Method for supporting dynamic model selection for wireless communication

Abstract

A wireless transmit / receive unit (WTRU) is disclosed that includes a processor configured to store one or more AI / ML model configurations. Each AI / ML model configuration may have an associated identifier. The WTRU may include a transmitter configured to transmit a message to the gNB indicating the AI / ML model reconfiguration. The AI / ML model reconfiguration may be based on NACK. The AI / ML model reconfiguration may be based on a trigger condition. The WTRU may transmit a message indicating the preferred AI / ML model. The preferred AI / ML model may be based on one or more of a configured threshold, an RDRP, a speed of the WTRU, a path loss, an area ID, a cell ID, or a TRP ID. The WTRU may evaluate one or more AI / ML model configurations. The WTRU may receive an acknowledgement message of AI / ML model reconfiguration from the gNB.

Description

[0001] Related applications This application claims the benefit and priority of U.S. Provisional Patent Application No. 63 / 445,452, filed February 14, 2023, entitled “Method for Supporting Dynamic Model Selection for Wireless Communication,” the entire contents of which are incorporated herein by reference. Background Technology

[0002] New implementations of wireless communication systems and protocols can deliver high-throughput, low-latency data exchange, but may require complex configurations to realize these benefits. For example, some implementations of communication systems use spatial filtering or beamforming to direct wireless signals to the target receiving device in order to efficiently increase signal strength without wasting power and / or to reduce interference or support additional user equipment (UE) or other devices. Establishing such guidance may require complex calculations, especially for mobile devices moving relative to each other, resulting in rapid changes in the transmission angle (and corresponding phase and amplitude adjustments for multi-antenna arrays). Similarly, communication channel configurations (e.g., timing, frequency, etc.) may need to be modified to avoid interference, meet application requirements, etc. Summary of the Invention

[0003] To deliver the benefits of next-generation wireless communication systems and to predict changing parameters and configurations so that they can be applied before congestion, interference, or other such conditions occur, the systems and methods discussed in this paper are implemented using artificial intelligence / machine learning (AI / ML) systems. These AI / ML systems can be based on trained models that enable the prediction of changing environmental conditions and the adjustment of beamforming parameters or channel state information, adjustments that can ideally be applied before communication equipment experiences any adverse effects. However, different AI / ML models can provide better performance depending on changing conditions. For example, when interference from a local airport radar system is present, an AI / ML model optimized for predicting beam management adjustments as equipment or other entities move within the area may not necessarily make accurate predictions.

[0004] Therefore, in some aspects, this disclosure relates to implementations of systems and methods for dynamically identifying or selecting AI / ML models or model types to provide optimized predictions based on changing environmental conditions, device movement, or combinations of these or other changes. In some implementations, the training model may be generated (and / or updated) by a UE, base station (gNB), or other network device (e.g., application server, AI / ML processing node, etc.) and may be provided to devices (e.g., UE and gNB or other devices) for performance, condition, or characteristic predictions and parameter adjustments. In some implementations, these predictions may enable devices to apply parameter adjustments, such as immediately changing retransmission window parameters in response to changes in packet loss rate, prior to requests or notifications from cooperating devices, without first exchanging configuration change details or other such modifications. In some implementations, devices (e.g., UE, gNB, or other devices) may monitor network conditions or performance and may dynamically determine whether to change the AI / ML model, and if so, which model to use. This can significantly reduce management overhead and enable faster mitigation or recovery from changing channel conditions or problems.

[0005] In some embodiments of the configuration for receiving AI / ML models with AI / ML model types, the UE is configured with one or more AI / ML models having model IDs, wherein each AI / ML model is associated with / configured with a set of CSI parameters (potentially for each CSI type or, for example, a CSI configuration for CSI and for BM). Alternatively, the UE indicates its capabilities (e.g., the CSI parameters required for each AI / ML model).

[0006] In some embodiments, the UE indicates a need for AI / ML model reselection. This is indicated if one or more of the following occur: the number of NACKs (e.g., number of consecutive NACKs > a threshold), RSRP (e.g., RSRP < a threshold), changes in UE speed (e.g., current UE speed - average UE speed within the measurement window > a threshold), changes in UE location (e.g., current UE location - average UE location within the measurement window > a threshold), changes in path loss (e.g., current path loss - average path loss within the measurement window > a threshold), assumed PDCCH BLER < a threshold, measured SINR, or differential SINR (e.g., from a scheduled DMRS port).

[0007] In some embodiments, the UE indicates the type of AI / ML model for reselection based on triggered conditions. If the measured RSRP > threshold and the number of consecutive NACKs > threshold, reselect the AI / ML model for CSI. If the measured RSRP < threshold, reselect the AI / ML model for CSI. If the number of failures of the measured hypothesized PDCCH BLER < threshold, reselect the AI / ML model for CSI. Otherwise, reselect the AI / ML for BM.

[0008] In some embodiments, the UE indicates the type of AI / ML model for reselection. The UE indicates the type of AI / ML model for reselection to an A gNB and receives one or more RSs associated with the reported type. If the AI / ML model for BM is requested, after reselecting the AI / ML model for BM, the AI / ML model for CSI may be reselected.

[0009] In some embodiments, the UE identifies a preferred model. The UE identifies a preferred model based on one or more of a threshold of a measurement configuration, RSRP, UE speed, path loss, area ID, cell ID, TRP ID, and the preferred model may have a model type indication (or CSI configuration ID). For example, if the UE speed < X and the path loss > Y (e.g., indoor), determine model #1. If the UE speed < X and the path loss < Y (e.g., walking outdoors), determine model #2. If the UE speed > X and the path loss < Y (e.g., high-speed outdoors), determine model #3.

[0010] In some embodiments, the UE indicates a preferred set of CSI parameters to the gNB for future reporting. For example, model #1: X beam IDs for current measurement and corresponding L1-RSRP (N1). Model #2: X1-1 beam IDs for current measurement and corresponding L1-RSRP, and X1-2 beam IDs for N1 + N2 and corresponding L1-RSRP. Model #3: X2-1, X2-2, X2-3, and X2-4 beam IDs for N1 / N2 / N3 / N4 and corresponding L1-RSRP.

[0011] In some embodiments, the UE evaluates the associated AI / ML model. Alternatively, the UE activates a set of associated AI / ML models in one or more AI / ML models having the currently identified AI / ML model / model type for reselection, thereby evaluating the prediction accuracy of the AI / ML model. Details of the evaluation, e.g., beam prediction accuracy / RSRP difference and RS transmission for evaluation, different RS transmissions for activating the model.

[0012] In some embodiments, the gNB confirms the AI / ML model type used for reselection. The UE receives confirmation of the UE request and activates the indicated model for the reported request type.

[0013] In some embodiments, the UE behavior changes after receiving gNB confirmation for reselecting the AI / ML model type. For example, the UE may deactivate a previously selected or activated model for evaluation, activate a newly indicated model, and apply the associated set of CSI parameters to the next CSI report, including a preferred size for the measurement / prediction window. Attached Figure Description

[0014] The invention can be understood in more detail from the following description given by way of example in conjunction with the accompanying drawings, wherein like reference numerals denote like elements, and wherein: Figure 1A This is a system diagram illustrating an example communication system that can implement one or more of the disclosed embodiments; Figure 1B This illustrates the possibility of implementation according to an embodiment. Figure 1A A system diagram of an exemplary wireless transmit / receive unit (WTRU) used in the communication system shown; Figure 1C This illustrates that, according to an embodiment, it is possible to Figure 1A The system diagram shows an example radio access network (RAN) and an example core network (CN) used within the communication system shown. Figure 1D This illustrates that, according to an embodiment, it is possible to Figure 1A The system diagram shows another example RAN and another example CN used within the communication system shown; Figure 2 This is a block diagram of a system for hybrid beamforming according to some implementation methods; Figure 3 It is a block diagram of a system for dynamic model selection for wireless communication according to some implementation methods; Figure 4 This is a flowchart of a method for dynamic model selection for wireless communication according to some implementation methods; and Figure 5 This is a logic diagram of an example implementation of dynamic model selection for wireless communication. Detailed Implementation

[0015] Figure 1AThis diagram illustrates an example communication system 100 in which one or more of the disclosed embodiments may be implemented. The communication system 100 may be a multiple access system providing content such as voice, data, video, messaging, and broadcasting to multiple wireless users. The communication system 100 enables multiple wireless users to access such content by sharing system resources, including wireless bandwidth. For example, the communication system 100 may employ one or more channel access methods, such as Code Division Multiple Access (CDMA), Time Division Multiple Access (TDMA), Frequency Division Multiple Access (FDMA), Orthogonal FDMA (OFDMA), Single Carrier FDMA (SC-FDMA), Zero-Tail Unique Word Discrete Fourier Transform Extended OFDM (ZT-UW-DFT-S-OFDM), Unique Word OFDM (UW-OFDM), Resource Block Filtered OFDM, Filter Bank Multicarrier (FBMC), etc.

[0016] like Figure 1A As shown, the communication system 100 may include wireless transmit / receive units (WTRUs) 102a, 102b, 102c, 102d, a radio access network (RAN) 104, a core network (CN) 106, a public switched telephone network (PSTN) 108, the Internet 110, and other networks 112. However, it should be understood that the disclosed embodiments contemplate any number of WTRUs, base stations, networks, and / or network elements. Each of the WTRUs 102a, 102b, 102c, and 102d may be any type of device configured to operate and / or communicate in a wireless environment. For example, WTRUs 102a, 102b, 102c, and 102d (any of which may be referred to as a station (STA)) may be configured to transmit and / or receive wireless signals and may include user equipment (UE), mobile stations, fixed or mobile subscriber units, subscription-based units, pagers, cellular phones, personal digital assistants (PDAs), smartphones, laptops, netbooks, personal computers, wireless sensors, hotspots or Mi-Fi devices, Internet of Things (IoT) devices, watches or other wearable devices, head-mounted displays (HMDs), vehicles, drones, medical devices and applications (e.g., remote surgery), industrial devices and applications (e.g., robots and / or other wireless devices operating in the context of industrial and / or automated processing chains), consumer electronics devices, devices operating on commercial and / or industrial wireless networks, and so on. Any of WTRUs 102a, 102b, 102c, and 102d may be interchangeably referred to as a UE.

[0017] The communication system 100 may also include base station 114a and / or base station 114b. Each of base stations 114a and 114b may be any type of device configured to wirelessly interface with at least one of WTRUs 102a, 102b, 102c, and 102d, for example, to facilitate access to one or more communication networks such as CN 106 / 115, Internet 110, and / or Network 112. As an example, base stations 114a and 114b may be base transceiver stations (BTS), Node-Bs, eNode Bs (eNBs), home Node-Bs, home eNode Bs, next-generation Node-Bs such as gNode Bs (gNBs), new radio (NR) Node-Bs, site controllers, access points (APs), wireless routers, etc. Although base stations 114a and 114b are each depicted as a single element, it will be understood that base stations 114a and 114b may include any number of interconnected base stations and / or network elements.

[0018] Base station 114a may be part of RAN 104, which may also include other base stations and / or network elements (not shown), such as base station controllers (BSCs), radio network controllers (RNCs), relay nodes, etc. Base station 114a and / or base station 114b may be configured to transmit and / or receive radio signals on one or more carrier frequencies, which may be referred to as cells (not shown). These frequencies may be licensed spectrum, unlicensed spectrum, or a combination of licensed and unlicensed spectrum. A cell may provide coverage of a specific geographic area, which may be relatively fixed or may change over time. A cell may be further divided into cell sectors. For example, the cell associated with base station 114a may be divided into three sectors. Thus, in one embodiment, base station 114a may include three transceivers, i.e., one transceiver per sector of the cell. In embodiments, base station 114a may employ multiple-input multiple-output (MIMO) technology and may use multiple transceivers for each sector of the cell. For example, beamforming may be used to transmit and / or receive signals in a desired spatial direction.

[0019] Base stations 114a and 114b can communicate with one or more of WTRUs 102a, 102b, 102c, and 102d via air interface 116, which can be any suitable wireless communication link (e.g., radio frequency (RF), microwave, centimeter wave, micrometer wave, infrared (IR), ultraviolet (UV), visible light, etc.). Air interface 116 can be established using any suitable radio access technology (RAT).

[0020] More specifically, as described above, the communication system 100 can be a multiple access system and can employ one or more channel access schemes, such as CDMA, TDMA, FDMA, OFDMA, SC-FDMA, etc. For example, base stations 114a and WTRUs 102a, 102b, and 102c in RAN 104 can implement radio technologies such as Universal Mobile Telecommunications System (UMTS) Terrestrial Radio Access (UTRA), which can use Wideband CDMA (WCDMA) to establish the air interface 116. WCDMA can include communication protocols such as High-Speed ​​Packet Access (HSPA) and / or Evolved HSPA (HSPA+). HSPA can include High-Speed ​​Downlink (DL) Packet Access (HSDPA) and / or High-Speed ​​Uplink (UL) Packet Access (HSUPA).

[0021] In one embodiment, base station 114A and WTRUs 102A, 102b, 102c may implement radio technologies such as Evolved UMTS Terrestrial Radio Access (E-UTRA), which may use Long Term Evolution (LTE) and / or LTE-A Advanced (LTE-A) and / or LTE-A Pro Advanced (LTE-A Pro) to establish air interface 116.

[0022] In one embodiment, base station 114a and WTRUs 102a, 102b, 102c may implement radio technologies such as NR wireless access, which can use NR to establish air interface 116.

[0023] In one embodiment, base station 114a and WTRUs 102a, 102b, and 102c can implement multiple radio access technologies. For example, base station 114a and WTRUs 102a, 102b, and 102c can implement both LTE and NR radio access together, for example, using a dual connectivity (DC) principle. Therefore, the air interface used by WTRUs 102a, 102b, and 102c can be characterized by multiple types of radio access technologies and / or transmissions sent to / from multiple types of base stations (e.g., eNBs and gNBs).

[0024] In other embodiments, base station 114a and wireless transmission / reception units 102a, 102b, and 102c may implement wireless technologies such as IEEE 802.11 (i.e., Wi-Fi), IEEE 802.16 (i.e., WiMAX), CDMA 2000, CDMA 2000 1X, CDMA 2000 EV-DO, Internet Standard 2000 (IS-2000), Internet Standard 95 (IS-95), Internet Standard 856 (IS-856), Global System for Mobile Communications (GSM), Enhanced Data Rate GSM Evolution (EDGE), GSM EDGE (GERAN), etc.

[0025] Figure 1A Base station 114B can be, for example, a wireless router, a home Node-B, a home eNode-B, or an access point, and can utilize any suitable RAT to facilitate wireless connectivity in a local area, such as a business premises, home, vehicle, campus, industrial facility, air corridor (e.g., for drone use), road, etc. In one embodiment, base station 114b and WTRUs 102c, 102d can implement radio technologies such as IEEE 802.11 to establish a wireless local area network (WLAN). In one embodiment, base station 114b and WTRUs 102c, 102d can implement radio technologies such as IEEE 802.15 to establish a wireless personal area network (WPAN). In yet another embodiment, base station 114b and WTRUs 102c, 102d can utilize cellular-based RATs (e.g., WCDMA, CDMA 2000, GSM, LTE-A Pro, NR, etc.) to establish picocells or femtocells. Figure 1A As shown, base station 114b can have a direct connection to Internet 110. Therefore, base station 114b does not need to access Internet 110 via CN 106.

[0026] RAN 104 can communicate with CN 106, which can be any type of network configured to provide voice, data, application, and / or Voice over Internet Protocol (VoIP) services to one or more WTRUs among WTRUs 102a, 102b, 102c, and 102d. Data may have varying Quality of Service (QoS) requirements, such as different throughput requirements, latency requirements, fault tolerance requirements, reliability requirements, data throughput requirements, mobility requirements, etc. CN 106 can provide call control, billing services, location-based services, prepaid calling, internet connectivity, video distribution, etc., and / or perform advanced security functions such as user authentication. Although in Figure 1AAlthough not shown, it should be understood that RAN 104 and / or CN 106 can communicate directly or indirectly with other RANs using the same RAT as RAN 104 or a different RAT. For example, in addition to connecting to RAN 104, which can utilize NR radio technology, CN 106 can also communicate with another RAN (not shown) using GSM, UMTS, CDMA 2000, WiMAX, E-UTRA, or WiFi radio technology.

[0027] CN 106 may also serve as a gateway for WTRUs 102a, 102b, 102c, and 102d to access PSTN 108, the Internet 110, and / or other networks 112. PSTN 108 may include a circuit-switched telephone network providing Common Old-Style Telephone Service (POTS). The Internet 110 may include a global system of interconnected computer networks and devices using common communication protocols, such as Transmission Control Protocol (TCP), User Datagram Protocol (UDP), and / or Internet Protocol (IP) from the TCP / IP Internet Protocol suite. Network 112 may include wired and / or wireless communication networks owned and / or operated by other service providers. For example, network 112 may include another CN connected to one or more RANs, which may use the same RAT as RAN 104 or a different RAT.

[0028] Some or all of the wireless transmission and reception units 102a, 102b, 102c, and 102d in the communication system 100 may include multi-mode capabilities (e.g., the wireless transmission and reception units 102a, 102b, 102c, and 102d may include multiple transceivers to communicate with different wireless networks via different wireless links). For example... Figure 1A The WTRU 102c shown can be configured to communicate with base station 114A, which can use cellular-based radio technology, and with base station 114b, which can use IEEE 802 radio technology.

[0029] Figure 1B This is a system diagram illustrating example WTRU 102. (See diagram below.) Figure 1B As shown, WTRU 102 may include a processor 118, a transceiver 120, a transmit / receive element 122, a speaker / microphone 124, a keyboard 126, a display / touchpad 128, non-removable memory 130, removable memory 132, a power supply 134, a Global Positioning System (GPS) chipset 136, and / or other peripheral devices 138, etc. It is understood that WTRU 102 may include any sub-combination of the foregoing elements while remaining consistent with the embodiments.

[0030] Processor 118 can be a general-purpose processor, a special-purpose processor, a conventional processor, a digital signal processor (DSP), multiple microprocessors, one or more microprocessors associated with a DSP core, a controller, a microcontroller, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), any other type of integrated circuit (IC), a state machine, etc. Processor 118 can perform signal encoding, data processing, power control, input / output processing, and / or any other function that enables WTRU 102 to operate in a wireless environment. Processor 118 can be coupled to transceiver 120, which can be coupled to transmitting / receiving element 122. Although Figure 1B While the processor 118 and transceiver 120 are depicted as separate components, it will be understood that the processor 118 and transceiver 120 may be integrated together in an electronic package or chip.

[0031] Transmitting / receiving element 122 can be configured to transmit signals to or receive signals from a base station (e.g., base station 114a) via air interface 116. For example, in one embodiment, transmitting / receiving element 122 can be an antenna configured to transmit and / or receive RF signals. In one embodiment, transmitting / receiving element 122 can be a transmitter / detector configured to transmit and / or receive, for example, IR, UV, or visible light signals. In yet another embodiment, transmitting / receiving element 122 can be configured to transmit and / or receive both RF and optical signals. It should be understood that transmitting / receiving element 122 can be configured to transmit and / or receive any combination of wireless signals.

[0032] Although the transmitting / receiving element 122 is in Figure 1B While described as a single element, WTRU 102 may include any number of transmitting / receiving elements 122. More specifically, WTRU 102 may use MIMO technology. Therefore, in one embodiment, WTRU 102 may include two or more transmitting / receiving elements 122 (e.g., multiple antennas) for transmitting and receiving wireless signals via air interface 116.

[0033] Transceiver 120 can be configured to modulate signals transmitted by transmitting / receiving element 122 and demodulate signals received by transmitting / receiving element 122. As described above, WTRU 102 can have multi-mode capability. Therefore, transceiver 120 can include multiple transceivers for enabling WTRU 102 to communicate via multiple RATs (e.g., NR and IEEE 802.11).

[0034] The processor 118 of WTRU 102 can be coupled to a speaker / microphone 124, a keyboard 126, and / or a display / touchpad 128 (e.g., a liquid crystal display (LCD) unit or an organic light-emitting diode (OLED) display unit) and can receive user input data from there. The processor 118 can also output user data to the speaker / microphone 124, keyboard 126, and / or display / touchpad 128. Additionally, the processor 118 can access and store information from any suitable type of memory, such as non-removable memory 130 and / or removable memory 132. Non-removable memory 130 may include random access memory (RAM), read-only memory (ROM), a hard disk, or any other type of memory storage device. Removable memory 132 may include a user identification module (SIM) card, memory stick, secure digital storage (SD) card, etc. In other embodiments, the processor 118 can access information from memory and store data in memory that is not physically located on WTRU 102, for example, on a server or home computer (not shown).

[0035] The processor 118 may receive power from the power supply 134 and may be configured to distribute power to other components in the WTRU 102 and / or control power to those other components. The power supply 134 may be any suitable device for powering the WTRU 102. For example, the power supply 134 may include one or more dry cell batteries (e.g., nickel-cadmium (NiCd), nickel-zinc (NiZn), nickel metal hydride (NiMH), lithium-ion (Li-ion), etc.), solar cells, fuel cells, etc.

[0036] The processor 118 may also be coupled to the GPS chipset 136, which may be configured to provide location information (e.g., longitude and latitude) about the current location of the WTRU 102. In addition to, or alternatively to, the information from the GPS chipset 136, the WTRU 102 may receive location information from base stations (e.g., base stations 114a, 114b) via the air interface 116, and / or determine its location based on the timing of signals received from two or more neighboring base stations. It should be understood that the WTRU 102 may acquire location information using any suitable location determination method while remaining consistent with the embodiments.

[0037] The processor 118 can also be connected to other peripheral devices 138, which may include one or more software and / or hardware modules that provide additional features, functions, and / or wired or wireless connectivity. For example, peripheral devices 138 may include accelerometers, electronic compasses, satellite transceivers, digital cameras (for photos and / or video), Universal Serial Bus (USB) ports, vibration devices, television transceivers, hands-free headsets, Bluetooth® modules, FM radio units, digital music players, media players, video game player modules, internet browsers, virtual reality and / or augmented reality (VR / AR) devices, activity trackers, etc. Peripheral devices 138 may include one or more sensors. Sensors may be one or more of the following: gyroscopes, accelerometers, Hall effect sensors, magnetometers, orientation sensors, proximity sensors, temperature sensors, time sensors; geolocation sensors, altimeters, light sensors, touch sensors, magnetometers, barometers, gesture sensors, biometric sensors, humidity sensors, etc.

[0038] WTRU 102 may include a full-duplex radio, for which the transmission and reception of some or all signals (e.g., signals associated with specific subframes for UL (e.g., for transmission) and DL (e.g., for reception)) may be concurrent and / or simultaneous. The full-duplex radio may include an interference management unit to reduce and / or substantially eliminate self-interference via hardware (e.g., chokes) or signal processing via a processor (e.g., a separate processor (not shown) or via processor 118). In embodiments, WTRU 102 may include a half-duplex radio, wherein the transmission and reception of some or all of the signals (e.g., signals associated with specific subframes of both the uplink (e.g., for transmission) and downlink (e.g., for reception)) may occur.

[0039] Figure 1C This is a system diagram illustrating RAN 104 and CN 106 according to an embodiment. As described above, RAN 104 may employ E-UTRA radio technology to communicate with WTRUs 102a, 102b, and 102c via air interface 116. RAN 104 may also communicate with CN 106.

[0040] RAN 104 may include eNode-Bs 160a, 160b, and 160c, but it should be understood that RAN 104 may include any number of eNode-Bs while remaining consistent with the embodiments. Each eNode-B 160a, 160b, and 160c may each include one or more transceivers to communicate with radio transmit / receive units 102a, 102b, and 102c via air interface 116. In one embodiment, eNode-Bs 160a, 160b, and 160c may implement MIMO technology. Therefore, for example, eNode-B 160a may use multiple antennas to transmit radio signals to and / or receive radio signals from WTRU 102a.

[0041] Each of the eNode-B 160a, 160b, and 160c can be associated with a specific cell (not shown) and can be configured to handle radio resource management decisions, handover decisions, user scheduling in the UL and / or DL, etc. Figure 1C As shown, eNode-B 160a, 160b, and 160C can communicate with each other via the X2 interface.

[0042] Figure 1C The CN 106 shown may include a Mobility Management Entity (MME) 162, a Serving Gateway (SGW) 164, and a Packet Data Network (PDN) Gateway (PGW) 166. Although the foregoing elements are depicted as part of CN 106, it will be understood that any of these elements may be owned and / or operated by an entity other than the CN operator.

[0043] The MME 162 can connect to each eNode-B162a, 162b, 162c in RAN 104 via the S1 interface and can be used as a control node. For example, the MME 162 can be responsible for authenticating users of WTRUs 102a, 102b, 102c, bearer activation / deactivation, selecting a specific serving gateway during the initial attachment of WTRUs 102a, 102b, 102c, etc. The MME 162 can provide control plane functions for handover between RAN 104 and other RANs (not shown) employing other radio technologies (such as GSM and / or WCDMA).

[0044] The SGW 164 can connect to each of the eNode-Bs 160a, 160b, and 160c in RAN 104 via the S1 interface. The SGW 164 can typically route and forward user data packets to / from WTRUs 102a, 102b, and 102c. The SGW 164 can perform other functions, such as anchoring the user plane during inter-eNode B handover, triggering paging when DL data is available for WTRUs 102a, 102B, and 102c, managing and storing the context of WTRUs 102a, 102B, and 102c, etc.

[0045] The SGW 164 can connect to the PGW 166, which can provide WTRU 102a, 102b, 102c with access to packet-switched networks such as the Internet 110 to facilitate communication between WTRU 102a, 102b, 102c and IP-enabled devices.

[0046] CN 106 can facilitate communication with other networks. For example, CN 106 can provide WTRU 102a, 102b, 102c with access to circuit-switched networks, such as PSTN 108, to facilitate communication between WTRU 102a, 102b, 102c and traditional landline communication equipment. For example, CN 106 may include an IP gateway (e.g., an IP Multimedia Subsystem (IMS) server), or can communicate with an IP gateway that serves as an interface between CN 106 and PSTN 108. Furthermore, CN 106 can provide WTRU 102a, 102b, 102c with access to other networks 112, which may include other wired and / or wireless networks owned and / or operated by other service providers.

[0047] Although WTRU is Figure 1A-1D While described as a wireless terminal, it is anticipated that in some representative embodiments, such a terminal may use (e.g., temporarily or permanently) a wired communication interface with a communication network.

[0048] In a representative embodiment, another network 112 may be a WLAN.

[0049] In an Infrastructure Basic Services Set (BSS) mode, a WLAN may have an Access Point (AP) for the BSS and one or more Stations (STAs) associated with the AP. The AP may have access to or an interface to a Distribution System (DS) or another type of wired / wireless network that loads traffic into and / or out of the BSS. Traffic originating from a STA outside the BSS can reach and be delivered to the AP. Traffic originating from a STA to a destination outside the BSS can be sent to the AP for delivery to the appropriate destination. Traffic between STAs within the BSS can be sent via the AP, for example, where a source STA can send traffic to the AP, and the AP can deliver traffic to a destination STA. Traffic between STAs within the BSS can be considered and / or referred to as peer-to-peer traffic. Peer-to-peer traffic can be sent between source and destination STAs (e.g., directly between source and destination STAs) using Direct Link Establishment (DLS). In some representative embodiments, the DLS may use 802.11e DLS or 802.11z Tunneled DLS (TDLS). A WLAN using the Standalone BSS (IBSS) mode may not have an access point (AP), and STAs within the IBSS or using the IBSS (e.g., all STAs) can communicate directly with each other. The IBSS communication mode may sometimes be referred to here as a "self-organizing ad-hoc" communication mode.

[0050] When operating in 802.11ac infrastructure mode or a similar mode, the AP can transmit beacons on a fixed channel (e.g., the primary channel). The primary channel can be of fixed width (e.g., a 20 MHz bandwidth) or dynamically configured. The primary channel can be the operating channel of the BSS and can be used by the STA to establish a connection with the AP. In some representative embodiments, such as in an 802.11 system, Carrier Sense Multiple Access - Collision Avoidance (CSMA / CA) can be implemented. For CSMA / CA, STAs including the AP (e.g., each STA) can sense the primary channel. If the primary channel is sensed / detected and / or determined to be busy by a particular STA, that particular STA can back off. A single STA (e.g., only one station) can transmit at any given time within a given BSS.

[0051] High-throughput (HT) STAs can communicate using a 40 MHz wide channel, for example, by combining a primary 20 MHz channel with adjacent or non-adjacent 20 MHz channels to form a 40 MHz wide channel.

[0052] Very High Throughput (VHT) STAs can support channels with widths of 20 MHz, 40 MHz, 80 MHz, and / or 160 MHz. 40 MHz and / or 80 MHz channels can be formed by combining adjacent 20 MHz channels. A 160 MHz channel can be formed by combining eight consecutive 20 MHz channels or by combining two non-consecutive 80 MHz channels; this is referred to as an 80+80 configuration. For the 80+80 configuration, after channel coding, the data can pass through a segmented parser that divides the data into two streams. Each stream can be processed separately using Inverse Fast Fourier Transform (IFFT) and time-domain processing. The streams can be mapped onto the two 80 MHz channels, and the data can be transmitted by the transmitting STA. At the receiver of the receiving STA, the operation of the 80+80 configuration can be reversed, and the combined data can be sent to the Media Access Control (MAC).

[0053] Operating modes below 1 GHz are supported by 802.11af and 802.11ah. The channel operating bandwidth and carrier are reduced in 802.11af and 802.11ah compared to those used in 802.11n and 802.11ac. 802.11af supports 5 MHz, 10 MHz, and 20 MHz bandwidths in the TV whitespace (TVWS) spectrum, while 802.11ah supports 1 MHz, 2 MHz, 4 MHz, 8 MHz, and 16 MHz bandwidths using non-TVWS spectrum. According to a representative embodiment, 802.11ah may support meter-type control / machine-type communication (MTC), such as MTC devices in macro coverage areas. MTC devices may have certain capabilities, such as limited capabilities including support for certain and / or limited bandwidths (e.g., only support). MTC devices may include batteries with a battery life exceeding a threshold (e.g., to maintain a very long battery life).

[0054] WLAN systems that can support multiple channels and channel bandwidths (e.g., 802.11n, 802.11ac, 802.11af, and 802.11ah) include a channel that can be designated as the primary channel. The primary channel can have a bandwidth equal to the maximum common operating bandwidth supported by all STAs in the BSS. The bandwidth of the primary channel can be set and / or limited by the STAs operating in the BSS, supporting minimum bandwidth operating modes. In the 802.11ah example, for STAs supporting (e.g., only supporting) 1MHz mode (e.g., MTC type devices), the primary channel can be 1 MHz wide, even if the AP and other STAs in the BSS support 2MHz, 4 MHz, 8 MHz, 16 MHz, and / or other channel bandwidth operating modes. Carrier Sense and / or Network Allocation Vector (NAV) settings can depend on the status of the primary channel. If the primary channel is busy, for example, because an STA (which only supports 1 MHz operating mode) is sending to the AP, all available bands can be considered busy, even if most available bands remain idle.

[0055] In the United States, the available frequency band for 802.11ah is from 902 MHz to 928 MHz. In South Korea, the available frequency band is from 917.5 MHz to 923.5 MHz. In Japan, the available frequency band is from 916.5 MHz to 927.5 MHz. Depending on the country code, the total bandwidth available for 802.11ah is 6 MHz to 26 MHz.

[0056] Figure 1D This is a system diagram illustrating RAN 104 and CN 106 according to an embodiment. As described above, RAN 104 can communicate with WTRUs 102a, 102b, and 102c via air interface 116 using NR wireless technology. RAN 104 can also communicate with CN 106.

[0057] RAN 104 may include gNBs 180a, 180b, and 180c; however, it should be understood that RAN 104 may include any number of gNBs while remaining consistent with the embodiments. Each of gNBs 180a, 180b, and 180c includes one or more transceivers for communicating with WTRUs 102a, 102b, and 102c via air interface 116. In one embodiment, gNBs 180a, 180b, and 180c may implement MIMO technology. For example, gNBs 180a and 180b may utilize beamforming to transmit signals to and / or receive signals from gNBs 180a, 180b, and 180c. Therefore, gNB 180a may, for example, use multiple antennas to transmit radio signals to and / or receive radio signals from WTRU 102a. In one embodiment, gNBs 180a, 180b, and 180c may implement carrier aggregation technology. For example, gNB 180a can transmit multiple component carriers (not shown) to WTRU 102a. A subset of these component carriers may be on unlicensed spectrum, while the remaining component carriers may be on licensed spectrum. In one embodiment, gNBs 180a, 180b, and 180c may implement Cooperative Multipoint (CoMP) technology. For example, WTRU 102a may receive coordinated transmissions from gNBs 180a and 180b (and / or gNB 180c).

[0058] WTRU 102a, 102b, and 102c can communicate with gNB 180a, 180b, and 180c using transmissions associated with a scalable digital architecture. For example, the OFDM symbol spacing and / or OFDM subcarrier spacing can vary for different transmissions, different cells, and / or different portions of the radio transmission spectrum. WTRU 102a, 102b, and 102c can communicate with gNB 180a, 180b, and 180c using subframes or transmission time intervals (TTIs) of various lengths or scalable lengths (e.g., containing a varying number of OFDM symbols and / or an absolute time of continuously varying lengths).

[0059] gNBs 180a, 180b, and 180c can be configured to communicate with WTRUs 102a, 102b, and 102c in standalone and / or non-standalone configurations. In standalone configuration, WTRUs 102a, 102b, and 102c can communicate with gNBs 180a, 180b, and 180c without needing to access other RANs (e.g., eNode-Bs 160a, 160b, and 160c). In standalone configuration, WTRUs 102a, 102b, and 102c can utilize one or more of gNBs 180a, 180b, and 180c as mobility anchors. In standalone configuration, WTRUs 102a, 102b, and 102c can communicate with gNBs 180a, 180b, and 180c using signals in unlicensed frequency bands. In a non-standalone configuration, WTRUs 102a, 102b, and 102c can communicate / connect with gNBs 180a, 180b, and 180c, and also with another RAN such as eNode-Bs 160a, 160b, and 160c. For example, WTRUs 102a, 102b, and 102c can implement the DC principle to communicate substantially simultaneously with one or more gNBs 180a, 180b, and 180c, and one or more eNode-Bs 160a, 160b, and 160c. In a non-standalone configuration, eNode-Bs 160a, 160b, and 160c can act as mobility anchors for WTRUs 102a, 102b, and 102c, and gNBs 180a, 180b, and 180c can provide additional coverage and / or throughput for serving WTRUs 102a, 102b, and 102c.

[0060] Each of gNBs 180a, 180b, and 180c can be associated with a specific cell (not shown) and can be configured to handle radio resource management decisions, handover decisions, user scheduling in UL and / or DL, network fragmentation support, interoperability between DC, NR, and E-UTRA, routing user plane data to User Plane Functions (UPF) 184a and 184b, and routing control plane information to Access and Mobility Management Functions (AMF) 182a and 182b, etc. Figure 1D As shown, gNB 180a, 180b, and 180c can communicate with each other via the Xn interface.

[0061] Figure 1DThe CN 106 shown may include at least one AMF 182a, 182b, at least one UPF 184a, 184b, at least one Session Management Function (SMF) 183a, 183b, and may include data networks (DN) 185a, 185b. Although the foregoing elements are depicted as part of CN 106, it will be understood that any of these elements may be owned and / or operated by an entity other than the CN operator.

[0062] AMF 182a and 182b can connect to one or more of the gNBs 180a, 180b, and 180c in RAN 104 via the N2 interface and can act as control nodes. For example, AMF 182a and 182b can be responsible for authenticating users of WTRU 102a, 102b, and 102c, supporting network segmentation (e.g., handling different Protocol Data Unit (PDU) sessions with different requirements), selecting specific SMF 183a and 183b, managing registration areas, terminating Non-Access Stratum (NAS) signaling, mobility management, and so on. AMF 182a and 182b can use network segmentation to customize CN support for WTRU 102a, 102b, and 102c based on the service types used by WTRU 102a, 102b, and 102c. For example, different network slices can be established for different use cases, such as services dependent on Ultra Reliable Low Latency (URLLC) access, services dependent on Enhanced Massive Mobile Broadband (eMBB) access, services for MTC access, etc. AMF 182A and 182b can provide control plane functions for switching between RAN 104 and other RANs (not shown) employing other radio technologies (such as LTE, LTE-A Pro, and / or non-3GPP access technologies (such as WiFi)).

[0063] SMFs 183a and 183b can connect to AMFs 182a and 182b in CN 106 via the N11 interface. SMFs 183a and 183b can also connect to UPFs 184a and 184b in CN 106 via the N4 interface. SMFs 183a and 183b can select and control UPFs 184a and 184b, and configure traffic routing through UPFs 184a and 184b. SMFs 183a and 183b can perform other functions, such as managing and allocating UE IP addresses, managing PDU sessions, controlling policy enforcement and QoS, and providing DL data notifications. PDU session types can be IP-based, non-IP-based, Ethernet-based, etc.

[0064] UPF 184a and 184b can connect to one or more of the gNBs 180a, 180b, and 180c in RAN 104 via the N3 interface. This provides WTRU 102a, 102b, and 102c with access to packet-switched networks such as the Internet 110, facilitating communication between WTRU 102a, 102b, and 102c and IP-enabled devices. UPF 184 and 184b can perform other functions such as routing and forwarding packets, enforcing user plane policies, supporting multi-homed PDU sessions, handling user plane QoS, buffering DL packets, providing mobility anchoring, and so on.

[0065] CN 106 can facilitate communication with other networks. For example, CN 106 may include an IP gateway (e.g., an IP Multimedia Subsystem (IMS) server), or may communicate with an IP gateway that serves as an interface between CN 106 and PSTN 108. Furthermore, CN 106 can provide WTRUs 102a, 102b, and 102c with access to other networks 112, which may include other wired and / or wireless networks owned and / or operated by other service providers. In one embodiment, WTRUs 102a, 102b, and 102c may be connected to local DNs 185a and 185b via the N3 interface to UPFs 184a and 184b and the N6 interface between UPFs 184a and 184b and DNs 185a and 185b.

[0066] Given Figure 1A-1D and Figure 1A-1D The functions described herein with respect to one or more of the following: WTRU 102a-d, Base Station 114a-B, eNode-B 160a-c, MME 162, SGW 164, PGW 166, gNB 180a-c, AMF 182a-B, UPF 184a-B, SMF 183a-B, DN 185a-B, and / or any other devices described herein may be performed by one or more emulation devices (not shown). An emulation device may be one or more devices configured to emulate one or more of the functions described herein. For example, an emulation device may be used to test other devices and / or simulate network and / or WTRU functions.

[0067] Simulation devices can be designed to perform one or more tests on other devices in laboratory and / or carrier network environments. For example, one or more simulation devices can perform one or more or all functions while being fully or partially implemented and / or deployed as part of a wired and / or wireless communication network to test other devices within the communication network. One or more simulation devices can perform one or more or all functions while being temporarily implemented / deployed as part of a wired and / or wireless communication network. Simulation devices can be directly coupled to another device to test and / or perform tests using over-the-air wireless communication.

[0068] One or more emulation devices may perform one or more functions, including all functions, rather than being implemented / deployed as part of a wired and / or wireless communication network. For example, emulation devices may be used in test scenarios in test laboratories and / or non-deployment (e.g., testing) wired and / or wireless communication networks to perform testing of one or more components. One or more emulation devices may be test devices. Emulation devices may transmit and / or receive data using direct RF connections and / or wireless communication via RF circuitry (e.g., which may include one or more antennas). The term Wireless Transmit / Receive Unit (WTRU) is used interchangeably with the term User Equipment (UE). Therefore, wherever UE is used, the term means WTRU, and vice versa. Additionally, the following abbreviations may be used in this specification. This list is not exhaustive, and other abbreviations may be used. Furthermore, the abbreviations may have alternatives or other uses that are obvious in the context.

[0069] Subcarrier spacing gNBNRNodeB AP is non-periodic; access point BFR beam fault recovery BFD-RS Beam Fault Detection - Reference Signal BLER block error rate BWP bandwidth portion CA carrier aggregation CB is based on contention (e.g., access, channel, resources). CDM code division multiplexing CG Community Group CoMP Cooperative Multipoint Transmission / Reception CP loop prefix CPE common phase error CP-OFDM (Traditional OFDM with Cyclic Prefix) CQI Channel Quality Indicator CN core network (e.g., LTE packet core or NR core) CRC Cyclic Redundancy Check CSI Channel Status Information CSI-RS Channel State Information - Reference Signal CU Central Unit D2D device-to-device transmission (e.g., LTE sidelink) DC Dual Connectivity DCI downlink control information DL downlink DM-RS demodulation reference signal DRB Data Radio Bearer DU Distributed Unit EN-DCE-UTRA–NR Dual Connectivity EPC Evolution Group Core FD-CDM Frequency Domain-Code Division Multiplexing FDD Frequency Division Duplex FDM (Frequency Division Multiplexing) ICI inter-cell interference ICIC Inter-cell Interference Cancellation IP Internet Protocol LBT listens before speaking LCH logical channel LCID (Logical Channel Identifier) LCP logical channel priority sorting LLC Low Latency Communication LoS line of sight LTE Long Term Evolution, for example, starting from 3GPP LTE R8 and above MAC Media Access Control MAC CE Media Access Control Element NACK negates ACK MBMS Multimedia Broadcasting System MCG main cell group MCS modulation and coding scheme MIMO (Multiple Input Multiple Output) MTC Machine-Type Communication MR-DC Multi-RAT Dual Connectivity NAS Non-Access Layer NCB-RS New Candidate Beam Reference Signal NE-DCNR-RAN–E-UTRA Dual Connectivity NR New Radio NR-DC features dual connectivity for new radios. OFDM (Orthogonal Frequency Division Multiplexing) Out-of-band (radiation) Total power available to the UE within a given transmission interval The primary cell in the Pcell primary cell group PCG main cell group PDU Protocol Data Unit PER grouping error rate PHY physical layer PLMN Public Land Mobile Network PLR packet loss rate PRACH Physical Random Access Channel PRB Physical Resource Block PRS positioning reference signal Primary and secondary cells in Pscell secondary cell group PSS master synchronization signal PT-RS Phase Tracking - Reference Signal QoS (Quality of Service) from the physical layer perspective RAB Wireless Access Bearer RANPA Radio Access Network Paging Area RACH (Random Access Channel or Procedure) RAR Random Access Response RAT wireless access technology RB resource blocks RCU Radio Access Network Central Unit RF front end RE Resource Elements RLF wireless link failure RLM radio link monitoring RNTI Wireless Network Identifier ROM read-only mode (for MBMS) RRC Radio Resource Control RRM Wireless Resource Management RS reference signal RTT round trip time SCG auxiliary community group SCMA Single Carrier Multiple Access SCS subcarrier spacing SDU Service Data Unit SOM Spectrum Operation Mode SP Semi-Continuous SpCell primary or secondary cell group primary cell SRB Signaling Radio Bearer SS synchronization signal SRS detection reference signal SSS auxiliary synchronization signal SUL supplements uplink SWG (Self-containing subframe) switching gap TB transport block TBS transport block size TCI Transport Configuration Index TDD Time Division Duplex TDM Time Division Multiplexing TI time interval (an integer multiple of one or more symbols) TTI transmission time interval (an integer multiple of one or more symbols) TRP Sending / Receiving Points TRPG Transmit / Receive Point Group TRS Tracking Reference Signal TRX transceiver UL uplink URC Ultra-Reliable Communication URLLC ultra-reliable and low-latency communication V2X vehicle-to-vehicle communication WLAN wireless local area network and related technologies (IEEE 802.xx domain).

[0070] To deliver the benefits of next-generation wireless communication systems and to predict changing parameters and configurations so that they can be applied before congestion, interference, or other such conditions occur, the systems and methods discussed in this paper are implemented using artificial intelligence / machine learning (AI / ML) systems. These AI / ML systems can be based on trained models that enable the prediction of changing environmental conditions and the adjustment of beamforming parameters or channel state information, adjustments that can ideally be applied before communication equipment experiences any adverse effects. However, different AI / ML models can provide better performance depending on changing conditions. For example, when interference from a local airport radar system is present, an AI / ML model optimized for predicting beam management adjustments as equipment or other entities move within the area may not necessarily make accurate predictions.

[0071] Therefore, in some aspects, this disclosure relates to implementations of systems and methods for dynamically identifying or selecting AI / ML models or model types to provide optimized predictions based on changing environmental conditions, device movement, or combinations of these or other changes. In some implementations, the training model may be generated (and / or updated) by a UE, base station (gNB), or other network device (e.g., application server, AI / ML processing node, etc.) and may be provided to devices (e.g., UE and gNB or other devices) for performance, condition, or characteristic predictions and parameter adjustments. In some implementations, these predictions may enable devices to apply parameter adjustments, such as immediately changing retransmission window parameters in response to changes in packet loss rate, prior to requests or notifications from cooperating devices, without first exchanging configuration change details or other such modifications. In some implementations, devices (e.g., UE, gNB, or other devices) may monitor network conditions or performance and may dynamically determine whether to change the AI / ML model, and if so, which model to use. This can significantly reduce management overhead and enable faster mitigation or recovery from changing channel conditions or problems.

[0072] In some embodiments disclosed herein, an AI / ML model selection / reselection method based on dynamic AI / ML model determination is disclosed. The UE is configured with one or more AI / ML models having model IDs, wherein each AI / ML model is associated with an AI / ML model type (e.g., BM or CSI), one or more RSs for measuring the quality of AI / ML model reselection, one or more RSs for reselecting the AI / ML model for BM, and one or more RSs for reselecting the AI / ML model for CSI. The UE identifies the need for AI / ML model reselection based on one or more of the following: the number of NACKs (e.g., number of consecutive NACKs > a threshold), RSRP (e.g., RSRP < a threshold), a change in UE speed (e.g., current UE speed - average UE speed within the measurement window > a threshold), a change in UE location (e.g., current UE location - average UE location within the measurement window > a threshold), a change in path loss (e.g., current path loss - average path loss within the measurement window > a threshold), an assumed PDCCH BLER < a threshold, a measured SINR or differential SINR (e.g., from a scheduled DMRS port), etc.

[0073] Then, the UE identifies the type of AI / ML model for reselection based on the triggered conditions, such as the following: If the measured RSRP > threshold and the number of consecutive NACKs > threshold, reselect the AI / ML model for CSI. If the measured RSRP < threshold, reselect the AI / ML model for CSI. If the number of failures of the measured hypothesized PDCCH BLER < threshold, reselect the AI / ML model for CSI. Otherwise, reselect the AI / ML for BM.

[0074] The UE identifies the AI / ML model according to the conditions. Alternatively, the UE activates the AI / ML model associated with the currently identified AI / ML model / model type from the one or more AI / ML models for reselection, so as to evaluate the prediction accuracy of the AI / ML model. Evaluation details such as: beam prediction accuracy / RSRP difference and the RS transmission for evaluation, different RS transmissions for the activated model, etc. The UE indicates the type of AI / ML model for reselection to the gNB and receives one or more RSs associated with the reported type. The UE measures the one or more RSs and indicates the preferred model ID to the gNB based on the measurement (if the AI / ML model for BM is requested, after the reselection of the AI / ML model for BM, the AI / ML model for CSI can also be reselected). The UE receives the confirmation of the UE request and activates the indicated model for the reported request type. Add UE behavior after activating / selecting / deciding the AI / ML model.

[0075] In some embodiments, the CSI parameter switching is based on the associated AI / ML model. The UE is configured with one or more AI / ML models having model IDs, where each AI / ML model is associated with a set of CSI parameters (potentially for each CSI type or CSI configuration, such as for CSI and for BM) / configured with the set of CSI parameters. Alternatively, the UE indicates its capabilities (e.g., the CSI parameters required for each AI / ML model). The UE identifies the preferred model based on measuring one or more configured thresholds, RSRP, UE speed, path loss, area ID, cell ID, TRP ID, etc., potentially using the model type indication (or CSI configuration ID). For example, if the UE speed is <X and the path loss > Y (e.g., indoor), then model #1 is determined. If the UE speed < X and the path loss < Y (e.g., walking outdoors), then model #2 is determined. If the UE speed > X and the path loss < Y (e.g., high-speed outdoor), then model #3 is determined.

[0076] Based on the determined model, the UE indicates the preferred model ID or set of CSI parameters to the gNB. The UE receives confirmation from the gNB (e.g., via CORESET / SS associated with the indication). The UE deactivates the previous model, activates the newly indicated model, and applies the associated set of CSI parameters to the next CSI report, which includes the preferred size of the measurement / prediction window. For example, Model #1: X beam IDs and corresponding L1-RSRPs (N1) for the current measurement. Model #2: X1-1 beam IDs and corresponding L1-RSRPs for the current measurement, and X1-2 beam IDs and corresponding L1-RSRPs for N1+N2. Model #3: X2-1, X2-2, X2-3, and X2-4 beam IDs and corresponding L1-RSRPs for N1 / N2 / N3 / N4.

[0077] As described above, the systems and methods used in this paper are implemented using artificial intelligence and / or machine learning systems. Artificial intelligence can be broadly defined as the behavior exhibited by machines. Such behavior can, for example, mimic cognitive functions such as perception, reasoning, adaptation, and action. Machine learning can refer to a type of algorithm based on learning through experience (“data”) rather than explicit programming (“configuration of a set of rules”) to solve problems. Machine learning can be considered a subset of AI. Different machine learning paradigms can be envisioned based on the nature of the data or feedback available to the learning algorithm. For example, supervised learning methods may involve learning a function that maps inputs to outputs based on labeled training examples, where each training example can be a pair consisting of an input and a corresponding output. For example, unsupervised learning methods may involve detecting patterns in data that do not have pre-existing labels. For example, reinforcement learning methods may involve performing a sequence of actions in an environment to maximize cumulative rewards. In some solutions, machine learning algorithms can be applied using combinations or interpolations of the above methods. For example, semi-supervised learning methods may use a combination of a small amount of labeled data and a large amount of unlabeled data during training. In this respect, semi-supervised learning falls between unsupervised learning (training data without labels) and supervised learning (training data with only labels).

[0078] Deep learning refers to a class of machine learning algorithms that use artificial neural networks (specifically, deep neural networks) loosely inspired by biological systems. Deep neural networks (DNNs) are a special category of machine learning models inspired by the human brain, where the input is linearly transformed and passed multiple times through a non-linear activation function. DNNs typically consist of multiple layers, each composed of a linear transformation and a given non-linear activation function. DNNs can be trained using training data via a backpropagation algorithm. Recently, DNNs have demonstrated state-of-the-art performance in various fields, such as speech, vision, natural language processing, etc., and have been used in a variety of supervised, unsupervised, and semi-supervised machine learning settings. The term AI / ML-based methods / processes can refer to the implementation of behaviors and / or compliance with requirements through data-driven learning without explicitly configured sequences of action steps. Such methods enable the learning of complex behaviors that might be difficult to specify and / or implement using traditional methods.

[0079] In Rel-15, the New Radio (NR) introduced Radio Access Technology (RAT) in Frequency Range 2 (FR2), where FR2 represents the frequency range of 24.25–52.6 GHz. One of the key challenges of using FR2 is the higher propagation loss. Since propagation loss increases with carrier frequency, FR2 experiences higher propagation loss compared to lower frequency range systems. To overcome this higher propagation loss, highly directional beamforming transmission and reception can be used to improve efficiency.

[0080] Beamforming gain can be achieved by adding or subtracting one signal from another. Since higher beamforming gain can be achieved as more signals are added or subtracted, a large number of antenna elements can be used for highly directional beamforming transmission. Controlling signal addition or subtraction can be accomplished by controlling the phase of the antenna elements.

[0081] Based on the phase control type, beamforming methods can generally be classified into three types (e.g., analog beamforming, digital beamforming, and hybrid beamforming). Figure 2 This is a block diagram of a system for hybrid beamforming according to some implementation methods. Figure 2The illustrated architecture can be implemented at a base station or WTRU / UE. Digital beamforming controls the signal phase by applying digital precoding, while analog beamforming controls the signal phase using phase shifters. Typically, digital beamforming offers good flexibility (e.g., applying different phases for different frequency resource blocks) but requires a more complex implementation. In contrast, analog beamforming offers a relatively simple implementation but has limitations (e.g., the same analog beam for all frequency resources). Given these trade-offs, hybrid beamforming is a good architecture for achieving large beamforming gains with reasonable implementation complexity. Hybrid beamforming provides sufficient flexibility with reasonable implementation complexity by combining analog and digital beamforming.

[0082] Because the beamwidth of a beam decreases as beamforming gain increases, a beam can only cover a limited area. Therefore, the BS and UE need to utilize multiple beams to cover the entire cell. For example, broadcast signals such as synchronization signal blocks (SSBs) can be transmitted in all directions (e.g., via beam scanning) to cover the entire cell. For unicast transmissions between the BS and UE, beam management (BM) provides a process for optimizing the beam direction reaching the UE. Beam management may include selecting and maintaining beam directions for unicast transmissions (including control channels and / or data channels) between the BS and UE or any other device.

[0083] Beam management processes can be categorized into beam determination, beam measurement and reporting, beam switching, beam indication, and beam recovery. In beam determination, the BS and UE, or other equipment, determine the beam direction to ensure good radio link quality for unicast control and data channel transmission. Once the link is established, the equipment (e.g., the UE) measures the link quality of multiple transmit (TX) and receive (RX) beam pairs and reports the results to other equipment (e.g., the BS). Furthermore, UE mobility, orientation, and channel congestion can alter the radio link quality of the TX and RX beam pairs. When the quality of the current beam pair deteriorates, the BS and UE can switch to another beam pair with better radio link quality. For this purpose, the BS and / or UE can monitor the quality of the current beam pair as well as some other beam pairs and perform a switchover when necessary. A beam indication procedure can be used when the BS assigns a TX beam to the UE via DL control signaling. Beam recovery requires a recovery process when the link between the BS and UE can no longer be maintained.

[0084] In some implementations, AI / ML systems can be used for one or more of the following use cases: CSI feedback enhancement, such as reduced overhead, improved accuracy, and prediction; beam management, such as beam prediction in the temporal and / or spatial domains to reduce system overhead and latency and improve beam selection accuracy; and positioning accuracy enhancement for various scenarios, including those with heavy NLOS conditions. However, the use of these systems may require additional implementation details discussed below.

[0085] First, it may be necessary to identify the appropriate AI / ML model type for selection / reselection. If the AI / ML model used for beam management is not working, accurate predictions from the AI / ML model used for CSI will not provide adequate performance because the optimal beam has not been properly selected. Furthermore, it may be necessary to identify suitable AI / ML models, including whether the model can be generalized to different conditions or measurements.

[0086] Consider the following scenarios to validate the generalization performance of AI / ML models across various scenarios / configurations as a starting point: • Scenario 1: Train an AI / ML model on a training dataset from a scenario #A / configuration #A, and then perform inference / testing on a dataset from the same scenario #A / configuration #A.

[0087] • Scenario 2: Train an AI / ML model on a training dataset from a scenario #A / configuration #A, and then perform inference / testing on a dataset different from scenario #A / configuration #A (e.g., scenario #B / configuration #B, scenario #A / configuration #B).

[0088] • Scenario 3: The AI / ML model is trained on a training dataset constructed by mixing datasets from multiple scenarios / configurations, including scenario #A / configuration #A, as well as datasets different from scenario #A / configuration #A (e.g., scenario #B / configuration #B, scenario #A / configuration #B). The AI / ML model then performs inference / testing on datasets from a single scenario / configuration (e.g., scenario #A / configuration #A, scenario #B / configuration #B, scenario #A / configuration #B) from multiple scenarios / configurations. Note that the company reports the ratio of dataset mixing, and the number of multiple scenarios / configurations can be greater than two.

[0089] In some implementations, the following scenarios for generalization verification may be optionally considered: • Scenario 2A: Train an AI / ML model on a training dataset from a scenario #A / configuration #A, then update the AI / ML model on datasets different from scenario #A / configuration #A (e.g., scenario #B / configuration #B, scenario #A / configuration #B). Subsequently, test the AI / ML model on datasets different from scenario #A / configuration #A, e.g., going through scenario #B / configuration #B, scenario #A / configuration #B. The company can report improvements in performance after fine-tuning the dataset settings (e.g., dataset size). The feasibility of fine-tuning on the UE / network side is for future research.

[0090] Based on the evaluation results, Case 3 (trained using a dataset with mixed scenarios / configurations) generally shows better generalization performance than Case 2 (trained using a dataset with different scenarios / configurations). Furthermore, Case 3's performance is generally lower than Case 1's; however, in some cases, Case 3 outperforms Case 1.

[0091] However, training an AI / ML model with all possible scenarios and configurations can be too complex, time-consuming, and resource-intensive. Therefore, in some implementations, multiple AI / ML models can be configured, and it may be necessary to select one AI / ML model for each AI / ML model type (e.g., BM or CSI).

[0092] Therefore, one problem addressed in this disclosure is how the UE identifies the need for selecting / reselecting an AI / ML model, the appropriate AI / ML model type, and the appropriate AI / ML model for the selected AI / ML model type.

[0093] This document provides various solution implementations for efficiently identifying selection / reselection needs. In some implementations, these solutions enable the UE to identify the need for AI / ML model reselection. In one solution, the UE can identify the need for AI / ML model reselection by measuring and evaluating parameters.

[0094] In other implementations, the solution can enable the UE to identify the appropriate AI / ML model type for reselection. In one solution, the UE can identify the AI / ML model type based on measurements and evaluated parameters.

[0095] In other implementations, the solution can enable the UE to identify the appropriate AI / ML model for the determined AI / ML model type. In one solution, the UE can identify the AI / ML model based on measurements and evaluated parameters.

[0096] In other implementations, the solution may enable the UE to indicate the AI / ML model type and AI / ML model used for reselection. In one solution, the UE may indicate the AI / ML model type ID and AI / ML model ID used for reselection as part of the CSI report.

[0097] In other implementations, the solution can determine CSI reporting parameters based on the determined AI / ML model type and AI / ML model ID. In one solution, the UE can determine the set of CSI parameters to be reported based on the determined AI / ML model type and AI / ML model ID.

[0098] These solutions can be implemented in any combination of settings without limitation.

[0099] Brief reference Figure 3 This diagram illustrates a block diagram of a system 300 for dynamic model selection in wireless communication according to some embodiments. System 300 may include a UE or WTRU 102, a base station 114, etc., as described above, or any other such device. System 300 may include one or more processors 118, memory devices 130, 132, an antenna or antenna array 122, and analog and / or digital beamformers, as combined above. Figure 2 The subject of discussion.

[0100] In some implementations, system 300 may include monitor 302. Monitor 302 may include applications, services, servers, daemons, routines, or other executable logic for measuring and / or monitoring channel characteristics (e.g., noise, received signal strength, interference, packet loss rate, block error rate, etc.) and / or physical or electrical characteristics (e.g., the location of system 300, changes in the location or velocity vector of system 300, changes in the velocity vector or acceleration of system 300, changes in the acceleration or jerks of system 300, orientation of system 300 or antenna array 122, beam arrival angle, beam departure angle, temperature, power usage, etc.). Monitor 302 may include hardware, software, or a combination of hardware and software, such as software implemented in an ASIC or FPGA, hardware sensors read by software processes, etc. Monitor 302 may be executed by one or more processors 118 and may perform measurements in a periodic, random, continuous manner, in response to application or device requests, or according to any other scheduling arrangement. Monitor 302 can store measurements in any suitable format (e.g., array, flat file, index list, bitmap, etc.) in log database 304 or in a file.

[0101] In some implementations, system 300 may include selector 306. Selector 306 may include an application, service, server, daemon, routine, or other executable logic for determining whether a new AI / ML model should be selected, and if so, which model should be selected. Selector 306 may be implemented in hardware, software, or a combination of hardware and software. In some implementations, selector 306 may utilize measurements of channel characteristics or conditions and / or physical or electrical characteristics measured by monitor 302 and / or stored in log 304. For example, in some implementations, selector 306 may compare measurements to one or more thresholds. In some implementations, selector 306 may include AI / ML systems such as DNN, decision trees, SVM, or any other type of AI / ML algorithm. For example, in some implementations, selector 306 may predict whether a channel is degrading and may become unavailable in the future based on past measurements of channel noise or signal strength, and if so, predict whether a new AI / ML model should be used for CSI configuration or beam management. In other implementations, other measurements and / or combinations of measurements with thresholds may be utilized. In many implementations, selector 306 and monitor 302 may be part of the same program or module. When determining to select a new AI / ML model, selector 306 may select an AI / ML model from models 308 stored in a database, file, or other storage device. For example, model 308 may include a set of hyperparameters, weights, biases, neuron layer configurations, or any other such parameters of the trained AI / ML model.

[0102] In some implementations, system 300 may execute one or more AI / ML engines 310. AI / ML engines 310 may include applications, services, servers, daemons, routines, or other executable logic for performing AI / ML algorithms using model 308. In some implementations, system 300 may execute multiple AI / ML engines 310 simultaneously. For example, when determining the selection of a new AI / ML model, in some implementations discussed in more detail below, system 300 may execute multiple models simultaneously or sequentially and compare predictions to identify the model with the highest accuracy, sensitivity, selectivity, or quality based on current conditions or changes in conditions. The system may then utilize the identified model and deactivate or disable other models until selector 306 determines a subsequent time when a new model is needed.

[0103] As described above, in some embodiments, one or more processors 118 of system 300 can execute applications and / or AI / ML algorithms. In some embodiments, system 300 may include one or more coprocessors, such as tensor processing units (TPUs), graphics processing units (GPUs), or other such processors, for executing AI / ML algorithms. Such coprocessors can be optimized to execute AI / ML algorithms or specific types of models efficiently or at high speed.

[0104] In some implementations, the UE can transmit or receive physical channels or reference signals based on at least one spatial domain filter. The term "beam" can be used herein to refer to a spatial domain filter. The UE can transmit physical channels or signals using the same spatial domain filter used to receive RS (e.g., CSI-RS) or SS blocks. The UE transmission can be referred to as the "target," and the received RS or SS block can be referred to as the "reference" or "source." In this context, it can be said that the UE transmits the target physical channel or signal based on its spatial relationship with the reference RS or SS block.

[0105] The UE can transmit the first physical channel or signal using the same spatial domain filter as the spatial domain filter used to transmit the second physical channel or signal. The first transmission and the second transmission can be referred to as the "target" and the "reference" (or "source"), respectively. In this case, it can be said that the UE transmits the first (target) physical channel or signal based on the spatial relationship of the reference second (reference) physical channel or signal.

[0106] Spatial relationships can be implicit, configured by the RRC, or signaled by the MAC CE or DCI. For example, a UE can implicitly transmit the DM-RS of PUSCH and PUSCH based on the same spatial domain filter as the SRI indicated by the DCI or the SRS configured by the RRC. In another example, spatial relationships can be configured by the RRC for the SRS Resource Indicator (SRI) or signaled by the MAC CE for the PUCCH. This spatial relationship can also be referred to as "beam indication".

[0107] The UE can receive a first (target) downlink channel or signal based on the same spatial domain filter or spatial reception parameters as the second (reference) downlink channel or signal. For example, such an association can exist between a physical channel such as PDCCH or PDSCH and its corresponding DM-RS. This association can exist at least when the first and second signals are reference signals, and when the UE is configured with a quasi-co-location (QCL) assumption type D between corresponding antenna ports. This association can be configured as a TCI (Transmission Configuration Indicator) state. The association between the CSI-RS or SS block and the DM-RS can be indicated to the UE via an index of the TCI state set configured by RRC and / or signaled by the MAC CE. This indication can also be referred to as a "beam indication."

[0108] In this disclosure, the term "RS resource set" or similar terms may be used interchangeably with resource configuration, RS resources, and / or beamgroups. Similarly, in this invention, the term "beam" may be used interchangeably with TCI status, TCI status group, and / or beam pair; and the term "beam report" or similar terms may be used interchangeably with CSI measurement, CSI reporting, and / or beam measurement. Likewise, "beam ID" may be used interchangeably with beam index and / or beam pair ID; and the reference signal (or beam reference signal) may be used interchangeably with one or more of the following: sounding reference signal (SRS), channel state information-reference signal (CSI-RS), demodulation reference signal (DM-RS), phase tracking reference signal (PT-RS), and / or synchronization signal block (SSB). A “channel” (or physical channel) may be used interchangeably with one or more of the following: PDCCH, PDSCH, Physical Uplink Control Channel (PUCCH), Physical Uplink Shared Channel (PUSCH), Physical Random Access Channel (PRACH), Physical Sidelink Control Channel (PSCCH), Physical Sidelink Shared Channel (PSSCH), Physical Sidelink Feedback Channel (PSFCH), and / or Physical Broadcast Channel (PBCH).

[0109] Now for reference Figure 4 A flowchart of a method 400 for dynamic model selection for wireless communication according to some embodiments is shown. Method 400 can be performed by a UE, WTRU, base station, or other network node or device. In some embodiments, the device can be configured with one or more models for AI / ML prediction, and / or can be configured with parameters for CSI and beam reporting. For example, in some embodiments, one or more of the following configurations can be used for CSI or beam reporting configuration, as described below.

[0110] The UE can be configured with one or more CSI reporting configurations.

[0111] Report configuration type (e.g., periodic, semi-persistent on PUCCH, semi-persistent on PUSCH, or non-periodic).

[0112] Number of reports (e.g., CRI-RI-PMI-CQI, CRI-RI-i1, CRI-RI-i1-CQI, CRI-RSRP, SSB-index-RSRP, CRI-RI-LI-PMI-CQI, CRI-SINR, SSB-index-SINR).

[0113] Report frequency configuration.

[0114] • CQI format indicator (wideband CQI or subband CQI).

[0115] • PMI format indicator (wideband PMI or subband PMI).

[0116] • CSI report band.

[0117] Time constraints for channel measurements.

[0118] Time constraints for interference measurements.

[0119] Codebook configuration.

[0120] Group-based beam reporting.

[0121] CQI form.

[0122] Sub-band size.

[0123] Non-PMI port indication.

[0124] Report a list of time slot configurations / offsets.

[0125] CSI reporting cycle and offset.

[0126] One or more PUCCH resources used for CSI reporting.

[0127] Port index.

[0128] In some embodiments, one or more of the following configurations may be used for the measurement configuration of beam reporting: • The UE can be configured with one or more CSI measurement configurations.

[0129] CSI measurement configurations may include one or more of the following: RS for channel measurements.

[0130] RS (zero power or non-zero power) used for interference measurement.

[0131] Report trigger size.

[0132] A list of non-periodic trigger states.

[0133] List of semi-persistent PUSCH trigger states.

[0134] Associated CSI resource configuration.

[0135] Associated CSI report configuration.

[0136] In some embodiments, one or more of the following configurations may be used for CSI resource configuration: • A UE can be configured with one or more CSI resource configurations.

[0137] CSI resource configuration may include one or more of the following: CSI Resource Configuration ID.

[0138] One or more RS resource sets used for channel measurements.

[0139] One or more RS resource sets used for interference measurement.

[0140] Bandwidth section ID.

[0141] Resource type (e.g., non-periodic, semi-persistent, or periodic).

[0142] In some embodiments, one or more of the following configurations may be used for the RS resource collection: • A UE can be configured with one or more RS resource sets.

[0143] RS resource set configuration may include one or more of the following: RS resource collection ID.

[0144] One or more RS resources used for an RS resource set.

[0145] Repeat (i.e., on or off).

[0146] Non-periodic trigger offset (e.g., one of slots 0-6).

[0147] TRS information (e.g., true or false).

[0148] In some embodiments, one or more of the following configurations may be used for RS resources: • A UE can be configured with one or more RS resources.

[0149] RS resource configuration may include one or more of the following: RS Resource ID.

[0150] Resource mapping (e.g., RE in PRB).

[0151] Power control offset (e.g., a value of -8, ..., 15).

[0152] Utilize the power control offset of SS (e.g., -3 dB, 0 dB, 3 dB, 6 dB).

[0153] Scrambling ID.

[0154] Periodicity and offset.

[0155] QCL information (e.g., based on TCI status).

[0156] In some embodiments, at step 402, the UE may indicate its capability for AI / ML models supported by the UE. For example, in the embodiment at 402, the UE may indicate one or more of the following: • Supported Model Types: The UE can indicate one or more types of AI / ML models that it supports. For example, the UE can indicate one or more of CSI, BM, positioning, etc., as the supported AI / ML model types (or AI / ML model functions).

[0157] • Supported RS Types: The UE can indicate one or more types of RSs that it supports for measurement. This indication can be specified for AI / ML model reselection detection (i.e., detecting the need for AI / ML model reselection), AI / ML model type evaluation (i.e., identifying the type of AI / ML model used for reselection), and AI / ML model reselection assessment (i.e., evaluating the quality of the AI / ML model). For example, one or more of the following RSs can be indicated as supported RS types: • SRS (e.g., from other UEs).

[0158] • CSI-RS.

[0159] · DM-RS.

[0160] PT-RS.

[0161] · SSB.

[0162] • Supported Measurement Types: The UE can indicate one or more measurement types supported by the UE. This indication can be specifically designated for AI / ML model reselection detection (i.e., detecting the need for AI / ML model reselection), AI / ML model type evaluation (i.e., identifying the type of AI / ML model used for reselection), and AI / ML model reselection assessment (i.e., evaluating the quality of the AI / ML model). For example, one or more of the following measurement types can be indicated as supported RS types: • RSRP or L1-RSRP.

[0163] • RSRQ or L1-RSRQ.

[0164] • SINR or L1-SINR.

[0165] • Hypothetical PDCCH BLER.

[0166] • Hypothetical PUCCH BLER.

[0167] • CQI.

[0168] • PDSCH ACK / NACK.

[0169] • Supported AI / ML Models: The UE can indicate one or more AI / ML models supported by the UE. For example, the UE can indicate one or more of the following: • BM used only for spatial domain prediction.

[0170] • BM used for both spatial domain prediction and temporal domain prediction.

[0171] • BM used only for time-domain prediction.

[0172] • CSI compression without time prediction.

[0173] • CSI compression with time prediction.

[0174] In some implementations, in step 402, the UE may indicate the type of the set of CSI parameters supported by the UE. For example, the UE may indicate one or more of the following: • Set #1 (BM for spatial domain prediction only): X beam IDs and corresponding L1-RSRP (N1) for the current measurement.

[0175] • Set #2 (BM for both spatial domain prediction and time domain prediction): X1-1 beam IDs and corresponding L1-RSRPs for the current measurement, and X1-2 beam IDs and corresponding L1-RSRPs for N1+N2.

[0176] • Set #3 (BM for time-domain prediction only): X2-1, X2-2, X2-3 and X2-4 beam IDs for N1 / N2 / N3 / N4 and their corresponding L1-RSRPs.

[0177] • Set #4 (CSI compression without time prediction): RI, X3-1 PMIs, and the corresponding CQI for each codeword.

[0178] • Set #5 (CSI compression with time prediction): RI#1 for the current measurement, X3-1 PMIs and corresponding CQI (N1) for each codeword, and RI#2, X3-2 PMIs and corresponding CQI for each codeword for N1+N3.

[0179] In various implementations, capability reports can be submitted via CSI reports or other management reports.

[0180] The term "AI / ML model functionality" can be used interchangeably with AI / ML model type, AI / ML model use case, and AI / ML model functionality or similar terms. Similarly, "AI / ML model reselection" can be used interchangeably with AI / ML model switching, AI / ML model determination, AI / ML model activation / deactivation, and AI / ML model rollback.

[0181] In some embodiments, the UE may pre-configure one or more AI / ML models with model IDs, or in other embodiments, one or more AI / ML models and their corresponding model IDs may be received in step 404. For example, in some embodiments, the AI / ML models may be pre-loaded into the memory of the UE or other devices. In other embodiments, the AI / ML models may be downloaded or received via broadcast or other transmissions (e.g., from a base station, network node, application server, etc.). For example, in some embodiments, the AI / ML models may be periodically updated by an application server performing continuous or periodic training on the AI / ML models based on new data from other network devices. Therefore, in some embodiments, in step 402, AI / ML models may be received periodically or in response to a capability transmission. In some embodiments, the received AI / ML configuration may be based on the reported UE capabilities (e.g., a subset of AI / ML models may be provided to the UE based on its capabilities, thereby reducing the network bandwidth and memory required to send and store models that are unavailable or beyond the UE's capabilities). The AI / ML model configuration may include one or more of the following: • Model Type ID: The model type ID indicates the type of AI / ML model. For example, a UE can be configured with one of CSI, BM, or positioning.

[0182] • Model ID: A model ID can indicate the ID of an AI / ML model. For example, a model ID can be the ID within an AI / ML model for each AI / ML model type. In another example, a model ID can be the ID within an AI / ML model for all AI / ML model types.

[0183] • Associated sets of CSI parameters / CSI parameter types: Each AI / ML model can be associated with a set of CSI parameters or CSI parameter types. For example, the following CSI parameter sets can be defined.

[0184] • Set #1 (BM for spatial domain prediction only): X beam IDs and corresponding L1-RSRP (N1) for the current measurement.

[0185] • Set #2 (BM for both spatial domain prediction and temporal domain prediction): X1-1 beam IDs and corresponding L1-RSRPs for the current measurement, and X1-2 beam IDs and corresponding L1-RSRPs for N1+N2.

[0186] • Set #3 (BM for time-domain prediction only): X2-1, X2-2, X2-3 and X2-4 beam IDs for N1 / N2 / N3 / N4 and their corresponding L1-RSRPs.

[0187] • Set #4 (CSI compression without time prediction): RI for each codeword, x3-1 PMIs, and corresponding CQI.

[0188] • Set #5 (CSI compression with time prediction): RI#1 for the current measurement, X3-1 PMIs and corresponding CQI (N1) for each codeword, and RI#2, X3-2 PMIs and corresponding CQI for each codeword for N1+N3.

[0189] The configuration of CSI parameter sets can be based on one or more of the following: CSI parameter set ID, CSI parameter indication (e.g., RI-PMI-CQI), etc. A set of CSI parameters associated with each AI / ML model can be predefined. For example, when the UE indication BM is used only for spatial domain prediction, set #1 can be predefined as the associated CSI parameter set.

[0190] • Associated sets of RS resources and / or RS resource sets: In some embodiments, each AI / ML model or AI / ML model type may be associated with a set of RS resources or RS resource sets. For example, a UE may be configured with a first set of RS resources / resource sets associated with a first AI / ML model and a second set of RS resources / resource sets associated with a second AI / ML model. In another example, a UE may be configured with a first set of RS resources / resource sets associated with a first AI / ML model type and a second set of RS resources / resource sets associated with a second AI / ML model type.

[0191] In one embodiment, the required RS resource type can be determined based on the AI / ML model type. For example, one or more RS resources can be configured for a first AI / ML model type (e.g., for CSI). One or more sets of RS resources can be configured for a second AI / ML model type (e.g., for BM or localization). In another embodiment, the configuration can be individually indicated for AI / ML model reselection detection (i.e., detecting the need for AI / ML model reselection), AI / ML model type evaluation (i.e., identifying the type of AI / ML model type used for reselection), and AI / ML model reselection assessment (i.e., evaluating the quality of the AI / ML model). For example, SSB, DMRS, or PT-RS can be configured for AI / ML model reselection detection, and SSB or CSI-RS can be configured for AI / ML model reselection assessment.

[0192] UEs may need to reselect their AI / ML models based on the performance of the AI / ML model they are currently using or based on the detection of changes to one or more parameters or measurements that can indicate or affect the performance of the AI / ML model. To this end, UEs can use one or more of the following solutions to determine the need for AI / ML model reselection.

[0193] In some embodiments, in step 406, the UE may measure one or more signals and channels to detect the need for AI / ML model reselection. The UE may use different types of signals and channels for AI / ML model reselection detection (i.e., detecting the need for AI / ML model reselection), AI / ML model type evaluation (i.e., identifying the type of AI / ML model used for reselection), and AI / ML model reselection evaluation (i.e., evaluating the quality of the AI / ML model). For example, the UE may use a first type of RS (e.g., DMRS or PT-RS) for AI / ML model reselection detection, a second type of RS (e.g., SSB) for AI / ML model type evaluation, and a third type of RS (e.g., CSI-RS) for AI / ML model reselection evaluation. In step 406, the UE may measure one or more of the following signals and channels to detect the need for AI / ML model reselection: • SRS (e.g., from other UEs).

[0194] • CSI-RS.

[0195] · DM-RS.

[0196] PT-RS.

[0197] · SSB.

[0198] · PDCCH .

[0199] · PDSCH.

[0200] • PUCCH (e.g., from other UEs).

[0201] • PUSCH (e.g., from other UEs).

[0202] • PRACH (e.g., from other UEs).

[0203] AI / ML model type or AI / ML model evaluation can be referred to as one or more of the following: • Monitor the performance of AI / ML model types or AI / ML models to determine if switching, activation / deactivation, or reselection of AI / ML model types (or AI / ML modes) is necessary.

[0204] • A process triggered by one or more conditions (e.g., throughput performance degradation, increased latency, a high number of HARQ-NACKs, packet failures, etc.).

[0205] In one embodiment, at step 408, the UE may determine that an AI / ML model reselection is needed when one or more NACKs (unsuccessful reception of a code block or transport block) are detected. For example, in some such embodiments, when the UE detects or identifies that the number of NACKs received within a pre-configured time window from the gNB (e.g., via RRC signaling or MAC-CE indication) exceeds a pre-configured / indicated threshold number (e.g., via RRC signaling or MAC-CE indication), at step 408, the UE may determine that an AI / ML model reselection is needed. The time window may be configured to begin from the time instance in which the UE evaluates the AI / ML model reselection or from the time instance in which the last measurement was performed in step 406, within a recently elapsed moving time window (e.g., the recently elapsed x ms, the recently elapsed y time slots) (i.e., as shown, steps 406-408 may be repeated iteratively or periodically until the measurement indicates that a reselection is needed). The duration of the time window may be configured via one or more of RRC signaling, MAC-CE indication, and / or DCI indication.

[0206] In another embodiment, in step 408, the UE may determine the need for reselection of the AI / ML model based on the number of consecutive NACKs. For example, if the UE detects or identifies that the number of consecutive NACKs exceeds a threshold number pre-configured by the gNB (e.g., via RRC signaling or MAC-CE indication), then in step 408, the UE may determine that AI / ML model reselection is required.

[0207] In one embodiment, at step 408, the UE may determine the need for AI / ML model reselection based on one or more measurements made in step 406 (e.g., measurements associated with one or more RSs) or by using the estimated number of one or more measurements (e.g., via a first AI / ML model, such as a predictive model predicting future channel conditions or possible measurements). The UE may be configured with one or more thresholds via RRC signaling, MAC-CE indication, or DCI indication to determine the need for AI / ML model reselection by comparing the number of measurements or estimates. For example, if the number of measurements or estimates... If the threshold is reached, then in step 408, the UE can determine that AI / ML model reselection is required. If the number of measurements or estimates is less than the threshold, then in step 408, the UE can determine that AI / ML model reselection is not required. The quality of the measurements or estimates may include, but is not limited to, one or more of the following: • RSRP (for example, RSRP is associated with one or more configured RS).

[0208] • Differential RSRP (e.g., differential RSRP is associated with one or more configured RSs). For example, a UE may consider measurements from two time measurement instances (e.g., the two last measurement instances) to calculate the differential RSRP (e.g., L1-RSRP) of an RS (e.g., beam failure detection RS). If the calculated differential RSRP... Based on the threshold pre-configured by the gNB, the UE can determine if a reselection of the AI / ML model is necessary. If the calculated differential RSRP is less than the threshold pre-configured by the gNB, the UE can determine if a reselection of the AI / ML model is not required.

[0209] • RSRQ (e.g., RSRQ is associated with one or more configured RS).

[0210] • SINR (for example, SINR is associated with one or more configured RS).

[0211] • Differential SINR. For example, a UE can calculate the differential SINR by using measurements associated with one or more configured RSs at two time instances. If the calculated differential SINR... Based on the threshold pre-configured by the gNB, the UE can determine whether AI / ML model reselection is required. If the calculated differential SINR is less than the threshold pre-configured by the gNB, the UE can determine that AI / ML model reselection is not required.

[0212] • UE speed.

[0213] • The location of the UE.

[0214] • Path loss.

[0215] • Hypothetical PDCCH BLER.

[0216] In another embodiment, in step 408, the UE can determine the need for AI / ML model reselection based on the change in the Loss condition. For example, if the UE determines that the current Loss condition is different from the Loss condition when the current AI / ML model is selected, the UE can determine that AI / ML model reselection is required.

[0217] In one embodiment, in step 408, the UE can determine the need for AI / ML model reselection by comparing one or more current measurements (e.g., measurements associated with one or more RSs) or current estimated quantities obtained by using one or more of the current measurements performed in step 406 with the corresponding average measurements or average estimated quantities within a pre-configured time window (e.g., in past iterations of step 406).

[0218] The time window can be configured to consider a moving time window that has just passed since the time instance of AI / ML model reselection or since the time instance of the last measurement (e.g., just x ms, just y time slots have passed). The duration of the time window can be configured via RRC signaling, MAC-CE indication, or DCI indication.

[0219] The UE can be configured with one or more thresholds via RRC signaling, MAC-CE indication and / or DCI indication.

[0220] In some implementations, the UE can calculate the difference between the measured or estimated quantity and the corresponding average measured or estimated quantity (the calculated difference), and then compare the calculated difference with a pre-configured threshold in step 408 to determine the need for AI / ML model reselection.

[0221] For example, if the difference between the measured quantity or estimated quantity and the corresponding average measured quantity or average estimated quantity If the threshold is not met, then in step 408, the UE can determine that AI / ML model reselection is required. If the difference between the number of measurements or estimates and the corresponding average number of measurements or average estimates is less than the threshold, then in step 408, the UE can determine that AI / ML model reselection is not required.

[0222] As mentioned above, the quality of a measurement or the quality of an estimate may include, but is not limited to, one or more of the following: • UE speed: For example, if the difference between the current UE speed and the average UE speed over the pre-configured measurement window If a pre-configured threshold is set, the UE can determine that AI / ML model reselection is necessary. Otherwise, the UE can determine that AI / ML model reselection is not required.

[0223] • UE location: For example, if the difference between the UE's current location and the UE's average location within a pre-configured measurement window If a pre-configured threshold is set, the UE can determine that AI / ML model reselection is necessary. Otherwise, the UE can determine that AI / ML model reselection is not required.

[0224] • Path loss: For example, the difference between the current path loss experienced by the UE and the average path loss of the UE over the pre-configured measurement window. If a pre-configured threshold is set, the UE can determine that AI / ML model reselection is necessary. Otherwise, the UE can determine that AI / ML model reselection is not required.

[0225] • Assumed PDCCH Block Error Rate (BLER): For example, the difference between the assumed BLER of the last received PDCCH and the average assumed BLER of PDCCHs received within a pre-configured measurement window. If a pre-configured threshold is set, the UE can determine that AI / ML model reselection is necessary. Otherwise, the UE can determine that AI / ML model reselection is not required.

[0226] • SINR: For example, if the difference between the current SINR (e.g., the measured SINR from the scheduled DMRS port) and the average SINR over the pre-configured measurement window... If a pre-configured threshold is set, the UE can determine that AI / ML model reselection is necessary. Otherwise, the UE can determine that AI / ML model reselection is not required.

[0227] • RSRP: For example, if the difference between the current RSRP (e.g., the L1-RSRP of the configured RS) and the average RSRP over the pre-configured measurement window If a pre-configured threshold is set, the UE can determine that AI / ML model reselection is necessary. Otherwise, the UE can determine that AI / ML model reselection is not required.

[0228] • LoS probability: For example, if the difference between the current LoS probability and the average LoS probability over the pre-configured measurement window is higher than a threshold, the UE can determine that the AI / ML model needs to be reselected.

[0229] When it is determined that an AI / ML model reselection is needed, in some implementations, such as when the selection of a new model is performed by the base station or gNB, application server, AI / ML server, or other network nodes, the UE may indicate to the gNB that a model reselection is needed. For this purpose, the UE may use one or a combination of the following embodiments.

[0230] In one embodiment, the UE can indicate to the gNB via PUCCH or PUSCH (MAC-CE) whether an AI / ML model reselection is required. For example, the UE can indicate the need for AI / ML model reselection via a single-bit transmission, where a bit value "1" indicates that AI / ML model reselection is required, and a bit value "0" indicates that AI / ML model reselection is not required.

[0231] In another embodiment, in step 418, the UE may indicate parameters, measurements, or soft information (e.g., the difference between the current or instantaneous RSRP and the average RSRP within a configured time window, the current RSRP, the current Loss condition) and its decision regarding the need for AI / ML model reselection. For example, if the indicated information (e.g., RSRP) is less than a threshold or the indicated information (e.g., path loss > a threshold), the UE may implicitly indicate the need for model reselection.

[0232] In another embodiment, the UE may be configured with one or more preambles. Each preamble corresponds to one or more pre-configured conditions that need to be met to determine the need for AI / ML model reselection (e.g., the L1-RSRP of the configured RS falls below a pre-configured threshold, the difference between the instantaneous L1-RSRP and the average L1-RSRP over the configured time window exceeds a pre-configured threshold, a change in the Loss condition, etc.). When it is detected in step 408 that an AI / ML model reselection is needed, in some embodiments, in step 418, the UE may send a preamble corresponding to one or more conditions used to determine that an AI / ML model reselection is needed.

[0233] In one embodiment, a UE's indication of a need to reselect an AI / ML model can trigger an additional measurement process, which includes RS transmission to measure or estimate additional measurements (e.g., RSRP, velocity, location, SINR, etc.) to support the AI / ML model reselection process. One or more RS resources can be associated RSs based on the AI / ML configuration.

[0234] In one embodiment, at step 420, the UE may monitor a confirmation indication or configuration (e.g., immediately following an indication to the gNB that an AI / ML model needs to be reselected, via a PDCCH indication possibly in the associated CORESETs / SearchSpace within a pre-configured monitoring window, or via a MAC-CE indication). The UE may receive an indication or selection for a new AI / ML model to use at 420, and may switch to using the new model at step 422.

[0235] In another embodiment, the UE may at least perform an initial selection. For example, in step 410, the UE may identify one or more model types (e.g., CSI, BM, positioning, etc.) for reselection of the AI / ML model. The UE may be configured with one or more thresholds (e.g., via RRC, MAC CE, and DCI, or more). For example, if the measured quality is greater than the threshold, then in step 410, the UE may determine a first type for reselection (e.g., AI / ML model for CSI). Otherwise, in step 410, the UE may determine a second type of AI / ML model for reselection (e.g., AI / ML model for BM). The measured quality may be one or more of the following: RSRP.

[0236] · RSRQ.

[0237] · SINR.

[0238] • Hypothetical PDCCH / PUCCH BLER.

[0239] • Path loss.

[0240] For example, and briefly refer to Figure 5 The diagram 500 illustrates a logic diagram of an example implementation of dynamic model selection for wireless communication. While a specific set of logic is shown, other configurations may be utilized in other implementations. This logic may be executed as a series of gates as shown, or via a decision tree or other structure. In the illustrated implementation, the measurement performed in step 406 can be compared with a threshold in step 408 to determine if model reselection is needed. If no measurement is above the threshold, steps 406-408 can be repeated. If one or more measurements are above the threshold, the logic can be evaluated in step 410. For example, if the measured RSRP > a first threshold and the number of consecutive NACKs (or assumed PDCCH BLERs) > a second threshold, then in step 410, the UE can determine a first type of AI / ML model (e.g., reselecting the AI / ML model 502 for CSI). If the measured RSRP < a first threshold (possibly with the number of consecutive NACKs (or assumed PDCCH BLERs) > a second threshold), then in step 410, the UE can determine a second type of AI / ML model (e.g., reselecting the AI / ML model 504 for BM).

[0241] In another embodiment, if the measured path loss is less than a first threshold and the number of consecutive NACKs (or hypothetical PDCCHBLERs) is greater than a second threshold, then in step 410, the UE may determine a first type of AI / ML model (e.g., reselecting the AI / ML model for CSI 502). If the measured path loss is greater than the first threshold (possibly the number of consecutive NACKs (or hypothetical PDCCHBLERs) is greater than the second threshold), then in step 410, the UE may determine a second type of AI / ML model (e.g., reselecting the AI / ML model for BM 504).

[0242] In another embodiment, if the measured RSRP > a first threshold and the measured RSRQ (or SINR) < a second threshold, then in step 410, the UE may determine a first type of AI / ML model (e.g., reselecting an AI / ML model for CSI 502). If the measured RSRP < the first threshold and the measured RSRQ (or SINR) < the second threshold, then in step 410, the UE may determine a second type of AI / ML model (e.g., reselecting an AI / ML model for BM 504).

[0243] In another embodiment, if the number of failures of the measured assumed PDCCH BLER is less than a first threshold, then in step 410, the UE may determine a first type of AI / ML model (e.g., AI / ML model 502 for CSI reselection). Otherwise (e.g., one or more of the following: measured RSRP < second threshold, path loss > third threshold, measured RSRQ (or SINR) < fourth threshold, etc.), then in step 410, the UE may determine a second type of AI / ML model (e.g., AI / ML 504 for BM reselection).

[0244] return Figure 4 In one embodiment, the UE may be configured with a measurement type for AI / ML model type identification used for AI / ML model reselection. Based on the indicated measurement type, the UE can use that measurement type to perform AI / ML model type identification. For example, one or more of the following may be indicated to the UE: RSRP.

[0245] · RSRQ.

[0246] · SINR.

[0247] • Hypothetical PDCCH / PUCCH BLER.

[0248] • Path loss.

[0249] In one embodiment, in step 412, the UE may (e.g., to the gNB) indicate one or more determined AI / ML model types for AI / ML model reselection. This indication may be based on one or more of the following: • PUCCH: In some implementations, the UE can indicate the determined AI / ML model type by using a PUCCH. For example, one or more PUCCH resources can be used to configure / indicate the UE (e.g., via one or more of RRC, MAC CE, and DCI). If a PUCCH resource is configured for all AI / ML model types, the PUCCH resource can be used for all AI / ML model types. In this case, the UE can indicate the AI / ML model type ID as part of the UCI. If a PUCCH resource is configured for each AI / ML model type, the UE can indicate the AI / ML model type in the PUCCH resource associated with the determined AI / ML model type or the currently active AI / ML model type. PUCCH transmissions can be one or more of a scheduling request (SR), a HARQ ACK / NACK report, and a CSI report.

[0250] • PUSCH: In some implementations, the UE can use a PUSCH to indicate the determined AI / ML model type. For example, one or more PUSCH resources can be used to configure / indicate the UE (e.g., via RRC (e.g., configured authorization), MAC CE, and DCI (e.g., dynamic authorization)). If the UE receives indications for both dynamic authorization and configured authorization, the UE can prioritize one authorization. For example, the UE can prioritize earlier PUSCH resources. In another example, the UE can prioritize based on a first type (e.g., dynamic) instead of a second type (e.g., configured).

[0251] • PRACH: In some implementations, the UE can indicate the determined AI / ML model type by using PRACH. For example, one or more PRACH resources can be used to configure / indicate the UE (e.g., via one or more of RRC, MAC CE, and DCI). If a PRACH resource is configured for all AI / ML model types, that PRACH resource is available for all AI / ML model types. In this case, a PRACH sequence can be associated with each AI / ML model type. For example, if the UE can determine a first AI / ML model type for reselection, the UE can send a first PRACH sequence. If the UE can determine a second AI / ML model type for reselection, the UE can send a second PRACH sequence. If a PRACH resource is configured for each AI / ML model type, the UE can indicate the AI / ML model type in the PRACH resource associated with the determined AI / ML model type or the currently active AI / ML model type.

[0252] In an embodiment, a UE indication of one or more AI / ML model types can trigger an additional measurement process including RS transmission. For example, if a first type of AI / ML model is indicated, a first set of RS resources can be sent to the UE. If a second type of AI / ML model is indicated, a second set of RS resources can be sent to the UE. The association between RS resources can be based on one or more of the following: • Predefined RS resources: For example, first RS resources and second RS resources can be predefined for the first AI / ML model type and the second AI / ML model type, respectively.

[0253] • Configured RS resources: For example, a first RS resource can be configured as an associated RS resource for a first AI / ML model type, and a second RS resource can be configured as an associated RS resource for the first AI / ML model type.

[0254] • Dynamic indication: For example, the UE can receive indications for RS resources and can send the indications for the indicated AI / ML model type. For example, the UE can receive indications for one or more RS resources from a configured set of RS resources (e.g., via RRC and / or MAC CE).

[0255] In one embodiment, one or more parameters of the RS resource can be determined dynamically, while other parameters are predefined and / or configured. For example, parameters can be dynamically indicated (e.g., candidate parameters from one or more configurations, such as those via RRC and / or MAC CE). In another example, the UE can determine parameters based on the indicated AI / ML model type. For example, if a first AI / ML model type is indicated, the UE can determine a first parameter; if a second AI / ML model type is indicated, the UE can determine a second parameter. The one or more parameters can be one or more of the following: • RS resource offset or RS resource set offset (e.g., between the UE indication of AI / ML model type and RS resource / RS resource set).

[0256] • RS density.

[0257] • Number of repetitions.

[0258] • Periodicity.

[0259] • Transmission type (e.g., periodic, semi-persistent, or dynamic).

[0260] In some such embodiments, in step 412, the UE may send to the gNB the results of its determination of the AI / ML model for (re)selection (e.g., model ID) and / or parameters for identifying the AI / ML model (e.g., measurements, performance monitoring parameters, static parameters about the model, any other parameters, etc.). The UE may send an indication of the preferred model for (re)selection and / or related measurements / parameters to the gNB via any of the following message types: • RRC signaling and / or NAS messages (e.g., SRB0, SRB1, SRB2, SRB3, SRB4).

[0261] • UL MAC CE (e.g., existing MAC CE, new MAC CE, regular BSR, periodic BSR, filler BSR, enhanced BSR, preemptive BSR, etc.).

[0262] • UCI (e.g., single-bit SR, multi-bit SR, feedback, ACK / NACK, CSI report).

[0263] · PUCCH.

[0264] · PUSCH。

[0265] · Application layer signaling / messages.

[0266] In one embodiment, the UE identity for the AI / ML model used for (re-)selection is measurement-based. The UE can be configured with resources for making measurements. The UE can compare the measurements with a threshold received from the gNB. If the measurement is less than or greater than the threshold, the UE can determine that it needs an AI / ML model and a preferred AI / ML model suitable for the UE. The measurements made by the UE for selecting an AI / ML model can include one or more of the following: · L1 or L3 measurements such as RSRP, RSSI, RSRQ, SINR, RI, CQI, PMI, LI.

[0267] · Doppler, Doppler spread, delay spread, number of multipaths.

[0268] · Channel coherence time, channel coherence bandwidth.

[0269] · UE speed, location, direction of movement, velocity vector.

[0270] · Interference / path loss measurements.

[0271] · Number of ACK / NACKs.

[0272] · Beam measurements such as L1-RSRP, beam direction, beam width, beam ID, number of beam IDs, corresponding L1-RSRP of the beam, etc.

[0273] · Whether the path is line-of-sight (LOS) or non-line-of-sight (NLOS), or the LoS probability.

[0274] · BLER.

[0275] · Throughput.

[0276] For example, if the UE speed < X and the path loss > Y (e.g., indoor), the UE can determine Model #1. If the UE speed < X and the path loss < Y (e.g., walking outdoors), the UE can determine Model #2. If the UE speed > X and the path loss < Y (e.g., high-speed outdoor), the UE can determine Model #3. If the SINR > X and / or the LoS probability > Y (e.g., relatively good channel environment), the UE can determine Model #4. Otherwise, it is Model #5. If the coherence time > Model #6, otherwise it is Model #7.

[0277] In one embodiment, the UE can maintain a number / percentage of NACKs over a period of time. If the number / percentage of NACKs is greater than X, the UE can determine model #4. This period of time can be determined dynamically (e.g., a sliding window).

[0278] In one embodiment, the UE may send a preferred set of beam measurements (e.g., beam ID and corresponding L1-RSRP) to the gNB, and the gNB may determine whether model #5 is suitable for the UE based on the set of beam measurements.

[0279] In one embodiment, the UE may send information about its location and / or mobility (e.g., UE speed, location, direction of movement, etc.) to the gNB, which can then determine whether model #6 is suitable for the UE.

[0280] In one embodiment, the UE can measure the channel coherence time. A smaller value of the channel coherence time < X indicates a fast fading channel. The UE can then determine model #7 accordingly.

[0281] In one embodiment, the UE may determine a set of CSI parameters (e.g., for CSI reporting associated with an AI / ML model and / or an AI / ML model type). For example, if the UE determines a first AI / ML model, the UE may determine a first set of CSI parameters for UE reporting. If the UE determines a second AI / ML model, the UE may determine a second set of CSI parameters for UE reporting. For example, one or more of the following may be used.

[0282] Model #1: X beam IDs and corresponding L1-RSRP (N1) for the current measurement.

[0283] Model #2: X1-1 beam IDs and corresponding L1-RSRPs for the current measurement, and X1-2 beam IDs and corresponding L1-RSRPs for N1+N2.

[0284] Model #3: X2-1, X2-2, X2-3 and X2-4 beam IDs and corresponding L1-RSRPs for N1 / N2 / N3 / N4.

[0285] Alternatively, the UE may indicate a preferred set of CSI parameters to the gNB for future reporting. For example, set 1: X beam IDs and corresponding L1-RSRPs (N1) for the current measurement. Set 2: X1-1 beam IDs and corresponding L1-RSRPs for the current measurement, and X1-2 beam IDs and corresponding L1-RSRPs for N1+N2. Set 3: X2-1, X2-2, X2-3, and X2-4 beam IDs and corresponding L1-RSRPs for N1 / N2 / N3 / N4.

[0286] In another embodiment, in step 416, the UE can execute multiple AI / ML models and evaluate their performance (e.g., accuracy, specificity, selectivity, etc.). In some embodiments, the UE can identify the AI / ML model for (re)selection based on the functionality of the AI / ML models. The gNB can have several AI / ML models for different functions. The UE can identify the AI / ML model for (re)selection based on the functionality of the AI / ML models, for example: • CSI compression.

[0287] • CSI prediction.

[0288] • Beam prediction.

[0289] • Obstacle prediction.

[0290] • Location prediction.

[0291] • UE speed prediction.

[0292] •UE rotation prediction.

[0293] For example, a UE might want to perform CSI compression and therefore choose an AI / ML model for CSI compression.

[0294] In some embodiments, the UE identifier for the (re)selected AI / ML model may be based on other parameters. The UE may identify the AI / ML model for (re)selection based on one or more of the following parameters: • UE Region ID, UE Cell ID, TRP / gNB ID: For example, a UE can request a model that has already been trained (possibly by a neighboring UE) in the same region / cell / TRP / gNB.

[0295] • UE capabilities: For example, the antenna configurations supported by the UE, whether the UE supports TDD / FDD / full-duplex / half-duplex, etc. The UE can determine whether the AI / ML model is suitable for at least one of its capabilities, such as whether the model was trained using the same antenna configuration, whether the ML model was trained on a full-duplex or half-duplex configuration, etc.

[0296] · Model type: such as Recurrent Neural Network (RNN), Convolutional Neural Network (CNN), Deep Neural Network (DNN), Long Short-Term Memory (LSTM), unilateral model and bilateral model, etc.). If the AI / ML model uses a model type supported by the UE, the UE can determine whether the AI / ML model is appropriate. For example, the UE may only support unilateral models and may not be able to support a two-sided model with an encoder at the UE and a decoder at the gNB.

[0297] · UE service type: For example, if the AI / ML model is applicable to the UE's service type (such as periodic / aperiodic, burst start / end / duration, throughput, etc.), the UE can determine whether the AI / ML model is appropriate. The UE can select a model only if the model has been trained for a similar type of service.

[0298] · System parameters: such as subcarrier spacing, CP length, waveform type, carrier ID, bandwidth part ID, slot number, frame number, and time index.

[0299] In some embodiments, the UE identification of the AI / ML model for (re)selection is based on model performance monitoring. The UE can identify the model based on performance metrics (such as Normalized Mean Square Error (NMSE), cosine similarity, etc.). The NMSE / cosine similarity can be measured, for example, by considering the difference between the output of the ML model and the traditional calculation method. The accuracy of the ML model can be inversely proportional to the NMSE value (calculated at the UE or gNB). At the gNB, the UE can request the performance of the available models. In one example, the UE can only select models with a low NMSE < X. In one example, the UE can only select models with a high cosine similarity coefficient > X.

[0300] In some embodiments, the UE identification of the AI / ML model for (re)selection is based on static information about the model. The UE can identify the AI / ML model for (re)selection based on static parameters of the model. The UE can send the identified / preferred model to the gNB and / or send the original parameters for selection (for example, the UE can send a request for model < X, where X can be reported in megabytes / megabits / kilobytes, etc.). The static model parameters for model identification / selection can include any one or more of the following: · Model size.

[0301] · Training metrics.

[0302] · Overhead associated with the model.

[0303] · Delay for the model output result.

[0304] In one example, the UE can identify models with a size < X (where X can be measured in megabytes / megabits / kilobytes, etc.). The UE can send the ID of the selected model based on the model size and / or the value of X. In one example, the UE can identify a model based on the bandwidth required to download the model. The UE can send the ID of the selected model based on the available bandwidth, or report the available bandwidth to the gNB. In one example, the UE can identify the model for (re)selection based on training metrics such as training time, number of iterations to achieve convergence, amount of data required for training, retraining frequency, etc.

[0305] In one embodiment, the UE's identification of the model for (re)selection can be a single-stage or multi-stage process. The (re)selection process for identifying the model to download can be either single-stage or multi-stage. In one example, when sending a request for a model to the gNB (e.g., via model ID), the UE may receive an ACK for model download, or it may receive the model directly in the DL. In one example, when sending parameters for model identification, the UE may receive several options of models suitable for the UE based on the sent parameters. For example, if the UE reports the SINR to the gNB, it may receive several model options (e.g., model #3 and model #4) from the gNB that fit the SINR range indicated by the UE. The UE may also receive other parameters from the gNB (e.g., the antenna configuration on which the model is trained). The UE can then make a downward selection based on these parameters and send a request for a preferred AI / ML model to the gNB.

[0306] In one example, the UE may need a model for CSI prediction. The UE can request a CSI prediction model available at the gNB and can receive one or more options for models for CSI prediction from the gNB. The UE can also send some additional parameters to the gNB (e.g., UE antenna configuration, SINR range, UE coherence bandwidth, etc.), which can allow the gNB to make an initial selection from the available models and only send UE models that may be suitable for the UE (e.g., matching the UE antenna configuration and trained for the SINR range measured at the UE). In the first step, the UE can only receive metadata about the AI / ML models available at the gNB (e.g., model features, configuration, model size, parameters / conditions for training the model, etc.). Once all rounds of the selection process are completed (at the UE and / or gNB), the UE can receive the selected AI / ML model in the DL.

[0307] Here, the terms "accuracy," "effectiveness," "predictive accuracy," or "predictive effectiveness" for AI / ML models are used interchangeably. Beam resources can include TCI states, CSI-RS, or SSBs for downlink, and SRS resources or TCI states for uplink.

[0308] In one embodiment, in step 416, the UE can select one or more AI / ML models and determine which AI / ML model is activated. The activated AI / ML model can be picked, determined, and / or selected from an associated set of AI / ML models. The activated AI / ML model can be selected from a list of candidate AI / ML models and / or AI / ML model types that the UE has already identified for AI / ML model reselection. In one example, the UE can determine whether to evaluate the predictive accuracy of the activated AI / ML model.

[0309] In some embodiments, a method for evaluating the accuracy of an AI / ML model is disclosed. The UE can determine the accuracy of predictions made by the AI / ML model. The UE can identify or configure one or more resources to measure one or more parameters, thereby determining the accuracy of the AI / ML model. Measurement resources may be one or more beam resources, reference signals, and time and frequency resources used to perform the measurements.

[0310] In one example, the AI / ML model can use one or more beam resources from a first beam resource set to perform beam prediction and / or beam selection for one or more beam resources in a second beam resource set. The UE can be configured with measurement resources from both the first and second beam resource sets to determine the accuracy of the AI / ML model. In the example, the UE can be configured with periodic, semi-persistent, or aperiodic measurement resources in the second beam resource set to measure and compare the predicted beam resources at the output of the AI / ML model (e.g., this could be based on the first beam resource set).

[0311] The UE may determine the accuracy of the AI / ML model based on one or more of the following metrics: measurement, transmission performance, and / or requested reference signals, each of which is discussed below.

[0312] First, referencing measurements, in one embodiment, the UE can determine accuracy based on one or more measurement parameters and corresponding thresholds, wherein the one or more measurement parameters are based on one or more reference signals. The UE can measure at least one of the following: RSRP, RSSI, RSRQ, CQI, PMI, RI, LI, LOS probability, Doppler shift, Doppler spread, average delay, delay spread, etc. The UE can determine the accuracy of the AI / ML model (e.g., beam prediction) by comparing the measured parameters with the corresponding thresholds.

[0313] In another example, the UE can compare measurements based on AI / ML models (e.g., beam prediction) with measurements based on conventional methods (e.g., non-AI / ML based) (e.g., beam selection). One or more of the following can be applied.

[0314] The UE may perform a first measurement on one or more parameters of beam resources (e.g., RSRP of the best N beams) (e.g., from a second set), wherein the one or more parameters are predicted by an AI / ML model (e.g., based on the first set).

[0315] The UE can perform a second measurement on one or more parameters (e.g., RSRP of the best N beams) selected based on conventional methods (e.g., non-AI / ML based, such as beam selection), which are based on received and / or configured reference signals (e.g., based on a second set).

[0316] The UE can measure the difference between the first and second measurements (e.g., compare the RSRP of the best N beams).

[0317] The UE can compare the measured difference with a threshold.

[0318] If the measured difference is below a threshold, the UE can determine that the AI / ML model has acceptable accuracy and / or performance.

[0319] Otherwise, if the measured difference is higher than the threshold, the UE can determine that the AI / ML model does not have acceptable accuracy and / or performance.

[0320] Now, referring to transmission performance, for example, the UE can determine the accuracy of the AIML model based on transmission performance (e.g., for predicted beam resources). Transmission performance can be determined based on at least one of BLER, assumed PDCCH BLER, HARQ-ACK / NACK performance (e.g., the ratio of consecutive ACKs or consecutive NACKs), latency, data transmission throughput, spectral efficiency, etc.

[0321] In another example, the UE can determine the accuracy of the AIML model based on the performance of the transport function (e.g., for predicted beam resources). The performance of the transport function can be determined based on one or more of the following: the rate or number of beam failure detections, the rate or number of radio link failures, the rate or number of link adaptation performance and ping-pong effects (e.g., back-and-forth movement in the selected optimal beam resources, the selected CQI, PMI, etc.), persistent LBT failures, persistent interference (e.g., cross-link interference (CLI)), the rate or number of (continuous) cell (re)selections, the failure rate or number of random access procedures, etc.

[0322] The UE can be configured with one or more parameters to measure and compare corresponding thresholds to determine the accuracy of the AI / ML model (e.g., RSRP measurement of the best N beams and comparison with RSRP measurement of the M actual best beams, etc.).

[0323] The UE can be configured with one or more counters and / or timers, along with corresponding limits and / or thresholds, to measure and compare the number of consecutive failures, the ratio or number of trigger events, or the duration of a process or event. Therefore, if the counter or timer is below the limit, the UE can determine that the accuracy of the AI / ML model is acceptable. Otherwise, if the counter or timer is above the limit, the UE can determine that the accuracy of the AI / ML model is unacceptable.

[0324] The UE can now determine the validity of the AIML model by referring to the requested reference signals, for example, by requesting resources (e.g., DL reference signals). The UE can indicate to the gNB the required resource type, AIML model (e.g., AIML model index), and associated functionality. Thus, the UE can determine the accuracy of the AI / ML model based on the number of requested RS transmissions used for evaluation and / or the number of different RS transmissions used to activate the model. If the number of requested reference signals used for verification is below a corresponding limit, the UE can determine that the accuracy of the AI / ML model is acceptable. Otherwise, if the number of requested reference signals used for verification exceeds a corresponding limit, the UE can determine that the accuracy of the AI / ML model is unacceptable.

[0325] In other embodiments, the UE may compare the prediction accuracy of active AI / ML models. The UE may compare the accuracy of one or more active AI / ML models. The UE may determine an accuracy metric for the active AI / ML models. The UE may sort the active AI / ML models (e.g., in descending order) based on the determined accuracy metric. The accuracy metric may be based on one or more measurement parameters, reference signals, counters, and / or timers determined or configured for evaluating accuracy (e.g., individual accuracy metrics, joint optimization metrics, or a combination of one or more measurement parameters).

[0326] In step 410, following step 416, in such an embodiment, the UE may report the accuracy or the order of accuracy for the active AI / ML models. Alternatively, when the UE selects and reports a model among the active AI / ML models at steps 410 and 412, the UE may (implicitly) report the AI / ML model with the highest accuracy level.

[0327] In some embodiments, in step 414, the gNB may confirm the AI / ML model type and / or AI / ML model for reselection. The UE may be configured to receive from the gNB an indication and / or configuration associated with AI / ML model reselection at the UE. In one embodiment, the UE may receive the indication for AI / ML model reselection in the DCI. For example, the UE may receive an indication in the DCI instructing the UE to apply model reselection based on a previous UE report / recommendation. In another example, the UE may receive an indication in the DCI configuring the UE to apply a specific AI / ML model. In one embodiment, the UE may receive the indication for AI / ML model reselection in the MAC CE. For example, the indication may include the target model identity that the UE should apply. For example, the indication may include the functions / use cases to which the AI / ML model is applicable. For example, the functions / use cases may be associated with CSI feedback, beam management, etc. For example, the indication may include sub-use cases associated with the target model identity and the functions / use cases. For example, the sub-use cases may be associated with spatial prediction, temporal prediction, compression only, compression and prediction, etc. In another embodiment, the UE may receive the configuration for AI / ML model reselection in an RRC message. For example, a UE may receive multiple configuration sets, each of which may include parameters for the AI / ML model (e.g., model ID, model type, etc.), applicable functions / use cases, and optionally sub-use cases. For example, a configuration set may include parameterizations for functions / use cases / sub-use cases. For example, a configuration set for a CSI feedback use case may include CSI-RS resources, measurement and / or reporting configurations, etc. For example, a configuration set for a beam management use case may include measurement windows, prediction windows, set A and / or set B parameters, etc. In one solution, the MAC CE can semi-statically activate / deactivate applicable configuration sets. In another solution, the DCI can dynamically instruct the UE to apply configurations for AI / ML model reselection (possibly within the active configuration set).

[0328] In some embodiments, UE behavior / actions can be defined based on gNB indications. The UE can be configured to perform one or more actions based on AI / ML model reselection indications from the gNB. The UE can apply AI / ML model reselection indications from the gNB even if they do not correspond to UE recommendations for AI / ML model reselection. If the UE does not receive any indications from the gNB, it can continue using the current AI / ML model. In one embodiment, the indication from the gNB can indicate an AI / ML model different from the AI / ML model currently used at the UE. Upon receiving such an indication, the UE can deactivate the currently used AI / ML model and activate the indicated AI / ML model. It is possible that the UE can discard any feedback / reports generated via the deactivated model that have not yet been sent. In embodiments, the UE can receive additional configurations of use case-specific parameters (e.g., CSI compression / prediction configuration, beam management / prediction configuration). For example, the UE can be configured to apply use case-specific parameters after reselecting to a new AI / ML model.

[0329] In one embodiment, the AI / ML model reselection indicator may carry an implicit use-case-specific configuration. For example, the UE may be configured to infer the use-case-specific configuration from an AI / ML model ID (or its configuration). In another solution, the AI / ML model reselection indicator may carry an implicit AI / ML model ID (or its configuration). For example, the UE may be configured to infer the AI / ML model ID from a use-case-specific configuration.

[0330] In one embodiment, in step 422, the UE may be configured to immediately activate the AI / ML model and / or apply a new CSI and / or beam management configuration upon receiving a gNB indication. For example, upon reselection to the indicated AI / ML model for beam management, the UE may update set A and / or set B and / or observation window and / or prediction window configurations to suit the reselected AI / ML model. In one embodiment, the UE may be configured to activate the AI / ML model and / or apply a new CSI / beam management configuration within a time offset T relative to the reception timing of the gNB indication. In another solution, the UE may be configured to activate the AI / ML model and / or apply a new CSI / beam management configuration before the next CSI and / or beam reporting opportunity relative to the reception timing of the gNB indication. For example, the UE may be configured to report the next CSI and / or beam report using the reselected AI / ML model and / or its associated configuration. In yet another embodiment, the UE may be configured to activate the AI / ML model and / or apply a new CSI / beam management configuration before N CSI and / or beam reporting opportunities. For example, the value of N can be pre-configured. For example, the value of time offset T and / or the value of N can be pre-configured and / or signaled in the gNB indication for AI / ML model reselection. For example, the value of T and / or the value of N can be a function within the UE capability. For example, different values ​​of T and / or N can be configured for different functions / use cases / sub-use cases / AI / ML models / parameter configurations.

[0331] In one embodiment, the UE may be pre-configured with rules for activating AI / ML models and / or applying new CSI and / or beam management (BM) configurations based on the type and / or number of AI / ML models associated with the gNB indication. For example, if the UE is configured to reselect AI / ML models for both CSI and BM, the UE may first reselect the AI / ML model for BM, and upon successful reselection of the BM AI / ML model, the UE may reselect the AI / ML model for CSI. Here, successful reselection may include activating a new AI / ML model and / or applying a new configuration associated with the function (e.g., CSI / BM) and / or performing measurements using the new configuration and / or performing inference and / or transmitting reports associated with the function using the new AI / ML model.

[0332] Although the features and elements have been described above in specific combinations, those skilled in the art will understand that each feature or element can be used alone or in any combination with other features and elements. Furthermore, the methods described herein can be implemented in a computer program, software, or firmware incorporated into a computer-readable medium for execution by a computer or processor. Examples of computer-readable media include electronic signals (transmitted via wired or wireless connections) and computer-readable storage media. Examples of computer-readable storage media include, but are not limited to, read-only memory (ROM), random access memory (RAM), registers, cache memory, semiconductor memory devices, magnetic media such as internal hard disks and removable disks, magneto-optical media, and optical media such as CD-ROMs and digital multifunction discs (DVDs). The processor associated with the software can be used to implement a radio frequency transceiver used in a WTRU, UE, terminal, base station, RNC, or any host computer.

Claims

1. A user equipment (UE), comprising: One or more processors, which are configured as follows: Measure one or more communication channel conditions. The reselection of an artificial intelligence / machine learning (AI / ML) model is triggered based on at least a first criterion related to one or more measured communication channel conditions, and In response to a triggered reselection, the AI / ML model type for AI / ML model reselection is determined based on at least a second criterion related to one or more communication channel conditions measured; and One or more transceivers are configured as follows: In response to the triggered reselection, a message indicating that the AI / ML model should be reselected is sent to the gNB, and In response to receiving the message, one or more reference signals are received from the gNB; The one or more processors are further configured to: Measure the one or more reference signals, and The AI / ML model is selected based on one or more measured reference signals to determine the AI / ML model type; and The one or more transceivers are configured to transmit the identifier of the selected AI / ML model to the gNB.

2. The UE according to any one of the preceding claims, wherein the first standard comprises: The number of negative acknowledgments (NACKs) exceeds the first threshold or the received signal strength is lower than the second threshold.

3. The UE according to any one of the preceding claims, wherein the second standard is different from the first standard.

4. The UE according to any one of the preceding claims, wherein, The selection of the AI / ML model is also based on the reference signal received power (RSRP), the UE's speed, path loss, area identifier (ID), cell ID, or transmit / receive point (TRP) ID.

5. The UE according to any one of the preceding claims, characterized in that, The one or more transceivers are further configured to receive an acknowledgment message for AI / ML model reconfiguration from the gNB.

6. The UE according to any one of the preceding claims, wherein, The one or more processors are further configured to: In response to the triggered reselection, execute multiple AI / ML models of the determined AI / ML type; For each of the plurality of AI / ML models, evaluate the predictive accuracy of subsequent measurements of the one or more communication channel conditions; and Choose the AI / ML model with the highest predictive accuracy.

7. The UE of claim 6, wherein the one or more processors are further configured to execute the plurality of AI / ML models using the measured one or more communication channel conditions or a subset of subsequent measurements of the one or more communication channel conditions as model input.

8. A method comprising: One or more communication channel conditions are measured by one or more processors of the user equipment (UE); The reselection of an artificial intelligence / machine learning (AI / ML) model is triggered by the one or more processors based on at least a first criterion related to the measured one or more communication channel conditions; The AI / ML model type for AI / ML model reselection is determined by the one or more processors in response to the triggered reselection, based at least on a second criterion related to one or more communication channel conditions measured. In response to the triggered reselection, one or more transceivers of the UE send a message to the gNB indicating that the AI / ML model should be reselected; In response to receiving the message, the one or more transceivers receive one or more reference signals from the gNB; The one or more reference signals are measured by the one or more processors; The AI / ML model of the determined AI / ML model type is selected by the one or more processors based on one or more measured reference signals; and The identifier of the selected AI / ML model is transmitted from one or more transceivers to the gNB.

9. The method of claim 8, wherein the first standard comprises: The number of negative acknowledgments (NACKs) exceeds the first threshold or the received signal strength is lower than the second threshold.

10. The method according to claim 8 or 9, wherein the second standard is different from the first standard.

11. The method according to any one of claims 8 to 10, wherein, The selection of the AI / ML model is also based on the reference signal received power (RSRP), the UE's speed, path loss, area identifier (ID), cell ID, or transmit / receive point (TRP) ID.

12. The method according to any one of claims 8 to 11, further comprising receiving an acknowledgment message for AI / ML model reconfiguration from the gNB by the one or more transceivers.

13. The method according to any one of claims 8 to 12, further comprising: In response to the triggered reselection, the one or more processors execute multiple AI / ML models of the determined AI / ML type; For each of the plurality of AI / ML models, the one or more processors evaluate the predictive accuracy of subsequent measurements of the one or more communication channel conditions; as well as The AI / ML model with the highest prediction accuracy is selected by one or more processors.

14. The method of claim 13, further comprising: The plurality of AI / ML models are executed using one or more communication channel conditions or a subset of subsequent measurements of the one or more communication channel conditions as model input.

15. A wireless transmit / receive unit (WTRU) configured to perform the method according to any one of claims 8 to 14.

16. One or more processors configured to perform the method according to any one of claims 8 to 14.

17. A wireless circuit configured to perform the method according to any one of claims 8 to 14.

18. A network device configured to perform the method according to any one of claims 8 to 14.

19. An electronic device configured to perform the method according to any one of claims 8 to 14.

20. A non-transitory computer-readable medium comprising instructions that, when executed by one or more processors, cause a device to perform the method according to any one of claims 8 to 14.

Images