Machine learning in wireless communications

Abstract

Certain aspects of the present disclosure provide techniques for machine learning in wireless communications. An example method performed by a user equipment includes receiving an indication that reports channel state information (CSI) associated with a first machine learning model; receiving a reference signal associated with the CSI; and transmitting a CSI report associated with the reference signal based at least in part on a machine learning capability associated with the user equipment.

Description

Background Art

[0001] Aspects of the present disclosure relate to wireless communications, and more particularly, to techniques for machine learning in wireless communications.

[0002] Wireless communication systems are widely deployed to provide various telecommunication services, such as telephony, video, data, messaging, broadcasts, or other similar types of services. These wireless communication systems may employ multiple-access technologies capable of supporting communication with multiple users by sharing available wireless communication system resources with the multiple users.

[0003] Despite the tremendous technological advancements in wireless communication systems over the years, challenges remain. For example, complex and dynamic environments can still attenuate or block signals between wireless transmitters and wireless receivers. Consequently, there is a continuing desire to improve the technical performance of wireless communication systems, including, for example, improving the speed and data carrying capacity of communications, improving the efficiency of shared communication media usage, reducing the power used by transmitters and receivers when performing communications, improving the reliability of wireless communications, avoiding redundant transmission and / or reception and related processing, improving the coverage area of ​​wireless communications, increasing the number and types of devices that can access wireless communication systems, increasing the ability of different types of devices to communicate with each other, and increasing the number and types of wireless communication media available for use. Consequently, there is a need for further improvements in wireless communication systems to overcome the aforementioned technical challenges and others. Summary of the Invention

[0004] One aspect provides a method for wireless communication by a user equipment. The method includes receiving an indication to report channel state information (CSI) associated with a first machine learning model; receiving a reference signal associated with the CSI; and sending a CSI report associated with the reference signal based at least in part on machine learning capabilities associated with the user equipment.

[0005] Another aspect provides an apparatus for wireless communication. The apparatus includes a memory and a processor coupled to the memory. The processor is configured to: receive an indication to report channel state information (CSI) associated with a first machine learning model; receive a reference signal associated with the CSI; and send a CSI report associated with the reference signal based at least in part on machine learning capabilities associated with a user equipment.

[0006] Another aspect provides an apparatus for wireless communication. The apparatus includes: means for receiving an indication to report channel state information (CSI) associated with a first machine learning model; means for receiving a reference signal associated with the CSI; and means for sending a CSI report associated with the reference signal based at least in part on machine learning capabilities associated with a user equipment.

[0007] Another aspect provides a non-transitory computer-readable medium having instructions stored thereon for: receiving an indication to report channel state information (CSI) associated with a first machine learning model; receiving a reference signal associated with the CSI; and sending a CSI report associated with the reference signal based at least in part on a machine learning capability associated with the user equipment.

[0008] Another aspect provides a method for wireless communication by a user equipment. The method includes receiving an indication to report CSI associated with a first machine learning model; determining a set of one or more active machine learning models in response to receiving the indication; receiving a reference signal associated with the CSI; and sending a CSI report based on the reference signal in response to at least satisfying a timing constraint, wherein the timing constraint is based at least in part on the set of one or more active machine learning models.

[0009] Another aspect provides an apparatus for wireless communication. The apparatus includes a memory and a processor coupled to the memory. The processor is configured to: receive an indication to report CSI associated with a first machine learning model; determine a set of one or more active machine learning models in response to receiving the indication; receive a reference signal associated with the CSI; and send a CSI report based on the reference signal in response to at least satisfying a timing constraint, wherein the timing constraint is based at least in part on the set of one or more active machine learning models.

[0010] Another aspect provides an apparatus for wireless communication. The apparatus includes: means for receiving an indication to report CSI associated with a first machine learning model; means for determining a set of one or more active machine learning models in response to receiving the indication; means for receiving a reference signal associated with the CSI; and means for sending a CSI report based on the reference signal in response to at least satisfying a timing constraint, wherein the timing constraint is based at least in part on the set of one or more active machine learning models.

[0011] Another aspect provides a non-transitory computer-readable medium having instructions stored thereon for: receiving an indication to report CSI associated with a first machine learning model; determining a set of one or more active machine learning models in response to receiving the indication; receiving a reference signal associated with the CSI; and sending a CSI report based on the reference signal at least in response to satisfying a timing constraint, wherein the timing constraint is based at least in part on the set of one or more active machine learning models.

[0012] Another aspect provides a method for wireless communication by a user equipment. The method includes: receiving an indication to report information associated with a first machine learning model; determining information using the first machine learning model while concurrently determining other information using a second machine learning model in response to satisfying a machine learning processing constraint; and sending a report indicating the information.

[0013] Another aspect provides an apparatus for wireless communication. The apparatus includes a memory and a processor coupled to the memory. The processor is configured to: receive an instruction to report information associated with a first machine learning model; determine the information using the first machine learning model while concurrently determining other information using a second machine learning model in response to satisfying a machine learning processing constraint; and transmit a report indicating the information.

[0014] Another aspect provides an apparatus for wireless communication. The apparatus includes: means for receiving an indication to report information associated with a first machine learning model; means for determining the information using the first machine learning model while concurrently determining other information using a second machine learning model in response to satisfying a machine learning processing constraint; and means for sending a report indicating the information.

[0015] Another aspect provides a non-transitory computer-readable medium having stored thereon instructions for: receiving an indication to report information associated with a first machine learning model; determining information using the first machine learning model while concurrently determining other information using a second machine learning model in response to satisfying a machine learning processing constraint; and sending a report indicating the information.

[0016] In another aspect, a method for wireless communication by a network entity is provided. The method includes outputting an indication to report CSI associated with a first machine learning model; and obtaining a CSI report associated with a reference signal based at least in part on machine learning capabilities associated with a user equipment.

[0017] Another aspect provides an apparatus for wireless communication. The apparatus includes a memory and a processor coupled to the memory. The processor is configured to: output an indication to report CSI associated with a first machine learning model; and obtain a CSI report associated with a reference signal based at least in part on machine learning capabilities associated with a user equipment.

[0018] Another aspect provides an apparatus for wireless communication. The apparatus includes: means for outputting an indication reporting CSI associated with a first machine learning model; and means for obtaining a CSI report associated with a reference signal based at least in part on machine learning capabilities associated with a user equipment.

[0019] Another aspect provides a non-transitory computer-readable medium having instructions stored thereon for: outputting an indication of a CSI report associated with a first machine learning model; and obtaining a CSI report associated with a reference signal based at least in part on a machine learning capability associated with a user equipment.

[0020] Another aspect provides a method for wireless communication by a network entity. The method includes: outputting an indication to report CSI associated with a first machine learning model; determining a set of one or more active machine learning models in response to outputting the indication; outputting a reference signal associated with the CSI; and obtaining a CSI report based on the reference signal in response to at least satisfying a timing constraint, wherein the timing constraint is based at least in part on the set of one or more active machine learning models.

[0021] Another aspect provides an apparatus for wireless communication. The apparatus includes a memory and a processor coupled to the memory. The processor is configured to: output an indication to report CSI associated with a first machine learning model; determine a set of one or more active machine learning models in response to outputting the indication; output a reference signal associated with the CSI; and obtain a CSI report based on the reference signal in response to at least satisfying a timing constraint, wherein the timing constraint is based at least in part on the set of one or more active machine learning models.

[0022] Another aspect provides an apparatus for wireless communication. The apparatus includes: means for outputting an indication to report CSI associated with a first machine learning model; means for determining a set of one or more active machine learning models in response to outputting the indication; means for outputting a reference signal associated with the CSI; and means for obtaining a CSI report based on the reference signal in response to at least satisfying a timing constraint, wherein the timing constraint is based at least in part on the set of one or more active machine learning models.

[0023] Another aspect provides a non-transitory computer-readable medium having instructions stored thereon for: outputting an indication to report CSI associated with a first machine learning model; determining a set of one or more active machine learning models in response to outputting the indication; outputting a reference signal associated with the CSI; and obtaining a CSI report based on the reference signal at least in response to satisfying a timing constraint, wherein the timing constraint is based at least in part on the set of one or more active machine learning models.

[0024] Another aspect provides a method for wireless communication by a network entity. The method includes: outputting an indication reporting information associated with a first machine learning model; and obtaining a report indicating the information based on satisfying a machine learning processing constraint.

[0025] Another aspect provides an apparatus for wireless communication. The apparatus includes a memory and a processor coupled to the memory. The processor is configured to: output an indication reporting information associated with a first machine learning model; and obtain a report indicating the information based on satisfying a machine learning processing constraint.

[0026] Another aspect provides an apparatus for wireless communication. The apparatus includes: means for outputting an indication reporting information associated with a first machine learning model; and means for obtaining a report indicating the information based on satisfying a machine learning processing constraint.

[0027] Another aspect provides a non-transitory computer-readable medium having stored thereon instructions for: outputting an indication reporting information associated with a first machine learning model; and obtaining a report indicating the information based on satisfying a machine learning processing constraint.

[0028] Other aspects provide: an apparatus operable to, configured to, or otherwise adapted to perform any one or more of the aforementioned methods and / or those described elsewhere herein; a non-transitory computer-readable medium comprising instructions that, when executed by a processor of the apparatus, cause the apparatus to perform the aforementioned methods and those described elsewhere herein; a computer program product embodied on a computer-readable storage medium comprising code for performing the aforementioned methods and those described elsewhere herein; and / or an apparatus comprising components for performing the aforementioned methods and those described elsewhere herein. By way of example, an apparatus may comprise a processing system, a device having a processing system, or processing systems cooperating through one or more networks.

[0029] For purposes of illustration, the following description and drawings set forth certain features. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The accompanying drawings depict certain features of the aspects described herein and should not be considered as limiting the scope of the disclosure.

[0031] Figure 1 An example wireless communication network is depicted.

[0032] Figure 2 An example disaggregated base station architecture is depicted.

[0033] Figure 3 Aspects of an example base station and example user equipment are depicted.

[0034] Figure 4A 、 Figure 4B 、 Figure 4C and Figure 4DVarious example aspects of data structures for a wireless communication network are described.

[0035] Figure 5A is a timing diagram illustrating example timing constraints for aperiodic channel state information (CSI).

[0036] Figure 5B is a timing diagram illustrating example timing constraints for periodic / semi-persistent CSI.

[0037] Figure 6 An example networking environment is illustrated in which a prediction model is used to determine CSI or channel estimates.

[0038] Figure 7 is a diagram of an example wireless communication network illustrating the use of a machine learning model.

[0039] Figure 8 This is a signaling flow illustrating an example of user equipment loading a machine learning model.

[0040] Figure 9 is a table illustrating example machine learning capability levels that may be supported by user equipment.

[0041] 10A to 10C It's about Figure 9 Figure 1.2 shows an example of machine learning capability levels 1.0, 1.1, and 1.2.

[0042] Figures 11A to 11C It's about Figure 9 Figure 2 shows an example of machine learning capability levels 2.0, 2.1, and 2.2.

[0043] 12A to 12C is a diagram illustrating various timelines associated with machine learning capability levels 1.0, 1.1, and 1.2.

[0044] 13A to 13C is a diagram illustrating various timelines associated with machine learning capability levels 2.0, 2.1, and 2.2.

[0045] Figure 14 is a timing diagram illustrating an example of updating the model state over time in response to a non-periodic CSI trigger and a corresponding timeline for processing CSI using a machine learning model.

[0046] Figure 15 is a timing diagram illustrating an example of using a machine learning model to handle timing constraints for periodic and semi-persistent CSI.

[0047] Figure 16 is a timing diagram illustrating an example protection for back-to-back model switching.

[0048] Figure 17 Depicted is a process flow for communications in a network between a base station and user equipment.

[0049] Figure 18 A method for wireless communication is described.

[0050] Figure 19 A method for wireless communication is described.

[0051] Figure 20 A method for wireless communication is described.

[0052] Figure 21 A method for wireless communication is described.

[0053] Figure 22 A method for wireless communication is described.

[0054] Figure 23 A method for wireless communication is described.

[0055] Figure 24 Aspects of an example communication device are depicted.

[0056] Figure 25 Aspects of an example communication device are depicted. DETAILED DESCRIPTION

[0057] Aspects of the present disclosure provide apparatus, methods, processing systems, and computer-readable media for machine learning in wireless communications.

[0058] A wireless communication network (e.g., a 5G NR system or other system) may use channel state information (CSI) feedback from a user equipment (UE) for adaptive communication. The network may adjust certain communication parameters in response to CSI feedback from the UE. For example, link adaptation (such as adaptive modulation and coding) with various modulation schemes and channel coding rates may be applied to certain communication channels. For the purpose of channel state estimation, the UE may be configured to measure a reference signal (e.g., a CSI reference signal (CSI-RS)) and estimate the downlink channel state based on the CSI-RS measurement. The UE may report the estimated channel state to the network in the form of CSI, which may be used in link adaptation. The CSI may indicate the channel properties of the communication link between the base station (BS) and the UE. The CSI may represent, for example, the effects of scattering, fading, and path loss of the signal across the communication link.

[0059] In some cases, a machine learning model may be used to determine various information associated with a wireless communication network. For example, a machine learning model may be used to determine CSI, beam management, UE positioning, and / or channel estimation at a UE. For beam management, a machine learning model may determine a beam to be used in future transmission opportunities, for example, to proactively avoid beam failures. A machine learning model may determine the channel quality associated with a narrow beam based on measurements associated with a wide beam. In some cases, a machine learning model may determine fine resolution measurements associated with a beam based on coarse resolution measurements associated with the beam. In some cases, a UE may use a machine learning model to perform positioning, for example, for drone positioning and / or vehicle-to-everything (V2X) communications. A machine learning model may be used for channel estimation.

[0060] In any of these applications, the machine learning model may be designed for certain scenarios, such as urban microcells, urban macrocells, indoor hotspots, payload resolution, antenna architecture, etc. For example, the UE may use a particular machine learning model for a microcell and a different machine learning model for a macrocell. As another example, the UE may use a particular machine learning model for high-resolution CSI (e.g., a pre-decoder represented by a few hundred bits) and a different machine learning model for low-resolution CSI (e.g., a pre-decoder represented by two to twelve bits). In some cases, the UE may use a particular machine learning model for UE positioning and a different machine learning model for beam management. Due to different UE architectures, the UE may support various capabilities related to the use of machine learning models. For example, the UE may store different numbers of models in its modem and / or memory, and the UE may use different amounts of time to switch to using a model stored in the modem or to switch to using a model stored in the memory. There is uncertainty about how to handle different UE architectures in the radio access network.

[0061] Various aspects of the present disclosure provide apparatus and techniques for using machine learning models in wireless communications. A UE may support different categories of machine learning capabilities. As an example, the machine learning capability may represent a maximum number of machine learning models that the UE is capable of processing or storing in its modem, a maximum number of machine learning models that the UE is capable of storing in memory (e.g., non-volatile memory and / or random access memory), or a minimum amount of time for switching to using a machine learning model stored in memory or in the modem. Different timelines may be supported for processing CSI using machine learning models. For example, if the UE is switching between machine learning models stored in the UE's modem, a faster timeline may be used, and if the UE is activating a new machine learning model (e.g., downloading a machine learning model from memory or loading a machine learning model), a slower timeline may be used. Certain criteria for concurrent processing of machine learning models may be supported. For example, when the UE is using a machine learning model, the machine learning model may occupy a certain number of processing units for a certain duration. The UE may support a maximum number of processing units associated with the machine learning model, such that the UE is capable of processing multiple machine learning models concurrently, up to the maximum number of processing units.

[0062] The machine learning model processes described herein may enable improved wireless communication performance (e.g., higher throughput, lower latency, and / or spectral efficiency). For example, different UE architectures may support different timelines for processing CSI (and / or other information) using a machine learning model. Different categories of machine learning capabilities may allow a radio access network to dynamically configure a machine learning model to a UE in response to the UE's specific machine learning capabilities. Such dynamic configuration may allow a UE to process CSI using a machine learning model under various conditions, such as, for example, high latency, low latency, ultra-low latency, high resolution CSI, or low resolution CSI.

[0063] Introduction to wireless communication networks

[0064] The techniques and methods described herein can be used in various wireless communication networks. Although various aspects may be described herein using terms typically associated with 3G, 4G, and / or 5G wireless technologies, various aspects of the present disclosure may also be applicable to other communication systems and standards not explicitly mentioned herein, such as future wireless communication technologies.

[0065] Figure 1 An example of a wireless communication network 100 is depicted in which various aspects described herein may be implemented.

[0066] In general, the wireless communication network 100 includes various network entities (alternatively, network elements or network nodes). A network entity is typically a communication device and / or a communication function performed by a communication device (e.g., user equipment (UE), base station (BS), a component of a BS, a server, etc.). For example, various functions of a network and various devices associated with and interacting with the network can be considered network entities. In addition, the wireless communication network 100 includes terrestrial aspects, such as ground-based network entities (e.g., BS 102), and non-terrestrial aspects, such as satellites 140 and aircraft 145, which may include airborne network entities (e.g., one or more BSs) capable of communicating with other network elements (e.g., ground BSs) and user equipment.

[0067] In the depicted example, the wireless communication network 100 includes a BS 102, a UE 104, and one or more core networks (such as an evolved packet core (EPC) 160 and a 5G core (5GC) network 190) that interoperate to provide communication services over various communication links (including wired and wireless links).

[0068] Figure 1 Various example UEs 104 are depicted, which may more generally include: a cellular phone, a smartphone, a Session Initiation Protocol (SIP) phone, a laptop computer, a personal digital assistant (PDA), a satellite radio, a global positioning system, a multimedia device, a video device, a digital audio player, a camera, a game console, a tablet computer, a smart device, a wearable device, a vehicle, an electric meter, a gas pump, a large or small kitchen appliance, a healthcare device, an implant, a sensor / actuator, a display, an Internet of Things (IoT) device, an always-on (AON) device, an edge processing device, or other similar devices. A UE 104 may also be more generally referred to as a mobile device, a wireless device, a wireless communication device, a station, a mobile station, a subscriber station, a mobile subscriber station, a mobile unit, a subscriber unit, a wireless unit, a remote unit, a remote device, an access terminal, a mobile terminal, a wireless terminal, a remote terminal, a handset, and others.

[0069] BS 102 wirelessly communicates with (e.g., transmits signals to or receives signals from) UE 104 via communication link 120. Communication link 120 between BS 102 and UE 104 may include uplink (UL) (also known as a reverse link) transmissions from UE 104 to BS 102 and / or downlink (DL) (also known as a forward link) transmissions from BS 102 to UE 104. In various aspects, communication link 120 may utilize multiple-input multiple-output (MIMO) antenna technology, including spatial multiplexing, beamforming, and / or transmit diversity.

[0070] BSs 102 may generally include: a NodeB, an enhanced NodeB (eNB), a next generation enhanced NodeB (ng-eNB), a next generation NodeB (gNB or gNodeB), an access point, a base transceiver station, a radio base station, a radio transceiver, transceiver functionality, a transmit / receive point, and / or the like. Each of BSs 102 may provide communication coverage for a respective geographic coverage area 110, which may sometimes be referred to as a cell and may overlap in some cases (e.g., a small cell 102' may have a coverage area 110' that overlaps with the coverage area 110 of a macro cell). For example, a BS may provide communication coverage for a macro cell (covering a relatively large geographic area), a pico cell (covering a relatively small geographic area, such as a stadium), a femto cell (covering a relatively small geographic area, such as a home), and / or other types of cells.

[0071] Although BS102 is depicted in various aspects as a single communication device, BS102 can be implemented in various configurations. For example, one or more components of the base station can be disaggregated, including a central unit (CU), one or more distributed units (DUs), one or more radio units (RUs), a near real-time (near-RT) RAN intelligent controller (RIC), or a non-real-time (non-RT) RIC, to name a few examples. In another example, various aspects of the base station can be virtualized. More generally, a base station (e.g., BS102) can include components located at a single physical location or components located at various physical locations. In examples where the base station includes components located at various physical locations, the various components can each perform a function such that the various components collectively implement functions similar to a base station located at a single physical location. In some aspects, a base station including components located at various physical locations can be referred to as a disaggregated radio access network architecture (such as an open RAN (O-RAN) or virtualized RAN (VRAN) architecture). Figure 2 An example decomposed base station architecture is depicted and described.

[0072] Different BSs 102 within the wireless communication network 100 may also be configured to support different radio access technologies (such as 3G, 4G, and / or 5G). For example, a BS 102 configured for 4G LTE (collectively referred to as the Evolved Universal Mobile Telecommunications System (UMTS) Terrestrial Radio Access Network (E-UTRAN)) may interface with the EPC 160 via a first backhaul link 132 (e.g., an S1 interface). A BS 102 configured for 5G (e.g., 5G NR or Next Generation RAN (NG-RAN)) may interface with the 5GC 190 via a second backhaul link 184. The BSs 102 may communicate with each other directly or indirectly (e.g., through the EPC 160 or the 5GC 190) over a third backhaul link 134 (e.g., an X2 interface), which may be wired or wireless.

[0073] The wireless communication network 100 may subdivide the electromagnetic spectrum into various categories, frequency bands, channels, or other characteristics. In some aspects, the subdivision is provided based on wavelength and frequency, where frequency may also be referred to as a carrier, subcarrier, channel, tone, or subband. The communication link 120 between the BS 102 and, for example, the UE 104 may be through one or more carriers, which may have different bandwidths (e.g., 5 MHz, 10 MHz, 15 MHz, 20 MHz, 100 MHz, 400 MHz, and / or other MHz) and may be aggregated in various aspects. The carriers may or may not be adjacent to each other. The allocation of carriers may be asymmetric with respect to DL and UL (e.g., more or fewer carriers may be allocated for DL ​​compared to UL).

[0074] Communications using higher frequency bands may have higher path loss and shorter range than communications at lower frequencies. Figure 1 180) may utilize beamforming 182 with UE 104 to improve path loss and range. For example, BS 180 and UE 104 may each include multiple antennas, such as antenna elements, antenna panels, and / or antenna arrays, to facilitate beamforming. In some cases, BS 180 may transmit beamformed signals to UE 104 in one or more transmit directions 182′. UE 104 may receive beamformed signals from BS 180 in one or more receive directions 182″. UE 104 may also transmit beamformed signals to BS 180 in one or more transmit directions 182″. BS 180 may also receive beamformed signals from UE 104 in one or more receive directions 182′. BS 180 and UE 104 may then perform beam training to determine the best receive direction and transmit direction for each of BS 180 and UE 104. Notably, the transmit direction and receive direction of BS 180 may be the same or different. Similarly, the transmit direction and receive direction of UE 104 may be the same or different.

[0075] Wireless communication network 100 also includes a Wi-Fi AP 150 that communicates with Wi-Fi stations (STAs) 152 via communication links 154 in, for example, the 2.4 GHz and / or 5 GHz unlicensed spectrum.

[0076] Some of the UEs 104 may communicate with each other using a device-to-device (D2D) communication link 158. The D2D communication link 158 may use one or more sidelink channels, such as a physical sidelink broadcast channel (PSBCH), a physical sidelink discovery channel (PSDCH), a physical sidelink shared channel (PSSCH), a physical sidelink control channel (PSCCH), and / or a physical sidelink feedback channel (PSFCH).

[0077] The EPC 160 may include various functional components, including a mobility management entity (MME) 162, other MMEs 164, a serving gateway 166, a multimedia broadcast multicast service (MBMS) gateway 168, a broadcast multicast service center (BM-SC) 170, and / or a packet data network (PDN) gateway 172, such as in the depicted example. The MME 162 may communicate with a home subscriber server (HSS) 174. The MME 162 is a control node that handles signaling between the UE 104 and the EPC 160. Generally speaking, the MME 162 provides bearer and connection management.

[0078] Generally speaking, user Internet Protocol (IP) packets are delivered through the Serving Gateway 166, which itself is connected to the PDN Gateway 172. The PDN Gateway 172 provides UE IP address allocation and other functions. The PDN Gateway 172 and the BM-SC 170 are connected to the IP services 176, which may include, for example, the Internet, an intranet, an IP Multimedia Subsystem (IMS), packet-switched (PS) streaming services, and / or other IP services.

[0079] BM-SC 170 may provide functionality for configuring and delivering MBMS user services. BM-SC 170 may serve as the entry point for content providers' MBMS delivery, may be used to authorize and initiate MBMS bearer services within a public land mobile network (PLMN), and / or may be used to schedule MBMS delivery. MBMS Gateway 168 may be used to distribute MBMS services to BSs 102 belonging to a Multicast Broadcast Single Frequency Network (MBSFN) area broadcasting a specific service, and / or may be responsible for session management (start / stop) and collecting eMBMS-related charging information.

[0080] 5GC 190 may include various functional components, including: access and mobility management function (AMF) 192, other AMFs 193, session management function (SMF) 194, and user plane function (UPF) 195. AMF 192 may communicate with unified data management (UDM) 196.

[0081] AMF 192 is a control node that handles signaling between UE 104 and 5GC 190. AMF 192 provides, for example, Quality of Service (QoS) flow and session management.

[0082] Internet Protocol (IP) packets are passed through UPF 195, which connects to IP services 197 and provides IP address allocation for UEs and other functions for 5GC 190. IP services 197 may include, for example, the Internet, intranet, IMS, PS streaming services, and / or other IP services.

[0083] The wireless communication network 100 includes a machine learning component 199 that can perform operations related to machine learning timelines and / or machine learning concurrent processing as described herein. The wireless network 100 also includes a machine learning component 198 that can perform operations related to machine learning timelines and / or machine learning concurrent processing as described herein.

[0084] In various aspects, a network entity or network node may be implemented as a converged base station, a decomposed base station, a component of a base station, an integrated access and backhaul (IAB) node, a relay node, a sidelink node, to name a few examples.

[0085] Figure 2 An example disaggregated base station 200 architecture is depicted. The disaggregated base station 200 architecture may include one or more central units (CUs) 210 that may communicate directly with a core network 220 via a backhaul link, or indirectly with the core network 220 through one or more disaggregated base station units, such as a near real-time (near-RT) RAN intelligent controller (RIC) 225 via an E2 link, or a non-real-time (non-RT) RIC 215 associated with a service management and orchestration (SMO) framework 205, or both. The CUs 210 may communicate with one or more distributed units (DUs) 230 via corresponding midhaul links, such as an F1 interface. The DUs 230 may communicate with one or more radio units (RUs) 240 via corresponding fronthaul links. The RUs 240 may communicate with corresponding UEs 104 via one or more radio frequency (RF) access links. In some implementations, a UE 104 may be served simultaneously by multiple RUs 240.

[0086] Each of the units (e.g., CU 210, DU 230, RU 240, and near-RT RIC 225, non-RT RIC 215, and SMO framework 205) may include or be coupled to one or more interfaces configured to receive or transmit signals, data, or information (collectively, signals) via a wired or wireless transmission medium. Each of the units, or an associated processor or controller that provides instructions to the communication interface of the unit, may be configured to communicate with one or more of the other units via the transmission medium. For example, the units may include a wired interface configured to receive or transmit signals to one or more of the other units via the wired transmission medium. Additionally or alternatively, the units may include a wireless interface that may include a receiver, transmitter, or transceiver (such as a radio frequency (RF) transceiver) configured to receive signals or transmit signals to one or more of the other units over the wireless transmission medium, or both.

[0087] In some aspects, the CU 210 may host one or more higher layer control functions. Such control functions may include radio resource control (RRC), packet data convergence protocol (PDCP), service data adaptation protocol (SDAP), etc. Each control function may be implemented using an interface that is configured to communicate signals with other control functions hosted by the CU 210. The CU 210 may be configured to handle user plane functionality (e.g., central unit-user plane (CU-UP)), control plane functionality (e.g., central unit-control plane (CU-CP)), or a combination thereof. In some specific implementations, the CU 210 may be logically split into one or more CU-UP units and one or more CU-CP units. When implemented in an O-RAN configuration, the CU-UP unit may communicate bidirectionally with the CU-CP unit via an interface (such as an E1 interface). As needed, the CU 210 may be implemented to communicate with the DU 230 for network control and signaling.

[0088] The DU 230 may correspond to a logical unit that includes one or more base station functions for controlling the operation of one or more RUs 240. In some aspects, the DU 230 may host one or more of a radio link control (RLC) layer, a medium access control (MAC) layer, and one or more higher physical (PHY) layers (such as modules for forward error correction (FEC) encoding and decoding, scrambling, modulation and demodulation, etc.) based at least in part on a functional split, such as those defined by the 3rd Generation Partnership Project (3GPP). In some aspects, the DU 230 may further host one or more lower PHY layers. Each layer (or module) may be implemented using an interface configured to communicate signals with other layers (and modules) hosted by the DU 230 or with control functions hosted by the CU 210.

[0089] Lower layer functionality may be implemented by one or more RUs 240. In some deployments, a RU 240 controlled by a DU 230 may correspond to a logical node that hosts RF processing functionality or low PHY layer functionality (such as performing Fast Fourier Transform (FFT), Inverse FFT (iFFT), digital beamforming, Physical Random Access Channel (PRACH) extraction and filtering, etc.), or both, based at least in part on a functional split, such as a lower layer functional split. In such an architecture, the RU 240 may be implemented to handle over-the-air (OTA) communications with one or more UEs 104. In some implementations, real-time and non-real-time aspects of control and user plane communications with the RU 240 may be controlled by the corresponding DU 230. In some scenarios, this configuration may enable the implementation of the DU 230 and CU 210 in a cloud-based RAN architecture, such as a vRAN architecture.

[0090] The SMO framework 205 can be configured to support RAN deployment and provisioning of both non-virtualized and virtualized network elements. For non-virtualized network elements, the SMO framework 205 can be configured to support the deployment of dedicated physical resources for RAN coverage requirements, which can be managed via an operations and maintenance interface (such as the O1 interface). For virtualized network elements, the SMO framework 205 can be configured to interact with a cloud computing platform (such as Open Cloud (O-Cloud) 290) to perform network element lifecycle management (such as instantiating virtualized network elements) via a cloud computing platform interface (such as the O2 interface). Such virtualized network elements may include, but are not limited to, the CU 210, DU 230, RU 240, and near-RT RIC 225. In some implementations, the SMO framework 205 can communicate with hardware aspects of the 4G RAN (such as Open eNB (O-eNB) 211) via the O1 interface. Additionally, in some implementations, the SMO framework 205 can communicate directly with one or more RUs 240 via the O1 interface. The SMO framework 205 may also include a non-RT RIC 215 configured to support the functionality of the SMO framework 205 .

[0091] The non-RT RIC 215 may be configured to include logic that enables non-real-time control and optimization of RAN elements and resources, artificial intelligence / machine learning (AI / ML) workflows including model training and updating, or policy-based guidance of applications / features in the near-RT RIC 225. The non-RT RIC 215 may be coupled to or in communication with the near-RT RIC 225 (e.g., via an A1 interface). The near-RT RIC 225 may be configured to include logic that enables near-real-time control and optimization of RAN elements and resources through data collection and actions over an interface (e.g., via an E2 interface) that connects one or more CUs 210, one or more DUs 230, or both, and the O-eNB with the near-RT RIC 225.

[0092] In some implementations, the non-RT RIC 215 can receive parameters or external enrichment information from an external server in order to generate an AI / ML model to be deployed in the near-RT RIC 225. Such information can be utilized by the near-RT RIC 225 and can be received from non-network data sources or from network functions at the SMO framework 205 or the non-RT RIC 215. In some examples, the non-RT RIC 215 or the near-RT RIC 225 can be configured to tune RAN behavior or performance. For example, the non-RT RIC 215 can monitor long-term trends and patterns in performance and employ AI / ML models to perform corrective actions through the SMO framework 205 (such as via reconfiguration of O1) or by creating RAN management policies (such as A1 policies).

[0093] Figure 3 Aspects of an example BS 102 and UE 104 are depicted.

[0094] In general, BS 102 includes various processors (e.g., 320, 330, 338, and 340), antennas 334a-334t (collectively 334), transceivers 332a-332t (collectively 332) including modulators and demodulators, and other aspects that enable wireless transmission of data (e.g., data source 312) and wireless reception of data (e.g., data sink 339). For example, BS 102 can transmit and receive data between BS 102 and UE 104. BS 102 includes a controller / processor 340 that can be configured to implement various functions related to wireless communication as described herein.

[0095] Base station 102 includes a controller / processor 340 that can be configured to perform various functions related to wireless communication. In the depicted example, controller / processor 340 includes a machine learning component 241 that can represent Figure 1 199 of the machine learning component. Notably, while depicted as one aspect of the controller / processor 340, in other implementations, the machine learning component 341 may additionally or alternatively be implemented in various other aspects of the base station 102.

[0096] In general, the UE 104 includes various processors (e.g., 358, 364, 366, and 380), antennas 352a-352r (collectively 352), transceivers 354a-354r (collectively 354) including modulators and demodulators, and other aspects that enable wireless transmission of data (e.g., retrieved from a data source 362) and wireless reception of data (e.g., provided to a data sink 360). The UE 104 includes a controller / processor 380 that can be configured to implement various functions described herein related to wireless communications.

[0097] The user equipment 104 includes a controller / processor 380 that can be configured to perform various functions related to wireless communication. In the depicted example, the controller / processor 380 includes a machine learning component 381 that can represent Figure 1 198 of the machine learning component. Notably, while depicted as one aspect of the controller / processor 380, in other implementations, the machine learning component 381 may additionally or alternatively be implemented in various other aspects of the user equipment 104.

[0098] Regarding example downlink transmissions, BS 102 includes a transmit processor 320 that can receive data from a data source 312 and control information from a controller / processor 340. The control information can be for a physical broadcast channel (PBCH), a physical control format indicator channel (PCFICH), a physical HARQ indicator channel (PHICH), a physical downlink control channel (PDCCH), a group common PDCCH (GC PDCCH), and / or other. In some examples, the data can be for a physical downlink shared channel (PDSCH).

[0099] The transmit processor 320 may process (e.g., encode and symbol map) data and control information to obtain data symbols and control symbols, respectively. The transmit processor 320 may also generate reference symbols (such as for a primary synchronization signal (PSS), a secondary synchronization signal (SSS), a PBCH demodulation reference signal (DMRS), and a channel state information reference signal (CSI-RS)).

[0100] The transmit (TX) multiple-input multiple-output (MIMO) processor 330 may perform spatial processing (e.g., precoding) on ​​data symbols, control symbols, and / or reference symbols, as applicable, and may provide an output symbol stream to a modulator (MOD) in transceivers 332a-332t. Each modulator in transceivers 332a-332t may process a corresponding output symbol stream to obtain an output sample stream. Each modulator may further process (e.g., convert to analog, amplify, filter, and frequency upconvert) the output sample stream to obtain a downlink signal. The downlink signals from the modulators in transceivers 332a-332t may be transmitted via antennas 334a-334t, respectively.

[0101] To receive downlink transmissions, UE 104 includes antennas 352a-352r that can receive downlink signals from BS 102 and provide received signals to demodulators (DEMODs) in transceivers 354a-354r, respectively. Each demodulator in transceivers 354a-354r can condition (e.g., filter, amplify, downconvert, and digitize) a corresponding received signal to obtain input samples. Each demodulator can further process the input samples to obtain received symbols.

[0102] A MIMO detector 356 may obtain received symbols from all demodulators in transceivers 354a-354r, perform MIMO detection on the received symbols if applicable, and provide detected symbols. A receive processor 358 may process (e.g., demodulate, deinterleave, and decode) the detected symbols, provide decoded data for UE 104 to a data sink 360, and provide decoded control information to a controller / processor 380.

[0103] With respect to example uplink transmissions, the UE 104 also includes a transmit processor 364 that can receive and process data from a data source 362 (e.g., for a PUSCH) and control information from a controller / processor 380 (e.g., for a physical uplink control channel (PUCCH)). The transmit processor 364 can also generate reference symbols for reference signals (e.g., for a sounding reference signal (SRS)). The symbols from the transmit processor 364 can be pre-decoded by a TX MIMO processor 366, if applicable, further processed by a modulator in the transceivers 354a-354r (e.g., for SC-FDM), and transmitted to the BS 102.

[0104] At BS 102, uplink signals from UE 104 may be received by antennas 334 a-334 t, processed by demodulators in transceivers 332 a-332 t, detected by MIMO detector 336 if applicable, and further processed by a receive processor 338 to obtain decoded data and control information transmitted by UE 104. Receive processor 338 may provide decoded data to a data sink 339 and decoded control information to a controller / processor 340.

[0105] Memories 342 and 382 may store data and program codes for BS 102 and UE 104, respectively.

[0106] A scheduler 344 may schedule UEs for data transmission on the downlink and / or uplink.

[0107] In various aspects, the BS 102 may be described as transmitting and receiving various types of data associated with the methods described herein. In these contexts, "transmitting" may refer to various mechanisms for outputting data, such as from a data source 312, a scheduler 344, a memory 342, a transmit processor 320, a controller / processor 340, a TX MIMO processor 330, transceivers 332a-332t, antennas 334a-334t, and / or other aspects described herein. Similarly, "receiving" may refer to various mechanisms for obtaining data, such as from antennas 334a-334t, transceivers 332a-332t, an RX MIMO detector 336, a controller / processor 340, a receive processor 338, a scheduler 344, a memory 342, and / or other aspects described herein.

[0108] In various aspects, the UE 104 may also be described as transmitting and receiving various types of data associated with the methods described herein. In these contexts, "transmitting" may refer to various mechanisms for outputting data, such as from a data source 362, memory 382, ​​a transmit processor 364, a controller / processor 380, a TX MIMO processor 366, transceivers 354a-354t, antennas 352a-352t, and / or other aspects described herein. Similarly, "receiving" may refer to various mechanisms for obtaining data, such as from antennas 352a-352t, transceivers 354a-354t, an RX MIMO detector 356, a controller / processor 380, a receive processor 358, memory 382, ​​and / or other aspects described herein.

[0109] In some aspects, the processor may be configured to perform various operations, such as those associated with the methods described herein, and send (output) data to or receive (obtain) data from another interface configured to send or receive data, respectively.

[0110] Figure 4A 、 Figure 4B 、 Figure 4C and Figure 4D Describes a method for use in wireless communication networks such as Figure 1 Various aspects of the data structure of the wireless communication network 100).

[0111] Specifically, Figure 4A is a diagram 400 illustrating an example of a first subframe within a 5G (e.g., 5G NR) frame structure, Figure 4B is a diagram 430 illustrating an example of a DL channel within a 5G subframe, Figure 4C is a diagram 450 illustrating an example of a second subframe within a 5G frame structure, and Figure 4D FIG480 is a diagram illustrating an example of UL channels within a 5G subframe.

[0112] Wireless communication networks can utilize Orthogonal Frequency Division Multiplexing (OFDM) with a Cyclic Prefix (CP) on both the uplink and downlink. Such systems can also support half-duplex operation using Time Division Duplex (TDD). OFDM and Single Carrier Frequency Division Multiplexing (SC-FDM) can be used to maximize the system bandwidth (e.g., Figure 4B and Figure 4D The RF spectrum is divided into multiple orthogonal subcarriers (as depicted in FIG). Each subcarrier can be modulated with data. Modulation symbols can be transmitted in the frequency domain using OFDM and / or in the time domain using SC-FDM.

[0113] The wireless communication frame structure may be frequency division duplex (FDD), where for a particular set of subcarriers, subframes within the set of subcarriers are dedicated to either DL or UL. The wireless communication frame structure may also be time division duplex (TDD), where for a particular set of subcarriers, subframes within the set of subcarriers are dedicated to both DL and UL.

[0114] exist Figure 4A and Figure 4C In the embodiment of the present invention, the wireless communication frame structure is TDD, where D is DL, U is UL, and X is flexibly used between DL / UL. The UE can be configured with a time slot format (dynamically configured by DL control information (DCI) or semi-statically / statically configured by radio resource control (RRC) signaling) through the received time slot format indicator (SFI). In the depicted example, the 10ms frame is divided into 10 equally sized 1ms subframes. Each subframe may include one or more time slots. In some examples, each time slot may include 7 or 14 symbols, depending on the time slot format. The subframe may also include micro slots, which generally have fewer symbols than a whole time slot. Other wireless communication technologies may have different frame structures and / or different channels.

[0115] In certain aspects, the number of slots within a subframe is based on the slot configuration and parameter set. For example, for slot configuration 0, different parameter sets (μ) 0 through 5 allow for 1, 2, 4, 8, 16, and 32 slots per subframe, respectively. For slot configuration 1, different parameter sets 0 through 2 allow for 2, 4, and 8 slots per subframe, respectively. Thus, for slot configuration 0 and parameter set μ, there are 14 symbols per slot and 2μ slots per subframe. The subcarrier spacing and symbol length / duration are a function of the parameter set. The subcarrier spacing may be equal to 2 μ × 15kHz, where μ is parameter set 0 to 5. Thus, parameter set μ=0 has a subcarrier spacing of 15kHz, and parameter set μ=5 has a subcarrier spacing of 480kHz. Symbol length / duration is inversely related to subcarrier spacing. Figure 4A 、 Figure 4B 、 Figure 4Cand Figure 4D An example is provided for slot configuration 0 with 14 symbols per slot and parameter set μ = 2 with 4 slots per subframe. The slot duration is 0.25 ms, the subcarrier spacing is 60 kHz, and the symbol duration is approximately 16.67 μs.

[0116] like Figure 4A 、 Figure 4B 、 Figure 4C and Figure 4D As depicted in FIG, a resource grid can be used to represent the frame structure. Each time slot includes a resource block (RB) (also called a physical RB (PRB)) extending over, for example, 12 consecutive subcarriers. The resource grid is divided into a plurality of resource elements (REs). The number of bits carried by each RE depends on the modulation scheme.

[0117] like Figure 4A As illustrated, some REs carry Figure 1 and Figure 3 The RS may include a demodulation RS (DMRS) and / or a channel state information reference signal (CSI-RS) for channel estimation at the UE. The RS may also include a beamforming RS (BRS), a beam refinement RS (BRRS), and / or a phase tracking RS (PT-RS).

[0118] Figure 4B Examples of various DL channels within a subframe of a frame are illustrated. The Physical Downlink Control Channel (PDCCH) carries DCI within one or more Control Channel Elements (CCEs), each CCE comprising, for example, nine RE Groups (REGs), each REG comprising, for example, four consecutive REs in an OFDM symbol.

[0119] The Primary Synchronization Signal (PSS) may be in symbol 2 of a particular subframe of a frame. The PSS is transmitted by a UE (e.g., Figure 1 and Figure 3 104) is used to determine subframe / symbol timing and physical layer identification.

[0120] A Secondary Synchronization Signal (SSS) may be within symbol 4 of a particular subframe of a frame. The SSS is used by the UE to determine the physical layer cell identity group number and radio frame timing.

[0121] Based on the physical layer identifier and the physical layer cell identifier group number, the UE can determine the physical cell identifier (PCI). Based on the PCI, the UE can determine the location of the aforementioned DMRS. The physical broadcast channel (PBCH) carrying the master information block (MIB) can be logically grouped with the PSS and SSS to form a synchronization signal (SS) / PBCH block. The MIB provides the system frame number (SFN) and the number of RBs in the system bandwidth. The physical downlink shared channel (PDSCH) carries user data, broadcast system information not sent via the PBCH (such as the system information block (SIB)), and / or paging messages.

[0122] As in Figure 4C As illustrated in , some REs carry DMRS for channel estimation at the base station (indicated as R for a specific configuration, but other DMRS configurations are possible). The UE may send DMRS for PUCCH and DMRS for PUSCH. The PUSCH DMRS may be sent, for example, in the first or first two symbols of the PUSCH. The PUCCH DMRS may be sent in different configurations depending on whether a short PUCCH or a long PUCCH is sent and depending on the specific PUCCH format used. The UE 104 may send a sounding reference signal (SRS). The SRS may be sent, for example, in the last symbol of the subframe. The SRS may have a comb structure, and the UE may send the SRS on one of the teeth of the comb. The SRS may be used by the base station for channel quality estimation to achieve frequency-dependent scheduling of the UL.

[0123] Figure 4D Examples of various UL channels within a subframe of a frame are illustrated. The PUCCH may be located at the position indicated in one configuration. The PUCCH carries uplink control information (UCI), such as scheduling requests, channel quality indicators (CQI), precoding matrix indicators (PMI), rank indicators (RI), and HARQ ACK / NACK feedback. The PUSCH carries data and may additionally be used to carry buffer status reports (BSRs), power headroom reports (PHRs), and / or UCI.

[0124] Introduction to mmWave Wireless Communications

[0125] In wireless communications, the electromagnetic spectrum is often subdivided into various categories, bands, channels, or other characteristics. The subdivision is typically provided based on wavelength and frequency, where frequency may also be referred to as a carrier, subcarrier, channel, tone, or subband.

[0126] In 5G NR, two initial operating frequency bands have been identified as frequency range designations FR1 (410 MHz - 7.125 GHz) and FR2 (24.25 GHz - 52.6 GHz). It should be understood that although a portion of FR1 is greater than 6 GHz, FR1 is often (interchangeably) referred to as the "sub-6 GHz" band in various documents and articles. A similar naming issue sometimes occurs with respect to FR2, which is often (interchangeably) referred to as the "millimeter wave" band in documents and articles, although it is different from the extremely high frequency (EHF) band (30 GHz to 300 GHz) identified as the "millimeter wave" band by the International Telecommunication Union (ITU).

[0127] Frequencies between FR1 and FR2 are generally referred to as mid-band frequencies. Recent 5G NR research has identified the operating bands for these mid-band frequencies as frequency range designation FR3 (7.125 GHz to 24.25 GHz). The frequency bands falling within FR3 can inherit FR1 characteristics and / or FR2 characteristics, and therefore can effectively extend the features of FR1 and / or FR2 to mid-band frequencies. In addition, higher frequency bands are currently being explored to extend 5G NR operation to more than 52.6 GHz. For example, three higher operating bands have been identified as frequency range designations FR4a or FR4-1 (52.6 GHz to 71 GHz), FR4 (52.6 GHz to 114.25 GHz), and FR5 (114.25 GHz to 300 GHz). Each of these higher frequency bands falls within the EHF band.

[0128] In view of the above aspects, unless otherwise specified, it should be understood that if the term "sub-6 GHz" or the like is used herein, it can be broadly referred to as a frequency that may be less than 6 GHz, may be within FR1, or may include mid-band frequencies. In addition, unless otherwise specified, it should be understood that if the term "millimeter wave" or the like is used herein, it can be broadly referred to as a frequency that may be less than 6 GHz, may be within FR1, or may include mid-band frequencies. In addition, unless otherwise specified, it should be understood that if the term "millimeter wave" or the like is used herein, it can be broadly referred to as a frequency that may include mid-band frequencies, may be within FR2, FR4, FR4-a or FR4-1 and / or FR5, or may be within the EHF band.

[0129] Communications using mmWave / near mmWave radio frequency bands (e.g., 3 GHz-300 GHz) may have higher path loss and shorter range than lower frequency communications. Figure 1 As described, a base station (eg, 180 ) configured to communicate using mmWave / near-mmWave radio frequency bands can utilize beamforming (eg, 182 ) with a UE (eg, 104 ) to improve path loss and range.

[0130] Furthermore, as described herein, the UE may use machine learning models to estimate channels or generate channel state information in mmWave bands and / or other frequency bands.

[0131] Channel State Information

[0132] The UE may report channel state information to the radio access network. In some cases, the CSI reporting configuration may indicate a codebook to be used for CSI feedback. For example, the codebook may include a precoding matrix that maps each layer (e.g., data stream) to a specific antenna port. In some cases, the codebook may include a precoding matrix that provides a linear combination of multiple input layers or beams. In some cases, the codebook may include a set of precoding matrices, one of which the UE may select for channel estimation.

[0133] For artificial intelligence (AI) based CSI feedback, for example, the UE may use a CSI encoder to generate CSI using a machine learning model. The encoder input may include a downlink channel matrix (H), a downlink precoder (V), and / or an interference covariance matrix (R nn ). A network entity (e.g., a base station) may use a decoder to convert the AI-encoded CSI into information indicating channel quality, such as a precoding matrix indicator (PMI) codeword. The encoder is similar to a PMI search algorithm, and the decoder is similar to a PMI codebook, which is used to convert CSI report bits into PMI codewords. The decoder output may include a downlink channel matrix (H), a transmit covariance matrix, a downlink precoder (V), an interference covariance matrix (R nn ), the original whitened downlink channel, or any combination thereof.

[0134] In some cases, the UE may report CSI aperiodically (e.g., in response to signaling from the radio access network). When the PDCCH triggers aperiodic CSI reporting, the UE may use certain computational resources to determine the CSI, and the UE may use a certain amount of time to perform the computation. In certain aspects, timing constraints may be applied to aperiodic CSI reporting.

[0135] Figure 5A5 is a timing diagram illustrating example timing constraints for aperiodic CSI. In this example, a UE may receive a PDCCH 502 indicating that aperiodic CSI is to be reported in a reporting opportunity 504 (e.g., a PUSCH). In some cases, the UE may receive a reference signal (e.g., a CSI-RS or SSB) or measure interference (e.g., interference measurement resources (IMR)) in a receive opportunity 506, and the UE may calculate CSI based on the received reference signal and / or IMR. If the timing constraints are met, the UE may provide an aperiodic CSI report. The first timing constraint may be met if the reporting opportunity 504 begins no earlier than a first timing reference 508, where the first timing reference 508 may be temporally located a first number 510 (Z) of OFDM symbols after the last symbol of the PDCCH 502 that triggered the aperiodic CSI report. That is, there may be at least Z symbols between the last symbol of the PDCCH 502 that triggers the aperiodic CSI report and the first symbol of the PUSCH carrying the CSI report (reporting opportunity 504). During this time, the UE may decode the PDCCH, perform possible CSI-RS / IM measurements (if the UE does not already have the latest previous channel / interference measurements stored in its memory), perform possible channel estimation, calculate the CSI report, and perform UCI multiplexing with the uplink shared channel.

[0136] If the UE measures reference signals (e.g., aperiodic CSI-RS) and / or IMR before reporting CSI, the first timing constraint alone may not guarantee that the UE has sufficient time to calculate CSI because the aperiodic CSI-RS may potentially be triggered close to the time of the PUSCH transmission. The second timing constraint may be satisfied if the reporting opportunity 504 begins no earlier than the second timing reference 512, where the second timing reference 512 may be temporally located at a second number 514 (Z') of OFDM symbols after the end of the last symbol in the time of the reception opportunity 506 (e.g., the latest of the following: aperiodic CSI-RS resources for channel measurement, aperiodic CSI-IMR for interference measurement, and aperiodic non-zero power (NZP) CSI-RS for interference measurement). That is, there may be at least Z' symbols between the last symbol of the aperiodic CSI-RS / IMR used to calculate the report and the first symbol of the PUSCH carrying the CSI report (reporting opportunity 504). In practice, relative to Z', Z may additionally include the DCI decoding time, so that Z is typically several symbols larger than the corresponding Z' value.

[0137] If the Z criterion (or Z' criterion) is not met (e.g., the gNB is too close to the PDCCH (or aperiodic CSI-RS / IM) triggering the PUSCH), the UE may simply ignore the scheduling DCI if the UE is also not scheduled for UL-SCH or HARQ-ACK, and the UE may refrain from sending anything. If UL-SCH or HARQ-ACK is scheduled to be multiplexed on PUSCH, the UE may send PUSCH with a CSI report, where the CSI report may be filled with dummy bits or outdated (old) information.

[0138] NR systems can support various values ​​of Z and Z'. For example, there can be separate values ​​of Z and Z' for each subcarrier spacing (SCS) for high-latency CSI, low-latency CSI, and ultra-low-latency CSI. The ultra-low-latency criteria can only be applied when a single low-latency CSI report is triggered without being multiplexed with UL-SCH or HARQ-ACK, and when all CSI processing units of the UE are not occupied. The UE can then allocate all its computing resources to calculate the ultra-low-latency CSI in a very short time.

[0139] In certain aspects, timing constraints may be used for periodic and / or semi-persistent CSI to provide the UE with sufficient time to measure the periodic reference signal and report the CSI. Figure 5B 5 is a timing diagram illustrating an example timing constraint for periodic / semi-persistent CSI. In this example, a UE may receive a periodic reference signal (e.g., CSI-RS and / or SSB) in a receive opportunity 520. In some cases, the UE may measure interference measurement resources in the receive opportunity 520. The UE may report CSI associated with the reference signal or interference measurement resource in a reporting opportunity 522 (e.g., PUCCH). The timing constraint may be satisfied if the receive opportunity 520 occurs no later than (e.g., occurs before) a third timing reference 524 (e.g., CSI reference resource), where the third timing reference 524 is temporally located a third number 526 (Y) of symbols (or time slots) before the receive opportunity 522. That is, there may be at least Y symbols (or time slots) between the receive opportunity 520 and the reporting opportunity 522. The UE is not expected to measure the reference signal or interference measurement resource after the third timing reference 524. After CSI reconfiguration, bandwidth part switching, serving cell activation, or semi-persistent CSI activation, the UE may drop CSI (eg, refrain from transmitting) if there is no CSI-RS before the third timing reference.

[0140] Adaptive learning

[0141] Non-adaptive algorithms are deterministic and depend on their input. If the algorithm is presented with the exact same input at different times, its output will be exactly the same. Adaptive algorithms (e.g., machine learning or artificial intelligence) are algorithms that change their behavior based on past experience. This means that different devices using an adaptive algorithm may end up with different algorithms over time.

[0142] According to certain aspects, channel estimation and CSI feedback processes may be performed using an adaptive learning-based algorithm (e.g., the machine learning module 712). Thus, over time, the channel estimation and CSI feedback algorithms change (e.g., adapt or update) based on new learning. The channel estimation and CSI feedback processes may be used to adapt to various characteristics of the communication link between the UE and the network entity, such as transmit power control, modulation and coding schemes, code rates, subcarrier spacing, etc. For example, adaptive learning may be used to determine channel estimation and / or CSI feedback.

[0143] In some examples, adaptive learning-based CSI / channel estimation involves training a model, such as a prediction model. The model can be used to determine CSI / channel estimates associated with a reference signal. The model can be trained based on training data (e.g., training information), which can include feedback, such as feedback associated with the CSI / channel estimate (e.g., measurements of the reference signal).

[0144] Figure 6 An example networking environment 600 is illustrated in which a prediction model 624 is used to determine CSI / channel estimates. Figure 6 As shown, networked environment 600 includes a node 620, a training system 630, and a training repository 615 that are communicatively connected via a network 605. Node 620 may be a UE (e.g., such as UE 104 in wireless communication network 100) or a BS (e.g., such as BS 102 in wireless communication network 100). Network 605 may be a wireless network, such as wireless communication network 100, which may be, for example, a 5G NR wireless network. Although training system 630, node 620, and training repository 615 are communicatively connected, Figure 6 Although illustrated as separate components in FIG, those skilled in the art will recognize that training system 630, nodes 620, and training repository 615 may be implemented on any number of computing systems, as one or more stand-alone systems, or in a distributed environment.

[0145] The training system 630 generally includes a prediction model training manager 632 that uses training data to generate a prediction model 624 for determining CSI and / or channel estimates based on signal measurements. The prediction model 624 may be determined based on information in the training repository 615.

[0146] The training repository 615 may include training data obtained before and / or after deployment of the node 620. Prior to deployment, the node 620 may be trained in a simulated communication environment (e.g., in a field test, a drive test, etc.). For example, various CSI and / or channel estimation metrics (e.g., channel quality indicator (CQI), precoding matrix indicator (PMI), reference signal received power (RSRP), signal-to-interference-plus-noise ratio (SINR), etc.) may be tested in various scenarios to obtain training information related to the CSI / channel estimation process. This information may be stored in the training repository 615. After deployment, the training repository 615 may be updated to include feedback associated with the CSI / channel estimation process performed by the node 620. The training repository may also be updated with information from other base stations and / or other UEs (e.g., based on experiences learned by those base stations and / or UEs, which may be associated with the CSI / channel estimation process performed by those base stations and / or UEs).

[0147] The prediction model training manager 632 can use the information in the training repository 615 to determine a prediction model 624 (e.g., an algorithm) for CSI / channel estimation, such as to determine CQI, PMI, RSRP, SINR, etc. As discussed in more detail herein, the prediction model training manager 632 can use various types of adaptive learning to form the prediction model 624, such as machine learning, deep learning, reinforcement learning, etc. The training system 630 can adapt (e.g., update / refine) the prediction model 624 over time. For example, when the training repository is updated with new training information (e.g., feedback), the model 624 is updated based on the new learning / experience.

[0148] Training system 630 may be located on node 620, on a BS in network 605, or on a different entity that determines prediction model 624. If located on a different entity, prediction model 624 is provided to node 620.

[0149] Training repository 615 may be a storage device, such as a memory. Training repository 615 may be located on node 620, training system 630, or another entity in network 605. For example, training repository 615 may be in cloud storage. Training repository 615 may receive training information from node 620, an entity in network 605 (e.g., a BS or UE of network 605), the cloud, or other sources.

[0150] As described above, node 620 is provided with (or generates, for example, if training system 630 is implemented in node 620) a prediction model 624. As illustrated, node 620 may include a CSI / channel estimation manager 622 configured to use prediction model 624 for CSI / channel estimation as described herein. In some examples, node 620 uses prediction model 624 to generate CSI and / or determine channel estimates based on received signal measurements. When training system 630 adapts prediction model 624 to new learnings, prediction model 624 is updated.

[0151] Thus, the CSI / channel estimation algorithm of node 620 using prediction model 624 is based on adaptive learning in that even after deployment, the algorithm used by node 620 changes over time based on experience / feedback gained by node 620 in deployment scenarios (and / or also utilizing training information provided by other entities).

[0152] According to certain aspects, adaptive learning can use any appropriate learning algorithm. As described above, the learning algorithm can be used by a training system (e.g., such as training system 630) to train a prediction model (e.g., such as prediction model 624) to obtain an adaptive learning-based CSI / channel estimation algorithm used by a device (e.g., such as node 620) to determine CSI / channel estimation based on received signal measurements, as further described herein. In some examples, the adaptive learning algorithm is an adaptive machine learning algorithm, an adaptive reinforcement learning algorithm, an adaptive deep learning algorithm, an adaptive continuous infinite learning algorithm, or an adaptive policy optimization reinforcement learning algorithm (e.g., a proximal policy optimization (PPO) algorithm, a policy gradient, a trust region policy optimization (TRPO) algorithm, etc.). In some examples, the adaptive learning algorithm is modeled as a partially observable Markov decision process (POMDP). In some examples, the adaptive learning algorithm is implemented by an artificial neural network (e.g., a deep Q network (DQN) including one or more deep neural networks (DNNs)).

[0153] In some examples, adaptive learning (e.g., used by training system 630) is performed using a neural network. Neural networks can be designed to have a variety of connection patterns. In a feedforward network, information is passed from lower layers to higher layers, where each neuron in a given layer communicates with neurons in a higher layer. A hierarchical representation can be constructed in successive layers of a feedforward network. Neural networks can also have loops or feedback (also known as top-down) connections. In a loop connection, the output from a neuron in a given layer can be conveyed to another neuron in the same layer. The loop architecture can help identify patterns that span more than one input data block in the input data blocks delivered sequentially to the neural network. The connection from a neuron in a given layer to a neuron in a lower layer is called a feedback (or top-down) connection. A network with many feedback connections can be helpful when the recognition of high-level concepts can assist in discerning specific low-level features of the input.

[0154] In some examples, adaptive learning (e.g., used by training system 630) is performed using a deep belief network (DBN). A DBN is a probabilistic model comprising multiple layers of hidden nodes. A DBN can be used to extract a hierarchical representation of a training data set. A DBN can be obtained by stacking layers of restricted Boltzmann machines (RBMs). RBMs are a type of artificial neural network that can learn probability distributions over a set of inputs. Because RBMs can learn probability distributions without information about the categories to which each input can be classified, RBMs are typically used for unsupervised learning. Using a hybrid unsupervised and supervised paradigm, the bottom RBM of a DBN can be trained in an unsupervised manner and can be used as a feature extractor, while the top RBM can be trained in a supervised manner (on the joint distribution of the input and target class from the previous layer) and can be used as a classifier.

[0155] In some examples, adaptive learning (e.g., used by training system 630) is performed using a deep convolutional network (DCN). A DCN is a network of convolutional networks configured with additional pooling layers and normalization layers. DCN has achieved state-of-the-art performance on many tasks. A DCN can be trained using supervised learning, where both the input and output targets are known for many examples and are used to modify the weights of the network using a gradient descent method. The DCN can be a feedforward network. In addition, as described above, the connection from a neuron in the first layer of the DCN to a group of neurons in the next higher layer is shared across each neuron in the first layer. The feedforward and shared connections of the DCN can be used for fast processing. For example, the computational burden of a DCN may be much smaller than that of a neural network of similar size that includes loops or feedback connections.

[0156] An artificial neural network, which can be composed of interconnected groups of artificial neurons (e.g., neuron models), is a computing device or represents a method performed by a computing device. These neural networks can be used in a variety of applications and / or devices, such as Internet Protocol (IP) cameras, Internet of Things (IoT) devices, autonomous vehicles, and / or service robots. Each node in an artificial neural network can simulate a biological neuron by taking input data and performing simple operations on the data. The results of the simple operations performed on the input data are selectively passed to other neurons. Weight values ​​are associated with each vector and node in the network, and these values ​​constrain how the input data is related to the output data. For example, the input data of each node can be multiplied by a corresponding weight value, and the products can be summed. The sum of the products can be adjusted by an optional bias, and an activation function can be applied to the result to produce an output signal or "output activation" for the node. The weight values ​​can be initially determined by iteratively passing training data through the network (e.g., the weight values ​​are established during a training phase when the network learns how to identify a specific class based on its typical input data characteristics).

[0157] Adaptive learning (e.g., used by training system 630) can be implemented using different types of artificial neural networks, such as recurrent neural networks (RNNs), multilayer perceptron (MLP) neural networks, and convolutional neural networks (CNNs). RNNs work by storing the output of a layer and feeding that output back into the input to help predict the outcome of that layer. In an MLP neural network, data can be fed into an input layer, and one or more hidden layers provide a level of abstraction for the data. A prediction can then be made for the output layer based on the abstracted data. MLPs are particularly well-suited for classification prediction problems where a class or label is assigned to an input. A convolutional neural network (CNN) is a type of feedforward artificial neural network. A convolutional neural network may include a collection of artificial neurons, each of which has a receptive field (e.g., a spatially localized region of the input space) and collectively forms an input space. Convolutional neural networks have numerous applications. In particular, CNNs have been widely used in the fields of pattern recognition and classification. In a layered neural network architecture, the outputs of artificial neurons in the first layer become the inputs to artificial neurons in the second layer, which in turn become the inputs to artificial neurons in the third layer, and so on. Convolutional neural networks can be trained to recognize hierarchical structures of features. The computation in a convolutional neural network architecture can be distributed across a set of processing nodes, which can be configured in one or more operation chains. These multi-layer architectures can be trained one layer at a time and can be fine-tuned using backpropagation.

[0158] In some examples, when using an adaptive machine learning algorithm, the training system 630 generates vectors from information in the training repository 615. In some examples, the training repository 615 stores the vectors. In some examples, the vectors map one or more features to labels. For example, these features may correspond to various deployment scenario modes discussed herein, such as frequency, subcarrier spacing, bandwidth, code rate, modulation and coding scheme, etc. The label may correspond to a CSI / channel estimate (e.g., CQI, PMI, RSRP, SINR, etc.) associated with the features used to perform CSI / channel estimation. The prediction model training manager 632 may use the vectors to train the prediction model 624 for the node 620. As discussed above, the vectors may be associated with weights in the adaptive learning algorithm. When the learning algorithm is adapted (e.g., updated), the weights applied to the vectors may also be changed. Therefore, when the CSI / channel estimation process is performed again, under the same characteristics (e.g., under the same set of conditions including frequency, subcarrier spacing, code rate, modulation and decoding scheme, etc.), the model may give different results (e.g., different CQI, PMI, RSRP, SINR, etc.) to node 620.

[0159] According to certain aspects, adaptive learning-based CSI / channel estimation allows for continuous and infinite learning. In some examples, this learning can be enhanced using federated learning. For example, while some machine learning methods use centralized training data on a single machine or in a data center, in the case of federated learning, learning can involve collaboration across multiple devices to form a predictive model. With federated learning, the training of the model can be done on the device through collaborative learning from multiple devices. For example, back to reference Figure 6 , node 620 can receive training information and / or updated training models from various different devices.

[0160] In certain aspects, the UE and / or the radio access network may train a collection of machine learning models, where each model may be designed and / or developed for a scenario (such as an urban microcell, an urban macrocell, or an indoor hotspot). These models may be associated with various bandwidth configurations. These models may be associated with various CSI payloads to adapt to different UE locations or wireless conditions. For example, if the UE is near the center of the cell, the UE may report the CSI payload with a larger resolution, and the UE may use a machine learning model associated with the specific CSI payload. If the UE is at the edge of the cell, the UE may report the CSI payload with a low resolution due to cell coverage, and the UE may use different machine learning associated with the low-resolution CSI payload. In some cases, these models may be associated with various antenna architectures at the UE and / or base station. In certain aspects, the UE may use the models to support beam management. For example, the UE may use the model to determine the beam to be used in future transmit opportunities or to determine a finer beam based on a coarse beam, and the UE may report the determined beam to the radio access network along with the RSRP associated with the beam. In certain aspects, the UE may perform positioning using a model, wherein the UE may use the model to determine, for example, the distance and / or angle of the UE's location relative to a base station.

[0161] After training the models, they can be registered with the radio access network. A model server (e.g., training system 630 and / or training repository 615) can test the models, compile the models into runtime images, and store the runtime images. The model server can indicate to the radio access network that the models are registered. When the radio access network deploys the model, the radio access network can configure the UE to use the model (e.g., by indicating a model identifier associated with the model), and the UE can download the runtime image of the model from the model server (e.g., training repository 615). Due to different UE architectures, the UE may support different numbers of models stored in its modem and / or memory (e.g., non-volatile memory and / or random access memory), and the UE may take different amounts of time to switch to using a model stored in the modem or to switch to using a model stored in memory.

[0162] Although this paper describes AI-based CSI or channel estimation Figure 6 The examples depicted in FIG. 5 are for ease of understanding, but aspects of the present disclosure may also be applied to other AI-generated information, including, for example, beam management information and / or UE positioning information.

[0163] Aspects related to machine learning in wireless communications

[0164] Various aspects of the present disclosure provide apparatus and techniques for using machine learning in wireless communications. Because UEs can have different architectures, machine learning capabilities can be divided into different categories or different aspects. As an example, the machine learning capabilities can represent the maximum number of active machine learning models that the UE can process or store (e.g., via the UE's modem), the maximum number of inactive machine learning models that the UE can store (e.g., in memory), the minimum amount of time for switching to using inactive machine learning models, or a combination of machine learning models that the UE can process or store concurrently. The UE can indicate to the radio access network the specific machine learning capabilities that the UE is capable of performing (e.g., the maximum number of machine learning models that the UE can process or store in its modem), and the radio access network can configure the UE with machine learning models based on the machine learning capabilities. For example, if the UE can only support storing and processing a single machine learning model in its modem, the radio access network can configure or schedule the UE to use only a single machine learning model for processing at any time.

[0165] Different timelines may be supported for processing CSI using a machine learning model. For example, if the UE is switching between active machine learning models (e.g., stored in the UE's modem), a faster timeline may be used, and if the UE is activating a new machine learning model, a slower timeline may be used. To activate the new machine learning model, the UE may download the machine learning model from the radio access network or load the machine learning model from memory (e.g., non-volatile memory or random access memory).

[0166] Certain criteria for concurrent processing of machine learning models may be supported. For example, a machine learning model may occupy a certain number of processing units. The UE may support a maximum number of processing units associated with the machine learning model, such that the UE can process multiple machine learning models concurrently, up to the maximum number of processing units. In another aspect, the criteria for concurrent processing may be supported by reporting machine learning model combinations and the number of concurrently processed tasks or inference tasks associated with the concurrently configured / scheduled combinations. In this case, the UE may report one or more model combinations, for example, a first combination including {models 1, 2, 3} and a second combination including {models 1 and 2}. The UE may report the total number of inference tasks that these combinations can process or report the number of inference tasks associated with each model in a combination. For example, the UE may report a total of three tasks for the {models 1 and 2} combination and a total of four tasks for the {models 1, 2, 3} combination.

[0167] The machine learning model process described herein may enable improved wireless communication performance (e.g., higher throughput, lower latency, and / or spectral efficiency). For example, different UE architectures may support different timelines for processing CSI (and / or other information) using a machine learning model. Different categories of machine learning capabilities may allow a radio access network to dynamically configure a machine learning model to a UE in response to the specific machine learning capabilities of the UE. Such dynamic configuration may allow a UE to process CSI (and / or beam management information, UE positioning information, channel estimation, etc.) using a machine learning model under various conditions (such as high latency, low latency, ultra-low latency, high resolution CSI, low resolution CSI, wide beam, narrow beam, etc.).

[0168] Figure 7 is a diagram illustrating an example wireless communication network 700 using a machine learning model according to certain aspects of the present disclosure. In this example, a model server 704 can perform model training, model testing, and / or compilation of a runtime image of a machine learning model, for example, as described herein with respect to Figure 6 As described. The model server 704 may include a training system 630 and / or a training repository 615. In some cases, the model server 704 may be integrated or co-located with the BS 102. In some cases, the model server 704 may communicate with the BS 102 and be accessed via network communications (such as the Internet). For example, the model server 704 may represent a cloud computing system operated by a UE vendor, a chipset vendor, and / or a third party. The model server 704 may include an over-the-top (OTT) server. In some aspects, the model server 704 may be accessed by the UE 104 via multiple radio access technologies (such as WiFi and 5G NR).

[0169] BS 102 may configure UE 104 with reference signals for data collection for machine learning training at model server 704. For example, BS 102 may configure UE 104 to measure reference signals (or interference measurement resources) for data collection and configure meta information associated with the reference signals. In this example, the reference signals (e.g., CSI-RS and / or SSB) may be associated with analog / digital beams 706 or a specific antenna layout equipped at the BS. For example, the meta information for data collection may include pre-decoder, beamforming, antenna setting information associated with the reference signals. In some cases, the meta information may be an ID that is considered a specific representative of those configurations. The UE may measure the reference signals and send these measurements to the model server 704 along with the meta information, Doppler information, delay spread information, channel quality information, and / or timestamps.

[0170] The model server 704 may perform model training based on measurements associated with reference signals acquired by the UE 104. After model training, the model server 704 may test the trained model and compile the trained model into a runtime image. After training or developing the model, the model server 704 may register a model identifier associated with the model with the radio access network. In the model deployment phase, the BS 102 may configure the UE with a model identifier associated with the model for AI-based processing. The UE may download the runtime image from the model server 704 according to the model identifier. In the inference phase, the BS 102 may send a reference signal associated with the beam 706 for AI-based reporting, such as CSI, beam management, and UE positioning.

[0171] In this example, BS 102 may send (e.g., transmit, output, or provide) machine learning model information 702 to UE 104. The machine learning model information 702 may include various information associated with AI-based reporting. In some cases, the machine learning model information 702 may include a configuration for AI-based reporting. For example, the machine learning model information 702 may include training data for one or more machine learning models, a runtime image associated with the machine learning model, and / or setting parameters associated with the machine learning model. The machine learning model information 702 may include a machine learning model configuration (e.g., a CSI reporting configuration associated with the machine learning model), configuration / reconfiguration of periodic / aperiodic CSI associated with the machine learning model, activation / deactivation of semi-persistent CSI associated with the machine learning model, and / or a trigger for reporting aperiodic CSI associated with the machine learning model.

[0172] As an example, UE 104 may monitor downlink reference signals (or interference measurement resources) from BS 102, such as CSI-RS and / or SSB associated with beam 706. UE 104 may determine AI-based information 708 (e.g., channel estimate) and / or AI-based report 710 (e.g., CSI) associated with beam 706 based at least in part on the received reference signals corresponding to beam 706. UE 104 may report AI-based report 710 to BS 102. AI-based report 710 may include various information, such as CSI, beam management information (e.g., preferred beam), UE positioning information, and / or channel estimate. In some cases, for example, UE 104 may use the channel estimate for transmit and / or receive beamforming.

[0173] In some cases, BS 102 may transmit a reference signal (e.g., CSI-RS and / or SSB) using a beam that may be obtained through artificial intelligence and / or machine learning (AI / ML) at BS 102 and / or UE 104. UE 104 may use a CSI encoder to compress the channel estimate to a small dimension and report the CSI to BS 102. BS 102 may use a CSI decoder to recover the full channel. In some cases, the CSI encoder and decoder are matched AI / ML modules, e.g., jointly trained AI / ML modules.

[0174] In some cases, the UE 104 may perform artificial intelligence (e.g., neural networks and / or machine learning) and / or regression analysis (e.g., linear minimum mean square error (LMMSE) operations) to determine the AI-based information 708 and / or the AI-based report 710. For example, the UE 104 may use a machine learning module 712 to determine the AI-based information 708 and / or the AI-based report 710. The machine learning module 712 may include one or more active machine learning models 718, and in some cases, the machine learning module may include one or more inactive machine learning models 720. The UE 104 may select a machine learning model from the active machine learning model 718 and / or the inactive machine learning model 720 to use for determining the AI-based information 708 and / or the AI-based report 710. The UE 104 may select a machine learning model in response to a configuration from the BS 102, such as a periodic or semi-persistent CSI configuration or a trigger for aperiodic CSI. The CSI encoder / decoder or AI-based channel estimator is similar to a CSI codebook, such as a type 1 codebook, a type 2 codebook, or an encrypted type 2 codebook. The model identifier may be associated / configured in the CSI reporting configuration like a codebook. In response to the BS 102 triggering or activating a specific CSI report, the UE may use the corresponding model indicated in the CSI reporting configuration.

[0175] An active machine learning model may be a machine learning model that is loaded into memory (e.g., random access memory) and available for processing, while an inactive machine learning model may be a machine learning model that is stored in a data storage device (e.g., non-volatile memory), for example, due to the size of the modem memory. The inactive model may be available for processing, with a certain amount of time being available or unavailable for loading the inactive machine learning model into memory. The time to load (e.g., switch to use) the inactive machine learning model may depend on the UE architecture, such that some UEs may support short loading times, while other UEs may support long loading times.

[0176] The processing system may execute a machine learning module 712, such as described herein with respect to Figure 24The processing system described herein. Input 714 to the machine learning module 712 may include measurements of reference signals (e.g., CSI-RS and / or SSB) or interference measurements received from the BS 102. For example, the input 714 may include received reference signals and / or interference measurements represented in the frequency domain. Output 716 of the machine learning module 712 may include AI-based information 708 and / or AI-based reports 710.

[0177] In certain aspects, timing criteria may be applied to loading or configuring new machine learning models at a UE. Assuming that the UE is capable of having a first number (M) of active machine learning models and a second number (N) of inactive machine learning models, the radio access network may configure (or deploy) to the UE a total of the sum of the first number and the second number (M+N) of machine learning models from a set of registered machine learning models. The M+N models may be used at any time for periodic, semi-persistent, or dynamic triggering, subject to certain processing timeline criteria and / or concurrent processing criteria, as further described herein. In some cases, after the UE receives a machine learning model configuration (e.g., a radio resource control (RRC) configuration), the UE may, for example, download the models / runtime images associated with the configuration from a model server (e.g., model server 704) and load the M models to the modem. In some cases, assuming that the UE has already downloaded all models / runtime images to a storage device, the UE may, for example, load the active models / runtime images associated with the configuration to the modem. The UE may send a confirmation to the radio access network that the models have been successfully loaded.

[0178] Figure 8 FIG8 is a signaling flow 800 illustrating an example of a UE loading a machine learning model. In this example, a UE 104 can communicate with a network entity (e.g., BS 102).

[0179] Optionally, at activity 802, UE 104 can obtain one or more machine learning models from model server 704. In some cases, UE 104 can communicate with model server 704 via BS 102. In some cases, UE 104 can communicate with model server 704 via another radio access technology, such as WiFi. For example, UE 104 can download runtime images associated with machine learning models from model server 704 (e.g., training system 630 and / or training repository 615). That is, UE 104 can pre-download machine learning models that are available from model server 704.

[0180] At activity 804, the UE 104 may receive a machine learning configuration that indicates a machine learning model identifier to be used for processing at the UE, such as CSI processing and / or channel estimation. The machine learning configuration may indicate an active machine learning model and / or an inactive machine learning model. The machine learning configuration may indicate scenarios in which certain machine learning models are used, such as radio conditions (e.g., frequency, subcarrier spacing, bandwidth, code rate, modulation and coding scheme, etc.), cell type (e.g., micro, macro, etc.), CSI resolution, antenna architecture, etc.

[0181] Optionally, at activity 806, UE 104 can obtain the machine learning model from model server 704 in response to receiving the configuration. For example, UE 104 can download a runtime image associated with the machine learning model indicated in the configuration from model server 704.

[0182] At activity 808, the UE 104 can load the active machine learning model associated with the configuration into the modem (e.g., the modulator and demodulator of the transceiver 354).

[0183] At activity 810, UE 104 may send a confirmation to BS 102 that the model has been successfully loaded at UE 104. In some cases, UE 104 may send a separate confirmation for each model. For example, UE 104 may send a first confirmation associated with a first model (Model 1) and a second confirmation associated with a second model (Model 2). In some cases, UE 104 may send a common confirmation for multiple models (e.g., the first model and the second model). The confirmation may be a new type of signaling or via RRC signaling.

[0184] The UE 104 may be expected to send an acknowledgment within a certain period of time from the time the machine learning configuration is received. For example, the UE 104 may send an acknowledgment within a time window 812 that begins when the machine learning configuration is received. The duration of the time window 812 may be based on the number of active machine learning models configured at the UE, the number of inactive machine learning models configured at the UE, and / or the size of the corresponding machine learning models. The duration of the time window 812 may be specific to loading the machine learning models and confirming such activities. For example, the duration of the time window 812 may be greater than some RRC timing criteria for confirming the RRC configuration.

[0185] If the UE 104 fails to send an acknowledgment within the time window 812, the radio access network may assume that the UE 104 failed to successfully load the machine learning model. In some cases, the UE 104 may fall back to using a non-AI-based codebook for CSI reporting, which may be signaled to the radio access network as a UE capability.

[0186] In certain aspects, machine learning capabilities may be divided into different categories (or levels) or different aspects. The UE may report its machine learning capabilities to the radio access network. For example, the machine learning capabilities associated with the UE may include a maximum number of active machine learning models (M), a maximum number of inactive machine learning models (N), and an (additional) time (T) for switching to (or activating) an inactive machine learning model. The UE may be able to run (process) up to M models concurrently. In some cases, the UE may be able to store up to N inactive machine learning models and activate them for a certain period of time with or without an additional time (T).

[0187] In certain aspects, the machine learning capabilities may be reported to the radio access network as a set of values ​​for M, N, and / or T (e.g., {M, N, T}). In some cases, the number of inactive machine learning models may be implicitly indicated by indicating the maximum number of active machine learning models supported by the UE (M) and the total number of active and inactive machine learning models (X). The time for switching to an inactive machine learning model may be indicated as an index associated with a set of values ​​for the duration, such as a set of two values ​​(e.g., {zero, T}, where the value of T may be predefined) or a set of two or more values ​​(e.g., {0, T1, T2, T3, etc.}, where the values ​​of T1, T2, T3, etc. may be predefined).

[0188] For certain aspects, the machine learning capabilities associated with a UE may include one or more combinations of active models supported by the UE and / or one or more combinations of inactive models supported by the UE. The number of active or inactive models in these combinations may be based on the size of the models or other model characteristics. For example, the UE may report that the UE is capable of processing / storing models {1,2} or models {1,3,4} as active / inactive models. In some cases, the UE may report the number of inference tasks associated with a combination of models and / or associated with each model in a combination, as further described herein.

[0189] In some cases, after downloading and / or loading the active models, the UE may report to the radio access network the machine learning models available for processing at the UE (e.g., the models that are active). The UE may indicate the models that support fast timeline as the initial set of active models, as further described herein.

[0190] Figure 9 is a table illustrating example machine learning capability levels that may be supported by a UE. In this example, for levels 1.0, 1.1, and 1.2, the UE may be able to have a single model (e.g., a runtime image of the model) in the modem (e.g., M=1). The UE may not be able to process multiple machine learning models concurrently.

[0191] For levels 1.0 and 2.0, the UE may not be able to have an inactive machine learning model (e.g., N=0). For example, the UE may not have the ability to store an inactive machine learning model such that the UE cannot (in a short timeline) load another model image or parameter setting from a memory (e.g., non-volatile memory or random access memory). In certain aspects, a model or parameter setting switch (to a model other than the configured M active models and / or N inactive models) may be performed via reconfiguration (e.g., RRC reconfiguration). For example, in response to a reconfiguration, the UE may replace the active model with another model.

[0192] For levels 1.1, 1.2, 2.1, and 2.2, the UE may be able to store up to N inactive machine learning models (e.g., runtime images and / or parameter settings) in memory (e.g., N>0). The UE may be able to perform dynamic model switching (e.g., activating an inactive model) via signaling (e.g., downlink control information (DCI) or medium access control (MAC) signaling). For levels 1.1 and 2.1, the UE may be able to activate an inactive model in a certain time frame that includes an additional time (T). For example, the processing timeline of the inactive model may be longer than the processing timeline of the active model. For levels 1.2 and 2.2, the UE may be able to activate an inactive model without the additional time used for levels 1.1 and 2.1. For example, the processing timeline of the inactive model may be the same as the processing timeline of the active model.

[0193] For levels 2.0, 2.1, and 2.2, the UE may be able to have multiple models (e.g., runtime images of models) in the modem (e.g., M > 1). The UE may be able to process multiple active machine learning models concurrently. Switching between the M active models may not use an additional timeline.

[0194] In certain aspects, a UE may report the level of machine learning capabilities it supports (e.g., as described herein with respect to Figure 9 The described levels 1.0-2.2) and / or the values ​​of {M, N, T} are used as machine learning capabilities. The radio access network may configure a UE with a sum of up to M and N (M+N) models, and the radio access network may schedule CSI reports based on whether the model is active or inactive to leave sufficient processing time for the UE to activate the inactive model and process the CSI.

[0195] 10A to 10C is a diagram of example machine learning capability levels 1.0, 1.1, and 1.2. In these examples, the UE may support up to a single active model (e.g., M=1) loaded in the modem 1002. Figure 10A, in a first state 1000A of Level 1.0, the UE may have a first model (Model 1) loaded in the modem 1002, wherein the model server 1004 may have the first model, the second model (Model 2), and the third model (Model 3) available to the UE. To load the active model in the modem 1002, the UE may download a runtime image associated with the first model from the model server 1004, which may be an over-the-top (OTT) server. For Level 1.0, the UE may not be able to process multiple models concurrently (e.g., M=1). The UE may be able to process measurements of CSI using the active model (e.g., Model 1) in a fast timeline, as described herein with respect to Figure 12A Further described.

[0196] In the second state 1000B, the UE may be reconfigured to have the second model loaded in the modem 1002. For example, the UE may receive an RRC reconfiguration indicating the use of the second model for CSI processing, and the UE may require a period of time (e.g., more than 100 ms to 1000 ms) to prepare the second model for processing (e.g., download the second model and load the second model into the modem).

[0197] like Figure 10B and Figure 10C As shown, for levels 1.1 and 1.2, the UE may have an active model (Model X) loaded in the modem 1002 and three inactive models (a first model, a second model, and a third model) loaded in the storage device 1006. The UE may not be able to process multiple models concurrently. The UE may be able to process CSI measurements using the active model (e.g., Model X) in a fast timeline, as described herein with respect to Figure 12B and Figure 12C Further described. For level 1.1, the UE may be able to use the inactive model to process CSI measurements in a certain period of time (e.g., a slower timeline or long duration) with an additional time (e.g., more than 10ms) for activating the inactive model. For level 1.2, the UE may be able to use the inactive model to process CSI measurements in the period of time without the additional time (e.g., a fast timeline or short duration). The fast timeline (or short duration) may be so called because it may span less time than the slower timeline (or long duration). Therefore, the terms fast timeline (or short duration) and slower timeline (or long duration) may be relative to each other.

[0198] Figures 11A to 11C is a diagram of example machine learning capability levels 2.0, 2.1, and 2.2. In these examples, the UE may support up to two active models loaded in the modem 1102. Figure 11A, in a first state 1100A of Level 2.0, the UE may have a first model (Model 1) and a second model (Model 2) loaded in a modem 1102, wherein a model server 1104 may have the first model, the second model, and a third model (Model 3) available to the UE. To load the active models in the modem 1102, the UE may download runtime images associated with the first model and the second model from the model server 1104. For Level 2.0, the UE may be able to process up to two active models concurrently (e.g., M=2). The UE may be able to process measurements of CSI using the active models (e.g., Model 1 and Model 2) in a fast timeline, as described herein with respect to Figure 13A Further described.

[0199] In the second state 1100B, the UE may be reconfigured to replace the second model with a third model loaded into the modem 1102. For example, the UE may receive an RRC reconfiguration indicating the use of the third model for CSI processing, and the UE may require a certain period of time (e.g., more than 100 ms to 1000 ms) to prepare the third model for processing (e.g., download the third model and load the third model into the modem).

[0200] like Figure 11B and Figure 11C As shown, for levels 2.1 and 2.2, the UE may have two active models (Model X1 and Model X2) loaded in the modem 1102 and three inactive models (a first model, a second model, and a third model) loaded in the storage device 1106. The UE may be able to process up to two active models concurrently. The UE may be able to process CSI measurements using the active models (e.g., Model X1 and Model X2) in a fast timeline, as described herein with respect to Figure 13B and Figure 13C As further described. For level 2.1, the UE may be able to use the inactive model to process CSI measurements in a certain period of time (e.g., a slower timeline) with additional time (e.g., more than 10 ms) for activating the inactive model. For level 2.2, the UE may be able to use the inactive model to process CSI measurements in the same period of time (e.g., a fast timeline) without the additional time.

[0201] In certain aspects, aperiodic CSI processing using machine learning may support multiple timelines for reporting CSI (e.g., a fast timeline and a slower timeline). The timeline may span the period from when a CSI trigger is received at the UE to when the CSI is reported to the radio access network. The specific duration associated with the timeline may depend on the machine learning capabilities of the UE. For example, for levels 1.1 and 2.1, the UE may be able to activate the inactive model within the slower timeline (e.g., using additional time relative to the fast timeline for the active model), while for levels 1.2 and 2.2, the UE may be able to activate the inactive model within the fast timeline (the same timeline as the active model).

[0202] The duration associated with this timeline may depend on the supported machine learning capability level (1.0-2.2) and / or one or more individual capabilities (e.g., {M, N, T}). These timelines may apply timing constraints for aperiodic CSI processing, such as those described herein with respect to Figure 5A The timing constraints Z and / or Z' described. The values ​​of Z and / or Z' may be configured for machine learning processing and corresponding capabilities of different UE architectures (e.g., levels 1.0-2.2). For example, the values ​​of Z and / or Z' may depend on the model complexity, an identifier associated with the machine learning model (wherein the identifier may indicate the model complexity and the processing time used), a rank indicator or maximum rank associated with one or more transmission layers, whether a CSI decoder is available for CSI processing, whether there are any other concurrent AI / ML models being processed, or any combination thereof. In some cases, the UE may report the values ​​of Z and / or Z' used for machine learning processing to the radio access network. If the UE fails to report CSI within the timeline for machine learning processing, the UE may report previously reported CSI (e.g., outdated CSI), or the UE may use a non-AI based codebook for CSI reporting.

[0203] 12A to 12C 1 is a diagram illustrating various timelines associated with machine learning capability levels 1.0, 1.1, and 1.2. In these examples, the UE may support having up to a single active model (e.g., Model 1 or parameter settings associated with Model 1). The UE may have a first model 1202a (Model 1) loaded as the active model.

[0204] refer to Figure 12A, the model server may have a second model 1202b (e.g., Model 2) and a third model 1202c (e.g., Model 3) available for download by the UE. For Level 1.0, the UE may be able to use the active model (e.g., Model 1) to process CSI measurements within the fast timeline 1204. For example, in response to an aperiodic CSI trigger, the UE may be able to use the active model (e.g., Model 1) to process CSI measurements and report CSI during the duration associated with the fast timeline 1204, as described herein with respect to Figure 14 As further described, the UE may switch to using the second model 1202b or the third model 1202c in response to a reconfiguration 1206 (eg, an RRC reconfiguration), which may require a longer duration than the fast timeline 1204 to execute.

[0205] refer to Figure 12B and Figure 12C , the UE may have the second model 1202b and the third model 1202c loaded in memory as inactive models. For levels 1.1 and 1.2, the UE may be able to process measurements of CSI using the active model (e.g., Model 1) within the fast timeline 1204. For level 1.1, the UE may be able to activate an inactive model (e.g., Model 2 or Model 3) within the slower timeline 1208 and process measurements of CSI using the inactive model. For example, in response to an aperiodic CSI trigger, the UE may be able to process measurements of CSI and report CSI using the inactive model (e.g., Model 2) for a duration associated with the slower timeline 1208, which may span the duration of the fast timeline plus an additional time (e.g., as described herein with respect to Figure 9 For level 1.2, the UE may be able to activate an inactive model (eg, Model 2 or Model 3) within the fast timeline 1204 and process CSI measurements using the inactive model.

[0206] 13A to 13C 2 is a diagram illustrating various timelines associated with machine learning capability levels 2.0, 2.1, and 2.2. In these examples, the UE may support having up to two active models (e.g., Model 1 and Model 2). The UE may have a first model 1202a (Model 1) and a second model 1202b loaded as active models, and the UE may be able to concurrently process CSI measurements using the first model 1202a and the second model 1202b.

[0207] refer to Figure 13A, the model server may have a third model 1202c (e.g., Model 3) available for download by the UE. For Level 2.0, the UE may be able to process CSI measurements using active models (e.g., Model 1 and Model 2) within the fast timeline 1204. For example, in response to an aperiodic CSI trigger, the UE may be able to process CSI measurements and report CSI using active models (e.g., Model 1 and / or Model 2) during the duration associated with the fast timeline 1204, as described herein with respect to Figure 14 As further described, the UE may switch to using the third model 1202c in response to a reconfiguration 1206 (eg, an RRC reconfiguration), which may require a longer duration than the fast timeline 1204 to execute.

[0208] refer to Figure 13B and Figure 13C , the UE may have a third model 1202c loaded in memory as an inactive model. For levels 2.1 and 2.2, the UE may be able to process measurements of CSI using active models (e.g., Model 1 and Model 2) within the fast timeline 1204. For level 2.1, the UE may be able to activate an inactive model (e.g., Model 3) within the slower timeline 1208 and process measurements of CSI using the inactive model. For example, in response to an aperiodic CSI trigger, the UE may be able to process measurements of CSI and report CSI using the inactive model (e.g., Model 3) for a duration associated with the slower timeline 1208, which may span the duration of the fast timeline plus an additional time (e.g., as described herein with respect to Figure 9 For level 2.2, the UE may be able to activate an inactive model (eg, model 3) within the fast timeline 1204 and process CSI measurements using the inactive model.

[0209] In certain aspects, a UE and / or a radio access network (e.g., a network entity) may track which model is the currently active model at the UE. The UE and the radio access network may track the currently active model. The UE and / or the radio access network may update the currently active model in response to an indication to load (or activate) an inactive model (e.g., switch from inactive to active) and / or an indication to deactivate an active model (e.g., switch from active to inactive). The indication may be an aperiodic CSI trigger, a semi-persistent CSI activation / deactivation, or a CSI reference resource location for periodic or semi-persistent CSI reporting. For example, the UE may track the currently active model using a list or set of identifiers associated with the active models. The list or set of identifiers associated with the active models may be referred to as a model state. The model state may be updated in response to an aperiodic CSI trigger, activation / deactivation of semi-persistent CSI, and / or configuration / reconfiguration of periodic CSI. The initial model state may be set to a default value (e.g., none) or reported by the UE to the radio access network.

[0210] The UE and / or radio access network may determine the timeline (e.g., fast timeline or slower timeline) for CSI processing based on the model state and UE capabilities. For aperiodic CSI triggering, if the triggered model identifier is in that model state, the fast timeline for CSI processing may be applied to levels 1.1 to 2.2. If the triggered model identifier is not in that model state, the fast timeline (e.g., the same timeline as the active model) may be used for levels 1.2 and 2.2, and additional time may be used for levels 1.1 and 2.1 (e.g., the fast timeline plus an additional time T, or a slower timeline). In some cases, the slower timeline may use longer Z and Z' values ​​than the fast timeline. In some cases, the slower timeline may use the same Z and Z' values ​​as the fast timeline and add the additional time T to the corresponding values.

[0211] In some cases, the model state can be an ordered series of model identifiers, where certain criteria can be applied to remove old entries and add new entries (e.g., last-in, first-out (LIFO) or first-in, first-out (FIFO)). The ordered series can enable the UE and / or radio access network to track the model state without requiring feedback from the UE regarding the status of the model state. In some cases, in response to receiving an indication to activate a new model (e.g., loading an inactive model or downloading a model from a model server), the first or last model identifier in the model state vector can be removed, and the new model identifier can be inserted in the first position of the model state. For example, assume the model state has the value {a, b, c} for the currently active model identifier, and CSI reporting is triggered for model d. The last model identifier in the model state can be removed, resulting in the removal of model c. The other models can be shifted, and model d can be inserted in the first position, resulting in the model state becoming {d, a, b}.

[0212] If multiple entries are to be added to the model state concurrently (e.g., when multiple CSI reports are triggered or scheduled in the same UCI by the same CSI request), the identifier with the lowest (or highest) value among the new model identifiers (for the new CSI report or model) can be inserted into the first position of the model state. For example, assume the model state has the values ​​{a, b, c} for the currently active model identifier, and CSI reports are triggered for models d and e. The last two entries can be removed, removing models b and c. Model a can be shifted to the last position, and models d and e can be inserted in the first two positions, so that the model state becomes {d, e, a}.

[0213] In some cases, the UE may report an indication of the current model state to the radio access network. Feedback regarding the status of the model state may allow the model state to be an unordered list or set. For example, the UE may report which model identifier was removed from the model state. In certain aspects, the UE may determine which model identifier to remove from the model state.

[0214] Figure 14 is a timing diagram illustrating an example of updating a model state over time in response to an aperiodic CSI trigger and a corresponding timeline for processing CSI using a machine learning model. In this example, the UE may have a first model (Model 1), a second model (Model 2), and a third model (Model 3) loaded as active models, such that the model state 1420 includes identifiers associated with Model 1, Model 2, and Model 3. The UE may also have a machine learning capability level of 2.1, such that the UE can have multiple active models, and the UE uses a slower timeline to activate inactive models. At activity 1402, the UE may report to the network entity that the model was successfully loaded, for example, as described herein with respect to Figure 8described.

[0215] At activity 1404, the UE receives a CSI request for CSI reporting using Model 1. Since Model 1 is in Model state 1420, Model state 1420 is not updated, and the UE applies fast timeline 1422 to report CSI at activity 1406. For example, in response to the CSI request associated with Model 1, the UE may process measurements and report CSI in fast timeline 1422 using Model 1 according to timing constraints Z and / or Z' associated with fast timeline 1422.

[0216] At activity 1408, the UE receives a CSI request for CSI reporting using a fourth model (Model 4). Since Model 4 is not in Model State 1420, Model State 1420 is updated to include Model 4, with Model 3 removed and Model 4 added. In response to Model 4 not being in Model State 1420, the UE applies a slower timeline 1424 to report CSI at activity 1410. For example, in response to the CSI request associated with Model 4, the UE may load Model 4 from memory in the slower timeline 1424 according to timing constraints Z and / or Z' (or additional time T) associated with the slower timeline 1424, process measurements using Model 4, and report CSI at activity 1410.

[0217] At activity 1412, the UE receives a CSI request for CSI reporting using model 2. While model 2 is in model state 1420, the order of the identifiers may be updated such that model 2 is moved to the first position. Since model 2 is in model state 1420, the UE applies fast timeline 1422 to report CSI at activity 1414.

[0218] For periodic and semi-persistent CSI, multiple timelines (e.g., a fast timeline and a slower timeline) may be used to report CSI, where these timelines may define the time position and Y duration of a timing reference (e.g., the third timing reference 524), as described herein with respect to Figure 5B The value of Y (e.g., the number of symbols or slots) may be associated with the timeline such that a slower timeline may have a longer Y duration than a faster timeline. The faster timeline may correspond to a short Y duration (e.g., the CSI resources may be located closer in time to the reporting opportunity), and the slower timeline may correspond to a long Y duration (e.g., the CSI resources may be located further in time from the reporting opportunity) or a short Y duration plus an additional time T.

[0219] In some cases, aspects described herein with respect to the duration of the timeline for aperiodic CSI may be applicable to the timeline associated with periodic and semi-persistent CSI. For example, the specific duration associated with the timeline (e.g., Y) may depend on the machine learning capabilities of the UE. For example, for levels 1.1 and 2.1, the UE may be able to activate an inactive model within a slower timeline (e.g., using additional time relative to the fast timeline), while for levels 1.2 and 2.2, the UE may be able to activate an inactive model within the fast timeline. The UE (and / or network entity) may determine the timeline for periodic / semi-persistent CSI processing based on the model state, where the model for periodic and semi-persistent CSI may occupy a position in the model state (next to the model for aperiodic CSI).

[0220] For certain aspects, the model for periodic / semi-persistent CSI may maintain a retained position in the model state from configuration / activation to reconfiguration / deactivation. In some cases, the model for periodic or semi-persistent CSI may not participate in model state updates, such that a fast timeline applies to periodic or semi-persistent CSI. For example, in response to periodic CSI configuration, the model for periodic CSI may be loaded as the active model, and thus, a fast timeline applies to periodic CSI (and similar behavior may apply to activation of semi-persistent CSI). In some cases, for semi-persistent CSI, the model state may be updated in response to receiving an activation / deactivation command such that a slower timeline applies to the initial (first) transmission of semi-persistent CSI, and a fast timeline applies to subsequent transmissions of semi-persistent CSI.

[0221] Figure 15 is a timing diagram illustrating an example of timing constraints for processing periodic and semi-persistent CSI using machine learning models. In this example, the UE may have a first model (Model 1) and a second model (Model 2) loaded as active models, such that the model state 1520 includes identifiers associated with Model 1 and Model 2. Model 1 may be used for aperiodic CSI, and Model 2 may be used for periodic CSI. The UE may also have a machine learning capability level of 2.1, such that the UE can have multiple active models, and the UE activates inactive models using a slower timeline. At activity 1502, the UE may report to the network entity that the model was successfully loaded, for example, as described herein with respect to Figure 8 described.

[0222] At activity 1504, the UE reports periodic CSI associated with Model 2 using a shorter reference resource 1522 corresponding to a fast timeline (e.g., a short Y duration). For example, the UE may receive a reference signal no earlier than a timing reference associated with the fast timeline (e.g., the third timing reference 524), which may have a short (default) Y duration.

[0223] At activity 1506, the UE receives an aperiodic CSI request for CSI reporting using a third model (Model 3). Since Model 3 is not in the model state 1520, the model state 1520 is updated to include Model 3, with Model 1 removed and Model 3 added. In response to Model 3 not being in the model state 1520, the UE applies the slower timeline 1524 to report CSI at activity 1508. For example, in response to the CSI request associated with Model 3, the UE may load Model 3 from memory as an inactive model in the slower timeline 1524 according to the timing constraints Z and / or Z' associated with the slower timeline 1524 (or the additional time T), process the measurement using Model 3, and report CSI at activity 1508.

[0224] At activity 1510 , the UE reports periodic CSI associated with Model 2 using the shorter reference resource 1522 .

[0225] At activity 1512, the UE receives an indication to activate semi-persistent CSI using a fourth model (Model 4). Since Model 4 is not in the Model State 1520, the Model State 1520 is updated to include Model 4, wherein Model 3 is removed and Model 4 is added.

[0226] At activity 1514, the UE reports semi-persistent CSI using Model 4. Since this is the initial transmission of semi-persistent CSI for Model 4, the UE may apply a slower timeline corresponding to the longer reference resource 1526. For example, the UE may receive the reference signal no earlier than the timing reference associated with the slower timeline (e.g., the third timing reference 524), which may have a longer duration Y (e.g., the default duration plus an additional time T).

[0227] At activity 1516 , the UE reports periodic CSI associated with Model 2 using the shorter reference resource 1522 .

[0228] At activity 1518 , the UE reports the semi-persistent CSI associated with Model 4 using the shorter reference resource 1522 , since this is a subsequent transmission of the semi-persistent CSI for Model 4 .

[0229] In certain aspects, protection may be applied to prevent or handle back-to-back switching between an active model and an inactive model. If a CSI report indicating a model switch (e.g., activation of an inactive model) is triggered, the UE may be expected not to perform certain actions during the time window (e.g., ending when the UE is scheduled to report CSI). For example, during this time window, the UE may be expected not to receive any CSI requests that trigger any other CSI reports (or AI-based reports) generated by a model that is not in a model state. The UE may be expected not to send any other CSI reports (or AI-based reports) generated by a model that is not in a model state. During this time window, the UE may ignore CSI requests using an inactive model outside of the model state, or the UE may send dummy CSI in response to a CSI request using an inactive model.

[0230] Figure 16 is a timing diagram illustrating an example protection for back-to-back model switching. In this example, the UE may have a first model (Model 1), a second model (Model 2), and a third model (Model 3) loaded as active models, such that the model state 1612 includes identifiers associated with Model 1, Model 2, and Model 3. The UE may be scheduled to report CSI using Model 1 at a reporting opportunity 1602 (T0). The UE may be expected not to perform certain actions during a time window 1604 spanning from a timing reference 1606 (T0-T_guard) to the reporting opportunity 1602. The duration of the time window 1604 may depend on a timeline (e.g., a fast timeline or a shorter timeline) applied to the CSI processing, for example, as described herein with respect to Figure 14 and Figure 15 As described. In some cases, the location of the timing reference 1606 may correspond to the short reference resource plus a certain number of symbols or time slots (T), where T may be reported as the UE capability. In certain aspects, the duration of the time window may depend on the type of CSI being reported. For the initial transmission of aperiodic CSI or semi-persistent CSI, the time window 1604 may start when the CSI request is received or when semi-persistent CSI is activated. For subsequent transmissions of periodic CSI or semi-persistent CSI, the time window 1604 may start a certain number of symbols (or time slots) before the CSI reference resource (e.g., before the receive opportunity 520). The duration of the time window supported by the UE may be reported as the UE capability.

[0231] In certain aspects, criteria for concurrently processing multiple machine learning models in an OFDM symbol may be supported. As used herein, concurrent processing may refer to performing processing at the same time. For example, a UE may be capable of processing a certain number of processing units associated with a machine learning model in a particular time instance, where the time instance may include one or more OFDM symbols, one or more time slots, or a predefined duration. The UE may indicate (to the radio access network) the number N of supported simultaneous machine learning model processing units in one or more component carriers, in one or more frequency bands, or across all component carriers. MPU In some cases, the UE may report the MPU occupied by a particular machine learning model in a time instance or duration (e.g., one or more OFDM symbols or time slots), in each one or more component carriers, in each one or more frequency bands, or across all component carriers. In some cases, the MPU occupied by each model may be predefined. The MPU for each model may be the same or different.

[0232] The UE may determine the MPU allocation for each AI-based inference, report, or measurement (e.g., CSI reporting, beam management, positioning, or channel estimation) according to specific rules for allocating unoccupied MPUs. MPU The UE may have N simultaneous machine learning computations for processing AI-based inference, reporting or measurement. MPU Model Processing Units (MPUs). If, in a given time instance (e.g., OFDM symbol), the computation of AI-based inference, reporting, or measurement occupies L MPUs, the UE may have N MPU - L unoccupied MPUs. For example, if S AI-based inferences, reports, or measurements currently occupy MPUs, and there are N MPU -L unoccupied MPUs, the UE can allocate N MPU - L MPUs are allocated to S' requested AI-based reports such that the sum of MPUs associated with the requested AI-based reports is less than or equal to N MPU -L unoccupied MPUs: in is the number of MPUs allocated to the i-th AI-based report. In some aspects, if the model is already active, the number of MPUs allocated to the i-th AI-based report may be equal to zero (e.g., ). In some aspects, the number of MPUs allocated to AI-based reporting It may depend on the model complexity, and the UE may report the number of MPUs associated with a particular model to the radio access network as the UE capability. The priority order for assigning MPUs may be based on various criteria, such as model identifier, carrier identifier, serving cell identifier, bandwidth part (subcarrier) identifier, report identifier, or any combination thereof. For example, the UE may assign an MPU to an AI-based report with a higher (or lower) model identifier than another AI-based report. The UE may not update the remaining S-S' AI-based reports. In some cases, the UE may report dummy information for AI-based reports that are not assigned an MPU. In some cases, the UE may report the budget of concurrent processing units (N MPU ). The UE may also report the identification number (or weight) of each specific supported model The allowed concurrency model combinations (set ) and the number of inferences allowed (k i ), the sum of inferences of each model satisfies the following expression: For model combinations UE can process up to k of model i i Inference. In this case, the priority order for allocating MPUs can be used. The supported model concurrency and its corresponding number of inference tasks are limited by Constraint. It is expected that no additional model inference tasks exceeding this limit will be assigned to the UE.

[0233] In certain aspects, the criteria for concurrently processing multiple machine learning models may be based on the model combinations supported by the UE and the number of inference tasks associated with the machine learning models of each combination. The UE may report the concurrent inference capability associated with the combination of models. For example, for each model combination, the UE may report the concurrent inference capability based on the total number of inference tasks that can be shared among all models in the corresponding combination and / or the number of inference tasks for each model in the corresponding combination. As an example, the UE may report a model combination including Model 1 and Model 2 (e.g., {Model 1, Model 2}) with up to three inference tasks, such that Model 1 occupies two inference tasks when Model 2 occupies one inference task, or Model 1 occupies one inference task when Model 2 occupies two inference tasks. As another example, the UE may report a model combination of Model 1 and Model 2, where Model 1 supports up to three inference tasks and Model 2 supports up to two inference tasks. The UE may be capable of concurrently processing up to three inference tasks for Model 1 and up to two inference tasks for Model 2. In certain aspects, the UE may report multiple model combinations and the number of inference tasks for each specific combination.

[0234] In some cases, if the inference opportunity is based on M measurement resources, then the triggered / configured inference report may use M inference tasks (or occupy M processing units). For example, for AI-based CSI feedback, if there are M CSI-RS resources for channel measurement, then the triggered aperiodic CSI report may use M inference tasks. If the triggered / configured inference uses an AI model, each with M i inference tasks, there are a total of Inference tasks. If the inference trigger indicates reporting multiple AI model inference reports that include more inference tasks than the reported UE capabilities, the UE may report only high-priority inferences and discard or not update (e.g., report dummy reports that include old or dummy information) low-priority reports that exceed the reported capabilities. The priority of the AI ​​model inference reports may be based on a model use case, a model identifier, a carrier identifier (e.g., a serving cell identifier), and / or a bandwidth part identifier.

[0235] An MPU associated with model (or AI-based report) and / or model inference task processing may occupy a certain amount of time (e.g., in symbols). In some cases, an MPU may be associated with an inference task or model. For the initial transmission of aperiodic CSI or semi-persistent CSI, the corresponding MPU may be occupied from the first symbol after the PDCCH that triggers the CSI report until the last symbol of the scheduled PUSCH carrying the report. For subsequent transmissions of periodic CSI and semi-persistent CSI, the corresponding MPU may be occupied from the first symbol of the earliest of the following until the last symbol of the configured PUSCH / PUCCH carrying the report: for each CSI-RS / CSI-IM / SSB resource used for channel or interference measurement, the corresponding latest CSI-RS / CSI-IM / SSB timing no later than the corresponding CSI reference resource (e.g., starting from the last timing of the reference signal for a certain period of time before the report) until the last symbol of the configured PUSCH / PUCCH carrying the report. In some cases, for aperiodic CSI, the corresponding MPU may be occupied from the first symbol after the PDCCH that triggers the CSI report until the last symbol of the scheduled PUSCH carrying the report. For periodic CSI and semi-persistent CSI, the corresponding MPU may be occupied from the activation / configuration of periodic or semi-persistent reporting until the deactivation / reconfiguration of periodic or semi-persistent reporting. It should be understood that the duration of the MPU described herein can be applied to inference task processing.

[0236] In certain aspects, the criteria for concurrently processing multiple machine learning models may be based on a process identifier associated with a particular model. For example, the UE may report a model process identifier associated with each model. Models with the same process identifier may not support concurrent processing with each other, and models with different process identifiers may support concurrent processing with each other. Models with different process identifiers may be processed concurrently with each other.

[0237] It should be understood that the aspects described herein with respect to AI-based CSI reporting are merely examples of AI-based processing for wireless communications. In addition to or as an alternative to the AI-based CSI described herein, the machine learning timelines and concurrent processing criteria described herein may be applied to other AI-generated information. As an example, in addition to or as an alternative to the AI-based CSI described herein, a UE may use a machine learning model to report beam management information, positioning information, channel estimation information, or any combination thereof. Accordingly, a UE may apply the various timelines and concurrent processing criteria described herein to generate AI-based CSI, beam management information, positioning information, channel estimation information, or any combination thereof.

[0238] Figure 17 A process flow 1700 is depicted for communicating in a network between a BS 102, a UE 104, and a model server 704. In some aspects, the BS 102 may be a Figure 1 and Figure 3 Base stations depicted and described with respect to or Figure 2 An example of a decomposed base station is depicted and described. Similarly, UE 104 may be a Figure 1 and Figure 3 However, in other aspects, UE 104 may be another type of wireless communication device and BS 102 may be another type of network entity or network node, such as those described herein.

[0239] At activity 1702, the UE 104 may receive information associated with one or more machine learning models, such as runtime images, setup parameters, and / or training data associated with the machine learning models, from the model server 704. In some cases, the UE 104 may initially load at least some of the machine learning models as active models, e.g., as described herein with respect to Figure 7 and Figure 8 described.

[0240] At activity 1704, the UE 104 may send machine learning capability information 1724 to the BS 102, the machine learning capability information indicating, for example, the UE's ability to process and / or store active and / or inactive machine learning models. For example, the machine learning capability information 1724 may include values ​​for M, N, and T (e.g., {M=3, N=2, T=0}), as described herein with respect to Figure 9 as described. The machine learning capability information 1724 may include combinations of machine learning models that the UE is capable of processing or storing concurrently, such as a first combination including a first model, a second model, and a fourth model (e.g., C1: {M1, M2, M4}), and a second combination including a third model and a fifth model (e.g., {M3, M5}). In some cases, the machine learning capability information 1724 may include the machine learning capability level supported by the UE, the maximum number of active machine learning models that the UE is capable of processing or storing (e.g., via a modem of the UE), the maximum number of inactive machine learning models that the UE is capable of storing (e.g., in memory), the minimum amount of time for activating an inactive machine learning model, the combination of machine learning models that the UE is capable of processing or storing concurrently (and their corresponding number of inference tasks), or any combination thereof. In certain aspects, the machine learning capability information 1724 may include any UE capability information described herein, such as the inference tasks supported by a combination of models or process identifiers associated with the models.

[0241] At activity 1706, the UE 104 may receive an indication of a machine learning configuration 1726 from the BS 102. The UE 104 may receive the machine learning configuration 1726 via RRC signaling (e.g., RRC configuration or reconfiguration), MAC signaling, and / or system information. The machine learning configuration 1726 may satisfy the machine learning capabilities associated with the UE 104. The machine learning configuration 1726 may indicate the use of a first model for aperiodic CSI, a second model for periodic CSI, and a fourth model for periodic beam management. The UE 104 may load the first model, the second model, and the fourth model as active machine learning models, and the UE 104 may load the third model and the fifth model as inactive machine learning models. As an example, the machine learning configuration 1726 may indicate the use of the first model for aperiodic CSI via a model identifier associated with the first model in the aperiodic CSI configuration. In some cases, the machine learning configuration 1726 may also include configuration / reconfiguration of periodic / aperiodic CSI associated with the machine learning model and / or activation / deactivation of semi-persistent CSI associated with the machine learning model.

[0242] At activity 1708, UE 104 may receive a trigger from BS 102 to report aperiodic CSI, for example, via a PDCCH carrying DCI indicating that CSI is to be reported at a reporting opportunity (e.g., reporting opportunity 504). Aperiodic CSI reporting may be associated with a first model. In some cases, the aperiodic CSI reporting trigger may indicate measuring a reference signal (e.g., a first reference signal) sent after the PDCCH triggers the CSI. It should be understood that the aperiodic CSI trigger is merely an example. The aperiodic trigger may indicate reporting of other information in addition to or in lieu of CSI, such as beam management information and / or UE positioning information.

[0243] At activity 1710, UE 104 may receive a first reference signal (e.g., SSB and / or CSI-RS) associated with an aperiodic CSI report from BS 102. At activity 1712, UE 104 may receive a second reference signal (e.g., SSB and / or CSI-RS) associated with a periodic CSI report from BS 102. In some cases, the first reference signal may be associated with a different transmission setting than the second reference signal (e.g., a narrower beam, a beam with a different angle of arrival (AoA), or a beam from a different cell or a different antenna layout (including antenna architecture, TxRU to antenna element mapping)). In some cases, the first reference signal and / or the second reference signal may represent interference measurement resources, where UE 104 may measure interference from other wireless devices (e.g., one or more UEs and / or base stations) in the reception opportunity. It should be understood that periodic CSI is merely an example. The UE may be configured to periodically report (e.g., via periodic reporting configuration or semi-persistent activation) other AI-based information, such as beam management information and / or UE positioning information.

[0244] At activity 1714, the UE 104 may determine CSI based on measurements associated with the first reference signal and the second reference signal using the first machine learning model and the second machine learning model. The UE 104 may determine the CSI based on measurements associated with the first reference signal using the first machine learning model, and the UE 104 may determine another CSI based on measurements associated with the second reference signal using the second machine learning model.

[0245] In some cases, the UE 104 may determine CSI using a first machine learning model while simultaneously determining another CSI using a second machine learning model if a machine learning processing constraint (e.g., as described herein) is satisfied. For example, assume that the MPU (or the number of concurrent inference tasks) associated with the first machine learning model is occupied for a first time period 1720 that begins when a PDCCH is received and ends when a CSI report is sent, and the MPU (or the number of concurrent inference tasks) associated with the second machine learning model is occupied for a second time period 1722 that begins when a second reference signal is received and ends when a CSI report is sent. If the MPUs (or the number of concurrent inference tasks) associated with the first machine learning model and the second machine learning model satisfy a threshold, the UE 104 may perform simultaneous machine learning model computations for the first and second machine learning models. For example, if the sum of the MPUs (or the number of concurrent inference tasks) associated with the first machine learning model and the second machine learning model is less than or equal to the number of MPUs (or the number of concurrent inference tasks) that are not occupied in the time period in which the first time period 1720 and the second time period 1722 overlap, the UE 104 may perform concurrent machine learning model operations using the first machine learning model and the second machine learning model.

[0246] At activity 1716, the UE 104 may send a CSI report including CSI associated with the first reference signal and the second reference signal to the BS 102, eg, via a PUSCH or a PUCCH.

[0247] At activity 1718, UE 104 can communicate with BS 102 via an adaptive communication link, where the link can be adapted based on the CSI report. For example, BS 102 can adjust the modulation and coding scheme (MCS), subcarrier spacing, frequency, bandwidth, and / or coding rate in response to the CSI report. As another example, BS 102 can change the beam used for communication in response to the CSI report.

[0248] It should be understood that AI-based CSI reporting is merely an example. The UE may be configured or triggered to report other AI-based information in addition to or instead of CSI, such as beam management information, UE positioning information, and / or (CSI-RS or DMRS) channel estimation. The UE may be configured to report AI-based information as an aperiodic report, a periodic report, or a semi-persistent report or a layer 3 report (e.g., an RRC report). For beam management, the UE may be configured with a reference signal associated with the beam for beam management, and the UE may report AI-based beam management information in response to an aperiodic trigger, a periodic reporting configuration, and / or a semi-persistent reporting activation. For example, the UE may report a preferred beam to be used in a future transmission opportunity. For UE positioning, the UE may be configured with a reference signal for determining UE positioning, and the UE may report UE positioning in response to an aperiodic trigger, a periodic reporting configuration, and / or a semi-persistent reporting activation. For channel estimation, the UE may be configured with a reference signal to perform channel estimation, and the UE may perform channel estimation in response to an aperiodic trigger, a periodic configuration, and / or a semi-persistent activation.

[0249] Example Operations of User Equipment

[0250] Figure 18 Shown is a diagram of a device such as a Figure 1 and Figure 3 An example of a method 1800 for performing wireless communications with a UE 104).

[0251] Method 1800 may optionally begin at step 1805, where the UE may receive an indication (e.g., AI-based report 710) to report CSI associated with a first machine learning model (e.g., one of the active machine learning model 718 or the inactive machine learning model 720). The indication to report CSI may include a configuration for periodic CSI, an activation for semi-persistent CSI, and / or a trigger for aperiodic CSI. The UE may receive the indication from a network entity (such as BS 102) via DCI, RRC signaling, MAC signaling, and / or system information. In some cases, the operation of this step refers to as described in reference to Figure 24 The circuits for receiving and / or the code for receiving are described, or can be performed by the circuits and / or the code.

[0252] Method 1800 then proceeds to step 1810, where the UE may receive a reference signal or interference measurement resource associated with the CSI from a network entity, such as described herein with respect to Figure 7 and Figure 17As described. For example, the reference signal may include DMRS, CSI-RS, SSB and / or phase tracking reference signal (PTRS). The interference measurement resource may include time-frequency resources for measuring interference from other wireless communication devices (such as other UEs or base stations) at the UE. The UE may determine the CSI based on measurements associated with the reference signal using a first machine learning model. The UE may receive the reference signal before and / or after receiving an indication to report the CSI. In some cases, the operation of this step refers to as described in reference Figure 24 The circuits for receiving and / or the code for receiving are described, or can be performed by the circuits and / or the code.

[0253] Method 1800 then proceeds to step 1815, where the UE may send a CSI report associated with a reference signal or interference measurement resource to a network entity based at least in part on a machine learning capability associated with the user equipment. For example, in certain aspects, certain timing criteria may be associated with the machine learning capability, such as described herein with respect to Figures 11A to 15 In some aspects, sending the CSI report at step 1815 includes sending the CSI report in response to satisfying a timing constraint, wherein the timing constraint may be based at least in part on a machine learning capability. In some aspects, a timing constraint may be satisfied if an event (e.g., sending a CSI report, receiving a reference signal, or measuring interference) occurs before (e.g., occurs no later than) or starts no earlier than (e.g., occurs after) a timing reference (e.g., the first timing reference 508, the second timing reference 512, or the third timing reference 524), for example, as described herein with respect to Figure 5A and Figure 5B As an example, for non-periodic CSI, the timing constraint may be satisfied if the event (e.g., sending a CSI report) does not start earlier than the timing reference, as described in conjunction with Figure 5A For periodic or semi-persistent CSI, the timing constraint may be satisfied if the event (e.g., receiving a reference signal or measuring an interference measurement resource) occurs before the timing reference, as described in conjunction with Figure 5B The time position of the timing reference may be determined based at least in part on machine learning capabilities. In some cases, the operation of this step refers to the operation of Figure 24 The described circuits for transmitting and / or code for transmitting may be performed by or be executable by the circuits and / or code.

[0254] In certain aspects, the machine learning capabilities can be based on certain capability parameters or combinations of models. For example, the machine learning capabilities can include: a first number (e.g., M) of one or more active machine learning models capable of being processed by the user equipment (capable of being processed at the user equipment); a second number (e.g., N) of one or more inactive machine learning models capable of being processed by the user equipment; a delay (e.g., additional time T) for switching between active and inactive machine learning models; a combination of machine learning models capable of being processed in an active state (e.g., active machine learning model 718) or an inactive state (e.g., inactive machine learning model 720); or any combination thereof.

[0255] In some aspects, machine learning capabilities can be classified by level or category, for example, as in this paper relative to Figure 9 In some aspects, a machine learning capability includes: a first capability (e.g., level 1.0) having at most a single active machine learning model and zero inactive machine learning models; a second capability (e.g., level 1.1) having at most a single active machine learning model, one or more inactive machine learning models, and a first duration (e.g., Z+T) of processing (for processing) the inactive machine learning models; and a third capability (e.g., level 1.2) having at most a single active machine learning model, one or more inactive machine learning models, and a second duration (e.g., Z) of processing (for processing) the inactive machine learning models, where the first duration can be longer than the second duration. Machine learning capabilities may also include a fourth capability (e.g., level 2.0) having one or more active machine learning models and zero inactive machine learning models; a fifth capability (e.g., level 2.1) having one or more active machine learning models, one or more inactive machine learning models, and a third duration (e.g., Z+T) of processing (for processing) inactive machine learning models; or a sixth capability (e.g., level 2.2) having one or more active machine learning models, one or more inactive machine learning models, and a fourth duration (e.g., Z) of processing (for processing) inactive machine learning models, where the third duration may be longer than the fourth duration.

[0256] In some aspects, method 1800 also includes sending an indication of machine learning capabilities (e.g., machine learning capability information 1724). In some cases, the operation of this step refers to as referring to Figure 24 The described circuits for transmitting and / or code for transmitting may be performed by or be executable by the circuits and / or code.

[0257] In some aspects, method 1800 also includes receiving a configuration (e.g., machine learning configuration 1726) indicating a first set of one or more active machine learning models (e.g., active machine learning models 718) and a second set of one or more inactive machine learning models (e.g., inactive machine learning models 720). The first total number of the first set of one or more active machine learning models is less than or equal to the first number of the one or more active machine learning models. The second total number of the second set of one or more inactive machine learning models is less than or equal to the second number of the one or more inactive machine learning models. In some cases, the operations of this step refer to as described with reference to Figure 24 The circuits for receiving and / or the code for receiving are described, or can be performed by the circuits and / or the code.

[0258] In certain aspects, the UE may load the active model at different times in response to a configuration indicating the active model, e.g., as described herein with respect to Figure 8 In some cases, the UE may download model information in response to receiving the configuration and load the model information. In some aspects, the method 1800 also includes receiving a configuration (e.g., machine learning configuration 1726) indicating that a first machine learning model is used for CSI reporting. In some cases, the configuration may indicate a first set of one or more active machine learning models and / or a second set of one or more inactive machine learning models. In some cases, the operation of this step refers to the operation of the method described in reference to FIG. Figure 24 In some aspects, the method 1800 further comprises receiving, in response to receiving the configuration, first information associated with a first set of one or more active machine learning models (e.g., training data, runtime images, and / or setup parameters) and second information associated with a second set of one or more inactive machine learning models (e.g., training data, runtime images, and / or setup parameters). In some cases, the operation of this step refers to the operation of the method described in the embodiment of the present invention. Figure 24 In some aspects, the method 1800 further includes loading the first information into a modem (e.g., a modulator and demodulator of the transceiver 354). In some cases, the operation of this step refers to the operation of the method described in the embodiment of the present invention. Figure 24 The described circuit for loading and / or code for loading may be executed by the circuit and / or the code.

[0259] In some cases, the UE may download the model information before receiving the configuration and load the model information in response to receiving the configuration. In some aspects, the method 1800 also includes receiving a configuration (e.g., machine learning configuration 1726) indicating that a first machine learning model is used for CSI reporting. In some cases, the configuration may indicate a first set of one or more active machine learning models and / or a second set of one or more inactive machine learning models. In some cases, the operation of this step refers to as described in reference to Figure 24 In some aspects, the method 1800 further comprises receiving first information associated with a first set of one or more active machine learning models and second information associated with a second set of one or more inactive machine learning models. In some cases, the operation of this step refers to the operation of the method described in the embodiment of the present invention. Figure 24 In some aspects, the method 1800 further comprises, in response to receiving the configuration, loading the first information into a modem (e.g., a modulator and demodulator of the transceiver 354). In some cases, the operation of this step is referred to as referring to Figure 24 The described circuit for loading and / or code for loading may be executed by the circuit and / or the code.

[0260] In certain aspects, the UE may report to the radio access network that the UE has successfully loaded the active model, e.g., as described herein with respect to Figure 8 In some aspects, method 1800 further includes, in response to receiving the configuration, sending a confirmation of successfully loading at least one active machine learning model in the first set of one or more active machine learning models (e.g., the confirmation at activity 810). In some cases, the operation of this step refers to the operation of the method described in reference to FIG. Figure 24 The described circuitry for sending and / or code for sending, or executable by the circuitry and / or code. In some aspects, sending the confirmation includes sending multiple confirmations (e.g., confirmations for each model), where each confirmation is for a (different) one of the first set of one or more active machine learning models. In some aspects, sending the confirmation includes sending the confirmation in a time window (e.g., time window 812) that begins when the configuration is received. In some aspects, the duration associated with the time window depends on the first number of the one or more active machine learning models or the second number of the one or more inactive machine learning models.

[0261] In some aspects, method 1800 further includes determining a CSI report based on the non-artificial intelligence codebook if an acknowledgment is not sent within a time window (e.g., time window 812) that begins when the configuration is received, wherein the acknowledgment indicates successful loading of at least one active machine learning model of the first set of one or more active machine learning models. In some cases, the operation of this step refers to as described in reference to Figure 24 The circuit for determining and / or the code for determining are described, or can be performed by the circuit and / or the code.

[0262] In some cases, after downloading and / or loading the active models, the UE may report to the network entity the machine learning models available for processing at the UE (e.g., the models that are active). The UE may indicate the models that support fast timelines as the initial set of active models, as described herein. In some aspects, the method 1800 further includes (in response to receiving the configuration) sending an indication of a timing constraint (e.g., Z, Z', and / or Y) associated with at least one active machine learning model in the first set of one or more active machine learning models. In some cases, the operations of this step refer to as described in reference to Figure 24 The described circuits for transmitting and / or code for transmitting may be performed by or be executable by the circuits and / or code.

[0263] In one aspect, method 1800 or any aspect related thereto may be performed by an apparatus such as Figure 24 The method 1800 is performed by a communication device 2400 comprising various components operable, configured, or adapted to perform the method 1800. The communication device 2400 is described in more detail below.

[0264] Please note that Figure 18 This is merely one example of a method, and other methods including fewer, additional, or alternative steps may also be consistent with the present disclosure.

[0265] Figure 19 Shown is a diagram of a device such as a Figure 1 and Figure 3 An example of a method 1900 for performing wireless communications with a UE 104).

[0266] Method 1900 may optionally begin at step 1905, where the UE may receive an indication to report CSI associated with a first machine learning model, e.g., as described herein with respect to Figure 18 In some cases, the operation of this step is referred to as Figure 24 The circuits for receiving and / or the code for receiving are described, or can be performed by the circuits and / or the code.

[0267] Method 1900 then proceeds to step 1910, where the UE may determine a set of one or more active machine learning models (e.g., model states 1420, 1520) in response to receiving the indication. In some aspects, the set includes a series of one or more identifiers (e.g., model identifiers, CSI report identifiers, etc.) associated with the one or more active machine learning models. In some cases, the operations of this step refer to the following as referenced in FIG. Figure 24 The circuit for determining and / or the code for determining are described, or can be performed by the circuit and / or the code.

[0268] Method 1900 may then proceed to step 1915, where the UE may receive a reference signal or interference measurement resource associated with the CSI, e.g., as described herein with respect to Figure 18 In some cases, the operation of this step is referred to as Figure 24 The circuits for receiving and / or the code for receiving are described, or can be performed by the circuits and / or the code.

[0269] Method 1900 then proceeds to step 1920, where the UE may send a CSI report based on a reference signal or interference measurement resource in response to at least satisfying a timing constraint, wherein the timing constraint is based at least in part on a set of one or more active machine learning models. The timing constraint may represent a value relative to the reference signal or interference measurement resource. Figure 5A 、 Figure 5B and Figures 12A to 15 In some aspects, the timing constraint may be satisfied if an event (e.g., sending a CSI report and / or receiving a reference signal) occurs before or starts no earlier than a timing reference (e.g., the first timing reference 508, the second timing reference 512, or the third timing reference 524), wherein the temporal position of the timing reference is determined based at least in part on a machine learning capability associated with the user equipment. In some cases, the operation of this step refers to as described with reference to Figure 24 The described circuits for transmitting and / or code for transmitting may be performed by or be executable by the circuits and / or code.

[0270] In certain aspects, the timing constraints may be associated with Z of the aperiodic CSI, as described herein with respect to Figure 5A In some aspects, the indication further indicates that the CSI is to be reported in a reporting opportunity (e.g., reporting opportunity 504). The event includes a reporting opportunity (sent in a reporting opportunity). The timing reference may be located in time at a first number (e.g., Z) of one or more symbols after the last symbol of the indication that triggered the CSI report.

[0271] For certain aspects, timing constraints may be associated with aperiodic Z', as described herein with respect to Figure 5AIn some aspects, the indication further indicates that the CSI is reported in a reporting opportunity (e.g., reporting opportunity 504). The event includes a reporting opportunity (sent in a reporting opportunity). The timing reference is positioned in time at a second number (e.g., Z') of one or more symbols after the last symbol of the reference signal or interference measurement resource.

[0272] In certain aspects, timing constraints may be associated with Y for periodic CSI and / or semi-persistent CSI, as described herein with respect to Figure 5B In some aspects, the indication further indicates reporting CSI in a reporting opportunity (e.g., reporting opportunity 522). The CSI report includes periodic CSI or semi-persistent CSI. The event includes a reception opportunity (e.g., reception opportunity 520) associated with a reference signal or interference measurement resource (received in the reception opportunity). The timing reference is temporally located at a third number (e.g., Y) of one or more symbols prior to the reporting opportunity.

[0273] In some aspects, method 1900 further includes: if the first machine learning model is in the set of one or more active machine learning models, selecting a first value for the third number of one or more symbols. In some aspects, method 1900 further includes: if the first machine learning model is not in the set of one or more active machine learning models, selecting a second value for the third number of one or more symbols, wherein the second value is different from the first value. For example, the first value may be less than the second value. In some cases, the operation of this step refers to as described in reference to Figure 24 The circuit for selecting and / or the code for selecting are described, or can be performed by the circuit and / or the code.

[0274] In some aspects, the method 1900 further includes: if the CSI report includes periodic CSI, selecting a first value for the third number of one or more symbols. In some aspects, the method 1900 further includes: if the CSI report is an initial transmission associated with semi-persistent CSI, selecting a second value for the third number of one or more symbols. In some aspects, the method 1900 further includes: if the CSI report is a subsequent transmission associated with semi-persistent CSI, selecting a third value for the third number of one or more symbols, wherein the second value is different from the third value (e.g., the second value may be greater than the third value). In some cases, the operations of this step refer to as described with reference to Figure 24 The circuit for selecting and / or the code for selecting are described, or can be performed by the circuit and / or the code.

[0275] In some aspects, the UE may report information associated with the timing constraints to the network entity. In some aspects, the method 1900 further includes sending an indication of the time location of the timing reference. For example, for a particular machine learning model, the UE may indicate the length or duration of certain timing constraints (such as Z, Z' and / or Y). In some cases, the operation of this step refers to the operation of the reference 1900 as described in the reference 1900. Figure 24 The described circuits for transmitting and / or code for transmitting, or executable by the circuits and / or code. In some aspects, the temporal position of the timing reference may depend in part on: an identifier associated with the first machine learning model (e.g., a CSI report identifier, a model identifier, etc.), a rank indicator associated with one or more transmission layers, whether a CSI feedback decoder is available for processing CSI, whether another machine learning model is active, or any combination thereof.

[0276] In some aspects, method 1900 further includes determining a temporal position of a timing reference based at least in part on a set of one or more active machine learning models. In some cases, the operation of this step is referred to as referring to Figure 24 Circuitry for determining and / or code for determining described herein, or executable by the circuitry and / or code. In some aspects, the timing reference is temporally positioned at: a first number of one or more symbols after the last symbol of an indication that triggered a CSI report (e.g., the timing reference is associated with Z), a second number of one or more symbols after the last symbol of a reference signal (e.g., the timing reference is associated with Z'), or a third number of one or more symbols before a reporting opportunity associated with the CSI report (e.g., the timing reference is associated with Y). In some aspects, determining the temporal location of the timing reference includes: if the first machine learning model is in a set of one or more active machine learning models, selecting a first value for the first number of one or more symbols, selecting a second value for the second number of one or more symbols, or selecting a third value for the third number of one or more symbols; and if the first machine learning model is not in the set of one or more active machine learning models, selecting a fourth value for the first number of one or more symbols, selecting a fifth value for the second number of one or more symbols, or selecting a sixth value for the third number of one or more symbols, wherein the second value is different from the first value (e.g., the first value is greater than the second value). In some aspects, the first value is less than the fourth value; the second value is less than the fifth value; and the third value is less than the sixth value.

[0277] In some aspects, method 1900 further includes sending an indication of machine learning capabilities associated with the user equipment, wherein the machine learning capabilities include a first number of one or more active machine learning models capable of being processed by the user equipment, and wherein the total number of machine learning models in the set is less than or equal to the first number. In some aspects, method 1900 further includes sending an indication of machine learning capabilities associated with the user equipment, wherein the machine learning capabilities include a combination of machine learning models capable of being processed in an active state, and wherein the one or more active machine learning models in the set are in the combination of machine learning models. In some cases, the operation of this step refers to as described with reference to Figure 24 The described circuits for transmitting and / or code for transmitting may be performed by or be executable by the circuits and / or code.

[0278] In some aspects, the set of one or more active machine learning models may comprise an ordered series, wherein certain rules may be used to add or remove entries in the series. In some aspects, determining the set comprises: if the first machine learning model is not in the set, updating the set to include the first machine learning model in response to receiving an indication to report CSI. In some aspects, updating the set comprises: removing a first or last identifier associated with the second machine learning model from the set; and adding an identifier associated with the first machine learning model to the set. In some aspects, updating the set comprises: removing the second machine learning model from the set; sending an indication to remove the second machine learning model from the set; and adding the first machine learning model to the set.

[0279] In some aspects, the indication further indicates reporting CSI associated with a first machine learning model and a second machine learning model in the same UCI; inserting the first machine learning model in a first position of the set if the first machine learning model is associated with an identifier having a smaller value than the second machine learning model; and inserting the second machine learning model in the first position of the set if the second machine learning model is associated with an identifier having a smaller value than the first machine learning model. In some cases, the operation of this step refers to as described in reference Figure 24 The described circuit for inserting and / or code for inserting may be performed by the circuit and / or the code.

[0280] In some aspects, certain rules may be applied to prevent or handle back-to-back instructions to use inactive machine learning models, for example, as described herein with respect to Figure 16In some aspects, method 1900 further includes, in response to satisfying the third timing constraint, omitting an indication of reporting CSI associated with a second machine learning model that is not in the set of one or more machine learning models. In some cases, the operation of this step refers to as described with reference to Figure 24 In some aspects, the method 1900 further comprises refraining from sending another CSI report associated with the second machine learning model in response to satisfying the third timing constraint. In some cases, the operation of this step refers to the operation of the method described in the embodiment of the present invention. Figure 24 Circuitry for suppression and / or code for suppression described, or executable by the circuitry and / or code. In some aspects, the third timing constraint is satisfied when an indication to report CSI associated with the second machine learning model is received in a time window that ends at a reporting opportunity for sending a CSI report, or another CSI report is scheduled to be reported in the time window. In some aspects, for an initial transmission of non-periodic CSI or semi-persistent CSI, the time window begins when an indication to report CSI associated with the first machine learning model is received; and for subsequent transmissions of periodic CSI or semi-persistent CSI, the time window begins at a third number of one or more symbols before a reference signal (or interference measurement resource). In some aspects, the third number of one or more symbols depends on whether the first machine learning model is in a set of one or more active machine learning models. In some aspects, method 1900 also includes sending an indication of a third number of one or more symbols. In some cases, the operation of this step refers to as described with reference to Figure 24 The described circuits for transmitting and / or code for transmitting may be performed by or be executable by the circuits and / or code.

[0281] In one aspect, method 1900 or any aspect related thereto may be performed by an apparatus such as Figure 24 The method 1900 is performed by a communication device 2400 comprising various components operable, configured, or adapted to perform the method 1900. The communication device 2400 is described in more detail below.

[0282] Please note that Figure 19 This is merely one example of a method, and other methods including fewer, additional, or alternative steps may also be consistent with the present disclosure.

[0283] Figure 20 Shown is a diagram of a device such as a Figure 1 and Figure 3 An example of method 2000 for performing wireless communications with a UE 104).

[0284] Method 2000 may optionally begin at step 2005, where the UE may receive an indication to report information associated with a first machine learning model (e.g., AI-based report 710), e.g., as described herein with respect to Figure 18 In some aspects, the information includes channel state information, beam management information, positioning information, channel estimation information, or any combination thereof. In some cases, the operation of this step refers to the operation of Figure 24 The circuits for receiving and / or the code for receiving are described, or can be performed by the circuits and / or the code.

[0285] Method 2000 then proceeds to step 2010, where the UE may determine the information using the first machine learning model while concurrently determining other information using the second machine learning model in response to satisfying the machine learning processing constraints (e.g., the concurrent processing criteria described herein). In some cases, the operations of this step refer to the following. Figure 24 The circuit for determining and / or the code for determining are described, or can be performed by the circuit and / or the code.

[0286] Method 2000 then proceeds to step 2015, where the UE may send a report indicating the information. In some cases, the operation of this step refers to the operation of Figure 24 The described circuits for transmitting and / or code for transmitting may be performed by or be executable by the circuits and / or code.

[0287] In some aspects, method 2000 further includes sending an indication of the machine learning capabilities, the machine learning capabilities comprising: a combination of machine learning models capable of processing in an active state or an inactive state; a first number of inference tasks that can be shared among the set of active machine learning models or models in the combination, a second number of one or more inference tasks for each model in the set of active machine learning models or the combination, a threshold associated with a machine learning processing constraint per one or more subcarriers, per one or more carriers, or per one or more frequency bands; or a combination thereof. In some cases, the operation of this step refers to as described in reference to Figure 24 The described circuits for transmitting and / or code for transmitting may be performed by or be executable by the circuits and / or code.

[0288] In certain aspects, machine learning processing constraints can be associated with inference tasks as described herein. In some aspects, a machine learning processing constraint can be satisfied if: a first total number of inference tasks associated with a first machine learning model and a second machine learning model is less than or equal to a first number; a second total number of inference tasks associated with the first machine learning model is less than or equal to a second number associated with the first machine learning model; and a third total number of inference tasks associated with the second machine learning model is less than or equal to the second number associated with the second machine learning model.

[0289] In some aspects, the machine learning processing constraints may be associated with model processing units as described herein. In some aspects, the machine learning processing constraints may be satisfied if the total number of processing units associated with the active machine learning model is less than or equal to a threshold. In some aspects, method 2000 further includes sending an indication of a first number of one or more processing units associated with the first machine learning model and a second number of one or more processing units associated with the second machine learning model. In some cases, the operations of this step refer to the steps described in reference to Figure 24 In some aspects, if the first machine learning model is active, the number of one or more processing units associated with the first machine learning model is equal to zero. In some aspects, method 2000 further includes allocating unoccupied processing units to the first machine learning model based at least in part on the priority associated with the active machine learning model. In some cases, the operation of this step refers to the operation of the first machine learning model as described in reference to FIG. Figure 24 The described circuits for allocating and / or code for allocating, or executable by the circuits and / or code, in some aspects, the priority associated with the active machine learning model is based at least in part on: a model identifier, a carrier identifier, a bandwidth portion identifier, a CSI report identifier, or any combination thereof.

[0290] In some aspects, if the report includes aperiodic CSI or an initial transmission associated with semi-persistent CSI, one or more processing units or one or more inference tasks associated with the first machine learning model are occupied from the time the indication is received until the report is sent; and if the report includes periodic CSI or a subsequent transmission associated with semi-persistent CSI, one or more processing units or one or more inference tasks associated with the first machine learning model are occupied from the time a last opportunity associated with a reference signal (or interference measurement resource) is received before a timing reference (e.g., a CSI reference resource such as a third timing reference 524) until the report is sent.

[0291] In some aspects, if the report includes non-periodic CSI, one or more processing units or one or more inference tasks associated with the first machine learning model are occupied from the time the indication is received until the report is sent; and if the report includes periodic CSI or semi-persistent CSI, one or more processing units or one or more inference tasks associated with the first machine learning model are occupied from the time activation or configuration of the report is received until deactivation or reconfiguration of the report is received, where the reconfiguration may disable periodic reporting.

[0292] In some aspects, the method 2000 further includes sending an indication of a process identifier for each of the plurality of machine learning models, wherein the process identifier indicates whether the corresponding machine learning model can be processed concurrently with another machine learning model. In some cases, the operation of this step refers to the process identifier of the machine learning model. Figure 24 The described circuits for sending and / or code for sending, or executable by the circuits and / or code. In some aspects, concurrent processing of machine learning models with different process identifiers is allowed.

[0293] In one aspect, method 2000 or any aspect related thereto may be performed by an apparatus such as Figure 24 The method 2000 is performed by a communication device 2400 comprising various components operable, configured, or adapted to perform the method 2000. The communication device 2400 is described in more detail below.

[0294] Please note that Figure 20 This is merely one example of a method, and other methods including fewer, additional, or alternative steps may also be consistent with the present disclosure.

[0295] Example Operations of Network Entities

[0296] Figure 21 shows that network entities such as Figure 1 and Figure 3 BS102 or as compared to Figure 2 An example of a method 2100 for performing wireless communication by a decomposed base station is discussed. The method 2100 may be complementary to the method 1800 performed by a UE.

[0297] Method 2100 may optionally begin at step 2105, where a network entity may output (e.g., output for transmission, transmit, or provide) an indication to a UE (e.g., UE 104) reporting CSI associated with a first machine learning model. For example, the network may send the indication to a UE (such as UE 104) via DCI, RRC signaling, MAC signaling, and / or system information. In some cases, the operation of this step refers to the operation described in reference to Figure 25The described circuit for outputting and / or code for outputting may be performed by the circuit and / or code.

[0298] Method 2100 then proceeds to step 2110, where the network entity may obtain a CSI report associated with a reference signal or interference measurement resource from the user equipment (e.g., UE 104) based at least in part on machine learning capabilities associated with the UE. For example, the network entity may receive the CSI report via a PUSCH or PUCCH. In some cases, the operation of this step refers to the operation of Figure 25 The described circuit for obtaining and / or code for obtaining may be performed by the circuit and / or code.

[0299] In some aspects, the method 2100 further includes outputting a reference signal associated with the CSI, e.g., as described herein with respect to Figure 7 and Figure 17 described.

[0300] In some aspects, obtaining the CSI report includes obtaining the CSI report in response to satisfying a timing constraint, wherein the timing constraint is based at least in part on a machine learning capability. In some aspects, the timing constraint is satisfied if the event occurs before or starts no earlier than a timing reference, wherein a temporal position of the timing reference is determined at least in part on the machine learning capability. In some cases, the operation of this step refers to as described with reference to Figure 25 The described circuit for outputting and / or code for outputting may be performed by the circuit and / or code.

[0301] In some aspects, the machine learning capabilities include: a first number of one or more active machine learning models capable of being processed by the user equipment; a second number of one or more inactive machine learning models capable of being processed by the user equipment; a delay for switching between active machine learning models and inactive machine learning models; a combination of machine learning models capable of being processed in an active state or an inactive state; or any combination thereof.

[0302] In some aspects, method 2100 further includes outputting a configuration indicating a first set of one or more active machine learning models and indicating a second set of one or more inactive machine learning models, wherein a first total number of the first set of one or more active machine learning models is less than or equal to a first number of the one or more active machine learning models, and wherein a second total number of the second set of one or more inactive machine learning models is less than or equal to a second number of the one or more inactive machine learning models. In some cases, the operation of this step refers to as described with reference to Figure 25 The described circuit for outputting and / or code for outputting may be performed by the circuit and / or code.

[0303] In some aspects, the machine learning capability includes: a first capability having at most a single active machine learning model and zero inactive machine learning models; a second capability having at most a single active machine learning model, one or more inactive machine learning models, and a first duration of processing (for processing) the inactive machine learning models; a third capability having at most a single active machine learning model, one or more inactive machine learning models, and a second duration of processing (for processing) the inactive machine learning models, wherein the first duration is longer than the second duration; a fourth capability having one or more active machine learning models and zero inactive machine learning models; a fifth capability having one or more active machine learning models, one or more inactive machine learning models, and a third duration of processing (for processing) the inactive machine learning models; or a sixth capability having one or more active machine learning models, one or more inactive machine learning models, and a fourth duration of processing (for processing) the inactive machine learning models, wherein the third duration is longer than the fourth duration.

[0304] In some aspects, method 2100 further includes obtaining an indication of machine learning capability. In some cases, the operation of this step is referred to as referring to Figure 25 The described circuit for obtaining and / or code for obtaining may be performed by the circuit and / or code.

[0305] In some aspects, the method 2100 further includes outputting a configuration indicating use of the first machine learning model for CSI reporting. In some aspects, the configuration may indicate a first set of one or more active machine learning models and / or a second set of one or more inactive machine learning models. In some cases, the operation of this step refers to the operation of Figure 25 The described circuit for outputting and / or code for outputting may be performed by the circuit and / or code.

[0306] In some aspects, the method 2100 further includes, in response to receiving the configuration, obtaining confirmation that at least one active machine learning model of the first set of one or more active machine learning models is successfully loaded at the user equipment. In some cases, the operation of this step refers to as described in reference Figure 25The described circuitry for obtaining and / or code for obtaining, or executable by the circuitry and / or code. In some aspects, obtaining the confirmation includes obtaining a plurality of confirmations, each confirmation for one active machine learning model in a first set of one or more active machine learning models. In some aspects, obtaining the confirmation includes obtaining the confirmation in a time window that begins when the configuration is output. In some aspects, the duration associated with the time window depends on a first number of one or more active machine learning models or a second number of one or more inactive machine learning models. In some aspects, method 2100 also includes outputting a configuration indicating use of the first machine learning model for CSI reporting, wherein if no confirmation is obtained in the time window that begins when the configuration is output, the CSI report is based on a non-artificial intelligence codebook, wherein the confirmation indicates successful loading of at least one active machine learning model in the first set of one or more active machine learning models at the user equipment. In some cases, the operation of this step refers to as described in reference to Figure 25 The described circuit for outputting and / or code for outputting may be performed by the circuit and / or code.

[0307] In some cases, after downloading and / or loading the active model, the UE may report to the network entity the machine learning models available for processing at the UE (e.g., the models that are active). The UE may indicate the models that support fast timeline as the initial set of active models, as described herein. In some aspects, the method 2100 further includes, in response to outputting the configuration, obtaining an indication of a timing constraint associated with at least one active machine learning model in the first set of one or more active machine learning models. In some cases, the operations of this step refer to as described in reference to Figure 25 The described circuit for obtaining and / or code for obtaining may be performed by the circuit and / or code.

[0308] In one aspect, method 2100 or any aspect related thereto may be performed by an apparatus such as Figure 25 The method 2100 is performed by a communication device 2500 comprising various components operable, configured, or adapted to perform the method 2100. The communication device 2500 is described in more detail below.

[0309] Please note that Figure 21 This is merely one example of a method, and other methods including fewer, additional, or alternative steps may also be consistent with the present disclosure.

[0310] Figure 22 shows that network entities such as Figure 1 and Figure 3 BS102 or as compared to Figure 2An example of a method 2200 for performing wireless communication by a decomposed base station is discussed. The method 2200 may be complementary to the method 1900 performed by a UE.

[0311] Method 2200 may optionally begin at step 2205, where the network entity may output an indication reporting CSI associated with the first machine learning model, e.g., as described herein with respect to Figure 21 In some cases, the operation of this step is referred to as Figure 25 The described circuit for outputting and / or code for outputting may be performed by the circuit and / or code.

[0312] The method 2200 then proceeds to step 2210, where the network entity may determine a set of one or more active machine learning models (e.g., model states 1420, 1520) in response to outputting the indication. In some cases, the operation of this step refers to the operation of Figure 25 The circuit for determining and / or the code for determining are described, or can be performed by the circuit and / or the code.

[0313] Method 2200 may then optionally proceed to step 2215, where the network entity may output a reference signal associated with the CSI. In some cases, the CSI report may be used for interference measurement at the user equipment, and the network entity may not output a reference signal. In some cases, the operation of this step refers to the reference signal. Figure 25 The described circuit for outputting and / or code for outputting may be performed by the circuit and / or code.

[0314] Method 2200 then proceeds to step 2220, where the network entity may obtain a CSI report based on a reference signal or interference measurement resource in response to at least satisfying a timing constraint, wherein the timing constraint is based at least in part on a set of one or more active machine learning models. In some cases, the operation of this step refers to as referring to Figure 25 Circuitry for obtaining and / or code for obtaining, or executable by the circuitry and / or code, as described herein, in some aspects satisfies a timing constraint if the event occurs before or starts no earlier than a timing reference, wherein a temporal position of the timing reference is determined based at least in part on machine learning capabilities associated with the user equipment.

[0315] In certain aspects, the timing constraints may be associated with Z of the aperiodic CSI, as described herein with respect to Figure 5A In some aspects, the indication further indicates reporting the CSI in a reporting opportunity; the event comprises the reporting opportunity; and the timing reference is positioned in time a first number of one or more symbols after a last symbol of the indication triggering the CSI reporting.

[0316] For certain aspects, timing constraints may be associated with aperiodic Z', as described herein with respect to Figure 5A In some aspects, the indication further indicates reporting the CSI in a reporting opportunity; the event comprises the reporting opportunity; and the timing reference is positioned in time a second number of one or more symbols after a last symbol of the reference signal.

[0317] In certain aspects, timing constraints may be associated with Y for periodic CSI and / or semi-persistent CSI, as described herein with respect to Figure 5B In some aspects, the indication further indicates reporting CSI in a reporting opportunity; the CSI report comprises periodic CSI or semi-persistent CSI; the event comprises a receive opportunity associated with a reference signal; and the timing reference is positioned in time a third number of one or more symbols before the reporting opportunity.

[0318] In some aspects, the method 2200 further includes: if the first machine learning model is in the set of one or more active machine learning models, selecting a first value for the third number of the one or more symbols. In some aspects, the method 2200 further includes: if the first machine learning model is not in the set of one or more active machine learning models, selecting a second value for the third number of the one or more symbols, wherein the second value is different from the first value. In some cases, the operation of this step refers to as described in reference to Figure 25 The circuit for selecting and / or the code for selecting are described, or can be performed by the circuit and / or the code.

[0319] In some aspects, the method 2200 further comprises: if the CSI report includes periodic CSI, selecting a first value for the third number of one or more symbols. In some aspects, the method 2200 further comprises: if the CSI report is an initial transmission associated with semi-persistent CSI, selecting a second value for the third number of one or more symbols. In some aspects, the method 2200 further comprises: if the CSI report is a subsequent transmission associated with semi-persistent CSI, selecting a third value for the third number of one or more symbols, wherein the second value is different from the third value. In some cases, the operations of this step refer to as described with reference to Figure 25 The circuit for selecting and / or the code for selecting are described, or can be performed by the circuit and / or the code.

[0320] In some aspects, the method 2200 further comprises obtaining an indication of a time position of a timing reference. In some cases, the operation of this step is referred to as referring to Figure 25The described circuitry for obtaining and / or code for obtaining, or executable by the circuitry and / or code. In some aspects, the temporal position of the timing reference depends in part on: an identifier associated with the first machine learning model, a rank indicator associated with one or more transmission layers, whether a CSI feedback decoder is available, whether another machine learning model is active, or any combination thereof.

[0321] In some aspects, the method 2200 further comprises determining a temporal position of a timing reference based at least in part on the set of one or more active machine learning models. In some cases, the operation of this step is referred to as referring to Figure 25 The circuit for determining and / or the code for determining are described, or can be performed by the circuit and / or the code.

[0322] In some aspects, the timing reference is positioned in time at: a first number of one or more symbols after a last symbol of an indication that triggered a CSI report, a second number of one or more symbols after a last symbol of a reference signal, or a third number of one or more symbols before a reporting opportunity associated with the CSI report. In some aspects, determining the temporal location of the timing reference comprises: if the first machine learning model is in a set of one or more active machine learning models, selecting a first value for the first number of one or more symbols, selecting a second value for the second number of one or more symbols, or selecting a third value for the third number of one or more symbols; and if the first machine learning model is not in the set of one or more active machine learning models, selecting a fourth value for the first number of one or more symbols, selecting a fifth value for the second number of one or more symbols, or selecting a sixth value for the third number of one or more symbols, wherein the second value is different from the first value. In some aspects, the first value is less than the fourth value; the second value is less than the fifth value; and the third value is less than the sixth value. In some aspects, the set comprises a series of one or more identifiers associated with the one or more active machine learning models.

[0323] In some aspects, method 2200 further includes obtaining an indication of machine learning capabilities associated with the user equipment, wherein the machine learning capabilities include a first number of one or more active machine learning models capable of being processed by the user equipment, and wherein a total number of machine learning models in the set is less than or equal to the first number. In some cases, the operation of this step refers to as described with reference to Figure 25 The described circuit for obtaining and / or code for obtaining may be performed by the circuit and / or code.

[0324] In some aspects, the method 2200 further includes obtaining an indication of a machine learning capability associated with the user equipment, wherein the machine learning capability includes a combination of machine learning models capable of processing in an active state, and wherein one or more active machine learning models in the set are in the combination of machine learning models. In some cases, the operation of this step refers to as described with reference to Figure 25 The described circuit for obtaining and / or code for obtaining may be performed by the circuit and / or code.

[0325] In some aspects, the set of one or more active machine learning models may comprise an ordered series, wherein certain rules may be used to add or remove entries in the series. In some aspects, determining the set comprises: if the first machine learning model is not in the set, updating the set to include the first machine learning model in response to receiving an indication to report CSI. In some aspects, updating the set comprises: removing a first or last identifier associated with the second machine learning model from the set; and adding an identifier associated with the first machine learning model to the set. In some aspects, updating the set comprises: removing the second machine learning model from the set; obtaining an indication to remove the second machine learning model from the set; and adding the first machine learning model to the set.

[0326] In some aspects, the indication further indicates reporting CSI associated with a first machine learning model and a second machine learning model in the same UCI; inserting the first machine learning model in a first position of the set if the first machine learning model is associated with an identifier having a smaller value than the second machine learning model; and inserting the second machine learning model in the first position of the set if the second machine learning model is associated with an identifier having a smaller value than the first machine learning model. In some cases, the operation of this step refers to as described in reference Figure 25 The described circuit for inserting and / or code for inserting may be performed by the circuit and / or the code.

[0327] In some aspects, certain rules may be applied to prevent or handle back-to-back instructions to use inactive machine learning models, for example, as described herein with respect to Figure 16 In some aspects, the method 2200 further includes suppressing outputting an indication of reporting CSI associated with a second machine learning model that is not in the set of one or more machine learning models during a time window ending at a reporting opportunity for obtaining a CSI report. In some cases, the operation of this step refers to the operation of Figure 25Circuitry for suppression and / or code for suppression as described, or executable by the circuitry and / or code. In some aspects, for an initial transmission of aperiodic CSI or semi-persistent CSI, the time window begins when an indication of CSI associated with the first machine learning model is output; and for subsequent transmissions of periodic CSI or semi-persistent CSI, the time window begins at a third number of one or more symbols before the reference signal. In some aspects, the third number of one or more symbols depends on whether the first machine learning model is in a set of one or more active machine learning models. In some aspects, the method 2200 further includes obtaining an indication of the third number of one or more symbols. In some cases, the operation of this step refers to as described with reference to Figure 25 The described circuit for obtaining and / or code for obtaining may be performed by the circuit and / or code.

[0328] In one aspect, method 2200 or any aspect related thereto may be performed by an apparatus such as Figure 25 The method 2200 is performed by a communication device 2500 comprising various components operable, configured, or adapted to perform the method 2200. The communication device 2500 is described in more detail below.

[0329] Please note that Figure 22 This is merely one example of a method, and other methods including fewer, additional, or alternative steps may also be consistent with the present disclosure.

[0330] Figure 23 shows that network entities such as Figure 1 and Figure 3 BS102 or as compared to Figure 2 An example of a method 2300 for performing wireless communications by a decomposed base station is discussed. The method 2300 may be complementary to the method 2000 performed by a UE.

[0331] Method 2300 may optionally begin at step 2305, where the network entity may output an indication to a UE (e.g., UE 104) to report information associated with the first machine learning model. For example, the information may include channel state information, beam management information, positioning information, channel estimation information, or any combination thereof. In some cases, the operation of this step refers to as described in reference to Figure 25 The described circuit for outputting and / or code for outputting may be performed by the circuit and / or code.

[0332] The method 2300 then proceeds to step 2310, where the network entity may obtain a report (e.g., a CSI report) from the UE indicating the information based on satisfying the machine learning processing constraints. In some cases, the operation of this step refers to the operation of Figure 25The described circuitry for obtaining and / or code for obtaining, or executable by the circuitry and / or code. In some aspects, the machine learning processing constraint allows concurrent processing with the first machine learning model and the second machine learning model.

[0333] In some aspects, method 2300 further includes obtaining an indication of a machine learning capability associated with the UE, the machine learning capability comprising: a combination of machine learning models capable of processing in an active state or an inactive state; a first number of inference tasks that can be shared among a set of active machine learning models or models in the combination, a second number of one or more inference tasks for each model in the set of active machine learning models or the combination, a threshold associated with a machine learning processing constraint per one or more subcarriers, per one or more carriers, or per one or more frequency bands; or a combination thereof. In some cases, the operation of this step refers to as referenced in Figure 25 The described circuit for obtaining and / or code for obtaining may be performed by the circuit and / or code.

[0334] In some aspects, a machine learning processing constraint may be satisfied if: a first total number of inference tasks associated with the first machine learning model and the second machine learning model is less than or equal to the first number; a second total number of inference tasks associated with the first machine learning model is less than or equal to the second number associated with the first machine learning model; and a third total number of inference tasks associated with the second machine learning model is less than or equal to the second number associated with the second machine learning model.

[0335] In some aspects, the machine learning processing constraint may be satisfied if the total number of processing units associated with the active machine learning model is less than or equal to a threshold. In some aspects, method 2300 further includes obtaining an indication of a first number of one or more processing units associated with the first machine learning model and a second number of one or more processing units associated with the second machine learning model. In some cases, the operations of this step refer to steps such as those described with reference to Figure 25 The described circuitry for obtaining and / or code for obtaining, or executable by the circuitry and / or code. In some aspects, if the first machine learning model is active, the number of one or more processing units associated with the first machine learning model is equal to zero.

[0336] In some aspects, if the report includes non-periodic CSI or an initial transmission associated with semi-persistent CSI, one or more processing units or one or more inference tasks associated with the first machine learning model are occupied from the time the indication is output until the report is obtained; and if the report includes periodic CSI or a subsequent transmission associated with semi-persistent CSI, one or more processing units or one or more inference tasks associated with the first machine learning model are occupied from the time a last opportunity associated with a reference signal is output before a timing reference (e.g., a CSI reference resource such as a third timing reference 524) until the report is obtained.

[0337] In some aspects, if the report includes non-periodic CSI, one or more processing units or one or more inference tasks associated with the first machine learning model are occupied from the time the indication is output until the report is obtained; and if the report includes periodic CSI or semi-persistent CSI, one or more processing units or one or more inference tasks associated with the first machine learning model are occupied from the time the report is activated or configured to be output until the report is deactivated or reconfigured to be output.

[0338] In some aspects, the method 2300 further includes obtaining an indication of a process identifier for each of the plurality of machine learning models, wherein the process identifier indicates whether the corresponding machine learning model can be processed concurrently with another machine learning model. In some cases, the operation of this step refers to the process identifier of the machine learning model. Figure 25 The described circuitry for obtaining and / or code for obtaining, or executable by the circuitry and / or code. In some aspects, concurrent processing of machine learning models with different process identifiers is allowed.

[0339] In one aspect, method 2300 or any aspect related thereto may be performed by an apparatus such as Figure 25 The method 2300 is performed by a communication device 2500 comprising various components operable, configured, or adapted to perform the method 2300. The communication device 2500 is described in more detail below.

[0340] Please note that Figure 23 This is merely one example of a method, and other methods including fewer, additional, or alternative steps may also be consistent with the present disclosure.

[0341] Example Communication Device

[0342] Figure 24 Depicted are aspects of an example communication device 2400. In some examples, the communication device 2400 is user equipment, such as described above with respect to Figure 1 and Figure 3 UE 104 is described.

[0343] The communication device 2400 includes a processing system 2405 coupled to a transceiver 2494 (e.g., a transmitter and / or a receiver). The transceiver 2494 is configured to transmit and receive signals for the communication device 2400, such as the various signals described herein, via an antenna 2496. The processing system 2405 can be configured to perform processing functions for the communication device 2400, including processing signals received and / or to be transmitted by the communication device 2400.

[0344] The processing system 2405 includes one or more processors 2410. In various aspects, such as with respect to Figure 3 As depicted, the one or more processors 2410 may represent one or more of the receive processor 358, the transmit processor 364, the TX MIMO processor 366, and / or the controller / processor 380. The one or more processors 2410 are coupled to the computer-readable medium / memory 2460 via the bus 2492. In certain aspects, the computer-readable medium / memory 2460 is configured to store instructions (e.g., computer-executable code) that, when executed by the one or more processors 2410, cause the one or more processors 2410 to: Figure 18 Method 1800 or any aspect thereof as described; Figure 19 Method 1900 as described or any aspect thereof; and / or with respect to Figure 20 The described method 2000 or any aspect related thereto. Note that references to a processor of the communication device 2400 performing a function may include one or more processors 2410 of the communication device 2400 performing that function.

[0345] In the depicted example, the computer-readable medium / memory 2460 stores code (e.g., executable instructions) such as code for receiving 2465, code for sending 2470, code for loading 2475, code for determining 2480, code for ignoring 2482, code for suppressing 2484, code for selecting 2486, code for allocating 2488, and code for inserting 2490. Processing of the code for receiving 2465, code for sending 2470, code for loading 2475, code for determining 2480, code for ignoring 2482, code for suppressing 2484, code for selecting 2486, code for allocating 2488, and code for inserting 2490 may cause the communication device 2400 to perform: Figure 18 Method 1800 or any aspect thereof as described; Figure 19 Method 1900 as described or any aspect thereof; and / or with respect to Figure 20The described method 2000 or any aspect related thereto.

[0346] The one or more processors 2410 include circuitry configured to implement (e.g., execute) code stored in the computer-readable medium / memory 2460, including circuitry such as circuitry for receiving 2415, circuitry for transmitting 2420, circuitry for loading 2425, circuitry for determining 2430, circuitry for ignoring 2435, circuitry for suppressing 2440, circuitry for selecting 2445, circuitry for allocating 2450, and circuitry for inserting 2455. Processing using the circuitry for receiving 2415, circuitry for transmitting 2420, circuitry for loading 2425, circuitry for determining 2430, circuitry for ignoring 2435, circuitry for suppressing 2440, circuitry for selecting 2445, circuitry for allocating 2450, and circuitry for inserting 2455 may cause the communication device 2400 to perform: Figure 18 Method 1800 or any aspect thereof as described; Figure 19 Method 1900 as described or any aspect thereof; and / or with respect to Figure 20 The described method 2000 or any aspect related thereto.

[0347] The various components of the communication device 2400 may provide means for performing the following: Figure 18 Method 1800 or any aspect thereof as described; Figure 19 Method 1900 as described or any aspect thereof; and / or with respect to Figure 20 The method 2000 described herein or any aspect thereof. For example, components for sending, transmitting, or outputting for sending may include Figure 3 The illustrated transceiver 354 and / or antenna 352 and / or Figure 24 The transceiver 2494 and antenna 2496 of the communication device 2400 in FIG. Components for receiving or obtaining may include Figure 3 The illustrated transceiver 354 and / or antenna 352 and / or Figure 24 The transceiver 2494 and antenna 2496 of the communication device 2400 are shown.

[0348] Figure 25 Aspects of an example communications device 2500 are depicted. In some aspects, communications device 2500 is a network entity such as Figure 1 and Figure 3 BS102 or as compared to Figure 2 The decomposed base station in question.

[0349] The communication device 2500 includes a processing system 2505 coupled to a transceiver 2575 (e.g., a transmitter and / or receiver) and / or a network interface 2582. The transceiver 2575 is configured to transmit and receive signals for the communication device 2500 via an antenna 2580, such as the various signals described herein. The network interface 2582 is configured to receive signals for the communication device 2500 via a communication link (e.g., such as the various signals described herein). Figure 2 The processing system 2505 may be configured to perform processing functions of the communication device 2500, including processing signals received and / or to be transmitted by the communication device 2500.

[0350] The processing system 2505 includes one or more processors 2510. In various aspects, such as with respect to Figure 3 As depicted, the one or more processors 2510 may represent one or more of the receive processor 338, the transmit processor 320, the TX MIMO processor 330, and / or the controller / processor 340. The one or more processors 2510 are coupled to the computer-readable medium / memory 2540 via the bus 2570. In certain aspects, the computer-readable medium / memory 2540 is configured to store instructions (e.g., computer-executable code) that, when executed by the one or more processors 2510, cause the one or more processors 2510 to: Figure 21 Method 2100 as described or any aspect thereof; Figure 22 Method 2200 or any aspect thereof as described; and / or with respect to Figure 23 The described method 2300 or any aspect related thereto. Note that references to a processor of the communication device 2500 performing a function may include one or more processors 2510 of the communication device 2500 performing that function.

[0351] In the depicted example, the computer-readable medium / memory 2540 stores code (e.g., executable instructions), such as code for outputting 2545, code for obtaining 2550, code for determining 2555, code for suppressing 2560, and code for selecting 2565. Processing of the code for outputting 2545, code for obtaining 2550, code for determining 2555, code for suppressing 2560, and code for selecting 2565 may cause the communication device 2500 to perform: Figure 21 Method 2100 as described or any aspect thereof; Figure 22 Method 2200 or any aspect thereof as described; and / or with respect to Figure 23 The described method 2300 or any aspect related thereto.

[0352] The one or more processors 2510 include circuits configured to implement (e.g., execute) code stored in the computer-readable medium / memory 2540, including circuits such as circuitry for outputting 2515, circuitry for obtaining 2520, circuitry for determining 2525, circuitry for suppressing 2530, and circuitry for selecting 2535. Processing using the circuitry for outputting 2515, circuitry for obtaining 2520, circuitry for determining 2525, circuitry for suppressing 2530, and circuitry for selecting 2535 may cause the communication device 2500 to perform: Figure 21 Method 2100 as described or any aspect thereof; Figure 22 Method 2200 or any aspect thereof as described; and / or with respect to Figure 23 The described method 2300 or any aspect related thereto.

[0353] The various components of the communication device 2500 may provide means for performing the following: Figure 21 Method 2100 as described or any aspect thereof; Figure 22 Method 2200 or any aspect thereof as described; and / or with respect to Figure 23 The described method 2300 or any aspect thereof. Means for sending, transmitting, or outputting for sending may include Figure 3 The illustrated transceiver 332 and / or antenna 334 of the BS 102 and / or Figure 25 The transceiver 2575 and antenna 2580 of the communication device 2500 in FIG. Components for receiving or obtaining may include Figure 3 The illustrated transceiver 332 and / or antenna 334 and / or Figure 25 The transceiver 2575 and antenna 2580 of the communication device 2500 are shown.

[0354] Sample Clauses

[0355] Specific implementation examples are described in the following numbered clauses:

[0356] Clause 1: A method of wireless communication by a user equipment, the method comprising: receiving an indication to report CSI associated with a first machine learning model; receiving a reference signal associated with the CSI; and sending a CSI report associated with the reference signal based at least in part on a machine learning capability associated with the user equipment.

[0357] Clause 2: The method of clause 1, wherein sending the CSI report comprises sending the CSI report in response to satisfying a timing constraint, wherein the timing constraint is based at least in part on the machine learning capability.

[0358] Clause 3: A method according to clause 2, wherein: the timing constraint is satisfied if the event occurs before or starts no earlier than a timing reference; and the time position of the timing reference is determined at least in part based on the machine learning capability.

[0359] Clause 4: A method according to any one of clauses 1 to 3, wherein the machine learning capability comprises: a first number of one or more active machine learning models capable of being processed by the user equipment; a second number of one or more inactive machine learning models capable of being processed by the user equipment; a delay for switching between active machine learning models and inactive machine learning models; a combination of machine learning models capable of being processed in an active state or an inactive state; or any combination thereof.

[0360] Clause 5: The method according to Clause 4, further comprising: receiving a configuration indicating a first set of one or more active machine learning models and indicating a second set of one or more inactive machine learning models, wherein a first total number of the first set of one or more active machine learning models is less than or equal to a first number of the one or more active machine learning models, and wherein a second total number of the second set of one or more inactive machine learning models is less than or equal to a second number of the one or more inactive machine learning models.

[0361] Clause 6: A method according to any one of clauses 1 to 5, wherein the machine learning capability includes: a first capability of having at most a single active machine learning model and zero inactive machine learning models; a second capability of having at most a single active machine learning model, one or more inactive machine learning models, and a first duration of processing the inactive machine learning model; a third capability of having at most a single active machine learning model, one or more inactive machine learning models, and a second duration of processing the inactive machine learning model, wherein the first duration is longer than the second duration; a fourth capability of having one or more active machine learning models and zero inactive machine learning models; a fifth capability of having one or more active machine learning models, one or more inactive machine learning models, and a third duration of processing the inactive machine learning model; or a sixth capability of having one or more active machine learning models, one or more inactive machine learning models, and a fourth duration of processing the inactive machine learning model, wherein the third duration is longer than the fourth duration.

[0362] Clause 7: A method according to any one of clauses 1 to 6, further comprising: sending an indication of the machine learning capability.

[0363] Clause 8: A method according to any one of clauses 1 to 7, the method further comprising: receiving a configuration indicating a first set of one or more active machine learning models and indicating a second set of one or more inactive machine learning models; receiving first information associated with the first set of one or more active machine learning models and second information associated with the second set of one or more inactive machine learning models in response to receiving the configuration; and loading the first information in a modem.

[0364] Clause 9: A method according to any one of clauses 1 to 8, the method further comprising: receiving a configuration indicating a first set of one or more active machine learning models and indicating a second set of one or more inactive machine learning models; receiving first information associated with the first set of one or more active machine learning models and second information associated with the second set of one or more inactive machine learning models; and loading the first information in a modem in response to receiving the configuration.

[0365] Clause 10: A method according to any one of clauses 1 to 9, further comprising: receiving a configuration indicating a first set of one or more active machine learning models and indicating a second set of one or more inactive machine learning models, and sending a confirmation of successful loading of at least one active machine learning model in the first set of the one or more active machine learning models in response to receiving the configuration.

[0366] Clause 11: The method of clause 10, wherein sending the confirmation comprises sending a plurality of confirmations, wherein each of the confirmations is for an active machine learning model in the first set of the one or more active machine learning models.

[0367] Clause 12: The method of clause 10 or 11, wherein sending the confirmation comprises sending the confirmation within a time window beginning when the configuration is received.

[0368] Clause 13: A method according to clause 12, wherein the machine learning capability includes a first number of one or more active machine learning models capable of being processed by the device and a second number of one or more inactive machine learning models capable of being processed by the device; and the duration associated with the time window depends on the first number of the one or more active machine learning models or the second number of the one or more inactive machine learning models.

[0369] Clause 14: A method according to any one of clauses 10 to 13, further comprising: sending an indication of a timing constraint associated with at least one active machine learning model in the first set of the one or more active machine learning models in response to receiving the configuration.

[0370] Clause 15: A method according to any one of clauses 1 to 14, further comprising: receiving a configuration indicating a first set of one or more active machine learning models and indicating a second set of one or more inactive machine learning models, and determining the CSI report based on a non-artificial intelligence codebook in the event that no confirmation is sent within a time window starting when the configuration is received, wherein the confirmation indicates successful loading of at least one active machine learning model in the first set of one or more active machine learning models.

[0371] Clause 16: A method of wireless communication by a user equipment, the method comprising: receiving an indication to report CSI associated with a first machine learning model; determining a set of one or more active machine learning models in response to receiving the indication; receiving a reference signal associated with the CSI; and sending a CSI report based on the reference signal at least in response to satisfying a timing constraint, wherein the timing constraint is based at least in part on the set of one or more active machine learning models.

[0372] Clause 17: A method according to clause 16, wherein: the timing constraint is satisfied if the event occurs before or starts no earlier than a timing reference; and the time position of the timing reference is determined at least in part based on machine learning capabilities associated with the user equipment.

[0373] Clause 18: A method according to clause 17, wherein: the indication further indicates reporting the CSI in a reporting opportunity; the event includes the reporting opportunity; and the timing reference is positioned in time at a first number of one or more symbols after the last symbol of the indication that triggers the CSI report.

[0374] Clause 19: A method according to clause 17 or 18, wherein: the indication further indicates reporting the CSI in a reporting opportunity; the event includes the reporting opportunity; and the timing reference is positioned in time at a second number of one or more symbols after the last symbol of the reference signal.

[0375] Clause 20: A method according to any one of clauses 17 to 19, wherein: the indication further indicates reporting the CSI in a reporting occasion; the CSI report comprises periodic CSI or semi-persistent CSI; the event comprises a reception occasion associated with the reference signal; and the timing reference is positioned in time at a third number of one or more symbols before the reporting occasion.

[0376] Clause 21: The method according to Clause 20, further comprising: selecting a first value for the third number of the one or more symbols when the first machine learning model is in the set of one or more active machine learning models, and selecting a second value for the third number of the one or more symbols when the first machine learning model is not in the set of one or more active machine learning models, wherein the second value is different from the first value.

[0377] Clause 22: A method according to clause 20, the method further comprising: selecting a first value for the third number of the one or more symbols when the CSI report includes periodic CSI; selecting a second value for the third number of the one or more symbols when the CSI report is an initial transmission associated with semi-persistent CSI; and selecting a third value for the third number of the one or more symbols when the CSI report is a subsequent transmission associated with semi-persistent CSI, wherein the second value is different from the third value.

[0378] Clause 23: A method according to any one of clauses 17 to 22, wherein the time position of the timing reference depends in part on: an identifier associated with the first machine learning model, a rank indicator associated with one or more transmission layers, whether a CSI feedback decoder is available, whether another machine learning model is active, or any combination thereof.

[0379] Clause 24: The method of any of clauses 17 to 23, further comprising sending an indication of the time position of the timing reference.

[0380] Clause 25: The method of any one of clauses 17 to 24, further comprising determining the temporal position of the timing reference based at least in part on the set of one or more active machine learning models.

[0381] Clause 26: A method according to clause 25, wherein: the timing reference is temporally positioned at: a first number of one or more symbols after the last symbol of the indication that triggers the CSI report, a second number of one or more symbols after the last symbol of the reference signal, or a third number of one or more symbols before a reporting opportunity associated with the CSI report; and determining the temporal position of the timing reference includes: if the first machine learning model is in the set of the one or more active machine learning models, selecting a first value for the first number of the one or more symbols, selecting a second value for the second number of the one or more symbols, or selecting a third value for the third number of the one or more symbols; and if the first machine learning model is not in the set of one or more active machine learning models, selecting a fourth value for the first number of the one or more symbols, selecting a fifth value for the second number of the one or more symbols, or selecting a sixth value for the third number of the one or more symbols, wherein the second value is different from the first value.

[0382] Clause 27: The method of clause 26, wherein: the first value is less than the fourth value; the second value is less than the fifth value; and the third value is less than the sixth value.

[0383] Clause 28: A method according to any one of clauses 16 to 27, wherein the set comprises a series of one or more identifiers associated with the one or more active machine learning models.

[0384] Clause 29: A method according to any one of clauses 16 to 28, the method further comprising: sending an indication of machine learning capabilities associated with the user equipment, wherein the machine learning capabilities include a first number of one or more active machine learning models capable of being processed by the user equipment, and wherein the total number of machine learning models in the set is less than or equal to the first number.

[0385] Clause 30: A method according to any one of clauses 16 to 29, the method further comprising: sending an indication of a machine learning capability associated with the user equipment, wherein the machine learning capability comprises a combination of machine learning models capable of processing in an active state, and wherein the one or more active machine learning models in the set are in the combination of machine learning models.

[0386] Clause 31: A method according to any one of clauses 16 to 30, wherein determining the set includes: in a case where the first machine learning model is not in the set, updating the set to include the first machine learning model in response to receiving the indication to report the CSI.

[0387] Clause 32: A method according to clause 31, wherein updating the set includes: removing the first identifier or the last identifier associated with the second machine learning model from the set; and adding the identifier associated with the first machine learning model to the set.

[0388] Clause 33: A method according to clause 31, wherein updating the set includes: removing a second machine learning model from the set; sending an indication to remove the second machine learning model from the set; and adding the first machine learning model to the set.

[0389] Clause 34: A method according to any one of clauses 31 to 33, wherein: the indication further indicates reporting CSI associated with the first machine learning model and the second machine learning model in the same UCI; in a case where the first machine learning model is associated with an identifier having a smaller value than that of the second machine learning model, inserting the first machine learning model in the first position of the set; and in a case where the second machine learning model is associated with an identifier smaller than that of the first machine learning model, inserting the second machine learning model in the first position of the set.

[0390] Clause 35: A method according to any one of clauses 16 to 34, the method further comprising: ignoring an indication of reporting CSI associated with a second machine learning model that is not in the set of one or more machine learning models in response to satisfying a third timing constraint, and suppressing sending another CSI report associated with the second machine learning model in response to satisfying the third timing constraint.

[0391] Clause 36: A method according to clause 35, wherein the third timing constraint is satisfied when: the indication to report CSI associated with the second machine learning model is received in a time window ending at the reporting opportunity for sending the CSI report, or the other CSI report is scheduled to be reported in the time window.

[0392] Clause 37: A method according to clause 36, wherein: for an initial transmission of non-periodic CSI or semi-persistent CSI, the time window begins when the indication to report the CSI associated with the first machine learning model is received; and for subsequent transmissions of periodic CSI or the semi-persistent CSI, the time window begins at a third number of one or more symbols before the reference signal.

[0393] Clause 38: The method of clause 37, wherein the third number of the one or more symbols depends on whether the first machine learning model is in a set of one or more active machine learning models.

[0394] Clause 39: A method as described in clause 37 or 38, further comprising: sending an indication of the third number of one or more symbols.

[0395] Clause 40: A method of wireless communication by a user equipment, the method comprising: receiving an indication to report information associated with a first machine learning model; determining the information using the first machine learning model and concurrently determining other information using a second machine learning model in response to satisfying a machine learning processing constraint; and sending a report indicating the information.

[0396] Clause 41: The method of clause 40, wherein the information comprises channel state information, beam management information, positioning information, channel estimation information, or any combination thereof.

[0397] Clause 42: A method according to any one of clauses 40 to 41, the method further comprising: sending an indication of machine learning capabilities, the machine learning capabilities comprising: a combination of machine learning models capable of processing in an active state or an inactive state; a first number of inference tasks that can be shared between a set of active machine learning models or models in the combination, a second number of one or more inference tasks for each model in the set of active machine learning models or the combination, a threshold associated with the machine learning processing constraint for each one or more subcarriers, each one or more carriers, or each one or more frequency bands; or a combination thereof.

[0398] Clause 43: A method according to clause 42, wherein the machine learning processing constraint is satisfied when: a first total number of inference tasks associated with the first machine learning model and the second machine learning model is less than or equal to the first number; a second total number of inference tasks associated with the first machine learning model is less than or equal to the second number associated with the first machine learning model; and a third total number of inference tasks associated with the second machine learning model is less than or equal to the second number associated with the second machine learning model.

[0399] Clause 44: A method according to any one of clauses 40 to 43, wherein the machine learning processing constraint is satisfied if the total number of processing units associated with the active machine learning model is less than or equal to a threshold.

[0400] Clause 45: The method of clause 44, further comprising sending an indication of a first number of one or more processing units associated with the first machine learning model and a second number of one or more processing units associated with the second machine learning model.

[0401] Clause 46: A method according to clause 44 or 45, wherein the number of one or more processing units associated with the first machine learning model is equal to zero when the first machine learning model is active.

[0402] Clause 47: The method of any one of clauses 44 to 46, further comprising allocating unoccupied processing units to the first machine learning model based at least in part on a priority associated with the active machine learning model.

[0403] Clause 48: The method of clause 47, wherein the priority associated with the active machine learning model is based at least in part on: a model identifier, a carrier identifier, a bandwidth portion identifier, a CSI report identifier, or any combination thereof.

[0404] Clause 49: A method according to any one of clauses 40 to 48, wherein: in the case where the report includes forward feedback of an initial transmission of non-periodic CSI or associated with semi-persistent CSI, one or more processing units or one or more inference tasks associated with the first machine learning model are occupied from the time the indication is received until the report is sent; and in the case where the report includes subsequent transmissions of periodic CSI or associated with semi-persistent CSI, the one or more processing units or the one or more inference tasks associated with the first machine learning model are occupied from the time the last opportunity associated with the reference signal is received before the timing reference until the report is sent.

[0405] Clause 50: A method according to any one of clauses 40 to 48, wherein: in the case where the report includes non-periodic CSI, one or more processing units or one or more inference tasks associated with the first machine learning model are occupied from the time the indication is received until the report is sent; and in the case where the report includes periodic CSI or semi-persistent CSI, the one or more processing units or the one or more inference tasks associated with the first machine learning model are occupied from the time activation or configuration of the report is received until deactivation or reconfiguration of the report is received.

[0406] Clause 51: A method according to any one of clauses 40 to 50, further comprising: sending an indication of a process identifier for each of a plurality of machine learning models, wherein the process identifier indicates whether the corresponding machine learning model can be processed concurrently with another machine learning model.

[0407] Clause 52: The method of clause 51, wherein concurrent processing of machine learning models with different process identifiers is allowed.

[0408] Clause 53: A method of wireless communication by a network entity, the method comprising: outputting an indication reporting CSI associated with a first machine learning model; and obtaining a CSI report associated with a reference signal based at least in part on machine learning capabilities associated with a user equipment.

[0409] Clause 54: The method of clause 53, further comprising: outputting a reference signal associated with the CSI, wherein obtaining the CSI report comprises obtaining the CSI report in response to satisfying a timing constraint, wherein the timing constraint is based at least in part on the machine learning capability.

[0410] Clause 55: A method according to clause 54, wherein: the timing constraint is satisfied if the event occurs before or starts no earlier than a timing reference; and the time position of the timing reference is determined at least in part based on the machine learning capability.

[0411] Clause 56: A method according to any one of clauses 53 to 55, wherein the machine learning capability comprises: a first number of one or more active machine learning models capable of being processed by the user equipment; a second number of one or more inactive machine learning models capable of being processed by the user equipment; a delay for switching between active machine learning models and inactive machine learning models; a combination of machine learning models capable of being processed in an active state or an inactive state; or any combination thereof.

[0412] Clause 57: The method according to Clause 56, further comprising: outputting a configuration indicating a first set of one or more active machine learning models and indicating a second set of one or more inactive machine learning models, wherein a first total number of the first set of one or more active machine learning models is less than or equal to a first number of the one or more active machine learning models, and wherein a second total number of the second set of one or more inactive machine learning models is less than or equal to a second number of the one or more inactive machine learning models.

[0413] Clause 58: A method according to any one of clauses 53 to 57, wherein the machine learning capability includes: a first capability of having at most a single active machine learning model and zero inactive machine learning models; a second capability of having at most a single active machine learning model, one or more inactive machine learning models, and a first duration of processing the inactive machine learning model; a third capability of having at most a single active machine learning model, one or more inactive machine learning models, and a second duration of processing the inactive machine learning model, wherein the first duration is longer than the second duration; a fourth capability of having one or more active machine learning models and zero inactive machine learning models; a fifth capability of having one or more active machine learning models, one or more inactive machine learning models, and a third duration of processing the inactive machine learning model; or a sixth capability of having one or more active machine learning models, one or more inactive machine learning models, and a fourth duration of processing the inactive machine learning model, wherein the third duration is longer than the fourth duration.

[0414] Clause 59: A method according to any one of clauses 53 to 58, further comprising: obtaining an indication of the machine learning capability.

[0415] Clause 60: A method according to any one of clauses 53 to 59, further comprising: outputting a configuration indicating a first set of one or more active machine learning models and indicating a second set of one or more inactive machine learning models, and obtaining confirmation of successful loading of at least one active machine learning model in the first set of one or more active machine learning models at the user equipment in response to receiving the configuration.

[0416] Clause 61: The method of clause 60, wherein obtaining the confirmation comprises obtaining a plurality of confirmations, wherein each of the confirmations is for an active machine learning model in the first set of one or more active machine learning models.

[0417] Clause 62: The method of clause 60 or 61, wherein obtaining the confirmation comprises obtaining the confirmation within a time window beginning when the configuration is output.

[0418] Clause 63: A method according to clause 62, wherein the machine learning capability includes: a first number of one or more active machine learning models capable of being processed by the device, and a second number of one or more inactive machine learning models capable of being processed by the device; and the duration associated with the time window depends on the first number of the one or more active machine learning models or the second number of the one or more inactive machine learning models.

[0419] Clause 64: A method according to any one of clauses 60 to 63, further comprising: obtaining an indication of a timing constraint associated with at least one active machine learning model in the first set of one or more active machine learning models in response to outputting the configuration.

[0420] Clause 65: A method according to any one of clauses 53 to 64, the method further comprising: outputting a configuration indicating a first set of one or more active machine learning models and indicating a second set of one or more inactive machine learning models, wherein in the event that no confirmation is obtained within a time window starting when the configuration is output, the CSI report is based on a non-artificial intelligence codebook, wherein the confirmation indicates that at least one active machine learning model of the first set of one or more active machine learning models is successfully loaded at the user equipment.

[0421] Clause 66: A method of wireless communication by a network entity, the method comprising: outputting an indication reporting CSI associated with a first machine learning model; determining a set of one or more active machine learning models in response to outputting the indication; outputting a reference signal associated with the CSI; and obtaining a CSI report based on the reference signal at least in response to satisfying a timing constraint, wherein the timing constraint is based at least in part on the set of one or more active machine learning models.

[0422] Clause 67: A method according to clause 66, wherein: the timing constraint is satisfied if the event occurs before or starts no earlier than a timing reference; and the time position of the timing reference is determined at least in part based on machine learning capabilities associated with the user equipment.

[0423] Clause 68: A method according to clause 67, wherein: the indication further indicates reporting the CSI in a reporting opportunity; the event includes the reporting opportunity; and the timing reference is positioned in time at a first number of one or more symbols after the last symbol of the indication that triggers the CSI report.

[0424] Clause 69: A method according to clause 67 or 68, wherein: the indication further indicates reporting the CSI in a reporting opportunity; the event includes the reporting opportunity; and the timing reference is positioned in time at a second number of one or more symbols after the last symbol of the reference signal.

[0425] Clause 70: A method according to any one of clauses 67 to 69, wherein: the indication further indicates reporting the CSI in a reporting occasion; the CSI report comprises periodic CSI or semi-persistent CSI; the event comprises a reception occasion associated with the reference signal; and the timing reference is positioned in time at a third number of one or more symbols before the reporting occasion.

[0426] Clause 71: The method according to Clause 70 further includes: selecting a first value for the third number of the one or more symbols when the first machine learning model is in the set of the one or more active machine learning models, and selecting a second value for the third number of the one or more symbols when the first machine learning model is not in the set of one or more active machine learning models, wherein the second value is different from the first value.

[0427] Clause 72: A method according to clause 70, the method further comprising: selecting a first value for the third number of the one or more symbols when the CSI report includes periodic CSI; selecting a second value for the third number of the one or more symbols when the CSI report is an initial transmission associated with semi-persistent CSI; and selecting a third value for the third number of the one or more symbols when the CSI report is a subsequent transmission associated with semi-persistent CSI, wherein the second value is different from the third value.

[0428] Clause 73: A method according to any one of clauses 67 to 72, wherein the time position of the timing reference depends in part on: an identifier associated with the first machine learning model, a rank indicator associated with one or more transmission layers, whether a CSI feedback decoder is available, whether another machine learning model is active, or any combination thereof.

[0429] Clause 74: The method of any of clauses 67 to 73, further comprising obtaining an indication of the temporal position of the timing reference.

[0430] Clause 75: The method of any one of clauses 67 to 74, further comprising determining the temporal position of the timing reference based at least in part on the set of one or more active machine learning models.

[0431] Clause 76: A method according to clause 75, wherein: the timing reference is temporally positioned at: a first number of one or more symbols after the last symbol of the indication that triggers the CSI report, a second number of one or more symbols after the last symbol of the reference signal, or a third number of one or more symbols before a reporting opportunity associated with the CSI report; and determining the temporal position of the timing reference includes: if the first machine learning model is in the set of the one or more active machine learning models, selecting a first value for the first number of the one or more symbols, selecting a second value for the second number of the one or more symbols, or selecting a third value for the third number of the one or more symbols; and if the first machine learning model is not in the set of one or more active machine learning models, selecting a fourth value for the first number of the one or more symbols, selecting a fifth value for the second number of the one or more symbols, or selecting a sixth value for the third number of the one or more symbols, wherein the second value is different from the first value.

[0432] Clause 77: The method of clause 76, wherein: the first value is less than the fourth value; the second value is less than the fifth value; and the third value is less than the sixth value.

[0433] Clause 78: A method according to any one of clauses 66 to 77, wherein the set comprises a series of one or more identifiers associated with the one or more active machine learning models.

[0434] Clause 79: A method according to any one of clauses 66 to 78, the method further comprising: obtaining an indication of machine learning capabilities associated with a user equipment, wherein the machine learning capabilities include a first number of one or more active machine learning models capable of being processed by the user equipment, and wherein the total number of machine learning models in the set is less than or equal to the first number.

[0435] Clause 80: A method according to any one of clauses 66 to 79, the method further comprising: obtaining an indication of machine learning capabilities associated with the user equipment, wherein the machine learning capabilities include a combination of machine learning models capable of processing in an active state, and wherein the one or more active machine learning models in the set are in the combination of machine learning models.

[0436] Clause 81: A method according to any one of clauses 66 to 80, wherein determining the set includes: in a case where the first machine learning model is not in the set, updating the set to include the first machine learning model in response to receiving the indication to report the CSI.

[0437] Clause 82: A method according to clause 81, wherein updating the set includes: removing the first identifier or the last identifier associated with the second machine learning model from the set; and adding the identifier associated with the first machine learning model to the set.

[0438] Clause 83: The method of clause 81, wherein updating the set comprises: removing a second machine learning model from the set; obtaining an indication to remove the second machine learning model from the set; and adding the first machine learning model to the set.

[0439] Clause 84: A method according to any one of clauses 81 to 83, wherein: the indication further indicates reporting CSI associated with the first machine learning model and the second machine learning model in the same UCI; in a case where the first machine learning model is associated with an identifier having a smaller value than that of the second machine learning model, inserting the first machine learning model in the first position of the set; and in a case where the second machine learning model is associated with an identifier smaller than that of the first machine learning model, inserting the second machine learning model in the first position of the set.

[0440] Clause 85: A method according to any one of clauses 66 to 84, further comprising: suppressing, in a time window ending at the reporting opportunity for obtaining the CSI report, an indication of the CSI associated with a second machine learning model that is not in the set of one or more machine learning models by outputting a report.

[0441] Clause 86: A method according to clause 85, wherein: for an initial transmission of non-periodic CSI or semi-persistent CSI, the time window begins when the output report indicates the CSI associated with the first machine learning model; and for subsequent transmissions of periodic CSI or the semi-persistent CSI, the time window begins at a third number of one or more symbols before the reference signal.

[0442] Clause 87: The method of clause 86, wherein the third number of the one or more symbols depends on whether the first machine learning model is in a set of one or more active machine learning models.

[0443] Clause 88: The method of clause 86 or 87, further comprising: obtaining an indication of the third number of one or more symbols.

[0444] Clause 89: A method of wireless communication by a network entity, the method comprising: outputting an indication reporting information associated with a first machine learning model; and obtaining a report indicating the information based on satisfying a machine learning processing constraint.

[0445] Clause 90: A method according to clause 89, wherein: the machine learning processing constraints allow concurrent processing with the first machine learning model and the second machine learning model; and the information includes channel state information, beam management information, positioning information, channel estimation information, or any combination thereof.

[0446] Clause 91: The method according to clause 90, further comprising: obtaining an indication of machine learning capabilities, the machine learning capabilities comprising: a combination of machine learning models capable of processing in an active state or an inactive state; a first number of inference tasks capable of being shared between a set of active machine learning models or models in the combination, a second number of one or more inference tasks for each model in the set of active machine learning models or the combination, a threshold value associated with the machine learning processing constraint for each one or more subcarriers, each one or more carriers, or each one or more frequency bands; or a combination thereof.

[0447] Item 92: A method according to Item 91, wherein the machine learning processing constraint is satisfied when: a first total number of inference tasks associated with the first machine learning model and the second machine learning model is less than or equal to the first number; a second total number of inference tasks associated with the first machine learning model is less than or equal to the second number associated with the first machine learning model; and a third total number of inference tasks associated with the second machine learning model is less than or equal to the second number associated with the second machine learning model.

[0448] Clause 93: A method according to any one of clauses 90 to 92, wherein the machine learning processing constraint is satisfied if the total number of processing units associated with the active machine learning model is less than or equal to a threshold.

[0449] Clause 94: The method of clause 93, further comprising obtaining an indication of a first number of one or more processing units associated with the first machine learning model and a second number of one or more processing units associated with the second machine learning model.

[0450] Clause 95: A method according to clause 93 or 94, wherein the number of one or more processing units associated with the first machine learning model is equal to zero when the first machine learning model is active.

[0451] Clause 96: A method according to any one of clauses 89 to 95, wherein: in the case where the report includes non-periodic CSI or an initial transmission associated with semi-persistent CSI, one or more processing units or one or more inference tasks associated with the first machine learning model are occupied from the time the indication is output until the report is obtained; and in the case where the report includes periodic CSI or a subsequent transmission associated with semi-persistent CSI, the one or more processing units or the one or more inference tasks associated with the first machine learning model are occupied from the time the last opportunity associated with the reference signal is output before the timing reference until the report is obtained.

[0452] Clause 97: A method according to any one of clauses 89 to 95, wherein: in the case where the report includes non-periodic CSI, one or more processing units or one or more inference tasks associated with the first machine learning model are occupied from the time the indication is output until the report is obtained; and in the case where the report includes periodic CSI or semi-persistent CSI, the one or more processing units or the one or more inference tasks associated with the first machine learning model are occupied from the time the report is activated or configured until the report is deactivated or reconfigured.

[0453] Clause 98: A method according to any one of clauses 89 to 97, the method further comprising: obtaining an indication of a process identifier for each of a plurality of machine learning models, wherein the process identifier indicates whether the corresponding machine learning model can be processed concurrently with another machine learning model.

[0454] Clause 99: A method according to clause 98, wherein concurrent processing of machine learning models with different process identifiers is allowed.

[0455] Clause 100: An apparatus, the apparatus comprising: a memory; and a processor, the processor coupled to the memory, the processor configured to perform the method of any of clauses 1 to 99.

[0456] Clause 101: An apparatus comprising: means for performing the method of any one of clauses 1 to 99.

[0457] Clause 102: A non-transitory computer-readable medium having stored thereon instructions for performing the method of any one of clauses 1 to 99.

[0458] Clause 103: A computer program product embodied on a computer-readable storage medium, the computer program product comprising: code for performing the method of any of clauses 1 to 99.

[0459] Additional Notes

[0460] The foregoing description is provided to enable any person skilled in the art to practice the various aspects described herein. The examples discussed herein do not limit the scope, applicability, or aspects set forth in the claims. Various modifications to these aspects will be apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects. For example, the functions and arrangements of the elements discussed may be changed without departing from the scope of this disclosure. Various examples may omit, replace, or add various processes or components as appropriate. For example, the methods described may be performed in an order different from that described, and various actions may be added, omitted, or combined. In addition, the features described with respect to some examples may be combined in some other examples. For example, a device or method may be implemented using any number of the aspects set forth herein. In addition, the scope of this disclosure is intended to cover such devices or methods that are implemented using other structures, functionalities, or structures and functionalities in addition to or different from the various aspects of the disclosure set forth herein. It should be understood that any aspect of the disclosure disclosed herein may be embodied by one or more elements of the claims.

[0461] The various illustrative logical blocks, modules, and circuits described in conjunction with the present disclosure may be implemented or performed using a general purpose processor, a digital signal processor (DSP), an ASIC, a field programmable gate array (FPGA) or other programmable logic device (PLD), discrete gate or transistor logic, discrete hardware components, or any combination thereof, designed to perform the functions described herein. Although a general purpose processor may be a microprocessor, in an alternative embodiment, the processor may be any commercially available processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, for example, a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, a system on a chip (SoC), or any other such configuration.

[0462] As used herein, a phrase referring to "at least one of" a list of items refers to any combination of those items (including single members). For example, "at least one of a, b, or c" is intended to cover a, b, c, ab, ac, bc, and abc, as well as any combination with multiples of the same element (e.g., aa, aaa, aab, aac, abb, acc, bb, bbb, bbc, cc, and ccc, or any other ordering of a, b, and c).

[0463] As used herein, the term "determining" encompasses a wide variety of actions. For example, "determining" may include calculating, computing, processing, deriving, investigating, searching (e.g., searching in a table, database, or another data structure), ascertaining, and the like. Additionally, "determining" may include receiving (e.g., receiving information), accessing (e.g., accessing data in a memory), and the like. Additionally, "determining" may include resolving, selecting, choosing, establishing, and the like.

[0464] The methods disclosed herein include one or more actions for implementing the methods. Method actions may be interchangeable with each other without departing from the scope of the claims. In other words, unless a specific order of actions is specified, the order and / or use of specific actions may be modified without departing from the scope of the claims. In addition, the various operations of the methods described above may be performed by any suitable component capable of performing the corresponding functions. The component may include various hardware and / or software components and / or modules, including but not limited to circuits, application specific integrated circuits (ASICs), or processors.

[0465] The following claims are not intended to be limited to the various aspects shown herein, but should be given the full scope consistent with the language of the claims. Within the claims, unless specifically stated otherwise, reference to elements in the singular is not intended to mean "one and only one", but "one or more". Unless otherwise specifically stated, the term "some" refers to one or more. No claim element is to be interpreted under the provisions of 35 U.S.C. § 112 (f) unless the element is explicitly stated using the phrase "parts for...". All structural and functional equivalents of the elements of the various aspects described throughout this disclosure that are known or will later be known to those of ordinary skill in the art are expressly incorporated herein by reference and are intended to be covered by the claims. In addition, regardless of whether such disclosure is explicitly stated in the claims, anything disclosed herein is not intended to be disclosed to the public.

Claims

1. A device for wireless communication, the device comprising: Memory; as well as a processor coupled to the memory, the processor being configured to: receiving an indication to report channel state information (CSI) associated with a first machine learning model, receiving a reference signal associated with the CSI, and A CSI report associated with the reference signal is sent based at least in part on a machine learning capability associated with the apparatus.

2. The apparatus of claim 1 , wherein, to send the CSI report, the processor is further configured to send the CSI report in response to satisfying a timing constraint, wherein the timing constraint is based at least in part on the machine learning capability.

3. The device according to claim 2, wherein: The timing constraint is satisfied if the event occurs before or starts no earlier than a timing reference; and The temporal position of the timing reference is determined based at least in part on the machine learning capability.

4. The apparatus of claim 1, wherein the machine learning capability comprises: a first number of one or more active machine learning models capable of being processed by the device; a second number of the one or more inactive machine learning models capable of being processed by the device; The delay used to switch between active and inactive machine learning models; A combination of machine learning models that can be processed in either active or inactive states; or Any combination of them.

5. The apparatus of claim 1, wherein the machine learning capabilities include: A first capability with at most a single active machine learning model and zero inactive machine learning models; having at most a single active machine learning model, one or more inactive machine learning models, and a second capability to process a first duration of the inactive machine learning models; having at most a single active machine learning model, one or more inactive machine learning models, and a third capability to process the inactive machine learning models for a second duration, wherein the first duration is longer than the second duration; a fourth capability of having one or more active machine learning models and zero inactive machine learning models; a fifth capability having one or more active machine learning models, one or more inactive machine learning models, and a third duration of processing the inactive machine learning models; or Having one or more active machine learning models, one or more inactive machine learning models, and a sixth capability for a fourth duration of processing the inactive machine learning models, wherein the third duration is longer than the fourth duration.

6. The apparatus of claim 1 , wherein the processor is further configured to send an indication of the machine learning capability.

7. The device according to claim 4, wherein: The processor is further configured to receive a configuration indicating a first set of one or more active machine learning models and indicating a second set of one or more inactive machine learning models, a first total number of the first set of one or more active machine learning models is less than or equal to a first number of the one or more active machine learning models, and The second total number of the second set of the one or more inactive machine learning models is less than or equal to the second number of the one or more inactive machine learning models.

8. The apparatus of claim 1, wherein the processor is further configured to: receiving a configuration indicating a first set of one or more active machine learning models and indicating a second set of one or more inactive machine learning models; and Sending a confirmation of successful loading of at least one active machine learning model from the first set of one or more active machine learning models in response to receiving the configuration.

9. An apparatus according to claim 8, wherein, in order to send the confirmation, the processor is further configured to send multiple confirmations, wherein each of the confirmations is for an active machine learning model in the first set of the one or more active machine learning models.

10. The apparatus of claim 8, wherein to send the confirmation, the processor is further configured to send the confirmation within a time window starting when the configuration is received.

11. The device according to claim 10, wherein: The machine learning capabilities include: a first number of one or more active machine learning models capable of being processed by the device, and a second number of the one or more inactive machine learning models capable of being processed by the device; and The duration associated with the time window depends on a first number of the one or more active machine learning models or a second number of the one or more inactive machine learning models.

12. The apparatus of claim 8, wherein the processor is further configured to send an indication of a timing constraint associated with at least one active machine learning model in a first set of the one or more active machine learning models in response to receiving the configuration.

13. The apparatus of claim 1, wherein the processor is further configured to: receiving a configuration indicating a first set of one or more active machine learning models and indicating a second set of one or more inactive machine learning models; and In the event that no confirmation is sent within a time window starting when the configuration is received, the CSI report is determined based on a non-artificial intelligence codebook, wherein the confirmation indicates successful loading of at least one active machine learning model from a first set of one or more active machine learning models.

14. An apparatus for wireless communication, the apparatus comprising: Memory; as well as a processor coupled to the memory, the processor being configured to: receiving an indication to report channel state information (CSI) associated with a first machine learning model, In response to receiving the indication, determining a set of one or more active machine learning models, receiving a reference signal associated with the CSI, and Sending a CSI report based on the reference signal at least in response to satisfying a timing constraint, wherein the timing constraint is based at least in part on the set of one or more active machine learning models.

15. The device according to claim 14, wherein: The timing constraint is satisfied if the event occurs before or starts no earlier than a timing reference; and The temporal position of the timing reference is determined based at least in part on machine learning capabilities associated with the apparatus.

16. The device according to claim 15, wherein: The indication further indicates reporting the CSI in a reporting opportunity; The event includes the reporting opportunity; and The timing reference is located in time a first number of one or more symbols after a last symbol of the indication triggering the CSI report.

17. The device according to claim 15, wherein: The indication further indicates reporting the CSI in a reporting opportunity; The event includes the reporting opportunity; and The timing reference is located in time a second number of one or more symbols after a last symbol of the reference signal.

18. The apparatus of claim 15, wherein: The indication further indicates reporting the CSI in a reporting opportunity; The CSI report includes periodic CSI or semi-persistent CSI; The event comprises a reception opportunity associated with the reference signal; and The timing reference is located in time a third number of one or more symbols before the reporting opportunity.

19. The apparatus of claim 15, wherein the time position of the timing reference is based in part on: an identifier associated with the first machine learning model, a rank indicator associated with one or more transmission layers, Whether the CSI feedback decoder is available, whether another machine learning model is active, or Any combination of them.

20. The apparatus of claim 15, wherein the processor is further configured to send an indication of the temporal position of the timing reference.

21. The apparatus of claim 15, wherein the processor is further configured to determine the temporal position of the timing reference based at least in part on the set of one or more active machine learning models.

22. The apparatus of claim 14, wherein the set comprises a series of one or more identifiers associated with the one or more active machine learning models.

23. The apparatus of claim 14, wherein the processor is further configured to: sending an indication of machine learning capabilities associated with the device, wherein the machine learning capabilities include a first number of one or more active machine learning models capable of being processed by the device; and Wherein the total number of machine learning models in the collection is less than or equal to the first number.

24. The apparatus of claim 14, wherein the processor is further configured to: sending an indication of machine learning capabilities associated with the device, wherein the machine learning capabilities include a combination of machine learning models that can be processed in an active state; and Wherein the one or more active machine learning models in the set are in the combination of machine learning models.

25. The apparatus of claim 21, wherein: The timing reference is located in time at: at a first number of one or more symbols after the last symbol of the indication triggering the CSI report and at a second number of one or more symbols after the last symbol of the reference signal, or at a third number of one or more symbols before a reporting opportunity associated with the CSI report; and To determine the time position of the timing reference, the processor is further configured to: selecting a first value for a first quantity of the one or more symbols, selecting a second value for a second quantity of the one or more symbols, or selecting a third value for a third quantity of the one or more symbols, if the first machine learning model is in the set of one or more active machine learning models, and In a case where the first machine learning model is not in the set of one or more active machine learning models, select a fourth value for the first quantity of the one or more symbols, select a fifth value for the second quantity of the one or more symbols, or select a sixth value for the third quantity of the one or more symbols, wherein the second value is different from the first value.

26. The apparatus of claim 25, wherein: The first value is less than the fourth value; The second value is less than the fifth value; and The third value is smaller than the sixth value.

27. An apparatus according to claim 14, wherein in order to determine the set, the processor is further configured to update the set to include the first machine learning model in response to receiving the indication to report the CSI if the first machine learning model is not in the set.

28. The apparatus of claim 27, wherein to update the set, the processor is further configured to: removing a second machine learning model from the set; sending an instruction to remove the second machine learning model from the set; and Add the first machine learning model to the collection.

29. The apparatus of claim 14, wherein the processor is further configured to: ignoring an indication to report CSI associated with a second machine learning model that is not in the set of one or more machine learning models in response to satisfying a third timing constraint; and In response to satisfying the third timing constraint, suppressing sending another CSI report associated with the second machine learning model.

30. The apparatus of claim 29, wherein the third timing constraint is satisfied if: receiving the indication to report CSI associated with the second machine learning model within a time window ending at a reporting opportunity for sending the CSI report, or The another CSI report is scheduled to be reported in the time window.

31. The apparatus of claim 30, wherein: For an initial transmission of aperiodic CSI or semi-persistent CSI, the time window begins upon receipt of the indication to report the CSI associated with the first machine learning model; and For subsequent transmissions of periodic CSI or the semi-persistent CSI, the time window starts at a third number of one or more symbols before the reference signal.

32. An apparatus according to claim 31, wherein the third number of the one or more symbols depends on whether the first machine learning model is in a set of one or more active machine learning models.

33. The apparatus of claim 18, wherein the processor is further configured to: In a case where the first machine learning model is in the set of one or more active machine learning models, selecting a first value for a third number of the one or more symbols; and In a case where the first machine learning model is not in the set of one or more active machine learning models, a second value is selected for a third quantity of the one or more symbols, wherein the second value is different from the first value.

34. The apparatus of claim 18, wherein the processor is further configured to: In a case where the CSI report includes periodic CSI, selecting a first value for a third number of the one or more symbols; selecting a second value for a third number of the one or more symbols in a case where the CSI report is an initial transmission associated with semi-persistent CSI; and In a subsequent transmission in which the CSI report is associated with semi-persistent CSI, a third value is selected for a third number of the one or more symbols, wherein the second value is different from the third value.

35. An apparatus for wireless communication, the apparatus comprising: Memory; as well as a processor coupled to the memory, the processor being configured to: receiving an instruction to report information associated with a first machine learning model, determining the information using the first machine learning model while concurrently determining other information using a second machine learning model in response to satisfying a machine learning processing constraint, and A report indicating the information is sent.

36. The apparatus of claim 35, wherein the information comprises channel state information, beam management information, positioning information, channel estimation information, or any combination thereof.

37. The apparatus of claim 35, wherein the processor is further configured to send an indication of a machine learning capability, the machine learning capability comprising: A combination of machine learning models that can be processed in active or inactive states; a first number of inference tasks that can be shared among the set of active machine learning models or models in the combination, a second number of one or more inference tasks for each model in the set of active machine learning models or the combination, a threshold associated with the machine learning processing constraint per one or more subcarriers, per one or more carriers, or per one or more frequency bands; or A combination of them.

38. The apparatus of claim 37, wherein the machine learning processing constraint is satisfied if: a first total number of inference tasks associated with the first machine learning model and the second machine learning model is less than or equal to the first number; a second total number of inference tasks associated with the first machine learning model is less than or equal to the second number associated with the first machine learning model; and A third total number of inference tasks associated with the second machine learning model is less than or equal to the second number associated with the second machine learning model.

39. An apparatus according to claim 35, wherein the machine learning processing constraint is satisfied when the total number of processing units associated with the active machine learning model is less than or equal to a threshold.

40. An apparatus according to claim 39, wherein the processor is further configured to send an indication of a first number of one or more processing units associated with the first machine learning model and a second number of one or more processing units associated with the second machine learning model.

41. An apparatus according to claim 39, wherein when the first machine learning model is active, the number of one or more processing units associated with the first machine learning model is equal to zero.

42. An apparatus according to claim 39, wherein the processor is further configured to allocate unoccupied processing units to the first machine learning model based at least in part on a priority associated with the active machine learning model.

43. The apparatus of claim 42, wherein the priority associated with the active machine learning model is based at least in part on: Model identifier, Carrier identifier, Bandwidth section identifier, CSI report identifier, or Any combination of them.

44. An apparatus according to claim 35, wherein the processor is further configured to send an indication of a process identifier for each of a plurality of machine learning models, wherein the process identifier indicates whether the corresponding machine learning model can be processed concurrently with another machine learning model.

45. An apparatus according to claim 44, wherein concurrent processing of machine learning models with different process identifiers is allowed.

46. ​​The apparatus of claim 35, wherein: In a case where the report includes aperiodic channel state information (CSI) or an initial transmission associated with semi-persistent CSI, occupying one or more processing units or one or more inference tasks associated with the first machine learning model from the time the indication is received until the report is transmitted; and In a case where the report includes periodic CSI or subsequent transmissions associated with semi-persistent CSI, the one or more processing units or the one or more inference tasks associated with the first machine learning model are occupied from the time the last opportunity associated with the reference signal is received before the timing reference until the report is sent.

47. The apparatus of claim 35, wherein: In a case where the report includes aperiodic channel state information (CSI), occupying one or more processing units or one or more inference tasks associated with the first machine learning model from the time the indication is received until the report is sent; and In a case where the report includes periodic CSI or semi-persistent CSI, the one or more processing units or the one or more inference tasks associated with the first machine learning model are occupied from the time activation or configuration of the report is received until deactivation or reconfiguration of the report is received.

48. An apparatus for wireless communication, the apparatus comprising: Memory; as well as a processor coupled to the memory, the processor being configured to: outputting an indication reporting channel state information (CSI) associated with the first machine learning model, and A CSI report associated with the reference signal is obtained based at least in part on a machine learning capability associated with the user equipment.

49. An apparatus for wireless communication, the apparatus comprising: Memory; as well as a processor coupled to the memory, the processor being configured to: outputting an indication reporting channel state information (CSI) associated with the first machine learning model, determining a set of one or more active machine learning models in response to outputting the indication, outputting a reference signal associated with the CSI, and A CSI report is obtained based on the reference signal at least in response to satisfying a timing constraint, wherein the timing constraint is based at least in part on the set of one or more active machine learning models.

50. An apparatus for wireless communication, the apparatus comprising: Memory; as well as a processor coupled to the memory, the processor being configured to: outputting instructions to report information associated with the first machine learning model, and A report indicating the information is obtained based on satisfying the machine learning processing constraints.

Images