Ai / ML for UE-to-UE communication

Abstract

This invention relates to Artificial Intelligence (Al) / Machine Learning (ML) for the radio air interface of a device in a wireless network such as 5G NR and 6G, in particular, it relates to wireless devices in roaming and / or using a device-to-device communication interface. In particular, this invention discloses A first communication device in a wireless network, comprising: - at least a transceiver for communicating with or through at least a second communication device and / or with one or more other third communication devices in device-to-device communication, wherein the first communication device is configured for: - at least one AI / ML function by or through the second communication device, and wherein the AI / ML function is adapted to assisting (a) the device-to-device communication and / or (b) AI / ML operation in roaming.

Description

[0001] AI / ML for UE-to-UE communication

[0002] This invention relates to Artificial Intelligence (Al) / Machine Learning (ML) for the radio air interface in a wireless network.

[0003] Background

[0004] In conventional cellular networks, a primary station serves a plurality of secondary stations located within a cell served by this primary station. Wireless communication from the primary station towards each secondary station is done on downlink channels. Conversely, wireless communication from each secondary towards the primary station is done on uplink channels. The wireless communication can include data traffic (sometimes referred to User Data), and control information (also referred sometimes as signalling). This control information typically comprises information to assist the primary station and / or the secondary station to exchange data traffic (e.g. resource allocation / requests, physical transmission parameters, information on the state of the respective stations).

[0005] In the context of cellular networks as standardized by 3GPP, the primary station is referred to a base station, or a gNodeB (or gNB) in 5G (NR) or an eNodeB (or eNB) in 4G (LTE). The eNB / gNB is part of the Radio Access Network (RAN), which interfaces to functions in the Core Network (CN). In the same context, the secondary station corresponds to a mobile station, or a User Equipment (or a UE) in 4G / 5G, which is a wireless client device or a specific role played by such device. The term "node" is also used to denote either a UE or a gNB / eNB.

[0006] Additionally, for example, in the case of PC5 interface or Sidelink communication, it is possible to have Direct communication between secondary stations, here UEs. It is then also possible for UEs to operate as Relays to allow for example out of coverage UEs to get an inter-mediate (or indirect) connection to the eNB or gNB. To be able to work as a relay, a UE may use discovery messages to establish new connections with other UEs.

[0007] 3GPP has completed a study FS_NR_AIML_Air (RP-221348) in Release-18 (R18) on AI / ML for the 5G New Radio (NR) radio air interface, and documented the study results in TR 38.843 V18.0.0. The study primarily focused on the Life Cycle Management (LCM) aspects of supporting AI / ML on the NR radio air interface, and a few carefully selected use cases were studied to assist in defining a generic 3GPP framework for AI / ML which can be extended to new AI / ML use cases to be supported in the future releases of 3GPP 5G and 6G standards. In the current 3GPP AI / ML framework for NR radio air interface, it only covers a centralized architecture of User Equipment (UE) and network nodes such as base station, core network functions (e.g. Location Management Function (LMF)), Operations Administration and Maintenance (0AM), and only considers collaboration between UEs and network nodes. In other words, the current 3GPP framework for AI / ML only applies to the R Uu radio air interface between the UE and the network, and the selected use cases to study, such as CSI compression / prediction, beam management, and positioning accuracy enhancements, all represent the examples and efforts of enhancing the Uu radio air interface with AI / ML functionality.However, while the use of AI / ML benefit for infrastructure devices, such network nodes, core networks or base stations, the improvement is still limited for terminals.

[0008] Another problem with the current 3GPP AI / ML framework is that it does not take country / region-specific requirements and regulations on AI / ML into account when a UE roams into a visiting Public Land Mobile Network (PLMN) in a country / region with different AI / ML regulations than the home PLMN. This may cause serious formal conformance issues if the framework for AI / ML does not consider this aspect.

[0009] Summary of the Invention

[0010] In view of the above illustrative problems, that there is a need to expand the framework and techniques to other communication modes such as device to device communication modes (e.g. sidelink / PC5 radio air interface in the example of 5G) in general, communication interfaces between devices, so that device-to-device communication-based use cases can benefit from the AI / ML framework.

[0011] It is an object of the invention to enable the usage of AI / ML techniques in device-to-device communication interfaces.

[0012] It is another object of the invention to overcome the limits of current 3GPP AI / ML frameworks, so that the AI / ML functionality can also be applied to sidel i nk / PC5 communication between UEs.

[0013] It is another object of the invention to overcome the limit of the current 3GPP AI / ML framework without considering the country / region specific requirements on AI / ML for a UE roaming from the home PLMN to a visiting PLMN, so that the UE roaming in a visiting PLMN in a country / region having different AI / ML related requirements and regulations can conform to those requirements and regulations of the visiting PLMN. Some or all of these objectives can be achieved by means of the methods, computer program product, communication devices and wireless systems as claimed in the appended set of claims.

[0014] More specifically, in accordance with a first aspect of the invention, it is proposed a method for operating a first communication device in a wireless network, comprising: the first communication device communicating with or through at least a second communication device; and / or the first communication device communicating with one or more other third communication devices in a device-to-device communication mode, the method further comprising receiving an AI / ML configuration from or relayed through the second communication device for executing at least one AI / ML function, and wherein the AI / ML function is adapted to assist

[0015] (a) the device-to-device communication mode and / or

[0016] (b) an AI / ML operation in roaming mode.

[0017] In a variant of the first aspect of the invention, the method comprises, before changing a serving network and / or radio access technology or upon detecting a change of a serving network and / or radio access technology, the first communication device performing one or more of the following: a. checking a configuration / policy determining whether an AI / ML function may be used; b. sending a request to receive a configuration containing criteria regarding the behavior of the first communication device regarding its AI / ML model and / or AI / ML functionality when / before changing the network and / or radio access technology; c. receiving a configuration determining conditions allowing and / or forbidding the reporting of measurements and data for training purposes of an AI / ML model wherein the conditions comprise at least the first communication device camping in a given network and / or the first communication device using a given radio access technology; d. sending a request to retrieve an AI / ML information; e. sending a request to update the AI / ML function; or f. sending measurements for training of an AI / ML model. In a second variant of the first aspect or of the first variant, the configuration provides at least one AI / ML function, to assist the device-to-device communication, and wherein the configuration is relayed through the second device from a fourth communication device.

[0018] Furthermore, the fourth communication device may be one of a base station, a core network function or an application function.

[0019] In a third variant of the first aspect or any of the previous variants, the AI / ML function includes at least one operation in the lifecycle of AI / ML functionality for one or more of the AI / ML model provision, configuration, training, transfer, update, inference, management and / or monitoring.

[0020] In a fourth variant of the first aspect or any of the previous variants, the AI / ML function's aim is optimizing one or more of the following procedures in device-to-device communication scenarios: a. device-based positioning and / or ranging measurements and / or estimations; b. relay (re-)selection in relay scenarios; c. path (re-)selection in multi-path scenarios; d. resource selection; and / or e. wireless sensing procedures.

[0021] In a fifth variant of the first aspect or any of the previous variants, the first communication device using the AI / ML function relying on an AI / ML model taking as input one or more first parameters: a. model identifier; b. signal strength; c. link quality; d. frequency band; and / or e. speed of the first and / or third communication devices.

[0022] Additionally, one or more of the first parameters may be explicitly exchanged between the first and the third communication devices and / or implicitly indicated to the first and the third communication devices, wherein the implicit indication is done by means of one or more of: relay service code; service code; and / or layer 2 address.

[0023] In accordance with a sixth variant of the first aspect or any of the previous variants, the method comprises sending a request for the AI / ML function to and / or through the second communication device. In accordance with a seventh variant of the first aspect or any of the previous variants, the method comprises forwarding a request for the AI / ML function to and / or through the second communication device, wherein the forwarded request is forwarded from a third communication device.

[0024] In a variant of the sixth or the seventh variants, the request includes one or more of: a. one or more requested AI / ML functions; b. the capabilities of the first and / or third communication devices; and / or c. request to provision and / or configure at least one AI / ML function.

[0025] In an eighth variant of the first aspect or any of the previous variants, the method comprises the first communication device monitoring by means of a monitoring function to determine whether a) the AI / ML function for device-to-device communication fulfils performance requirements and / or b) whether the current AI / ML parameters are valid in the current serving network.

[0026] In accordance with a second aspect of the invention, it is proposed a method for operating a first communication device capable of AI / ML RAN communication, wherein the method is adapted to receiving at least one AI / ML function from or through a second communication device, and wherein the AI / ML function is adapted to assisting the first communication device in device-to-device communication and / or AI / ML operation in roaming.

[0027] In accordance with a fourth aspect of the invention, it is proposed a computer program product comprising instructions which, when executed, cause a communication device to implement the method of any of the previous claims.

[0028] In accordance with a fifth aspect of the invention, it is proposed a first communication device, comprising: a transceiver configured to communicate with or through at least a second communication device and / or communicate with one or more other third communication devices in a device-to-device communication mode, wherein the receiver is adapted to receive an AI / ML configuration from or relayed through the second communication device the communication device comprising executing at least one AI / ML function, and wherein the AI / ML function is adapted to assist

[0029] (a) the device-to-device communication mode and / or (b) an AI / ML operation in roaming mode.

[0030] In accordance with a sixth aspect of the invention, it is proposed a communication device in a wireless network, comprising: a transceiver for communications with one or more access devices, (such as base stations, repeaters, network-controlled repeaters, IAB node, relay, or NTN payload), a sidelink transceiver for the sidelink communications with one or more other communication devices, wherein the communication device is configured for at least one AI / ML function by the access device, and the AI / ML function being adapted to assist the sidelink communication, and wherein the AI / ML function includes at least one operation in the lifecycle of AI / ML functionality for one or more of the AI / ML model provision, configuration, training, transfer, update, inference, management and / or monitoring.

[0031] In accordance with a sixth aspect of the invention, it is proposed a second communication device , comprising: a transceiver for communications with one or more first communication devices, a controller configured to provision and configure at least one AI / ML function at the communication device, and the AI / ML function is adapted to assist a sidelink communication, and wherein the AI / ML function includes at least one operation in a lifecycle of AI / ML functionality for one or more of AI / ML model provision, configuration, training, transfer, update, inference, management and / or monitoring.

[0032] In accordance with a seventh aspect of the invention, it is proposed a wireless communication system, comprising: one or more base stations, a plurality of user devices, UEs, configured for sidelink communication, wherein the plurality of UEs comprises at least one AI / ML function, and at least one function for sidelink communication, the AI / ML function is configured to assist the sidelink communication, wherein the base station is configured to connect to the core network, and provide at least one AI / ML function, to assist the sidelink communication, and wherein the AI / ML function consists of at least one operation in the lifecycle of AI / ML functionality for AI / ML model provision, configuration, training, transfer, update, inference, management and monitoring. In accordance with an eighth aspect of the invention, it is proposed a second communication device in a wireless network comprising: one transceiver for communications with one or more first communication device, wherein the second communication device is configured to a) provision and / or forward a configuration of at least one AI / ML function to at the first communication device, and the AI / ML function is adapted to assist the device-to-device communication and / or b) assist AI / ML operation in roaming.

[0033] In accordance with a ninth aspect of the invention, it is proposed a wireless communication system, comprising: at least one second communication device, at least one first communication device, wherein the first communication device comprises at least one AI / ML function to assist AI / ML operation in roaming and / or the second communication device is configured to assist AI / ML operation in roaming.

[0034] In accordance with a tenth aspect of the invention, it is proposed a wireless communication system, comprising: at least one second communication device, at least one first communication device, at least one third communication device, wherein the first and third communication devices comprise at least one AI / ML function, at least one communication unit adapted for device-to-device communication, the AI / ML function configured to assist the device-to-device communication.

[0035] It shall be understood that a preferred embodiment of the invention can also be any combination of the dependent claims or above embodiments with the respective independent claim.

[0036] These and other aspects of the invention will be apparent from and elucidated with reference to the embodiments described hereinafter.

[0037] Brief Description of the Drawings Fig. 1 schematically represents the overall cellular system including UEs, RAN, and core network;

[0038] Fig. 2 provides a schematic representation of a UE and its components; and

[0039] Fig. 3 schematically represents different entities involved in a non-terrestrial network;

[0040] Fig. 4 schematically represents a random-access procedure in a wireless network;

[0041] Fig. 5 schematically represents a signalling procedure by an access device;

[0042] Fig. 6 schematically represents the periodic transmission of SSB bursts;

[0043] Fig. 7 schematically represents examples of wireless devices according to some embodiments;

[0044] Fig. 8 is a diagram illustrating the current 3GPP AI / ML framework applied to the selected use cases, which are over the Uu radio air interface, in TR 38.843 V18.0.0.;

[0045] Fig. 9 is a diagram illustrating an embodiment of the invention proposing to extend the 3GPP AI / ML framework to sidelink / PC5 radio air interface for the new AI / ML use cases for the communication between UEs; and

[0046] Fig. 10 is the diagram illustrating the AI / ML framework for sidelink / PC5 communication between UEs.

[0047] Detailed Description of the Invention

[0048] Embodiments of the present invention will now be described in the context of a cellular communication network environment, for example 5G or 6G. However, the present invention and its embodiments may also be used in connection with other wireless technologies.

[0049] Throughout the present disclosure, the abbreviation "gNB" (5G terminology) or "BS" (base station) or the term "access device" is intended to mean a wireless access device such as a cellular base station or a Wi-Fi access point or a ultrawide band (UWB) personal area network (PAN) coordinator. The gNB may consist of a centralized control plane unit (gNB-CU-CP), multiple centralized user plane units (gNB-CU-UPs) and / or multiple distributed units (gNB-DUs). The gNB is part of a radio access network (RAN), which provides an interface to functions in the core network (CN). The RAN is part of a wireless communication network. It implements a radio access technology (RAT). Conceptually, it resides between a communication device such as a mobile phone, a computer, or any remotely controlled machine and provides connection with its CN. The CN is the communication network's core part, which offers numerous services to customers who are interconnected via the RAN. More specifically, it directs communication streams over the communication network and possibly other networks.

[0050] Furthermore, the terms "base station" (BS) and "network" may be used as synonyms in this disclosure. This means for example that when it is written that the "network" performs a certain operation it may be performed by a CN function of a wireless communication network, or by one or more base stations that are part of such a wireless communication network, and vice versa. It can also mean that part of the functionality is performed by a CN function of the wireless communication network and part of the functionality by the base station.

[0051] It is further noted that throughout the present disclosure only those blocks, components and / ordevices that are relevant are shown in the accompanying drawings. Other blocks have been omitted for reasons of brevity. Furthermore, blocks designated by same reference numbers are intended to have the same or at least a similar function, so that their function is not described again later.

[0052] A cellular system is a wireless communication system that consists of three main components: user equipment (UE), radio access network (RAN), and core network (CN). These components work together to provide voice and data services to mobile users over a large geographic area.

[0053] In conventional cellular networks, a primary station serves a plurality of secondary stations located within a cell served by this primary station. Wireless communication from the primary station towards each secondary station is done on downlink channels. Conversely, wireless communication from each secondary towards the primary station is done on uplink channels. The wireless communication can include data traffic (sometimes referred to User Data), and control information (also referred sometimes as signalling). This control information typically comprises information to assist the primary station and / or the secondary station to exchange data traffic (e.g. resource allocation / requests, physical transmission parameters, information on the state of the respective stations). In the context of cellular networks as standardized by 3GPP, the primary station is referred to a base station, or a gNodeB (or gNB) in 5G (NR) or an eNodeB (or eNB) in 4G (LTE). The eNB / gNB is part of the Radio Access Network (RAN), which interfaces to functions in the Core Network (CN). In the same context, the secondary station corresponds to a mobile station, or a User Equipment (or a UE) in 4G / 5G, which is a wireless client device or a specific role played by such device. The term "node" is also used to denote either a UE or a gNB / eNB.

[0054] Additionally, for example, in the case of PC5 interface or Sidelink communication, it is possible to have Direct communication between secondary stations, here UEs. It is then also possible for UEs to operate as Relays to allow for example out of coverage UEs to get an inter-mediate (or indirect) connection to the eNB or gNB. To be able to work as a relay, a UE may use discovery messages to establish new connections with other UEs. Certain UEs may communicate with each other by using device-to-device communication, also known as sidelink communication using the PC5 interface that may rely on physical sidelink (PS) broadcast channel, PS shared channel, PS control channel, etc. Furthermore, the role of a relay node has been introduced in 3GPP. This relay node is a wireless communication station that includes functionalities for relaying communication between a primary station, e.g. a gNB and a secondary station, e.g. a UE. This relay function for example allows to extend the coverage of a cell to an out-of-coverage (OoC) secondary station. This relay node may be a mobile station or could be a different type of device. In the specifications for 4G, the Proximity Services (ProSe) functions are defined inter alia in TS 23.303, and TS 24.334 to enable - amongst others -connectivity for the cellular User Equipment (UE) that is temporarily not in coverage of the cellular network base station (eNB) serving the cell. This particular function is called ProSe UE-to-network relay, or Relay UE for short. The Relay UE relays application and network traffic in two directions between the OoC UE and the eNB. The local communication between the Relay UE and the OoC UE is called device-to-device (D2D) communication or Sidelink (also known as PC5) communication in TS 23.303 and TS 24.334. Once the relaying relation is established, the OoC-UE is, e.g., IP-connected via the Relay UE and acts in a role of "Remote UE". This situation means the Remote UE has an indirect network connection to selected functions of the Core Network as opposed to a direct network connection to all Core Network functions that is the normal case. Furthermore, it has been introduced the role of a UE-to-UE relay node, i.e., a relay node re-laying the communication between two UE devices. The relay node relays the communications between UE devices. UEs may connect to the core network through a base station when in-coverage. In such relay scenarios, the relay devices may receive and store some information for some time before forwarding it towards the target device. This information that may be stored and forwarded may be discovery messages received from a source UE whereby the relay UE may release them at some point of time later. This information that may be stored and forwarded may be a SIB that may contain a timestamp.

[0055] User equipment (UE) is the device that a user uses to access the cellular system, such as a smartphone, a tablet, a laptop, loT device, or a wearable device. A UE typically may contain the following components:

[0056] - A universal integrated circuit card (UICC), which stores the user's identification and authentication information, such as the subscription permanent identifier (SUPI) or credentials. - A transceiver, which converts the digital signals from the processor into analog signals for transmission and reception over the air interface. The transceiver also performs modulation, demodulation, coding, decoding, and other signal processing functions.

[0057] - A processor, which controls the operation of the UE and executes the applications and services that the user requests. The processor also communicates with the RAN and the CN using various protocols.

[0058] - A display, which shows the user the information and feedback from the UE, such as the signal strength, the battery level, the call status, the messages, the contacts, the menu, etc.

[0059] - A microphone and a speaker, which enable the user to make and receive voice calls, as well as use other audio features, such as voice mail, voice recognition, etc.

[0060] - A keyboard and / or a touch screen, which allow the user to enter and select commands, text, numbers, etc.

[0061] - A camera and / or a video recorder, which enable the user to capture and send images and videos, as well as use other multimedia features, such as video calling, video streaming, etc.

[0062] - A memory, which stores the data and programs that the user needs, such as the phone book, the messages, the photos, the videos, the applications, etc as well as a computer program to perform the operations of the RAN and CN protocols.

[0063] - A battery, which provides the power supply for the UE.

[0064] Fig. 2 provides a schematic representation of a UE and its components, e.g., UICC (201), processor (202), transceiver (203), memory (204), input devices (205) such as camera, microphone, etc and output devices (206) such as display, speaker, etc. Fig. 7 schematically represents wireless devices that may include the capabilities of a UE and / or a STA. Fig. 7a) represents AR / VR glasses; Fig. 7b) represents a connected vehicle; and Fig. 7c) represents a mobile phone. In these devices, a reflective intelligent surface (RIS) may be embedded, e.g., by covering and / or under the whole a part of the UE surface. This may be used, e.g., to better deal with interferences or improve wireless sensing.

[0065] A UE may access the cellular network via the radio access network, as described below. Certain UEs may communicate with each other by using device-to-device communication, also known as sidelink communication using the PC5 interface that may rely on physical sidelink (PS) broadcast channel, PS shared channel, PS control channel, etc.

[0066] A UE may receive a configuration by means of different procedures: Downlink control information (DO) is a type of control information that is sent from the BS to the UE on the physical downlink control channel (PDCCH). DO contains various parameters that instruct the UE how / when to decode and transmit data on the physical downlink shared channel (PDSCH) and the physical uplink shared channel (PUSCH), such as the resource allocation, the modulation and coding scheme. The UE needs to monitor the PDCCH in each subframe to detect and decode the DO that is addressed to it.

[0067] Uplink control information (UO) is a type of control information that is sent from the UE to the BS on the physical uplink control channel (PUCCH) or the physical uplink shared channel (PUSCH). UO contains various feedback signals that inform the BS about the status and quality of the downlink transmission, such as the HARQ acknowledgments (ACKs), the channel state information (CSI), and the scheduling requests (SRs). The UE needs to encode and transmit the UCI according to the configuration and timing indicated by the BS.

[0068] Sidelink control information (SCI) is a type of control information that is sent from the UE to another UE on the physical sidelink control channel (PSCCH) in device-to-device (D2D) communication scenarios. The main functions of SCI include resource allocation, synchronization, channel quality reporting,

[0069] Medium access control (MAC) control element (MAC CE) is a type of control information that is sent from the BS to the UE or vice versa on the MAC layer. MAC CE contains various commands or indications that regulate the MAC layer functions, such as the buffer status report (BSR), the timing advance command (TAC), the discontinuous reception (DRX) command, etc. The UE needs to process the MAC CE according to the MAC protocol and the configuration provided by the BS.

[0070] Radio resource control (RRC) command is a type of control information that is exchanged between the BS and the UE on the RRC layer. RRC Command contains various messages that modify / configure RRC parameters and / or initiate, modify, or release the RRC connection or the radio bearers between the UE and the BS, such as the RRC connection setup, the RRC connection reconfiguration, the RRC connection release, the security mode command, the mobility from E-UTRA command, the handover from E-UTRA preparation request, etc. The UE needs to respond to the RRC Command according to the RRC protocol and the configuration provided by the BS.

[0071] Non-access stratum (NAS) messages are used for signalling between UE and core network

[0072] (CN) on the non-access stratum (NAS) layer. NAS messages enable functionality such as registration, session establishment, security, and mobility management. The UE needs to respond to the NAS Command according to the NAS protocol and the configuration provided by the CN. UE parameter update (UPU) is a procedure between the UE and the home network that enables the home network to update configuration parameters in mobile phones and / or USIM using the UDM control plane procedure (TS 23.502). The UE can receive Parameters Update Data from the UDM after the UE has registered in the 5G network.

[0073] Steering of Roaming (SoR) enables the home network to guide the user equipment (UE) when registering on a visited network. For detailed information about the interfaces and registration in the 5G System, refer to 3GPP TS.23.501 (Release 15) and 3GPP TS 24.501 (Release 15). The 5G CP-SOR is activated during or after registration to update the UE's "Operator Controlled PLMN Selector with Access Technology" list via secure NAS messages, as directed by the home PLMN based on specific operator policies, such as preferred networks or UE location.

[0074] UE configuration update (UCU) is used to update configuration parameters as per TS 23.502 that may include Access and Mobility Management related parameters decided and provided by the AMF, UE Policy provided by the PCF. When AMF wants to change the UE configuration for access and mobility management related parameters the AMF initiates the procedure defined in clause 4.2.4.2. When the PCF wants to change or provide new UE Policies in the UE, the PCF initiates the procedure defined in clause 4.2.4.3. If the UE Configuration Update procedure requires the UE to initiate a Registration procedure, the AMF indicates this to the UE explicitly. The procedure in clause 4.2.4.2 may be triggered also when the AAA Server that performed Network Slice-Specific Authentication and Authorization for an S-NSSAI revokes the authorization.

[0075] Radio access network (RAN) is the part of the cellular system that connects the UEs to the CN via the air interface. The RAN consists of base stations (BSs). A base station (BS) is a fixed or mobile transceiver that covers a certain geographic area, called a cell. In 5G, a BS is also called a gNB (next generation node B). A BS can serve multiple UEs simultaneously within its cell, by using different frequencies, time slots, codes, or beams. A BS also performs functions such as power control, handover control, channel allocation, interference management, etc. A base station can be divided into two units: a central unit (CU) and a distributed unit (DU). The CU performs the higher layer functions, such as RLC, PDCP, RRC, etc. The DU performs the lower layer functions, such as PHY and MAC. The CU and the DU can be co-located or separated, depending on the network architecture and deployment. In cellular systems, a base station may be denoted, based on context, as a cell, or gNB.

[0076] The cell may also refer to the coverage area of a base station. A BS may have different coverage areas such as a macro cell (e.g. several kilometres wide), a pico cell (e.g., for a given location such as a stadium) or a femto cell for a small location (e.g., a home or part of it). A base station may communicate with the core network. Since there can be base stations for different cellular systems, different interfaces are required. For instance, a base station, eNB, in a 4G Long Term Evolution (LTE) system (also known as Evolved Universal Mobile Telecommunications Systems (UMTS) Terrestrial Radio Access Network (E-UTRAN)) may interface with the 4G CN known as EPC through the corresponding interface. For instance, a base station, gNB, in a 5G system (i.e., 5G New Radio or Next Generation RAN) may communicate with the 5GC through a different interface. 4G and 5G base stations may communicate with each other directly or through their corresponding core networks.

[0077] The main protocols used between the UEs and the RAN are:

[0078] - The physical layer (PHY), which defines the characteristics of the air interface, such as the frequency bands, the modulation schemes, the coding rates, the frame structure, the synchronization, etc.

[0079] - The medium access control (MAC) layer, which regulates the access of the UEs to the shared radio channel, by using techniques such as orthogonal frequency division multiple access (OFDMA), time division duplex (TDD), frequency division duplex (FDD), etc.

[0080] - The radio link control (RLC) layer, which provides reliable data transmission over the radio channel, by using techniques such as segmentation, reassembly, error detection, error correction, retransmission, etc.

[0081] - The packet data convergence protocol (PDCP) layer, which compresses and decompresses the headers of the data packets, encrypts and decrypts the data, and performs data integrity protection.

[0082] - The radio resource control (RRC) layer, which establishes, maintains, and releases the radio bearers between the UEs and the RAN, as well as exchanges the signalling messages for functions such as connection setup, handover, measurement reporting, security activation, etc.

[0083] A transmission / reception communication unit or transceiver may be used by BS and UE to transmit / receive data. Control data may be required for a physical broadcast channel, physical downlink control channel, etc. Data may be for the physical downlink shared channel.

[0084] Data may be encoded by the UE and / or BS to obtain data symbols and / or control symbols that may be exchanged over the wireless interface. The conversion from digital data into analog symbols may be done by the transmission / reception communication unit.

[0085] A medium access control control-element (MAC-CE) is a MAC layer communication element that is used to control the communication between wireless devices. A MAC-CE may be exchanged in a shared channel, e.g., the physical downlink / uplink / sidelink shared channel. The communication between a UE and a base station or the communication between UEs (when sidelink is used) may involve the exchange of reference signals. Reference signals may include primary synchronization signal (PSS), a secondary synchronization signal (SSS), a physical broadcast channel demodulation reference signal (DMRS), a channel state information reference signal (CSI-RS). Core network (CN) is the part of the cellular system that connects the RAN to other networks, such as the Internet, or other cellular systems. The CN consists of two main (control / user) domains. The control domain is responsible for providing signalling and control functions for the UEs, such as authentication, authorization, mobility management, session management, etc. The control plane consists of several network functions (NFs), such as the access and mobility management function (AMF), the session management function (SMF), the unified data management (UDM), the policy control function (PCF), the network exposure function (NEF), and the authentication server function (AUSF). The access and mobility management function (AMF) is a NF that handles the registration, deregistration, connection management, and mobility management for the UEs. The session management function (SMF) is a NF that handles the establishment, modification, and release of the sessions for the UEs. The SMF also communicates with the user plane devices to perform functions such as IP address allocation, tunneling, QoS, etc. The unified data management (UDM) is a NF that stores and manages the user data, such as the SUPI, the service profile, the subscription status, etc. The policy control function (PCF) is a NF that provides the policy rules and charging information for the UEs, such as the access type, the service level, the data rate, the quota, etc. The network exposure function (NEF) is a NF that exposes the network capabilities and services to external applications and devices, such as the IMS, the Internet of Things (loT), etc. The authentication server function (AUSF) is a NF that performs the primary authentication with the by using credentials and the SUPI. The user domain is responsible for providing data and multimedia services to the UEs, by using packets and IP addresses. The user plane consists of two main functions: the user plane function (UPF) and the data network (DN). The user plane function (UPF) is a device that forwards the data packets between the UEs and the DNs, as well as performs functions such as tunneling, firewall, QoS, charging, etc. The data network (DN) is a network that provides access to the services and applications that the UEs request, such as the Internet, the IMS, etc.

[0086] A residential gateway (RG) is a device that connects a home network to an external network, such as the Internet or a cellular system. An RG typically provides functions such as routing, switching, firewall, NAT, DHCP, DNS, VPN, etc. An RG can also support various types of interfaces, such as Ethernet, Wi-Fi, Bluetooth, USB, etc. A cellular-capable RG is an RG that has a cellular interface, such as a UICC slot, a cellular modem, or an antenna, that enables it to access the cellular system as a backup or an alternative to the wired or wireless broadband connection. A cellular-capable RG can provide benefits such as: (1) Enhanced reliability, by switching to the cellular connection in case of a failure or a degradation of the broadband connection; (2) Increased bandwidth, by aggregating the cellular connection and the broadband connection to achieve higher data rates or QoS.

[0087] A multi-SIM subscription is a subscription that allows a user to have multiple SIMs (or eSIMs) that are linked to the same account and service profile. A user can use the multi-SIM subscription to access the cellular system from different devices, such as a smartphone, a tablet, a laptop, or a wearable device, without having to switch the SIM card or the device.

[0088] Overall system: Fig. 1 provides an overall description of a wireless system wherein devices 100, 102, and 128 can play the role of UEs. Device 102 is part of a cellular-capable RG providing connectivity to a home network 129 e.g., by means of a local area network and / or wireless local area network. Device 102 is served by base station 104.

[0089] The RAN 127 comprises base station 103 and serves UE 128. UE 128 may also be a UE to Network relay given access to remote UE 136 that is out of coverage of base station 103. UEs 134 and 136 also communicate with each other via a UE-to-UE relay 135. UE to UE communication via relays is enabled by means of sidelink communication / PC5 interface.

[0090] Within the RAN, the range of base station 103 is extended via smart repeater 137 and reflective intelligent surface (RIS) 138. Smart repeater 137 and RIS 138 give access to UE 142.

[0091] The RAN 143 includes base station 104 and serves as wireless access infrastructure for the home network. Base station 104 also serves a mobile access device and / or UE as a UAV 139. UAV 139 may provide connectivity to remote UE 136.

[0092] Furthermore, a satellite gateway 141 is shown that connects to satellite 140 and may provide connectivity services to remote UE 136 or UE 100.

[0093] In Fig. 1, the 5G core network 133 may include one or more an AMF 121, SMF 123, UPF 122, AUSF 124, UDM 125, PCF 131, NEF 132 and allows the connection to a data network 130.

[0094] In Fig. 1, a second core network 142, e.g., a legacy core network as a 4G core network, is also shown that may interface with the 5G core network 133, interface with base stations denoted eNB in 4G, and provide a connection to the data network 130. The legacy 4G core network is denoted EPC and may include one or more mobility management entities (MME), a serving gateway, a multimedia broadcast multicast service gateway, a broadcast multicast service center, a packet data network gateway, etc. The mobility management entity may handle the signalling between UE and the 4G CN and may interact with the home subscriber server (HSS) in charge of the storage and management of subscriber data and secrets. The MME may provide connection management, similar to the AMF in 5G. The serving gateway may be used to exchange user internet protocol messages whereby the serving gateway may interact with the packet data network gateway that is connected to IP services. Multiple protocols in 4G and 5G have similar features. For example, the 5G network registration and 4G attach registration message are initially sent by the UE to establish a connection between the UE and the CN, which involves sending an initial request from the UE with its identity and capabilities, receiving an authentication request from the CN with a challenge, sending an authentication response from the UE with a response, receiving an authentication result from the CN with an indication of success or failure, and sending a security mode command from the CN with the selected security algorithms. As a result of this connection establishment procedure, NAS and AS keys are derived from the K_AMF (5G) and K_ASME (4G) where K_AMF is managed by the AMF and K_ASME is managed by the MME. A UE may connect to a serving network or serving Public Land Mobile Network (PLMN). A UE may have a subscription with a home PLMN, and during the registration procedure, the (AMF of the) serving PLMN may forward the registration request to the (AUSF of the) home PLMN that may perform an initial authentication procedure between home PLMN and UE. If the authentication procedure is successful, keys are derived and the home PLMN may share derived credentials with the serving PLMN, including K_SEAF, that may be used to derive K_AMF, from which NAS keys and AS keys are derived. The registration request sent by the UE includes an identifier that can be used by the home PLMN to identify the UE. To prevent privacy vulnerabilities, the long-term subscriber's identifier known as Subscriber Permanent Identifier (SUPI) may not be exchanged in the clear, but instead, either a Subscription Concealed Identifier (SUCI) or a pseudonym known as GUTI are exchanged with the AMF of the serving PLMN. The AMF of the PLMN may then forward the SUCI to the home PLMN so that the home PLMN decrypts / verifies it.

[0095] Satellite access: Fig. 1 depicts satellite 140 providing access to one or more UEs. Satellite access can be performed by means of non-terrestrial devices at different altitudes such as Low Earth Orbit (LEO), Medium Earth Orbit (MEO) or Geosynchronous Equatorial Orbit (GEO) satellites. Other types of nonterrestrial devices may include high-altitude platform station (HAPS) or unmanned aerial vehicle (UAVs) that may comprise a base station. Fig. 3 illustrates different elements including a GEO satellite 302, a MEO satellite 303, a LEO satellites 304 and 304', a UAV 305, all of them potential non-terrestrial mobile access devices giving coverage to wireless device (e.g., a UE) 301. GEO satellite 302 remains static over a given earth position while MEO and LEO satellites move. MEO satellites 303 have a slower moving vector 306 in relation to the earth compared with LEO satellites 304 / 304' that have a faster moving vector 307 / 307'. A non-terrestrial gateway 308 is included that provides connectivity to the mobile access device via a feeder link 310. A mobile access device provides service to the wireless device via a service link 311. Two mobile access devices in the same orbit may communicate with each other via an intra-orbit-satellite link 312 while two mobile access devices in different orbits may communicate with each other via an inter- orbit-satellite link 313. Fig. 3 finally also includes a terrestrial access device 309 that may also provide connectivity to wireless device 301. The terrestrial access device 309, the wireless device 301, and nonterrestrial gateway are on the earth surface 314.

[0096] Non-terrestrial devices such as satellites distribute system information in specific SIBs, in particular, SIB31 in 4G and SIB19 in 5G. SIB19 information element as defined in TS 38.331 18.2.0.

[0097]

[0098] A UE in a cellular system performs an initial random-access procedure to connect an access device. The 5G random access procedure is illustrated by means of Fig. 4 wherein 401 represents a user equipment and 402 represents an access device. The access device distributes signals 402. Signals 402 can be distributed periodically or on demand. Signals 402 may comprise the Master Information Block (MIB) transmitted together with / in the physical broadcast channel (PBCH) and the synchronization signals

[0099] The MIB comprises:

[0100] MIB ::= SEQUENCE { systemFrameNumber BIT STRING (SIZE (6)), subCarrierSpacingCommon ENUMERATED {scsl5or60, scs30orl20}, ssb-SubcarrierOffset INTEGER (0..15), dmrs-TypeA-Position ENUMERATED {pos2, pos3}, pdcch-ConfigSIBl INTEGER (0..255), cellBarred ENUMERATED {barred, notBarred}, intraFreqReselection ENUMERATED {allowed, notAllowed}, spare BIT STRING (SIZE (1)) }

[0101] MIB and PBCH are transmited as part of a Synchronization Signal Block, and the access device may transmit multiple SSBs through different beams, allowing the user equipment to determine the preferred beam, and once the preferred beam is obtained, retrieve the MIB, and use the information in the MIB to atempt to retrieve System Information Block 1 (SIB1) that may also be distributed periodically. The UE can use the information in SIB1 to perform the random-access procedure selecting a preamble to indicate its intention to access the cell by means of message 404, e.g., preamble transmission. This message may be used to derive a random-access radio network temporary identifier (RA-RNTI). Upon reception of message 404, access device 402 replies with message 405, e.g., a random access response. This message may include a time advance field to adapt the transmission timing, a value matching the preamble used by wireless device 401, and a grant (communication resources) for the wireless device. The access device also assigns a temporary cell radio network temporary identifier (TC-RNTI). Prior to this message 405, the access device may send a PDCCH DCI message assigning resources (a communication grant). This message may be addressed using the RA-RNTI. Upon reception of message 405, wireless device uses the initial grant received in the previous message and the RA-RNTI to transmit a subsequent message 406, e.g, an RRCSetupRequest or PHY layer. This message may include a Contention Resolution Identifier (CRI). This message may be sent in the PUSCH. As a response, access device replies with message 407, e.g., RRCSetup, that includes / repeats the received CRI confirming that the access device has identified the access device. This message includes a Cell RNTI (C-RNTI). Next, wireless device replies with message 408, e.g., an RRCSetupComplete that includes the RegistrationRequest message, and UE capabilities.

[0102] MIB and PBCH are transmitted as part of a Synchronization Signal Block, and the access device may transmit multiple SSBs through different beams. Multiple SSBs transmitted through multiple beams form an SSB burst. The multiple SSBs in an SSB burst are transmitted sequentially in the first part of a frame. SSB bursts are transmitted periodically, typically every 20 ms, or more.

[0103] Fig. 5 schematically illustrates an access device 500 transmitting four beams, each of them transmitting an SSB, namely 501, 502, 503, and 504. A wireless device 505 can measure the signal strength, i.e., RSRP (Reference Signal Received Power), of the beams. This is illustrated by means of the graph in Fig. 5 where 501', 502', 503', and 504' represent the RSRP of beams 501, 502, 503, and 504, respectively, as measured by wireless device 505. Wireless device 505 can use this information to determine which one of the beams is the preferred beam for further communication, e.g., to perform the random access procedure.

[0104] Fig. 6 further schematically illustrates SSB bursts transmitted periodically. In this case, each SSB burst comprises four SSBs transmitted in the first part / half of every second frame. In this figure, frames are denoted as f, f+1, f+2, f+3, ... A frame has a typical duration of 10 ms.

[0105] Resource grid: in a cellular network, such as a 5G network, the resource grid is a structured framework used to allocate and manage communication resources efficiently. It is characterized by a timefrequency matrix where each element, known as a resource element, is defined by its position in both time and frequency domains. The vertical axis represents frequency, segmented into subcarriers, which are spaced at intervals. The subcarrier spacing can vary depending on the deployment scenario, with common spacings being 15 kHz, 30 kHz, 60 kHz, 120 kHz, 240 kHz, and 480 kHz (corresponding to mu equal to 0, 1, 2, 3, 4, and 5, respectively). The horizontal axis of the grid represents time and is divided into frames, subframes, and slots, each frame has a duration of 10 ms and each subframe has a duration of 1 millisecond. Within these subframes, the time is further divided into slots. For mu, there are 2Λmu slots per subframe. Each slot typically spans 14 OFDM symbols. Each resource element in the grid, defined by the intersection of a time symbol and a frequency subcarrier, can carry a small portion of data, control information, or reference signals. These resource elements are grouped into larger units called Resource Blocks (RBs), which span 12 subcarriers in frequency and one slot in time. The allocation of these RBs is dynamically managed.

[0106] Reflective intelligent surfaces (RIS): may be used as part of the wireless infrastructure or as part of the wireless devices. RIS, often referred to as metasurfaces, are advanced materials engineered with sub-wavelength structures that can manipulate electromagnetic waves in a controlled manner. These surfaces consist of an array of unit cells, each capable of adjusting its electromagnetic response through electronic control, thus enabling dynamic alteration of the wavefront of the incident signal. The wireless device can utilize the RIS to fine-tune the reflection properties of the wireless sensing signal, such as phase, amplitude, and polarization. By dynamically adjusting these parameters, the RIS can enhance signal strength, directivity, and overall signal quality. For instance, the RIS can focus the reflected signal towards the transmitter, significantly improving signal reception. This capability is particularly advantageous in urban environments where obstacles and interference are prevalent. Technical details of the RIS involve the implementation of tunable elements, such as varactor diodes or microelectromechanical systems (MEMS), in each unit cell. These elements allow real-time reconfiguration of the surface's electromagnetic properties in response to control signals from the wireless device. The control signals can be generated based on real-time analysis of the received signal's quality and contextual parameters, ensuring optimal reflection under varying conditions. The RIS can operate in various frequency bands, including sub-6 GHz and millimeter-wave (mmWave) frequencies, making it versatile for different wireless applications. Additionally, the RIS can incorporate sensing capabilities to monitor the environment and further refine the reflection parameters. For example, integrated sensors can detect changes in temperature, humidity, or the presence of obstacles, and adjust the reflection properties accordingly to maintain high signal quality. Quality of Service: a wireless system may be used to transport data belonging to different types of applications such as Machine Type Communication (MTC), Critical Machine Type Communication (CMTC), Enhanced Mobile Broadband (EMB), or Fixed Wireless Access (FWA). MTC (e.g., smart meters, tracking,...) requires low bandwidth and non-latency critical, CMTC (e.g., industrial applications) has strict throughput, latency, and availability needs, EMB (VR / AR, 4K UDH, ...) and FWA (e.g., in the home) require high data rate, with low latency, and low end-to-end response time. In wireless network such as 5G the Quality of Service has to accommodate different applications such as EMB, MTC, ultra-reliable low latency communications. QoS is influenced by the entities involved in the communication, UE, RAN, UPF, and DN. Data exchanges between UE and DN are mapped to QoS flows, and each QoS flow is mapped to a 5G QoS Identifier (5QI) in TS 23.501 (Table 5.7.4-1) that describes resource types, priority, packet delay budget, packet error rate, maximum data burst volume. Network is configured to configure RAN and core network interfaces to achieve the requirements of a 5QI. QoS is applied to a data stream from the wireless physical layer to the core network. Between RAN and UPF, QoS is applied in terms of a QoS flow. QoS in the RAN is managed by means of Data Radio Bearers (DRB). A QoS flow on core network side is created by means of a PDU session establishment accept. The mapping between a QoS flow and a DRB is done by means of SDAP configuration in an RRC message (RRCSetup or RRCReconfiguration). The indication or identifier that connects the whole QoS pipe is called QoS flow identifier. Downlink traffic requires mapping IP messages and the QoS pipe, and this is done by the UPF. For each IP message or packet, the UPF checks (by means of a packet QoS assignment / detection rule) the packet information (source / destination / protocol / type of service / ...) and directs the IP packet to a QoS flow. The packet QoS assignment / detection rule is provided by SMF interacting with PCF. In the uplink, the UE performs a similar task by applying QoS rules provided in NAS messages (e.g., PDU session establishment) by the SMF or are pre-configured / derived by the UE.

[0107] Discontinuous reception (DRX) in cellular networks such as 5G is in two types, Idle mode DRX and Connected mode DRX. In Idle mode DRX, the UE wakes up to monitor for paging messages. If no paging message is detected, it sleeps further. In Connected DRX mode, the UE enters in sleep mode periodically and during the sleep period the UE is not required to monitor the Physical Download Control Channel. The access device configures the UE device with C-DRX parameters. Connected DRX approach reduces energy consumption of the device because it does not require monitoring the PDCCH periodically and it also reduces the transmissions of CSI or SRS signals, that also has a positive effect in the network / access devices load. There are two types of DRX cycles, long and short. A long DRX cycle consists of an on period and an off period. The on duration is in terms of milliseconds. The long DRC cycle may be configured or the long DRX cycle and short DRX cycles may be configured. The access device can configure the time (drx-onDurationTimer) during which the UE is awake and goes back to sleep if there is no PDCCH received. The access device can also configure a given drx-LongCycleStartOffiset to start to awake period at a subframe boundary and / or drx-SlotOffset relative to the subframe boundary. If there is activity in an awake period, the UE may remain awake some more time determined by the drx-lnactivityTimer. Furthermore, the access device can configure long DRX cycle together with additional DRX cycle which is shorter than long DRX cycle. Configurable parameters include the drx-ShortCycle (duration of the short cycle) and drx- ShortCycleTImer that determines how many short cycles before the device should apply.

[0108] Data scheduling in a cellular network such as a 5G cellular network may be performed by means of a scheduler wherein the scheduler takes as input information such as measurements of UE / network, buffer status report, QoS requirements, associated radio bearers, or a scheduling request. In the downlink, data scheduling may be performed by means of dynamic scheduling and semi persistent scheduling (SPS). In dynamic scheduling, every data exchange in the Physical Downlink Shared Channel (PDSCH) is scheduled by means of a downlink control information (DCI) message in the Physical Downlink Control Channel (PDCCH). In SPS, the scheduling is done by means of an RRC message. In the uplink, scheduling can be performed by means of dynamic scheduling and configured scheduling (CS). In dynamic scheduling each Physical Uplink Shared Channel (PUSCH) is scheduled over DCI. In CS, the PUSCH transmission is scheduled via RRC message. Furthermore, a Scheduling Request message may be sent over the PUCCH (Physical Uplink Control Channel) or in an Uplink Control Information (UCI) in the PUSCH (Physical Uplink Shared Channel). An SR may be sent by a UE device when it has data to transmit. Upon reception, the access device can allocate resources (Uplink Grant by means of the Physical Downlink Control Channel. Upon resource allocation, the UE device can transmit data in the Physical Uplink Shared Channel.

[0109] Wireless sensing and integrated wireless sensing and communication: wireless systems are evolving to include wireless sensing capabilities. These wireless sensing capabilities may be implemented e.g. by a radar functionality in wireless communication involving one or more access devices (e.g., base stations (BS)) and / or one or more terminal devices (e.g., UEs). As an example, Frequency Modulated Continuous Wave (FMCW) mmWave radar systems can measure range, velocity, and angle of arrival (if two receivers are available) of objects in the scene which reflect radio waves. Such radar systems emit a chirp signal, e.g., a sine wave that increases in frequency over time. The chirp signal (e.g., a continuous wave pulse) has a bandwidth and a frequency increase rate. Generally, a continuous series of such chirps are emitted. The transmitted and received analogue chirp signals are mixed to generate an intermediate frequency (IF) signal which corresponds to the difference in frequencies of the two signals (outbound and inbound) and whose output phase corresponds to the difference in the phases of the two signals. Each surface of a scene or environment will therefore produce a constant frequency IF signal whose frequency relates to the distance to the surface (i.e., a first distance from the transmitter of the chirp signal to the surface plus a second distance from the surface to the receiver of the chirp signal). To resolve two surfaces at different distances, the two IF signals can be frequency resolved. A longer time window of the IF signal results in greater resolution. As the chirp time is related to its bandwidth (with constant chirp frequency change) the resolution of the radar is related to the chirp bandwidth. The IF signal may then be band pass filtered (to remove signals below some minimal range and frequencies above the maximum frequency for a subsequent analogue-to-digital converter (ADC)) and digitized prior to further processing. The upper frequency sensing range of the bandpass filter and ADC sets the maximum range that can be detected (i.e., IF frequencies increase with range). To detect vibrations, the phase of the IF signal is important, since the phase (i.e., the difference in phases of the transmitted and received chirp signals) is a sensitive measure of small changes in the distance of a surface. Small distance changes can be detected in the phase signal but may be indiscernible in the frequency signal. Moreover, phase difference measures between two consecutive chirp signals can be used to determine the velocity of the surface. As an example, a fast Fourier transform (FFT) processing can be performed across multiple chirp signals to enable separation of objects with the same range but moving at different velocities. A Fourier transform converts a signal from a space or time domain into the frequency domain. In the frequency domain the signal is represented by a weighted sum of sine and cosine waves. A discrete digital signal with N samples can be represented exactly by a sum of N waves. FFT provides a faster way of computing a discrete Fourier transform by using the symmetry and repetition of waves to combine samples and reuse partial results. This method can save a huge amount of processing time, especially with real-world signals that can have many thousands or even millions of samples. As a further example, angle estimation can be performed by using the phase difference between the received chirp signal at two separated receivers.

[0110] As another option, a channel state information (CSI) can be used, which is a measure of the phases and amplitudes of many frequencies detected at a receiver, thereby forming a complex 'map' of the radio environment, including effects of objects within that environment. CSI characterizes how wireless signals propagate from the transmitter to the receiver at certain carrier frequencies. CSI amplitude and phase are impacted by multi-path effects including amplitude attenuation and phase shift, e.g., by the displacements and movements of the transmitter, receiver, and surrounding objects and humans. In other words, CSI captures the wireless characteristics of the nearby environment. These characteristics, assisted by mathematical modeling or machine learning algorithms, can be used for different sensing applications. A radio channel may be divided into multiple subcarriers, as is done e.g. in 5G communication systems (using e.g. orthogonal frequency division multiplexing (OFDM)). To measure CSI, the transmitter may send long training symbols (LTFs), which contain pre-defined symbols for each subcarrier, e.g., in a packet preamble. When those LTFs are received, the receiver can estimate a CSI matrix using the received signals and the original LTFs. For each subcarrier, the channel can be modeled by y = Hx + n, where y is the received signal, x is the transmitted signal, H is the CSI matrix, and n is the noise vector. The receiver estimates the CSI matrix H using a pre-defined signal x and the received signal y after signal processing such as removing cyclic prefix, de-mapping and demodulation. The estimated CSI is then a three-dimensional matrix of complex values and this matrix represents an 'image' of the radio environment at that time. By processing a time series of such 'images' information on movements, locations and vibrations of objects can be extracted. Such a processing of a CSI matrix can be used for vital signs monitoring, presence detection, and human movement recognition. As an example, neural network like recognition techniques can be used to process the CSI matrix to perform such kinds of recognition.

[0111] It is noted that systems using channel state information (CSI) are somehow related to systems with FMCW mmWave radar. In a CSI-based system, the input signal X may be defined and the receiver may use the received signal Y to obtain H, i.e., as H = (Y - N) / X . In a FMCW mmWave radar, the transmitted signal Chirp X may also be predefined, and the receiver may use the received signal Y to obtain a transfer function as H = Y / X . This last step is in fact somehow related to multiplying the locally computed chirp signal and the received chirp signal and applying a bandpass filter. According to various embodiments in this invention, the above-described wireless sensing techniques are implemented in a mobile communication system (e.g. 5G or 6G or other cellular or Wi-Fi communication systems), while the functional coexistence of radar and communication operating in the same frequency bands is configured to avoid interference bandwidths. Thereby, radio sensing can be integrated into large-scale mobile networks to create perceptive mobile networks.

[0112] As another example, the sensing signal may consist of a number of pulses sent, e.g., at specific frequencies and timing (sensing signal parameter information) by a sensing transmitter. The sensing receiver may include a number of bandpass filters that allow identifying the sensing signal parameter information, e.g, timing and frequency of the received pulses. In particular, if the transmitter determines a given pseudo-random sequence of frequency / timing pulses and beams it, e.g., by means of beamforming, in a specific direction, and if the transmitter communicates to the receiver the timing / frequency, in general, the sensing signal parameter information, of the transmitted sensing signal, the receiver can use its bandpass filters to identify the reception of the same transmitted pulses, i.e., sensing signal, based on the received sensing signal parameter information.

[0113] The wireless sensing signal may be part of the synchronization signal block. For instance, the wireless sensing signal may be a reference signal included in the primary synchronization signal or in the secondary synchronization signal. It may consist of a number of reference signals and / or it may be a wide band signal. This wireless sensing signal can allow the access devices to determine the presence of a wireless device. The wireless device may also use this wireless sensing signal to determine the access device that is more suitable to (re-)select.

[0114] Wireless local area network technologies such as Wi-Fi allow devices to connect to the Internet or to each other without using cables. Wi-Fi is based on radio waves that are transmitted and received by a device called a wireless access point (AP). The AP acts as a hub that connects Wi-Fi enabled devices, such as laptops, smartphones, tablets, smart TVs, etc., to a wired network, such as a local area network (LAN) or the Internet.

[0115] The term Wi-Fi is a trademark of the Wi-Fi Alliance, an industry association that certifies products that comply with the IEEE 802.11 standards for wireless local area networks (WLANs). These standards define the physical and data link layers of the communication protocol, such as the frequency bands, modulation schemes, encryption methods, authentication mechanisms, and data rates used by WiFi devices. The most common Wi-Fi standards are 802.11a, 802.11b, 802.11g, 802. lln, 802.11ac, and 802.11ax, which operate in different frequency bands (2.4 GHz, 5 GHz, or both) and offer different levels of performance and compatibility.

[0116] To use Wi-Fi, a device needs to have a wireless network interface card (NIC) that can send and receive radio signals. The NIC scans the available wireless channels and detects the presence of nearby APs. The device then selects an AP to connect to, based on factors such as signal strength, security settings, and network name (SSID). The device and the AP exchange information, such as the MAC address, IP address, encryption key, and password, to establish a connection. This process is called association. After the connection is established, the device can communicate with the AP and other devices on the same network, or access the Internet through the AP.

[0117] IEEE 802. lln (Wi-Fi 4) provided new features such as MIMO and frame aggregation to increase throughput. IEEE 802.11ac (Wi-Fi 5) introduced wider bandwidth and MU-MIMO. IEEE 802.11ax (Wi-Fi-6) included OFDMA and BSS color or spatial reuse to use spectrum resources more efficiently. IEEE 802. llah introduced target wake time (TWT) to support low power loT applications by allowing STAs to go into sleep when not in a wake period after negotiation with AP. IEEE 802.11be (Wi-Fi 7) aims at improving throughput and latency operating in unlicensed bands between 1GHz and 7.125 GHz. Wi-Fi 7 increases bandwidths up to 320 MHz, 4096 QAM modulation, and supporting up to 16 spatial streams in MU-MIMO with an improved sounding procedure. Wi-Fi 7 also enables multiple resource units to be assigned to a single device. Furthermore, it includes an enhanced preamble with a universal SIG filed indicating the PHY version. It also extends the negotiated ack buffer size to 1024 bits. It also enables multilink operation (MLO) enabling multiple links between a station and an access point, for instance an AP can have two radios 2.4 and 5 GHz and use both of them for simultaneous transmission and / or reception with a multi-link capable device (MLD) capable station. Wi-Fi 7 also includes a restricted TWT providing predictable latency by assigning STAs to different rTWT types and making sure that other STAs do not transmit if they do not belong to a given rTWT type. Wi-Fi 7 also includes multi-AP coordination performing, e.g., coordinated transmission, beamforming, or joint transmission.

[0118] For instance, in references to Fig. 1, devices 100, 101 and 102 can be Wi-Fi access points and device 106 can be a wireless station. Station 106 and access point 101 are MLD and communicate with two links 126. Device 102 is a cellular capable residential gateway.

[0119] This invention is described in the context of recent developments wherein NR has been greatly expanded in support of sidelink / PC5 communication between UEs since Release 16 (R16). A number of device-to-device use cases, e.g., wireless device-to-wireless device use cases, e.g., sidelinkbased use cases can also potentially benefit from AI / ML functionality, such as sidelink CSI compression / prediction, sidelink resource sensing for autonomous scheduling, sidelink positioning etc. Therefore, there is a need to extend the current 3GPP AI / ML framework for the radio air interface to the sidelink / PC5 interface to support AI / ML functionality for UE-to-UE communications.

[0120] Fig. 8 schematically describes the 3GPP AI / ML framework for selected use cases including (1-1) channel state information (CSI) compression; (1-2) CSI prediction; (2-1) spatial-domain downlink beam prediction; (2-2) temporal-domain downlink beam prediction, (3-1) direct AI / ML positioning and (3- 2) assisted AI / ML positioning. In the figure, a wireless device such as a UE includes an AI / ML model to fulfil one or more of previous use cases. The wireless device communicates with an access device (base station) by means of a communication interface, e.g., Uu interface. Multiple functions of a network are shown including a mobility function such as the 5G AMF and / or the LMF. The figure also shows the Operations, Administration, and Maintenance (0AM) functionality.

[0121] In the current framework, all devices seem to make use of the AI / ML model and / or functionality when they are in the same network, e.g., home network or home PLMN. However, devices may roam, changing the network either within a same country or between countries. When a device moves to a different network, e.g., a network in a different country, the country legislation may be of a different nature, and collected data may not be used for training purposes and / or an AI / ML model trained using data of a first country may not be used in a different country. Thus, it is an aim of the invention to address such a shortcoming of the current AI / ML RAN framework.

[0122] In the sidelink / PC5 communication, UEs may not be in coverage of a base station, or only some UEs of a group of UEs are in the coverage of the base station (i.e. partial coverage). The current 3GPP AI / ML framework relies on the Uu radio air interface for the collaboration between UE-side AI / ML model and Network-side AI / ML model, and does not support federated learning between UEs, therefore it is not sufficient to cover the cases where UE is out of the coverage of the network, or support federated learning and collaboration between UEs, or support federated learning and collaboration between UEs and network nodes jointly. Thus, it is an aim of the invention to describe how the existing framework can be improved by using a local communication interface such as the PC5 / slidelink communication interface.

[0123] As a matter of illustration, Fig. 9 describes a potential extension of the current 3GPP AI / ML framework including the sidelink / PC5 communication interface, in general, a device-to-device communication interface, e.g., Wi-Fi based. Fig. 9 shows three devices, e.g., three UEs, each of them configured with an AI / ML model and capable of communicating and / or collaborating with each other through the PC5 / sidelink communication interface.

[0124] This embodiment is described in the context of wireless devices such as User Equipments (UEs), e.g., using sidelink as the device-to-device communication technology and / or in roaming. However, embodiments of the invention apply to other wireless devices in roaming and / or other device-to-device wireless communication technologies.

[0125] Section: AI / ML model for the sidelink / PC5 radio air interface

[0126] In one embodiment, as illustrated in Fig. 10, two or more UEs may participate in a UE-to-

[0127] UE communication use case, e.g. metaverse Extended Reality (XR), where high throughput low latency data streams are exchanged between UEs over the local device-to-device interface, e.g., sidelink / PC5 interface, in the local area, and AI / ML functionality may be used for one or more AI / ML functionality, e.g., to compress the data streams, perform real-time media content analysis for the data exchanged between UEs, etc. AI / ML functionality may also be used to perform certain RAN or networking functions such as beam steering or PC5 / sidelink specific functions such as UE-to-Network or UE-to-UE relay selection, path switch triggering, etc. Fig. 10 describes an access device 1000 such as a base station that orchestrates or helps in the orchestration (e.g., by forwarding (configuration / request) messages between wireless devices and a core network) of the enhanced AI / ML communication / collaboration between several devices 1001, 1002, 1003 in the local network by using of a local communication interface 1004, e.g., PC5. The access device and / or devices may comprise an AI / ML model 1005, e.g., as described in different embodiments, and an AI / ML model function 1006. The AI / ML model function 1006 may be, e.g., one or more of the AI / ML model provision, configuration, training, transfer, update, inference, management and / or monitoring.

[0128] In an embodiment, UE may have its own AI / ML model for a particular (sidelink / PC5) AI / ML use case / functionality, and each AI / ML model may be identified by its functionality and / or a model identifier. The AI / ML model may be provisioned from the network / through the access device (base station) when the UE is in coverage of the network, or the model may be provisioned by the vendor in an upgraded firmware over the top (i.e., using application layer protocols) by using the interface out of the scope of 3GPP, or the model may be provisioned by the other UE in the UE-to-UE communication.

[0129] For AI / ML model provisioning, in one embodiment, UE may be in coverage of the base station, and UE may receive the AI / ML model directly from the network.

[0130] In one embodiment, a first UE may not be in coverage of the base station, while the other UE in the UE-to-UE communication may be in coverage of the base station, then the first UE may be provisioned with the AI / ML model by the network through the other UE as a relay. The first UE may send a request message via the relay UE to the network: (1) for the particular AI / ML model it may be interested in, (2) for the particular AI / ML functionality with the AI / ML parameters associated to first UE ID (entity), capability of the first UE (including but not limited to radio capability, UE capability, AI / ML capability), and the AI / ML functionality interested. Upon receiving the AI / ML model request via a relay UE, the network may respond with the selected AI / ML model and / or the assistance information of the selected AI / ML model (e.g. testing performance of the model such as accuracy, latency etc., hardware requirements to operate the model such as RAM, CPU, GPU) to the UE via the relay UE. The UE may further send a response to the network via the relay UE on confirmation of the reception of the model, such as model accepted and deployed, model not accepted / out of capability of the UE, or simply model not accepted. The network may further respond with a new selected AI / ML model and the assistance information of the selected AI / ML model, then the UE may further respond with a confirmation of the model reception similarly as before.

[0131] In a variant of this embodiment, the first UE may send a request including and / or referring to one or more AI / ML models. The network may then perform a selection.

[0132] In a variant of the embodiment, the network may respond to the first UE with a list of candidate AI / ML models and the assistance information of each model (e.g. hardware requirements to operate the model such as RAM, CPU, GPU). Then the first UE may reply with a selection of the AI / ML model from the list of candidate models based on capability match of the first UE with the assistance information of the candidate AI / ML model provided by the network. The first UE may also reply that none of the candidate AI / ML models fits. Then the network may respond with a new set of candidate AI / ML models and assistance information accordingly.

[0133] For AI / ML model training and update, in one embodiment, the UE may use its locally collected data to train and update the model, or the data collected by the other UE and transferred to the first UE over the UE-to-UE communication interface, or both the data locally collected by the UE and the data collected by the other UE.

[0134] For AI / ML model inference, in one embodiment, the first UE may use its local AI / ML model to perform inference operation.

[0135] In another embodiment, the first UE may use the AI / ML model of another UE or network node for AI / ML model inference operation.

[0136] In another embodiment, a federated inference scheme may be used, where the inference may be performed on more than one UE each with configured / customized and trained AI / ML model, then a joint inference process may be used to reach the final inference result. In a variant of embodiment, the joint inference process may select the inference result preferred by the majority of the UEs with AI / ML model participating in the inference. In a variant of embodiment, the joint inference process may combine the inference result of each UE with AI / ML model participating the inference according to a confidence level estimate from the AI / ML model inference, using a method of weighted sum of confidence level.

[0137] For AI / ML model monitoring, one node in the network may be selected to have the monitoring function, which may use a set of data to validate the performance of the AI / ML model during the lifetime of the model, e.g., to validate whether the AI / ML model achieves a desired performance level. When the performance of the AI / ML model degrades, the AI / ML model monitoring function may send a management command to the AI / ML model to deactivate the current active model and select a new model for the AI / ML functionality. Furthermore, the AI / ML model monitoring function may trigger the training and update of the AI / ML model with degrading performance. The AI / ML model may be reselected / reactivated for the AI / ML model inference when the performance test results exceed the required performance value. In a variant of embodiment, the AI / ML model monitoring function may deactivate the currently active performance degrading AI / ML model, and may not select any new AI / ML model, then the system may fall back to an operation mode without any AI / ML functionality. In a variant of embodiment, the management command sent by the AI / ML model monitoring function to activate / deactivate the AI / ML model on a UE may be a PC5-RRC message and / or PC5-S message. In a variant of embodiment, the fallback management command sent by the AI / ML model monitoring function to the UE to fall back on the operation without AI / ML functionality may be a PC5-RRC message and / or PC5-S message.

[0138] In one embodiment, the AI / ML model monitoring function may co-locate with the AI / ML model to be monitored.

[0139] In another embodiment, the AI / ML model monitoring function may locate on another UE in the local area, where AI / ML model monitoring related parameters and data may be exchanged from the UE with the AI / ML model to be monitored to the UE with AI / ML model monitoring function.

[0140] In a variant of embodiment, the first UE may have direct sidelink / PC5 connection with the other UE.

[0141] In another variant of embodiment, the first UE may have indirect connection with the other UE, such as through one-hop or multi-hop UE-to-UE (U2U) relay.

[0142] In another embodiment, the AI / ML model monitoring function may locate on a network node, e.g. base station, a node in the core network such as AMF, LMF, 0AM, where AI / ML model monitoring related parameters and data may be exchanged from the UE with the AI / ML model to be monitored to the network node with AI / ML model monitoring function.

[0143] In a variant of embodiment, the first UE may have a direct connection with the network node.

[0144] In another variant of embodiment, the first UE may have an indirect connection with the network node, such as through one-hop or multi-hop UE-to-Network (U2N) relay. For the AI / ML model provision, configuration, training, transfer, update, inference, management and / or monitoring related signalling over the Uu radio air interface, pre-configuration may be used, or configuration messages such as SIBs broadcasting (in RRC_IDLE / RRC_INACTIVE state), dedicated RRC signalling (in RRC_CONNECTED state), Downlink Control Information (DCI) on PDCCH, or MAC Control Elements (CE) commands between the network and the first UE. Furthermore, for the AI / ML model provision, configuration, training, transfer, update, inference, and / or monitoring related signalling over the sidelink / PC5 radio air interface, pre-configuration may be used, or PC5-S and / or PC5-RRC signalling between UEs.

[0145] In another embodiment, multiple UEs may participate in a UE-to-UE communication use case, e.g. sidelink relay service involving UEs such as a sidelink source UE, a sidelink relay UE, and a sidelink destination UE, where sidelink data packets are exchanged between UEs, and AI / ML functionality may be used to improve sidelink relay selection or path selection by methods of either centralized AI / ML sidelink relay or path selection, or distributed AI / ML sidelink relay or path selection.

[0146] In an example, in the centralized AI / ML sidelink relay or path selection, the network, e.g., serving PLMN (e.g. base station or core network node), may distribute an AI / ML model to all the UEs participating in the sidelink relay service, where the AI / ML model may take into account multiple input parameters such as signal quality, battery level, mobility pattern, traffic load, and / or service priority of the communication link between UEs, and output the optimal relay UE or path for each source-destination pair. The UEs may then use the AI / ML model to make decisions on relay selection or path selection based on their local observations and measurements.

[0147] In the distributed AI / ML sidelink relay or path selection, each UE may have its own AI / ML model that is / maybe further trained with its own data and / or data from other UEs, where the AI / ML model may take into account multiple input parameters such as signal quality, battery level, mobility pattern, traffic load, and / or service priority of the communication link between UEs, and output the optimal relay UE or path for each source-destination pair. The UEs may then use the AI / ML model to make decisions on relay selection or path selection based on their local observations and measurements, and communicate their decisions to other UEs via PC5-S signalling.

[0148] In another embodiment that may be combined with other embodiments or used independently, as illustrated in Fig. 10, multiple UEs may participate in a UE-to-UE communication use case, e.g. V2X communication involving UEs such as Vehicle to Vehicle (V2V), Vehicle to Infrastructure (V2I), and Vehicle to Network (V2N), where V2X data packets are exchanged between UEs, and AI / ML functionality may be used to improve V2X communication performance by methods of either centralized AI / ML V2X resource allocation, or distributed AI / ML V2X resource allocation.

[0149] In the centralized AI / ML V2X resource allocation, the PLMN, e.g., the serving PLMN (e.g. base station or core network node) may distribute and / or indicate an AI / ML model to all the UEs participating in the V2X communication service, where the AI / ML model may take into account multiple input parameters such as traffic density, vehicle speed, channel quality, latency requirement, and service type of the V2X communication link between UEs, and output the optimal resource allocation scheme for each V2X pair. The UEs may then use the AI / ML model to make decisions on resource allocation based on their local observations and measurements. In this case, only UEs whose home PLMN is the serving PLMN may participate in the learning process whereby UEs may report measurements to the network to train the AI / ML model that may then be distributed. This may also be configurable so that any UE may participate in the learning process (e.g., by reporting measurements). This may depend on agreements between PLMNs, country location of the PLMNs, as well as the local regulations. This may be controlled by means of a policy configured in the UE, e.g., distributed by the home PLMN. This embodiment has been illustrated in the context of V2X, however, such a policy may also be applicable to other use cases, e.g., UEs communicating via the Uu interface with an access device.

[0150] Section: AI / ML for sidelink positioning

[0151] In one embodiment, as illustrated in Fig. 10, multiple UEs may participate in a UE-to-UE communication use case, e.g. sidelink positioning involving UEs such as sidelink target UE, sidelik anchor UE, and sidelink positioning server UE. In such a use case, e.g., sidelink positioning, related measurements and assistance data may need to be exchanged between UEs, and AI / ML functionality may be used to improve the performance of the protocol / use cases, e.g., sidelink positioning accuracy, by methods of either directed AI / ML sidelink positioning, or assisted AI / ML sidelink positioning.

[0152] In the directed AI / ML sidelink positioning, sidelink positioning measurements and assistance data from multiple anchor UEs are taken as input to AI / ML model running on sidelink positioning server UE and / or LMF or jointly for AI / ML model inference operation to output the exact and / or estimated location information of a sidelink target UE. In one embodiment, the direct AI / ML sidelink positioning model may be trained with sidelink positioning measurements and / or assistance data from multiple anchor UEs at a known location, e.g. Positioning Reference Unit (PRU), where PRU may locate near the target UE with location information unknown and anchor UEs participating in the sidelink positioning operation.

[0153] In another embodiment, the direct AI / ML sidelink positioning model may be trained with sidelink positioning measurements and assistance data from multiple anchor UEs as the input features for positioning a target UE, which may be in coverage of the base station so that Uu positioning may be used to obtain the location information as the desired target location information to train the AI / ML model.

[0154] In another embodiment, the target UE may be equipped with GNSS and may receive the (real-time) location information from GNSS, then the location information may be used as the desired target location information together with sidelink positioning measurements and assistance data from multiple anchor UEs to train the AI / ML model. Additionally or alternatively, the location information and sidelink positioning measurements and an AI / ML model may be used for inference of an improved positioning estimation.

[0155] In the assisted AI / ML sidelink positioning, sidelink positioning measurements and assistance data are taken as input to the AI / ML model running on sidelink positioning server UE and / or LMF or jointly for AI / ML model inference operation to output the intermediate positioning measurements which may be sent to sidelink positioning server UE and / or LMF or jointly for further sidelink positioning location estimate. In such a case of assisted AI / ML positioning, AI / ML model may be used to help improve the quality of direct measurements, i.e. using AI / ML model to take direct measurements as input to output the intermediate measurements for positioning estimate.

[0156] Section: AI / ML for sidelink sensing

[0157] In one embodiment, wireless sensing may be employed to determine, e.g., the features of an object and wireless sensing may use AI / ML functionalities. Wireless sensing may be utilized between two wireless devices, such as two cars, and may rely on a local device-to-device communication link. A device may be configured with an AI / ML functionality and / or model and the device may perform wireless sensing based on it.

[0158] To initiate the process, the wireless devices may establish a communication link, e.g., a sidelink communication link, which serves as the foundational channel for transmitting and receiving data. The local communication link may ensure low latency and high reliability, essential for accurate sensing operations. Once the communication link is established, the devices may perform initial data exchange, sharing preliminary information about their relative positions and motion parameters and / or other parameters of interest, such as, e.g., device type that may give information, e.g., about the device form and / or composition. Data may also include speed, direction, and distance, which can be gathered through integrated sensors such as radar or LiDAR.

[0159] Wireless sensing may be performed, e.g., using above data to increase accuracy.

[0160] AI / ML models may also be employed to enhance the sensing accuracy. The models may be selected, e.g., based on the exchanged data. For further accuracy, the devices may utilize advanced AI / ML algorithms to filter noise and predict potential movements or changes in the object's features. These algorithms may leverage historical data and pattern recognition techniques to provide more reliable and precise sensing outcomes.

[0161] These models may be (have been) trained with extensive datasets containing various object features and environmental conditions. The training process may involve federated learning, where each device trains a local model with its specific data and sends the model parameters to a central entity for global aggregation.

[0162] The aggregated global model may be redistributed to all participating devices, ensuring consistency and robustness in the sensing operation. The AI / ML models may be capable of real-time inference, continuously updating the object's features based on the incoming data from the communication link.

[0163] Section: AI / ML model roaming

[0164] Different AI / ML use cases may have different requirements and regulations. For example, AI / ML for the use cases of radio resource management, such as CSI compression, CSI prediction, beam management, modulation and coding scheme selection, may have less restrictions than AI / ML for the use cases of positioning accuracy enhancement, multimedia content compression for extended Reality (XR). In order to take into account of these different levels of restrictions, the AI / ML model may have an associated restriction level in the form of numeric values or categorical values which may indicate the appropriateness of the use of the model in a use case.

[0165] In one embodiment, when an AI / ML model is selected for a particular AI / ML functionality, the model's restriction level may be checked by the initiator initiating the AI / ML functionality to verify whether the model's restriction level conforms to the required restriction level by the particular AI / ML functionality, which may be configured by the network or the upper layer. If the model's restriction level does not conform to the network or upper layer configured restriction level for a particular AI / ML functionality, the operation may fall back to the non-AI / ML operation mode

[0166] Different countries / regions may have different regulations for the use of AI / ML, therefore requirements on the AI / ML model may be different for the PLMNs operating in different countries / regions. In another embodiment, when a UE roams from a home PLMN to a visiting PLMN, the two PLMNs may have different regulations and requirements on the use of AI / ML functionality. The AI / ML model from the home PLMN (either UE-sided or network-sided) may not conform to the local regulations and requirements of visiting PLMN. In order to use the AI / ML functionality in the visiting PLMN, the UE may send a request (including the current AI / ML model information and assistance data of the model and its hardware capabilities) to the network to request a compatible AI / ML model for a particular AI / ML functionality from the visiting PLMN, then the visiting PLMN may provide one or multiple compatible AI / ML models and the associated assistance data (such as testing performance, running time, hardware requirements like CPU, RAM, GPU, latency etc.) for the use of a particular AI / ML functionality. Then the UE may select one local AI / ML model and confirm the selection with the visiting PLMN, or simply reject all candidate local AI / ML models from the visiting PLMN. Upon reception of the confirmation at the visiting PLMN, if the UE rejects all candidate local AI / ML models and requests a new set of candidate local AI / ML models, the visiting PLMN may provide another set of candidate local AI / ML models and the associated assistance data to the UE.

[0167] In an embodiment that may be combined with other embodiments or used independently, an AI / ML function and / or model may be enabled / disabled / adjusted before determining the network may change and / or once it has been determined that the network has changed. In some cases, the network selection may be based on whether certain AI / ML models / functions are available and / or may be used. Allowed / supported AI / ML models / functions may be signalled to a device, e.g., by means of a broadcast message including system information, and / or by means of a unicast control message (e.g., RRC command). A communication / wireless device may be configured with a configuration containing criteria to enable / disable / adjust the usage of AI / ML models / functions. The configuration may include a list of networks / radio access technologies for which certain AI / ML functions may be enable / disabled.

[0168] Section: AI / ML model identification

[0169] For an AI / ML use case for sidelink / PC5 radio air interface, one or more UEs may have a specific AI / ML functionality or model for sidelink / PC5. However, it may be problematic to identify the right model. Thus, in an embodiment that may be combined with other embodiments or used independently, model identification may include the information of AI / ML functionality, and / or an identifier of the group of one or more UEs participating in the AI / ML use case. For example, it could include destination address and / or Layer2 ID for groupcast communication, or a destination address and / or Layer2 ID for broadcast communication.

[0170] Thus, in accordance with a general definition of the invention it is proposed a method for operating a wireless device, comprising the wireless device communicating (receiving / transmitting) a model identification of an AI / ML functionality and / or a participant identifier indicative of one or more devices using the AI / ML functionality, and the wireless device using the AI / ML functionality identified in the model identification.

[0171] In an example, the information of AI / ML functionality in the Model identification may be (or include) an identifier of the application / service of AI / ML functionality, such as Relay Service Code (RSC) defined for sidelink relay communications. In some cases, the RSC may include one or more attribute describing the AI / ML functionalities associated with the RSC.

[0172] In another example, the model identification may include and / or refer to assistance information, such as additional data related to the requirements of types of data collected for training, testing and performance monitoring / validation, conditions to start / pause / resume / stop the data collection, and / or performance related parameters for model updates / selection / reselection.

[0173] In another embodiment that may be combined with other embodiments or used independently, a UE may use the assistance information provided in the model identification to monitor its testing or performance of the AI / ML functionality. The assistance information may for example include criteria or thresholds for evaluating the testing or performance results, such as accuracy, precision, recall, Fl-score, etc. The assistance information may also or alternatively include instructions or policies for updating or reselecting the model, for example based on the evaluation. For example, the assistance information may specify that if the accuracy of the model falls below a certain value, the UE should request a new model from the network or from another UE.

[0174] Alternatively, the assistance information may specify that if the precision of the model exceeds a certain value, the UE should share its model with the network or with another UE. Sharing the model may involve sending an indication of the achieved performance first and / or sending it to a common repository and / or distributing it upon agreement with / request from another device. The assistance information may be generic and applicable to multiple models, or it may be specific to a particular model. The assistance information may be provided by the network as part of the model identification, for example at the initial configuration, and / or it may be requested by the UE separately, for example in the course of operation or at initialization. In this way, the UE may use the assistance information to improve its AI / ML functionality and participate in the AI / ML use case more effectively.

[0175] In another embodiment that may be combined with other embodiments or used independently, the model identification may also contain information regarding the requirements to use the model, such as hardware requirements, software requirements, memory requirements, processing power requirements, etc. A UE that wants to participate in the AI / ML use case may interact with the network providing the model and provide its capabilities, such as device type, hardware specifications, software version, memory capacity, processing speed, etc. Based on the capabilities of the UE, the network may provide a suitable model that the UE can use for the AI / ML functionality. Alternatively, the network may provide a model to the UE and the UE may check whether it is able to execute it or not. If not, the UE may request another model from the network or decline to participate in the AI / ML use case.

[0176] Section: Federated learning

[0177] Federated learning enables multiple devices to collaboratively learn from their local data without sharing the data itself. Instead, the devices share the model parameters or updates, such as gradients, weights, or features, with a central server or with each other. Federated learning can be done in two main ways: centrally or decentrally.

[0178] In the centralized approach, a central server coordinates the learning process and aggregates the model updates from the devices. The central server may also provide the initial model or the global model to the devices, and monitor their performance and feedback. The devices may communicate with the central server periodically or on-demand, depending on the availability of resources and network conditions. The centralized approach may be suitable for scenarios where the devices have similar data distributions and capabilities, and where the communication cost is low.

[0179] In the decentralized approach, the devices communicate with each other directly or via intermediate nodes. The devices may exchange their model updates or parameters with their neighbours, and use consensus algorithms or distributed optimization methods to reach a global agreement. It may also be applicable to situations in which communication is costly or not possible, e.g., when devices are out of coverage.

[0180] Federated learning may involve devices of different capabilities, in terms of CPU, memory, battery, etc affecting the quality of the learning process. Multiple strategies may apply: - Selecting a subset of devices to participate in each round of communication, based on their availability, reliability, or performance. For instance, in sidelink communication it may be needed to identify which UEs may need to share information about their models, either centrally or decentrally, and if so in which order.

[0181] - Adapting the model size or complexity to the device capabilities, such as using model compression, pruning, or quantization techniques. For instance, device capabilities may be exchanged in sidelink and depending on the capabilities a different number of nodes in the model may be shared.

[0182] - Adapting the learning rate or the number of local iterations to the device capabilities or constraints, such as using adaptive or asynchronous methods. For instance, in sidelink depending on the network topology or the types of devices, more or less learning rounds may be applicable. This may require identifying and exchanging the learning parameters over sidelink. To enable these strategies, it is important to identify and exchange identification information either with a central server, e.g., a NF in the core network or other devices about the federated learning model and purpose, the version of the model (e.g., since the model may have only be updated in certain devices), the group of devices that have been involved when updating the model, e.g., when done in a decentralized manner, the order in which the learning process is to be executed in a decentralized manner with a number of devices, e.g., as a list of Layer2 IDs or UserlnfolDs and capabilities of a number of UEs performing sidelink communication.

[0183] In general, in the federated learning, one or more UE participating in the AI / ML use case may have different AI / ML model, the model identification may further include the pairing information of the AI / ML models on one or more UEs, and example of pairing information is a list of Layer2 ID of UEs.

[0184] In one embodiment, for AI / ML model training and update, UEs may use a federated learning / training scheme, where one AI / ML model is trained on one UE first, then the model may be transferred to the other UEs for the further training, e.g., in a sequential order, then the jointly trained model may be distributed to one or multiple or all UEs in the group for the AI / ML model inference operation.

[0185] In another embodiment, the AI / ML model trained by federated learning may be based on local training and global aggregation scheme, where each UE in the group trains a local AI / ML model with its own data, then sends the local model or the model parameters to a central entity (e.g. sidelink positioning server UE, base station, or core network entity) for the aggregation of a global AI / ML model. The global AI / ML model may be further distributed to one or multiple or all UEs in the group for the AI / ML model inference operation or for further local training iterations. Signalling of results of further local training iterations may be done by sending the change of model parameters rather than the new model parameters.

[0186] In a variant of embodiment, the AI / ML model trained by federated learning may be associated with an identifier related to the group of UEs participating in the training, in addition to the AI / ML model identifier or AI / ML model functionality identification.

[0187] In another embodiment, for AI / ML model training and update, an entity in the network, named as AI / ML model federated learning manager, (e.g. a selected UE among a group of UEs, such as sidelink positioning server UE in the sidelink positioning accuracy enhancement use case, or base station, or an entity in the core network) may select which UEs may participate the federated training / learning of the federated AI / ML model according to some criteria, such as how much training time is allowed for the training of the model, the required performance of the federated learning trained AI / ML model from the model monitoring / validation performance result, computing / communication / energy resource available for the training etc.

[0188] In another embodiment, for AI / ML model monitoring, the AI / ML model monitoring function may validate the performance of the AI / MLmodel in each iteration of federated learning / training, and may decide to accept the updated model from the current iteration or revert back to the model before the current iteration of the federated learning / training.

[0189] In another embodiment, for AI / ML model monitoring, the AI / ML model monitoring function may monitor the performance of a model and may determine which device (e.g. wireless device) updates and which device shares its model. This may involve the execution of a protocol wherein the involved devices, e.g., two devices, negotiate the model update, e.g., one or more devices may indicate the achieved performance; e.g., one or more devices may request the achieved performance; e.g., one or more devices may request a model. This may also involve the configuration of thresholds and / or conditions and / or criteria to either offer a model that has been locally trained and / or accept a model that has been trained by a peer device. The conditions and / or criteria may be received by means of a configuration message, e.g., from the home network or the serving network.

[0190] In another embodiment, for AI / ML model training and update, the entity in the network named as manager, e.g., AI / ML model federated learning manager, may perform the aggregation tasks. For instance, when multiple communication devices have a model that needs to be further improved, the multiple communication devices may agree which of the devices works as manager and will perform the aggregation of the tasks. The selection may aim at reducing energy consumption, communication overhead, latency, ... so that the devices may execute a negotiation procedure to determine the manager. The negotiation procedure may involve exchanging certain parameters such as achieved performance, computational capabilities,.... The negotiation procedure may also involve a voting / consensus procedure wherein one or more devices vote the preferred manager.

[0191] In another embodiment that may be combined with other embodiments or used independently, AI / ML model, AI / ML model updates information, data collected for AI / ML training, testing and performance monitoring / validation, control signalling may be transmitted to or between one or more UEs participating in the AI / ML use case in the mode of unicast, groupcast and broadcast. For instance,

[0192] (1) A network function (NF) in the core network may send an AI / ML model to a UE via a dedicated radio bearer that provides reliable and secure transmission. The NF may include the model identification information in the header or the payload of the message carrying the model. The UE may use the model identification information to verify the authenticity and integrity of the model, as well as to check whether it meets the requirements and capabilities of the UE.

[0193] (2) A UE that has already received an AI / ML model from the network or generated its own model may share its model with other UEs in the same group via device-to-device (D2D) communication. The sharing UE may broadcast or multicast the model to the group members using a common control channel or a shared data channel. The sharing UE may also attach the model identification information to the model, such as the source, version, size, format, and pairing information of the model. The receiving UEs may use the model identification information to decide whether to accept or reject the model, as well as to pair the model with other models if needed. Only send differences with current model?

[0194] (3) A network function in the core network may send an AI / ML model to multiple UEs via a multicast or broadcast service. The NF may use a multicast or broadcast channel that covers the area where the UEs are located. The NF may also encode the model identification information in a service announcement message or a service descriptor message that informs the UEs about the availability and characteristics of the service. The UEs may use the service announcement or descriptor message to tune to the appropriate channel and receive the model, as well as to extract the model identification information from the model.

[0195] Furthermore, this invention can be applied to various types of UEs or terminal devices, such as mobile phone, vital signs monitoring / telemetry devices, smartwatches, detectors, vehicles (for vehicle-to-vehicle (V2V) communication or more general vehicle-to-everything (V2X) communication), V2X devices, Internet of Things (loT) hubs, loT devices, including low-power medical sensors for health monitoring, medical (emergency) diagnosis and treatment devices, for hospital use or first-responder use, virtual reality (VR) headsets, etc.

[0196] Other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed invention, from a study of the drawings, the disclosure and the appended claims. In the claims, the word "comprising" does not exclude other elements or steps, and the indefinite article "a" or "an" does not exclude a plurality. A single processor or other unit may fulfil the functions of several items recited in the claims. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage. The foregoing description details certain embodiments of the invention. It will be appreciated, however, that no matter how detailed the foregoing appears in the text, the invention may be practiced in many ways, and is therefore not limited to the embodiments disclosed. It should be noted that the use of particular terminology when describing certain features or aspects of the invention should not be taken to imply that the terminology is being re-defined herein to be restricted to include any specific characteristics of the features or aspects of the invention with which that terminology is associated. Additionally, the expression "at least one of A, B, and C" is to be understood as disjunctive, i.e., as "A and / or B and / or C".A single unit or device may fulfil the functions of several items recited in the claims. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage.

[0197] The described operations like those indicated in the above embodiments may be implemented as program code means of a computer program and / or as dedicated hardware of the related network device or function, respectively. The computer program may be stored and / or distributed on a suitable medium, such as an optical storage medium or a solid-state medium, sup-plied together with or as part of other hardware, but may also be distributed in other forms, such as via the Internet or other wired or wireless telecommunication systems.

Claims

1. A method for operating a first communication device in a wireless network, comprising: the first communication device communicating with or through at least a second communication device and / or the first communication device communicating with one or more other third communication devices in a device-to-device communication mode, the method further comprising receiving an AI / ML configuration from or relayed through the second communication device for executing at least one AI / ML function, and wherein the AI / ML function is adapted to assist(a) the device-to-device communication mode and / or(b) an AI / ML operation in roaming mode.

2. The method of claim 1, comprising, before changing a serving network and / or radio access technology or upon detecting a change of a serving network and / or radio access technology, the first communication device performing one or more of the following: a. checking a configuration / policy determining whether an AI / ML function may be used; b. sending a request to receive a configuration containing criteria regarding the behavior of the first communication device regarding its AI / ML model and / or AI / ML functionality when / before changing the network and / or radio access technology; c. receiving a configuration determining conditions allowing and / or forbidding the reporting of measurements and data for training purposes of an AI / ML model wherein the conditions comprise at least the first communication device camping in a given network and / or the first communication device using a given radio access technology; d. sending a request to retrieve an AI / ML information; e. sending a request to update the AI / ML function; or f. sending measurements for training of an AI / ML model.

3. The method of claim 1 or 2, wherein the configuration provides at least one AI / ML function, to assist the device-to-device communication, and wherein the configuration is relayed through the second device from a fourth communication device.

4. The method of claim 3, wherein the fourth communication device is one of a base station, a core network function or an application function.

5. The method of any of the previous claims, wherein the AI / ML function includes at least one operation in the lifecycle of AI / ML functionality for one or more of the AI / ML model provision, configuration, training, transfer, update, inference, management and / or monitoring.

6. The method of any of the previous claims, wherein the AI / ML function's aim is optimizing one or more of the following procedures in device-to-device communication scenarios: a. device-based positioning and / or ranging measurements and / or estimations; b. relay (re-)selection in relay scenarios; c. path (re-)selection in multi-path scenarios; d. resource selection; and / or e. wireless sensing procedures.

7. The method of any of the previous claims, comprising the first communication device using the AI / ML function relying on an AI / ML model taking as input one or more first parameters: a. model identifier; b. signal strength; c. link quality; d. frequency band; and / or e. speed of the first and / or third communication devices.

8. The method of claim 7, wherein one or more of the first parameters are explicitly exchanged between the first and the third communication devices and / or implicitly indicated to the first and the third communication devices, wherein the implicit indication is done by means of one or more of: relay service code; service code; and / orlayer 2 address.

9. The method of any of the previous claims, comprising sending a request for the AI / ML function to and / or through the second communication device.

10. The method of any of the previous claims, comprising forwarding a request for the AI / ML function to and / or through the second communication device, wherein the forwarded request is forwarded from a third communication device.

11. The method of claim 9 or 10, wherein the request includes one or more of: a. one or more requested AI / ML functions; b. the capabilities of the first and / or third communication devices; and / or c. request to provision and / or configure at least one AI / ML function.

12. The method of any of the previous claims, comprising the first communication device monitoring by means of a monitoring function to determine whether a) the AI / ML function for device-to-device communication fulfils performance requirements and / or b) whether the current AI / ML parameters are valid in the current serving network.

13. A method for operating a first communication device capable of AI / ML RAN communication, wherein the method is adapted to receiving at least one AI / ML function from or through a second communication device, and wherein the AI / ML function is adapted to assisting the first communication device in device-to-device communication and / or AI / ML operation in roaming.

14. A computer program product comprising instructions which, when executed, cause a communication device to implement the method of any of the previous claims.

15. A first communication device, comprising: a transceiver configured to communicate with or through at least a second communication device and / orcommunicate with one or more other third communication devices in a device-to-device communication mode, wherein the receiver is adapted to receive an AI / ML configuration from or relayed through the second communication device the communication device comprising executing at least one AI / ML function, and wherein the AI / ML function is adapted to assist(a) the device-to-device communication mode and / or(b) an AI / ML operation in roaming mode.

16. A communication device in a wireless network, comprising: a transceiver for communications with one or more access devices, (such as base stations, repeaters, network-controlled repeaters, IAB node, relay, or NTN payload), a sidelink transceiver for the sidelink communications with one or more other communication devices, wherein the communication device is configured for at least one AI / ML function by the access device, and the AI / ML function being adapted to assist the sidelink communication, and wherein the AI / ML function includes at least one operation in the lifecycle of AI / ML functionality for one or more of the AI / ML model provision, configuration, training, transfer, update, inference, management and / or monitoring.

17. A second communication device , comprising: a transceiver for communications with one or more first communication devices, a controller configured to provision and configure at least one AI / ML function at the communication device, and the AI / ML function is adapted to assist a sidelink communication, and wherein the AI / ML function includes at least one operation in a lifecycle of AI / ML functionality for one or more of AI / ML model provision, configuration, training, transfer, update, inference, management and / or monitoring.

18. A wireless communication system, comprising: one or more base stations, a plurality of user devices, UEs, configured for sidelink communication,wherein the plurality of UEs comprises at least one AI / ML function, and at least one function for sidelink communication, the AI / ML function is configured to assist the sidelink communication, wherein the base station is configured to connect to the core network, and provide at least one AI / ML function, to assist the sidelink communication, and wherein the AI / ML function consists of at least one operation in the lifecycle of AI / ML functionality for AI / ML model provision, configuration, training, transfer, update, inference, management and monitoring.

19. A second communication device in a wireless network comprising: one transceiver for communications with one or more first communication device, wherein the second communication device is configured to a) provision and / or forward a configuration of at least one AI / ML function to at the first communication device, and the AI / ML function is adapted to assist the device-to-device communication and / or b) assist AI / ML operation in roaming.

20. A wireless communication system, comprising: at least one second communication device, at least one first communication device, wherein the first communication device comprises at least one AI / ML function to assist AI / ML operation in roaming and / or the second communication device is configured to assist AI / ML operation in roaming.

21. A wireless communication system, comprising: at least one second communication device, at least one first communication device, at least one third communication device, wherein the first and third communication devices comprise at least one AI / ML function, at least one communication unit adapted for device-to-device communication, the AI / ML function configured to assist the device-to-device communication.

Images