performing the at least one prediction of the at least one communication parameter, receiving a configuration for the at least one communication parameter to be used for communication with the network node, monitoring predicted quality or utilization of the at least one communication parameter for a pre-defined duration, or an amount of correlation between the at least one prediction and at least one observed value of the at least one communication parameter.

4. The apparatus according to any preceding claim, further being caused to: determine whether the at least one prediction correlates with at least one observed value of the at least one communication parameter; and based on determining that the at least one prediction correlates with the at least one observed value, update the machine learning model by rewarding the machine learning model.

5. The apparatus according to any of claims 1-3, further being caused to: determine whether the at least one prediction correlates with at least one observed value of the at least one communication parameter; and based on determining that the at least one prediction does not correlate with the at least one observed value, update the machine learning model by penalizing the machine learning model.

6. The apparatus according to any preceding claim, further being caused to: attempt to determine whether the at least one prediction correlates with at least one observed value of the at least one communication parameter; and based on not being able to determine whether the at least one prediction correlates with the at least one observed value, refrain from updating the machine learning model until being able to determine whether the at least one prediction correlates with the at least one observed value.

7. An apparatus comprising at least one processor, and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to: transmit, 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; transmit, 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 to update the machine learning model; and receive, from the at least one user equipment, an updated version of the machine learning model.

8. The apparatus according to claim 7, further being caused to: transmit an indication indicating support for provisioning one or more machine learning models for performing the at least one prediction of the at least one communication parameter in the at least one cell; receive, from the at least one user equipment, an indication indicating a capability for at least one of: predicting the at least one communication parameter, or training at least one of the one or more machine learning models; and determine, based on the capability, the machine learning model to be transmitted to the at least one user equipment.

9. The apparatus according to any of claims 7-8, further being caused to: transmit, to the at least one user equipment, a configuration for predicting the at least one communication parameter; receive, from the at least one user equipment, information indicating the at least one prediction of the at least one communication parameter; monitor the at least one prediction of the at least one user equipment based on communication quality between the apparatus and the at least one user equipment; and transmit, based on the monitoring, to the at least one user equipment, the indication to update the machine learning model.

10. The apparatus according to any of claims 7-9, further being caused to: determine whether the at least one prediction correlates with at least one observed value of the at least one communication parameter; and based on determining that the at least one prediction correlates with the at least one observed value, transmit, to the at least one user equipment, the indication to update the machine learning model by rewarding the machine learning model.

11. The apparatus according to any of claims 7-9, further being caused to: determine whether the at least one prediction correlates with at least one observed value of the at least one communication parameter; and based on determining that the at least one prediction does not correlate with the at least one observed value, transmit, to the at least one user equipment, the indication to update the machine learning model by penalizing the machine learning model.

12. The apparatus according to any of claims 7-11, further being caused to: determine whether the updated version of the machine learning model provides a performance improvement compared to a previous version of the machine learning model; and based on determining that the updated version provides the performance improvement, transmit the updated version of the machine learning model to one or more other user equipments for performing at least one subsequent prediction of the at least one communication parameter in the at least one cell.

13. The apparatus according to any of claims 7-11, further being caused to: determine whether the updated version of the machine learning model provides a performance improvement compared to a previous version of the machine learning model; and based on determining that the updated version does not provide the performance improvement, discard the updated version of the machine learning model.

14. A method comprising: receiving, from a network node, a machine learning model for performing at least one prediction of at least one communication parameter in at least one cell; receiving, from the network node, at least one of: an indication to update the machine learning model, or a configuration indicating at least one condition for determining whether to update the machine learning model; performing the at least one prediction of the at least one communication parameter in the at least one cell using the machine learning model; determining, based on the indication or the at least one condition, whether to update the machine learning model; obtaining, based on determining to update the machine learning model, an updated version of the machine learning model by updating the machine learning model based on the at least one prediction; and transmitting, to the network node, the updated version of the machine learning model.

15. A 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 to update the machine learning model; and receiving, from the at least one user equipment, an updated version of the machine learning model.