Post deployment adaptation of artificial intelligence (AI) / machine learning (ML) models in wireless communication networks

Abstract

Apparatuses, systems, and methods for adapting an Artificial Intelligence (AI)ZMachine Learning (ML)-based model to an environment after the AI / ML-based model is deployed are described including systems, methods, and mechanisms for a user equipment device (UE) to fine-tune a pre-trained AI / ML-based model. In some examples, transfer learning may be applied. In some examples, meta learning may be applied.

Description

POST DEPLOYMENT ADAPTATION OF ARTIFICIAL INTELLIGENCE (AI) / MACHINE LEARNING (ML) MODELS IN WIRELESS COMMUNICATION NETWORKS FIELD

[0001] The invention relates to wireless communications, and more particularly to apparatuses, systems, and methods for post deployment adaptation of AI / ML-based models, including systems, methods, and mechanisms for data collection, training, and performance monitoring of AI / ML-based models. DESCRIPTION OF THE RELATED ART

[0002] Wireless communication systems are rapidly growing in usage. In recent years, wireless devices such as smart phones and tablet computers have become increasingly sophisticated. In addition to supporting telephone calls, many mobile devices now provide access to the internet, email, text messaging, and navigation using the global positioning system (GPS), and are capable of operating sophisticated applications that utilize these functionalities. Additionally, there exist numerous different wireless communication technologies and standards.

[0003] Long Term Evolution (LTE), also referred to as the Evolved Universal Terrestrial Radio Access Network (E-UTRAN, has been the technology of choice for the majority of wireless network operators worldwide, providing mobile broadband data and high-speed Internet access to their subscriber base. LTE was first proposed in 2004 and was first standardized in 2008. Since then, as usage of wireless communication systems has expanded exponentially, demand has risen for wireless network operators to support a higher capacity for a higher density of mobile broadband users. Thus, in 2015 study of a new radio access technology began and, in 2017, a first release of the Third Generation Partnership Project (3GPP) Fifth Generation New Radio (5G NR) was standardized. 5th generation mobile networks or 5th generation wireless systems, referred to as 3GPP NR (otherwise known as 5G-NR or NR-5G for 5G New Radio, also simply referred to as NR). NR proposes a higher capacity for a higher density of mobile broadband users, also supporting device-to-device, ultra-reliable, and massive machine communications, as wellas lower latency and lower battery consumption, than LTE standards.

[0004] 5G-NR provides, as compared to LTE, a higher capacity for a higher density of mobile broadband users, while also supporting device-to-device, ultra-reliable, and massive machine type communications with lower latency and / or lower battery consumption. Further, NR may allow for more flexible UE scheduling as compared to current LTE. Consequently, efforts are being made in ongoing developments of 5G-NR to take advantage of higher throughputs possible at higher frequencies.

[0005] One aspect of wireless communication systems, e.g., systems for NR cellular wireless communications, is the transmission and measurement of reference signals, including channel state information reference signals (CSI-RS). SUMMARY

[0006] Embodiments relate to wireless communications, and more particularly to apparatuses, systems, and methods to perform training or performance monitoring for an artificial intelligence (AI) / machine learning (ML) model.

[0007] Embodiments relate to wireless communications, and more particularly to apparatuses, systems, and methods for a device configured for communicating in a wireless communication network, comprising: one or more processors, coupled to a memory, configured to employ a pre-trained model for a task, determine performance of the pre- trained model has degraded based on a set of samples, and fine-tune the pretrained model.

[0008] Other embodiments relate to a user equipment comprising: one or more processors, coupled to a memory, configured to: employ a pre-trained model for a task, determine performance of the pre-trained model has degraded based on a set of samples, and fine-tune the pretrained model.

[0009] The techniques described herein may be implemented in and / or used with a number of different types of devices, including but not limited to base stations, access points, cellular phones, tablet computers, wearable computing devices, portable media players, vehicles, and any of various other computing devices.

[0010] This Summary is intended to provide a brief overview of some of the subject matter described in this document. Accordingly, it will be appreciated that the above- described features are merely examples and should not be construed to narrow the scope orspirit of the subject matter described herein in any way. Other features, aspects, and advantages of the subject matter described herein will become apparent from the following Detailed Description, Figures, and Claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] A better understanding of the present subject matter can be obtained when the following detailed description of various embodiments is considered in conjunction with the following drawings, in which:

[0012] Figure 1A illustrates an example wireless communication system according to some embodiments.

[0013] Figure 1B illustrates an example of a base station and an access point in communication with a user equipment (UE) device, according to some embodiments.

[0014] Figure 2 illustrates an example block diagram of a base station, according to some embodiments.

[0015] Figure 3 illustrates an example block diagram of a server according to some embodiments.

[0016] Figure 4 illustrates an example block diagram of a UE according to some embodiments.

[0017] Figure 5 illustrates an example block diagram of cellular communication circuitry, according to some embodiments.

[0018] Figure 6A illustrates an example of a 5G network architecture that incorporates both 3GPP (e.g., cellular) and non-3GPP (e.g., non-cellular) access to the 5G CN, according to some embodiments.

[0019] Figure 6B illustrates an example of a 5G network architecture that incorporates both dual 3GPP (e.g., LTE and 5G NR) access and non-3GPP access to the 5G CN, according to some embodiments.

[0020] Figure 7 illustrates an example of a baseband processor architecture for a UE, according to some embodiments.

[0021] Figure 8 illustrates an example of a device in accordance with someembodiments.

[0022] Figure 9 illustrates an example baseband circuitry in accordance with some embodiments.

[0023] Figure 10 illustrates an example of a control plane protocol stack in accordance with some embodiments.

[0024] Figure 11 illustrates an example timing diagram of utilizing CSI feedback, according to some embodiments.

[0025] Figure 12 illustrates an example of a two-sided AI / ML-based model for CSI compression, according to some embodiments.

[0026] Figure 13 illustrates an example of a one-sided AI / ML-based model for CSI prediction, according to some embodiments.

[0027] Figure 14 illustrates an example of an AI / ML-based model for joint CSI prediction and compression, according to some embodiments.

[0028] Figure 15 illustrates an example of layers of an AI / ML-based model according to some embodiments

[0029] Figure 16 is a conceptual diagram illustrating fine-tuning an AI / ML-based model for different target domains according to some embodiments.

[0030] Figure 17 illustrates a block diagram of an example of a method to perform training or performance monitoring of an AI / ML-based model, according to some embodiments.

[0031] Figure 18 illustrates a block diagram of an example of a method to perform training or performance monitoring of an AI / ML-based model, according to some embodiments.

[0032] Figure 19 illustrates a block diagram of an example of a method to perform life cycle management (LCM) of an AI / ML-based model, according to some embodiments.

[0033] While the features described herein may be susceptible to various modifications and alternative forms, specific embodiments thereof are shown by way of example in the drawings and are herein described in detail. It should be understood, however, that the drawings and detailed description thereto are not intended to be limiting to the particular form disclosed, but on the contrary, the intention is to cover all modifications, equivalentsand alternatives falling within the spirit and scope of the subject matter as defined by the appended claims. DETAILED DESCRIPTION

[0034] The following is a glossary of terms used in this disclosure:

[0035] Memory Medium – Any of various types of non-transitory memory devices or storage devices. The term “memory medium” is intended to include an installation medium, e.g., a CD- ROM, floppy disks, or tape device; a computer system memory or random- access memory such as DRAM, DDR RAM, SRAM, EDO RAM, Rambus RAM, etc.; a non-volatile memory such as a Flash, magnetic media, e.g., a hard drive, or optical storage; registers, or other similar types of memory elements, etc. The memory medium may include other types of non-transitory memory as well or combinations thereof. In addition, the memory medium may be located in a first computer system in which the programs are executed, or may be located in a second different computer system which connects to the first computer system over a network, such as the Internet. In the latter instance, the second computer system may provide program instructions to the first computer for execution. The term “memory medium” may include two or more memory mediums which may reside in different locations, e.g., in different computer systems that are connected over a network. The memory medium may store program instructions (e.g., embodied as computer programs) that may be executed by one or more processors.

[0036] Carrier Medium – a memory medium as described above, as well as a physical transmission medium, such as a bus, network, and / or other physical transmission medium that conveys signals such as electrical, electromagnetic, or digital signals.

[0037] Programmable Hardware Element - includes various hardware devices comprising multiple programmable function blocks connected via a programmable interconnect. Examples include FPGAs (Field Programmable Gate Arrays), PLDs (Programmable Logic Devices), FPOAs (Field Programmable Object Arrays), and CPLDs (Complex PLDs). The programmable function blocks may range from fine grained (combinatorial logic or look up tables) to coarse grained (arithmetic logic units or processor cores). A programmable hardware element may also be referred to as "reconfigurable logic”.

[0038] Computer System (or Computer) – any of various types of computing or processing systems, including a personal computer system (PC), mainframe computer system, workstation, network appliance, Internet appliance, personal digital assistant (PDA), television system, grid computing system, or other device or combinations of devices. In general, the term "computer system" can be broadly defined to encompass any device (or combination of devices) having at least one processor that executes instructions from a memory medium.

[0039] User Equipment (UE) (or “UE Device”) – any of various types of computer systems devices which are mobile or portable and which performs wireless communications. Examples of UE devices include mobile telephones or smart phones (e.g., iPhone™, Android™-based phones), portable gaming devices (e.g., Nintendo DS™, PlayStation Portable™, Gameboy Advance™, iPhone™), laptops, wearable devices (e.g., smart watch, smart glasses), PDAs, portable Internet devices, music players, data storage devices, other handheld devices, unmanned aerial vehicles (UAVs) (e.g., drones), UAV controllers (UACs), and so forth. In general, the term “UE” or “UE device” can be broadly defined to encompass any electronic, computing, and / or telecommunications device (or combination of devices) which is easily transported by a user and capable of wireless communication.

[0040] Base Station – The term "Base Station" has the full breadth of its ordinary meaning, and at least includes a wireless communication station installed at a fixed location and used to communicate as part of a wireless telephone system or radio system.

[0041] Processing Element (or Processor) – refers to various elements or combinations of elements that are capable of performing a function in a device, such as a user equipment or a cellular network device. Processing elements may include, for example: processors and associated memory, portions or circuits of individual processor cores, entire processor cores, processor arrays, circuits such as an ASIC (Application Specific Integrated Circuit), programmable hardware elements such as a field programmable gate array (FPGA), as well any of various combinations of the above.

[0042] Channel - a medium used to convey information from a sender (transmitter) to a receiver. It should be noted that since characteristics of the term “channel” may differ according to different wireless protocols, the term “channel” as used herein may be considered as being used in a manner that is consistent with the standard of the type ofdevice with reference to which the term is used. In some standards, channel widths may be variable (e.g., depending on device capability, band conditions, etc.). For example, LTE may support scalable channel bandwidths from 1.4 MHz to 20MHz. In contrast, WLAN channels may be 22MHz wide while Bluetooth channels may be 1Mhz wide. Other protocols and standards may include different definitions of channels. Furthermore, some standards may define and use multiple types of channels, e.g., different channels for uplink or downlink and / or different channels for different uses such as data, control information, etc.

[0043] Band - The term "band" has the full breadth of its ordinary meaning, and at least includes a section of spectrum (e.g., radio frequency spectrum) in which channels are used or set aside for the same purpose.

[0044] Wi-Fi – The term "Wi-Fi" (or WiFi) has the full breadth of its ordinary meaning, and at least includes a wireless communication network or RAT that is serviced by wireless LAN (WLAN) access points and which provides connectivity through these access points to the Internet. Most modern Wi-Fi networks (or WLAN networks) are based on IEEE 802.11 standards and are marketed under the name “Wi-Fi”. A Wi-Fi (WLAN) network is different from a cellular network.

[0045] 3GPP Access – refers to accesses (e.g., radio access technologies) that are specified by the Third Generation Partnership Project (3GPP) standards. These accesses include, but are not limited to, GSM / GPRS, LTE, LTE-A, and / or 5G NR. In general, 3GPP access refers to various types of cellular access technologies.

[0046] Non-3GPP Access – refers any accesses (e.g., radio access technologies) that are not specified by 3GPP standards. These accesses include, but are not limited to, WiMAX, CDMA2000, Wi-Fi, WLAN, and / or fixed networks. Non-3GPP accesses may be split into two categories, "trusted" and "untrusted": Trusted non-3GPP accesses can interact directly with an evolved packet core (EPC) and / or a 5G core (5GC) whereas untrusted non-3GPP accesses interwork with the EPC / 5GC via a network entity, such as an Evolved Packet Data Gateway and / or a 5G NR gateway. In general, non-3GPP access refers to various types on non-cellular access technologies.

[0047] Automatically – refers to an action or operation performed by a computer system (e.g., software executed by the computer system) or device (e.g., circuitry, programmable hardware elements, ASICs, etc.), without user input directly specifyingor performing the action or operation. Thus, the term "automatically" is in contrast to an operation being manually performed or specified by the user, where the user provides input to directly perform the operation. An automatic procedure may be initiated by input provided by the user, but the subsequent actions that are performed “automatically” are not specified by the user, i.e., are not performed “manually”, where the user specifies each action to perform. For example, a user filling out an electronic form by selecting each field and providing input specifying information (e.g., by typing information, selecting check boxes, radio selections, etc.) is filling out the form manually, even though the computer system can update the form in response to the user actions. The form may be automatically filled out by the computer system where the computer system (e.g., software executing on the computer system) analyzes the fields of the form and fills in the form without any user input specifying the answers to the fields. As indicated above, the user may invoke the automatic filling of the form, but is not involved in the actual filling of the form (e.g., the user is not manually specifying answers to fields but rather they are being automatically completed). The present specification provides various examples of operations being automatically performed in response to actions the user has taken.

[0048] Approximately - refers to a value that is almost correct or exact. For example, approximately may refer to a value that is within 1 to 10 percent of the exact (or desired) value. It should be noted, however, that the actual threshold value (or tolerance) may be application dependent. For example, in some embodiments, “approximately” may mean within 0.1% of some specified or desired value, while in various other embodiments, the threshold may be, for example, 2%, 3%, 5%, and so forth, as desired or as used by the particular application.

[0049] Concurrent – refers to parallel execution or performance, where tasks, processes, or programs are performed in an at least partially overlapping manner. For example, concurrency may be implemented using “strong” or strict parallelism, where tasks are performed (at least partially) in parallel on respective computational elements, or using “weak parallelism”, where the tasks are performed in an interleaved manner, e.g., by time multiplexing of execution threads.

[0050] Various components may be described as “configured to” perform a task or tasks. In such contexts, “configured to” is a broad recitation generally meaning “having structure that” performs the task or tasks during operation. As such, the component can beconfigured to perform the task even when the component is not currently performing that task (e.g., a set of electrical conductors may be configured to electrically connect a module to another module, even when the two modules are not connected). In some contexts, “configured to” may be a broad recitation of structure generally meaning “having circuitry that” performs the task or tasks during operation. As such, the component can be configured to perform the task even when the component is not currently on. In general, the circuitry that forms the structure corresponding to “configured to” may include hardware circuits.

[0051] Various components may be described as performing a task or tasks, for convenience in the description. Such descriptions should be interpreted as including the phrase “configured to.” Reciting a component that is configured to perform one or more tasks is expressly intended not to invoke 35 U.S.C. § 112(f) interpretation for that component. Figures 1A and 1B: Communication Systems

[0052] Figure 1A illustrates a simplified example wireless communication system, according to some embodiments. It is noted that the system of Figure 1A is merely one example of a possible system, and that features of this disclosure may be implemented in any of various systems, as desired.

[0053] As shown, the example wireless communication system includes a base station 102A which communicates over a transmission medium with one or more user devices 106A, 106B, etc., through 106N. The user devices may be referred to herein as a “user equipment” (UE). Thus, the user devices 106 are referred to as UEs or UE devices.

[0054] The base station (BS) 102A may be a base transceiver station (BTS) or cell site (a “cellular base station”) and may include hardware that enables wireless communication with the UEs 106A through 106N.

[0055] The communication area (or coverage area) of the base station may be referred to as a “cell.” The base station 102A and the UEs 106 may be configured to communicate over the transmission medium using any of various radio access technologies (RATs), also referred to as wireless communication technologies, or telecommunication standards, such as GSM, UMTS (associated with, for example, WCDMA or TD-SCDMA air interfaces), LTE, LTE-Advanced (LTE-A), 5G new radio (5G NR), HSPA, 3GPP2CDMA2000 (e.g., 1xRTT, 1xEV-DO, HRPD, eHRPD), etc. Note that if the base station 102A is implemented in the context of LTE (E-UTRAN), it may alternately be referred to as an 'eNodeB' or ‘eNB’. Note that if the base station 102A is implemented in the context of 5G NR, it may alternately be referred to as ‘gNodeB’ or ‘gNB’.

[0056] As shown, the base station 102A may also be equipped to communicate with a network (NW) 100 (e.g., a core network of a cellular service provider, a telecommunication network such as a public switched telephone network (PSTN), and / or the Internet, among various possibilities). Thus, the base station 102A may facilitate communication between the user devices and / or between the user devices and the network 100. In particular, the cellular base station 102A may provide UEs 106 with various telecommunication capabilities, such as voice, SMS and / or data services.

[0057] Base station 102A and other similar base stations (such as base stations 102B…102N) operating according to the same or a different cellular communication standard may thus be provided as a network of cells, which may provide continuous or nearly continuous overlapping service to UEs 106A-N and similar devices over a geographic area via one or more cellular communication standards.

[0058] Thus, while base station 102A may act as a “serving cell” for UEs 106A-N as illustrated in Figure 1A, each UE 106 may also be capable of receiving signals from (and possibly within communication range of) one or more other cells (which might be provided by base stations 102B-N and / or any other base stations), which may be referred to as “neighboring cells”. Such cells may also be capable of facilitating communication between user devices and / or between user devices and the network 100. Such cells may include “macro” cells, “micro” cells, “pico” cells, and / or cells which provide any of various other granularities of service area size. For example, base stations 102A-B illustrated in Figure 1A might be macro cells, while base station 102N might be a micro cell. Other configurations are also possible.

[0059] In some embodiments, base station 102A may be a next generation base station, e.g., a 5G New Radio (5G NR) base station, or “gNB”. In some embodiments, a gNB may be connected to a legacy evolved packet core (EPC) network and / or to a NR core (NRC) network. In addition, a gNB cell may include one or more transmission and reception points (TRPs). In addition, a UE capable of operating according to 5G NR may be connected to one or more TRPs within one or more gNBs.

[0060] Note that a UE 106 may be capable of communicating using multiple wireless communication standards. For example, the UE 106 may be configured to communicate using a wireless networking (e.g., Wi-Fi) and / or peer-to-peer wireless communication protocol (e.g., Bluetooth, Wi-Fi peer-to-peer, etc.) in addition to at least one cellular communication protocol (e.g., GSM, UMTS (associated with, for example, WCDMA or TD-SCDMA air interfaces), LTE, LTE-A, 5G NR, HSPA, 3GPP2 CDMA2000 (e.g., 1xRTT, 1xEV-DO, HRPD, eHRPD), etc.). The UE 106 may also or alternatively be configured to communicate using one or more global navigational satellite systems (GNSS, e.g., GPS or GLONASS), one or more mobile television broadcasting standards (e.g., ATSC-M / H or DVB-H), and / or any other wireless communication protocol, if desired. Other combinations of wireless communication standards (including more than two wireless communication standards) are also possible.

[0061] Figure 1B illustrates user equipment 106 (e.g., one of the devices 106A through 106N) in communication with a base station 102 and an access point 112, according to some embodiments. The UE 106 may be a device with both cellular communication capability and non-cellular communication capability (e.g., Bluetooth, Wi-Fi, and so forth) such as a mobile phone, a hand- held device, a computer or a tablet, or virtually any type of wireless device.

[0062] The UE 106 may include a processor that is configured to execute program instructions stored in memory. The UE 106 may perform any of the method embodiments described herein by executing such stored instructions. Alternatively, or in addition, the UE 106 may include a programmable hardware element such as an FPGA (field-programmable gate array) that is configured to perform any of the method embodiments described herein, or any portion of any of the method embodiments described herein.

[0063] The UE 106 may include one or more antennas for communicating using one or more wireless communication protocols or technologies. In some embodiments, the UE 106 may be configured to communicate using, for example, CDMA2000 (1xRTT / 1xEV- DO / HRPD / eHRPD), LTE / LTE-Advanced, or 5G NR using a single shared radio and / or GSM, LTE, LTE- Advanced, or 5G NR using the single shared radio. The shared radio may couple to a single antenna, or may couple to multiple antennas (e.g., for MIMO) for performing wireless communications. In general, a radio may include any combination of a baseband processor, analog RF signal processing circuitry (e.g., including filters,mixers, oscillators, amplifiers, etc.), or digital processing circuitry (e.g., for digital modulation as well as other digital processing). Similarly, the radio may implement one or more receive and transmit chains using the aforementioned hardware. For example, the UE 106 may share one or more parts of a receive and / or transmit chain between multiple wireless communication technologies, such as those discussed above.

[0064] In some embodiments, the UE 106 may include separate transmit and / or receive chains (e.g., including separate antennas and other radio components) for each wireless communication protocol with which it is configured to communicate. As a further possibility, the UE 106 may include one or more radios which are shared between multiple wireless communication protocols, and one or more radios which are used exclusively by a single wireless communication protocol. For example, the UE 106 might include a shared radio for communicating using either of LTE (E-UTRAN) or 5G NR (or LTE or 1xRTTor LTE or GSM), and separate radios for communicating using each of Wi-Fi and Bluetooth. Other configurations are also possible. Figure 2: Block Diagram of a Base Station

[0065] Figure 2 illustrates an example block diagram of a base station 102, according to some embodiments. It is noted that the base station of Figure 2 is merely one example of a possible base station. As shown, the base station 102 may include processor(s) 204 which may execute program instructions for the base station 102. The processor(s) 204 may also be coupled to memory management unit (MMU) 240, which may be configured to receive addresses from the processor(s) 204 and translate those addresses to locations in memory (e.g., memory 260 and read only memory (ROM) 250) or to other circuits or devices.

[0066] The base station 102 may include at least one network port 270. The network port 270 may be configured to couple to a telephone network and provide a plurality of devices, such as UE devices 106, access to the telephone network as described above in Figures 1 and 2.

[0067] The network port 270 (or an additional network port) may also or alternatively be configured to couple to a cellular network, e.g., a core network of a cellular service provider. The core network may provide mobility related services and / or other services to aplurality of devices, such as UE devices 106. In some cases, the network port 270 may couple to a telephone network via the core network, and / or the core network may provide a telephone network (e.g., among other UE devices serviced by the cellular service provider).

[0068] In some embodiments, base station 102 may be a next generation base station, e.g., a 5G New Radio (5G NR) base station, or “gNB”. In such embodiments, base station 102 may be connected to a legacy evolved packet core (EPC) network and / or to a NR core (NRC) network. In addition, base station 102 may be considered a 5G NR cell and may include one or more transition and reception points (TRPs). In addition, a UE capable of operating according to 5G NR may be connected to one or more TRPs within one or more gNBs.

[0069] The base station 102 may include at least one antenna 234, and possibly multiple antennas. The at least one antenna 234 may be configured to operate as a wireless transceiver and may be further configured to communicate with UE devices 106 via radio 230. The antenna 234 communicates with the radio 230 via communication chain 232. Communication chain 232 may be a receive chain, a transmit chain or both. The radio 230 may be configured to communicate via various wireless communication standards, including, but not limited to, 5G NR, LTE, LTE-A, GSM, UMTS, CDMA2000, Wi-Fi, etc.

[0070] The base station 102 may be configured to communicate wirelessly using multiple wireless communication standards. In some instances, the base station 102 may include multiple radios, which may enable the base station 102 to communicate according to multiple wireless communication technologies. For example, as one possibility, the base station 102 may include an LTE radio for performing communication according to LTE as well as a 5G NR radio for performing communication according to 5G NR. In such a case, the base station 102 may be capable of operating as both an LTE base station and a 5G NR base station. As another possibility, the base station 102 may include a multi-mode radio which is capable of performing communications according to any of multiple wireless communication technologies (e.g., 5G NR and Wi-Fi, LTE and Wi-Fi, LTE and UMTS, LTE and CDMA2000, UMTS and GSM, etc.).

[0071] As described further subsequently herein, the BS 102 may include hardware and software components for implementing or supporting implementation of features described herein. The processor 204 of the base station 102 may be configured to implement or support implementation of part or all of the methods described herein, e.g.,by executing program instructions stored on a memory medium (e.g., a non-transitory computer-readable memory medium). Alternatively, the processor 204 may be configured as a programmable hardware element, such as an FPGA (Field Programmable Gate Array), or as an ASIC (Application Specific Integrated Circuit), or a combination thereof. Alternatively (or in addition) the processor 204 of the BS 102, in conjunction with one or more of the other components 230, 232, 234, 240, 250, 260, 270 may be configured to implement or support implementation of part or all of the features described herein.

[0072] In addition, as described herein, processor(s) 204 may be comprised of one or more processing elements. In other words, one or more processing elements may be included in processor(s) 204. Thus, processor(s) 204 may include one or more integrated circuits (ICs) that are configured to perform the functions of processor(s) 204. In addition, each integrated circuit may include circuitry (e.g., first circuitry, second circuitry, etc.) configured to perform the functions of processor(s) 204.

[0073] Further, as described herein, radio 230 may be comprised of one or more processing elements. In other words, one or more processing elements may be included in radio 230. Thus, radio 230 may include one or more integrated circuits (ICs) that are configured to perform the functions of radio 230. In addition, each integrated circuit may include circuitry (e.g., first circuitry, second circuitry, etc.) configured to perform the functions of radio 230. Figure 3: Block Diagram of a Server

[0074] Figure 3 illustrates an example block diagram of a server 104, according to some embodiments. It is noted that the server of Figure 3 is merely one example of a possible server. As shown, the server 104 may include processor(s) 344 which may execute program instructions for the server 104. The processor(s) 344 may also be coupled to memory management unit (MMU) 374, which may be configured to receive addresses from the processor(s) 344 and translate those addresses to locations in memory (e.g., memory 364 and read only memory (ROM) 354) or to other circuits or devices.

[0075] The server 104 may be configured to provide a plurality of devices, such as base station 102 and UE devices 106 access to network functions, e.g., as further described herein.

[0076] In some embodiments, the server 104 may be part of a radio access network, such as a 5G New Radio (5G NR) radio access network. In some embodiments, the server 104 may be connected to a legacy evolved packet core (EPC) network and / or to a NR core (NRC) network.

[0077] As described further subsequently herein, the server 104 may include hardware and software components for implementing or supporting implementation of features described herein. The processor 344 of the server 104 may be configured to implement or support implementation of part or all of the methods described herein, e.g., by executing program instructions stored on a memory medium (e.g., a non-transitory computer-readable memory medium). Alternatively, the processor 344 may be configured as a programmable hardware element, such as an FPGA (Field Programmable Gate Array), or as an ASIC (Application Specific Integrated Circuit), or a combination thereof. Alternatively (or in addition) the processor 344 of the server 104, in conjunction with one or more of the other components 354, 364, and / or 374 may be configured to implement or support implementation of part or all of the features described herein.

[0078] In addition, as described herein, processor(s) 344 may be comprised of one or more processing elements. In other words, one or more processing elements may be included in processor(s) 344. Thus, processor(s) 344 may include one or more integrated circuits (ICs) that are configured to perform the functions of processor(s) 344. In addition, each integrated circuit may include circuitry (e.g., first circuitry, second circuitry, etc.) configured to perform the functions of processor(s) 344. Figure 4: Block Diagram of a UE

[0079] Figure 4 illustrates an example simplified block diagram of a communication device 106, according to some embodiments. It is noted that the block diagram of the communication device of Figure 4 is only one example of a possible communication device. According to embodiments, communication device 106 may be a user equipment (UE) device, a mobile device or mobile station, a wireless device or wireless station, a desktop computer or computing device, a mobile computing device (e.g., a laptop, notebook, or portable computing device), a tablet, an unmanned aerial vehicle (UAV), a UAV controller (UAC) and / or a combination of devices, among other devices. As shown, the communication device 106 may include a set of components 400 configured to performcore functions. For example, this set of components may be implemented as a system on chip (SOC), which may include portions for various purposes. Alternatively, this set of components 400 may be implemented as separate components or groups of components for the various purposes. The set of components 400 may be coupled (e.g., communicatively; directly or indirectly) to various other circuits of the communication device 106.

[0080] For example, the communication device 106 may include various types of memory (e.g., including NAND flash 410), an input / output interface such as connector I / F 420 (e.g., for connecting to a computer system; dock; charging station; input devices, such as a microphone, camera, keyboard; output devices, such as speakers; etc.), the display 460, which may be integrated with or external to the communication device 106, and cellular communication circuitry 430 such as for 5G NR, LTE, GSM, etc., and short to medium range wireless communication circuitry 429 (e.g., Bluetooth™ and WLAN circuitry). In some embodiments, communication device 106 may include wired communication circuitry (not shown), such as a network interface card, e.g., for Ethernet.

[0081] The cellular communication circuitry 430 may couple (e.g., communicatively; directly or indirectly) to one or more antennas, such as antennas 435 and 436 as shown. The short to medium range wireless communication circuitry 429 may also couple (e.g., communicatively; directly or indirectly) to one or more antennas, such as antennas 437 and 438 as shown. Alternatively, the short to medium range wireless communication circuitry 429 may couple (e.g., communicatively; directly or indirectly) to the antennas 435 and 436 in addition to, or instead of, coupling (e.g., communicatively; directly or indirectly) to the antennas 437 and 438. The short to medium range wireless communication circuitry 429 and / or cellular communication circuitry 430 may include multiple receive chains and / or multiple transmit chains for receiving and / or transmitting multiple spatial streams, such as in a multiple-input multiple output (MIMO) configuration.

[0082] In some embodiments, as further described below, cellular communication circuitry 430 may include dedicated receive chains (including and / or coupled to, e.g., communicatively; directly or indirectly. dedicated processors and / or radios) for multiple RATs (e.g., a first receive chain for LTE and a second receive chain for 5G NR). In addition, in some embodiments, cellular communication circuitry 430 may include a single transmit chain that may be switched between radios dedicated to specific RATs. For example, a first radio may be dedicated to a first RAT, e.g., LTE, and may be in communication with adedicated receive chain and a transmit chain shared with an additional radio, e.g., a second radio that may be dedicated to a second RAT, e.g., 5G NR, and may be in communication with a dedicated receive chain and the shared transmit chain.

[0083] The communication device 106 may also include and / or be configured for use with one or more user interface elements. The user interface elements may include any of various elements, such as display 460 (which may be a touchscreen display), a keyboard (which may be a discrete keyboard or may be implemented as part of a touchscreen display), a mouse, a microphone and / or speakers, one or more cameras, one or more buttons, and / or any of various other elements capable of providing information to a user and / or receiving or interpreting user input.

[0084] The communication device 106 may further include one or more smart cards 445 that include SIM (Subscriber Identity Module) functionality, such as one or more UICC(s) (Universal Integrated Circuit Card(s)) cards 445. Note that the term “SIM” or “SIM entity” is intended to include any of various types of SIM implementations or SIM functionality, such as the one or more UICC(s) cards 445, one or more eUICCs, one or more eSIMs, either removable or embedded, etc. In some embodiments, the UE 106 may include at least two SIMs. Each SIM may execute one or more SIM applications and / or otherwise implement SIM functionality. Thus, each SIM may be a single smart card that may be embedded, e.g., may be soldered onto a circuit board in the UE 106, or each SIM 410 may be implemented as a removable smart card. Thus, the SIM(s) may be one or more removable smart cards (such as UICC cards, which are sometimes referred to as “SIM cards”), and / or the SIMs 410 may be one or more embedded cards (such as embedded UICCs (eUICCs), which are sometimes referred to as “eSIMs” or “eSIM cards”). In some embodiments (such as when the SIM(s) include an eUICC), one or more of the SIM(s) may implement embedded SIM (eSIM) functionality; in such an embodiment, a single one of the SIM(s) may execute multiple SIM applications. Each of the SIMs may include components such as a processor and / or a memory; instructions for performing SIM / eSIM functionality may be stored in the memory and executed by the processor. In some embodiments, the UE 106 may include a combination of removable smart cards and fixed / non-removable smart cards (such as one or more eUICC cards that implement eSIM functionality), as desired. For example, the UE 106 may comprise two embedded SIMs, two removable SIMs, or a combination of one embedded SIMs and one removable SIMs. Various other SIM configurations are also contemplated.

[0085] As noted above, in some embodiments, the UE 106 may include two or more SIMs. The inclusion of two or more SIMs in the UE 106 may allow the UE 106 to support two different telephone numbers and may allow the UE 106 to communicate on corresponding two or more respective networks. For example, a first SIM may support a first RAT such as LTE, and a second SIM 410 support a second RAT such as 5G NR. Other implementations and RATs are of course possible. In some embodiments, when the UE 106 comprises two SIMs, the UE 106 may support Dual SIM Dual Active (DSDA) functionality. The DSDA functionality may allow the UE 106 to be simultaneously connected to two networks (and use two different RATs) at the same time, or to simultaneously maintain two connections supported by two different SIMs using the same or different RATs on the same or different networks. The DSDA functionality may also allow the UE 106 to simultaneously receive voice calls or data traffic on either phone number. In certain embodiments the voice call may be a packet switched communication. In other words, the voice call may be received using voice over LTE (VoLTE) technology and / or voice over NR (VoNR) technology. In some embodiments, the UE 106 may support Dual SIM Dual Standby (DSDS) functionality. The DSDS functionality may allow either of the two SIMs in the UE 106 to be on standby waiting for a voice call and / or data connection. In DSDS, when a call / data is established on one SIM, the other SIM is no longer active. In some embodiments, DSDx functionality (either DSDA or DSDS functionality) may be implemented with a single SIM (e.g., a eUICC) that executes multiple SIM applications for different carriers and / or RATs.

[0086] As shown, the SOC 400 may include processor(s) 402, which may execute program instructions for the communication device 106 and display circuitry 404, which may perform graphics processing and provide display signals to the display 460. The processor(s) 402 may also be coupled to memory management unit (MMU) 440, which may be configured to receive addresses from the processor(s) 402 and translate those addresses to locations in memory (e.g., memory 406, read only memory (ROM) 450, NAND flash memory 410) and / or to other circuits or devices, such as the display circuitry 404, short to medium range wireless communication circuitry 429, cellular communication circuitry 430, connector I / F 420, and / or display 460. The MMU 440 may be configured to perform memory protection and page table translation or set up. In some embodiments, the MMU 440 may be included as a portion of the processor(s) 402.

[0087] As noted above, the communication device 106 may be configured tocommunicate using wireless and / or wired communication circuitry. The communication device 106 may be configured to perform methods for post deployment adaptation of AI / ML-based models, including systems, methods, and mechanisms for data collection, training, and performance monitoring of AI / ML-based models, as further described herein.

[0088] As described herein, the communication device 106 may include hardware and software components for implementing the above features for a communication device 106 to communicate a scheduling profile for power savings to a network. The processor 402 of the communication device 106 may be configured to implement part or all of the features described herein, e.g., by executing program instructions stored on a memory medium (e.g., a non-transitory computer-readable memory medium). Alternatively (or in addition), processor 402 may be configured as a programmable hardware element, such as an FPGA (Field Programmable Gate Array), or as an ASIC (Application Specific Integrated Circuit). Alternatively (or in addition) the processor 402 of the communication device 106, in conjunction with one or more of the other components 400, 404, 406, 410, 420, 429, 430, 440, 445, 450, 460 may be configured to implement part or all of the features described herein.

[0089] In addition, as described herein, processor 402 may include one or more processing elements. Thus, processor 402 may include one or more integrated circuits (ICs) that are configured to perform the functions of processor 402. In addition, each integrated circuit may include circuitry (e.g., first circuitry, second circuitry, etc.) configured to perform the functions of processor(s) 402.

[0090] Further, as described herein, cellular communication circuitry 430 and short to medium range wireless communication circuitry 429 may each include one or more processing elements. In other words, one or more processing elements may be included in cellular communication circuitry 430 and, similarly, one or more processing elements may be included in short to medium range wireless communication circuitry 429. Thus, cellular communication circuitry 430 may include one or more integrated circuits (ICs) that are configured to perform the functions of cellular communication circuitry 430. In addition, each integrated circuit may include circuitry (e.g., first circuitry, second circuitry, etc.) configured to perform the functions of cellular communication circuitry 430. Similarly, the short to medium range wireless communication circuitry 429 may include one or more ICs that are configured to perform the functions of short to medium range wirelesscommunication circuitry 429. In addition, each integrated circuit may include circuitry (e.g., first circuitry, second circuitry, etc.) configured to perform the functions of short to medium range wireless communication circuitry 429. Figure 5: Block Diagram of Cellular Communication Circuitry

[0091] Figure 5 illustrates an example simplified block diagram of cellular communication circuitry, according to some embodiments. It is noted that the block diagram of the cellular communication circuitry of Figure 5 is only one example of a possible cellular communication circuit. According to embodiments, cellular communication circuitry 530, which may be cellular communication circuitry 430, may be included in a communication device, such as communication device 106 described above. As noted above, communication device 106 may be a user equipment (UE) device, a mobile device or mobile station, a wireless device or wireless station, a desktop computer or computing device, a mobile computing device (e.g., a laptop, notebook, or portable computing device), a tablet and / or a combination of devices, among other devices.

[0092] The cellular communication circuitry 530 may couple (e.g., communicatively; directly or indirectly) to one or more antennas, such as antennas 435a-b and 436 as shown (in Figure 4). In some embodiments, cellular communication circuitry 530 may include dedicated receive chains (including and / or coupled to, e.g., communicatively; directly or indirectly. dedicated processors and / or radios) for multiple RATs (e.g., a first receive chain for LTE and a second receive chain for 5G NR). For example, as shown in Figure 5, cellular communication circuitry 530 may include a modem 510 and a modem 520. Modem 510 may be configured for communications according to a first RAT, e.g., such as LTE or LTE- A, and modem 520 may be configured for communications according to a second RAT, e.g., such as 5G NR.

[0093] As shown, modem 510 may include one or more processors 512 and a memory 516 in communication with processors 512. Modem 510 may be in communication with a radio frequency (RF) front end 530. RF front end 530 may include circuitry for transmitting and receiving radio signals. For example, RF front end 530 may include receive circuitry (RX) 532 and transmit circuitry (TX) 534. In some embodiments, receive circuitry 532 may be in communication with downlink (DL) front end 550, which may include circuitry for receiving radio signals via antenna 335a.

[0094] Similarly, modem 520 may include one or more processors 522 and a memory 526 in communication with processors 522. Modem 520 may be in communication with an RF front end 540. RF front end 540 may include circuitry for transmitting and receiving radio signals. For example, RF front end 540 may include receive circuitry 542 and transmit circuitry 544. In some embodiments, receive circuitry 542 may be in communication with DL front end 560, which may include circuitry for receiving radio signals via antenna 335b.

[0095] In some embodiments, a switch 570 may couple transmit circuitry 534 to uplink (UL) front end 572. In addition, switch 570 may couple transmit circuitry 544 to UL front end 572. UL front end 572 may include circuitry for transmitting radio signals via antenna 336. Thus, when cellular communication circuitry 530 receives instructions to transmit according to the first RAT (e.g., as supported via modem 510), switch 570 may be switched to a first state that allows modem 510 to transmit signals according to the first RAT (e.g., via a transmit chain that includes transmit circuitry 534 and UL front end 572). Similarly, when cellular communication circuitry 530 receives instructions to transmit according to the second RAT (e.g., as supported via modem 520), switch 570 may be switched to a second state that allows modem 520 to transmit signals according to the second RAT (e.g., via a transmit chain that includes transmit circuitry 544 and UL front end 572).

[0096] In some embodiments, the cellular communication circuitry 530 may be configured to perform methods for post deployment adaptation of AI / ML-based models, including systems, methods, and mechanisms for data collection, training, and performance monitoring of AI / ML-based models, as further described herein.

[0097] As described herein, the modem 510 may include hardware and software components for implementing the above features or for time division multiplexing UL data for NSA NR operations, as well as the various other techniques described herein. The processors 512 may be configured to implement part or all of the features described herein, e.g., by executing program instructions stored on a memory medium (e.g., a non-transitory computer-readable memory medium). Alternatively (or in addition), processor 512 may be configured as a programmable hardware element, such as an FPGA (Field Programmable Gate Array), or as an ASIC (Application Specific Integrated Circuit). Alternatively (or in addition) the processor 512, in conjunction with one or more of the other components 530, 532, 534, 550, 570, 572, 335 and 336 may be configured to implement part or all of the features described herein.

[0098] In addition, as described herein, processors 512 may include one or more processing elements. Thus, processors 512 may include one or more integrated circuits (ICs) that are configured to perform the functions of processors 512. In addition, each integrated circuit may include circuitry (e.g., first circuitry, second circuitry, etc.) configured to perform the functions of processors 512.

[0099] As described herein, the modem 520 may include hardware and software components for implementing the above features for performing methods for post deployment adaptation of AI / ML models, including systems, methods, and mechanisms for data collection, training, and performance monitoring of AI / ML models, as further described herein, as well as the various other techniques described herein. The processors 522 may be configured to implement part or all of the features described herein, e.g., by executing program instructions stored on a memory medium (e.g., a non-transitory computer- readable memory medium). Alternatively (or in addition), processor 522 may be configured as a programmable hardware element, such as an FPGA (Field Programmable Gate Array), or as an ASIC (Application Specific Integrated Circuit). Alternatively (or in addition) the processor 522, in conjunction with one or more of the other components 540, 542, 544, 550, 570, 572, 335 and 336 may be configured to implement part or all of the features described herein.

[0100] In addition, as described herein, processors 522 may include one or more processing elements. Thus, processors 522 may include one or more integrated circuits (ICs) that are configured to perform the functions of processors 522. In addition, each integrated circuit may include circuitry (e.g., first circuitry, second circuitry, etc.) configured to perform the functions of processors 522. Figures 6A, 6B, and 7: 5G Core Network Architecture – Interworking with Wi-Fi

[0101] In some embodiments, the 5G core network (CN) may be accessed via (or through) a cellular connection / interface (e.g., via a 3GPP communication architecture / protocol) and a non- cellular connection / interface (e.g., a non-3GPP access architecture / protocol such as Wi-Fi connection). Figure 6A illustrates an example of a 5G network architecture that incorporates both 3GPP (e.g., cellular) and non-3GPP (e.g., non- cellular) access to the 5G CN, according to some embodiments. As shown, a user equipment device (e.g., such as UE 106) may access the 5G CN through both a radio accessnetwork (RAN, e.g., such as gNB 604, which may be a base station 102) and an access point, such as AP 612. The AP 612 may include a connection to the Internet 600 as well as a connection to a non-3GPP inter-working function (N3IWF) 603 network entity. The N3IWF may include a connection to a core access and mobility management function (AMF) 605 of the 5G CN. The AMF 605 may include an instance of a 5G mobility management (5G MM) function associated with the UE 106. In addition, the RAN (e.g., gNB 604) may also have a connection to the AMF 605. Thus, the 5G CN may support unified authentication over both connections as well as allow simultaneous registration for UE 106 access via both gNB 604 and AP 612. As shown, the AMF 605 may include one or more functional entities associated with the 5G CN (e.g., network slice selection function (NSSF) 620, short message service function (SMSF) 622, application function (AF) 624, unified data management (UDM) 626, policy control function (PCF) 628, and / or authentication server function (AUSF) 630). Note that these functional entities may also be supported by a session management function (SMF) 606a and an SMF 606b of the 5G CN. The AMF 605 may be connected to (or in communication with) the SMF 606a. Further, the gNB 604 may in communication with (or connected to) a user plane function (UPF) 608a that may also be communication with the SMF 606a. Similarly, the N3IWF 603 may be communicating with a UPF 608b that may also be communicating with the SMF 606b. Both UPFs may be communicating with the data network (e.g., DN 610a and 610b) and / or the Internet 600 and Internet Protocol (IP) Multimedia Subsystem / IP Multimedia Core Network Subsystem (IMS) core network 610.

[0102] Figure 6B illustrates an example of a 5G network architecture that incorporates both dual 3GPP (e.g., LTE and 5G NR) access and non-3GPP access to the 5G CN, according to some embodiments. As shown, a user equipment device (e.g., such as UE 106) may access the 5G CN through both a radio access network (RAN, e.g., such as gNB 604 or eNB 602, which may be a base station 102) and an access point, such as AP 612. The AP 612 may include a connection to the Internet 600 as well as a connection to the N3IWF 603 network entity. The N3IWF may include a connection to the AMF 605 of the 5G CN. The AMF 605 may include an instance of the 5G MM function associated with the UE 106. In addition, the RAN (e.g., gNB 604) may also have a connection to the AMF 605. Thus, the 5G CN may support unified authentication over both connections as well as allow simultaneous registration for UE 106 access via both gNB 604 and AP 612. In addition, the 5G CN may support dual-registration of the UE on both a legacy network (e.g.,LTE via eNB 602) and a 5G network (e.g., via gNB 604). As shown, the eNB 602 may have connections to a mobility management entity (MME) 642 and a serving gateway (SGW) 644. The MME 642 may have connections to both the SGW 644 and the AMF 605. In addition, the SGW 644 may have connections to both the SMF 606a and the UPF 608a. As shown, the AMF 605 may include one or more functional entities associated with the 5G CN (e.g., NSSF 620, SMSF 622, AF 624, UDM 626, PCF 628, and / or AUSF 630). Note that UDM 626 may also include a home subscriber server (HSS) function and the PCF may also include a policy and charging rules function (PCRF). Note further that these functional entities may also be supported by the SMF606a and the SMF 606b of the 5G CN. The AMF 605 may be connected to (or in communication with) the SMF 606a. Further, the gNB 604 may in communication with (or connected to) the UPF 608a that may also be communication with the SMF 606a. Similarly, the N3IWF 603 may be communicating with a UPF 608b that may also be communicating with the SMF 606b. Both UPFs may be communicating with the data network (e.g., DN 610a and 610b) and / or the Internet 600 and IMS core network 610.

[0103] Note that in various embodiments, one or more of the above-described network entities may be configured to perform methods for post deployment adaptation of AI / ML models, including systems, methods, and mechanisms for data collection, training, and performance monitoring of AI / ML models, as further described herein.

[0104] Figure 7 illustrates an example of a baseband processor architecture for a UE (e.g., such as UE 106), according to some embodiments. The baseband processor architecture 700 described in Figure 7 may be implemented on one or more radios (e.g., radios 429 and / or 430 described above) or modems (e.g., modems 510 and / or 520) as described above. As shown, the non-access stratum (NAS) 710 may include a 5G NAS 720 and a legacy NAS 750. The legacy NAS 750 may include a communication connection with a legacy access stratum (AS) 770. The 5G NAS 720 may include communication connections with both a 5G AS 740 and a non-3GPP AS 730 and Wi-Fi AS 732. The 5G NAS 720 may include functional entities associated with both access stratums. Thus, the 5G NAS 720 may include multiple 5G MM entities 726 and 728 and 5G session management (SM) entities 722 and 724. The legacy NAS 750 may include functional entities such as short message service (SMS) entity 752, evolved packet system (EPS) session management (ESM) entity 754, session management (SM) entity 756, EPS mobility management (EMM) entity 758, and mobility management (MM) / GPRS mobilitymanagement (GMM) entity 760. In addition, the legacy AS 770 may include functional entities such as LTE AS 772, UMTS AS 774, and / or GSM / GPRS AS 776.

[0105] Thus, the baseband processor architecture 700 allows for a common 5G-NAS for both 5G cellular and non-cellular (e.g., non-3GPP access). The baseband processor architecture 700 can be in communication with one or more UICC(s) 745. Note that as shown, the 5G MM may maintain individual connection management and registration management state machines for each connection. Additionally, a device (e.g., UE 106) may register to a single PLMN (e.g., 5G CN) using 5G cellular access as well as non-cellular access. Further, it may be possible for the device to be in a connected state in one access and an idle state in another access and vice versa. Finally, there may be common 5G-MM procedures (e.g., registration, de-registration, identification, authentication, as so forth) for both accesses.

[0106] Note that in various embodiments, one or more of the above-described functional entities of the 5G NAS and / or 5G AS may be configured to perform methods for perform methods for post deployment adaptation of AI / ML models, including systems, methods, and mechanisms for data collection, training, and performance monitoring of AI / ML models, as further described herein. Figures 8 and 9: Device components

[0107] Figure 8 illustrates example components of a device 800 in accordance with some embodiments. In some embodiments, the device 800 may include application circuitry 802, baseband circuitry 804, Radio Frequency (RF) circuitry 806, front-end module (FEM) circuitry 808, one or more antennas 810, and power management circuitry (PMC) 812 coupled together at least as shown. The components of the illustrated device 800 may be included in a UE or a RAN node. In some embodiments, the device 800 may include less elements (e.g., a RAN node may not utilize application circuitry 802, and instead include a processor / controller to process IP data received from an EPC). In some embodiments, the device 800 may include additional elements such as, for example, memory / storage, display, camera, sensor, or input / output (I / O) interface. In other embodiments, the components described below may be included in more than one device (e.g., said circuitries may be separately included in more than one device for Cloud-RAN (C-RAN) implementations).

[0108] The application circuitry 802 may include one or more application processors. For example, the application circuitry 802 may include circuitry such as, but not limited to, one or more single-core or multi-core processors. The processor(s) may include any combination of general-purpose processors and dedicated processors (e.g., graphics processors, application processors, etc.). The processors may be coupled with or may include memory / storage and may be configured to execute instructions stored in the memory / storage to enable various applications or operating systems to run on the device 800. In some embodiments, processors of application circuitry 802 may process IP data packets received from an EPC.

[0109] The baseband circuitry 804 may include circuitry such as, but not limited to, one or more single-core or multi-core processors. The baseband circuitry 804 may include one or more baseband processors or control logic to process baseband signals received from a receive signal path of the RF circuitry 806 and to generate baseband signals for a transmit signal path of the RF circuitry 806. Baseband processing circuity 804 may interface with the application circuitry 802 for generation and processing of the baseband signals and for controlling operations of the RF circuitry 806. For example, in some embodiments, the baseband circuitry 804 may include a third generation (3G) baseband processor 804A, a fourth generation (4G) baseband processor 804B, a fifth generation (5G) baseband processor 804C, or other baseband processor(s) 804D for other existing generations, generations in development or to be developed in the future (e.g., second generation (2G), sixth generation (6G), etc.). The baseband circuitry 804 (e.g., one or more of baseband processors 804A-D) may handle various radio control functions that enable communication with one or more radio networks via the RF circuitry 806. In other embodiments, some or all of the functionality of baseband processors 804A-D may be included in modules stored in the memory 804G and executed via a Central Processing Unit (CPU) 804E. The radio control functions may include, but are not limited to, signal modulation / demodulation, encoding / decoding, radio frequency shifting, etc. In some embodiments, modulation / demodulation circuitry of the baseband circuitry 804 may include Fast-Fourier Transform (FFT), precoding, or constellation mapping / demapping functionality. In some embodiments, encoding / decoding circuitry of the baseband circuitry 804 may include convolution, tail-biting convolution, turbo, Viterbi, or Low Density Parity Check (LDPC) encoder / decoder functionality. Embodiments of modulation / demodulation and encoder / decoder functionality are not limited to these examples and may include othersuitable functionality in other embodiments.

[0110] In some embodiments, the baseband circuitry 804 may include one or more audio digital signal processor(s) (DSP) 804F. The audio DSP(s) 804F may be include elements for compression / decompression and echo cancellation and may include other suitable processing elements in other embodiments. Components of the baseband circuitry may be suitably combined in a single chip, a single chipset, or disposed on a same circuit board in some embodiments. In some embodiments, some or all of the constituent components of the baseband circuitry 804 and the application circuitry 802 may be implemented together such as, for example, on a system on a chip (SOC).

[0111] In some embodiments, the baseband circuitry 804 may provide for communication compatible with one or more radio technologies. For example, in some embodiments, the baseband circuitry 804 may support communication with an evolved universal terrestrial radio access network (EUTRAN) or other wireless metropolitan area networks (WMAN), a wireless local area network (WLAN), a wireless personal area network (WPAN). Embodiments in which the baseband circuitry 804 is configured to support radio communications of more than one wireless protocol may be referred to as multi-mode baseband circuitry.

[0112] RF circuitry 806 may enable communication with wireless networks using modulated electromagnetic radiation through a non-solid medium. In various embodiments, the RF circuitry 806 may include switches, filters, amplifiers, etc. to facilitate the communication with the wireless network. RF circuitry 806 may include a receive signal path which may include circuitry to down-convert RF signals received from the FEM circuitry 808 and provide baseband signals to the baseband circuitry 804. RF circuitry 806 may also include a transmit signal path which may include circuitry to up-convert baseband signals provided by the baseband circuitry 804 and provide RF output signals to the FEM circuitry 808 for transmission.

[0113] In some embodiments, the receive signal path of the RF circuitry 806 may include mixer circuitry 806a, amplifier circuitry 806b and filter circuitry 806c. In some embodiments, the transmit signal path of the RF circuitry 806 may include filter circuitry 806c and mixer circuitry 806a. RF circuitry 806 may also include synthesizer circuitry 806d for synthesizing a frequency for use by the mixer circuitry 806a of the receive signal path and the transmit signal path. In some embodiments, the mixer circuitry 806a of the receive signalpath may be configured to down-convert RF signals received from the FEM circuitry 808 based on the synthesized frequency provided by synthesizer circuitry 806d. The amplifier circuitry 806b may be configured to amplify the down-converted signals and the filter circuitry 806c may be a low-pass filter (LPF) or band-pass filter (BPF) configured to remove unwanted signals from the down-converted signals to generate output baseband signals. Output baseband signals may be provided to the baseband circuitry 804 for further processing. In some embodiments, the output baseband signals may be zero-frequency baseband signals, although this is not a requirement. In some embodiments, mixer circuitry 806a of the receive signal path may comprise passive mixers, although the scope of the embodiments is not limited in this respect.

[0114] In some embodiments, the mixer circuitry 806a of the transmit signal path may be configured to up-convert input baseband signals based on the synthesized frequency provided by the synthesizer circuitry 806d to generate RF output signals for the FEM circuitry 808. The baseband signals may be provided by the baseband circuitry 804 and may be filtered by filter circuitry 806c.

[0115] In some embodiments, the mixer circuitry 806a of the receive signal path and the mixer circuitry 806a of the transmit signal path may include two or more mixers and may be arranged for quadrature downconversion and upconversion, respectively. In some embodiments, the mixer circuitry 806a of the receive signal path and the mixer circuitry 806a of the transmit signal path may include two or more mixers and may be arranged for image rejection (e.g., Hartley image rejection). In some embodiments, the mixer circuitry 806a of the receive signal path and the mixer circuitry 806a may be arranged for direct downconversion and direct upconversion, respectively. In some embodiments, the mixer circuitry 806a of the receive signal path and the mixer circuitry 806a of the transmit signal path may be configured for super-heterodyne operation.

[0116] In some embodiments, the output baseband signals and the input baseband signals may be analog baseband signals, although the scope of the embodiments is not limited in this respect. In some alternate embodiments, the output baseband signals and the input baseband signals may be digital baseband signals. In these alternate embodiments, the RF circuitry 806 may include analog-to-digital converter (ADC) and digital-to-analog converter (DAC) circuitry and the baseband circuitry 804 may include a digital baseband interface to communicate with the RF circuitry 806.

[0117] In some dual-mode embodiments, a separate radio IC circuitry may be provided for processing signals for each spectrum, although the scope of the embodiments is not limited in this respect.

[0118] In some embodiments, the synthesizer circuitry 806d may be a fractional-N synthesizer or a fractional N / N+1 synthesizer, although the scope of the embodiments is not limited in this respect as other types of frequency synthesizers may be suitable. For example, synthesizer circuitry 806d may be a delta-sigma synthesizer, a frequency multiplier, or a synthesizer comprising a phase-locked loop with a frequency divider.

[0119] The synthesizer circuitry 806d may be configured to synthesize an output frequency for use by the mixer circuitry 806a of the RF circuitry 806 based on a frequency input and a divider control input. In some embodiments, the synthesizer circuitry 806d may be a fractional N / N+1 synthesizer.

[0120] In some embodiments, frequency input may be provided by a voltage controlled oscillator (VCO), although that is not a requirement. Divider control input may be provided by either the baseband circuitry 804 or the applications processor 802 depending on the desired output frequency. In some embodiments, a divider control input (e.g., N) may be determined from a look-up table based on a channel indicated by the applications processor 802.

[0121] Synthesizer circuitry 806d of the RF circuitry 806 may include a divider, a delay- locked loop (DLL), a multiplexer and a phase accumulator. In some embodiments, the divider may be a dual modulus divider (DMD) and the phase accumulator may be a digital phase accumulator (DPA). In some embodiments, the DMD may be configured to divide the input signal by either N or N+1 (e.g., based on a carry out) to provide a fractional division ratio. In some example embodiments, the DLL may include a set of cascaded, tunable, delay elements, a phase detector, a charge pump and a D-type flip-flop. In these embodiments, the delay elements may be configured to break a VCO period up into Nd equal packets of phase, where Nd is the number of delay elements in the delay line. In this way, the DLL provides negative feedback to help ensure that the total delay through the delay line is one VCO cycle.

[0122] In some embodiments, synthesizer circuitry 806d may be configured to generate a carrier frequency as the output frequency, while in other embodiments, the output frequency may be a multiple of the carrier frequency (e.g., twice the carrier frequency, four times the carrier frequency) and used in conjunction with quadrature generator and divider circuitry togenerate multiple signals at the carrier frequency with multiple different phases with respect to each other. In some embodiments, the output frequency may be a LO frequency (fLO). In some embodiments, the RF circuitry 806 may include an IQ / polar converter.

[0123] FEM circuitry 808 may include a receive signal path which may include circuitry configured to operate on RF signals received from one or more antennas 810, amplify the received signals and provide the amplified versions of the received signals to the RF circuitry 806 for further processing. FEM circuitry 808 may also include a transmit signal path which may include circuitry configured to amplify signals for transmission provided by the RF circuitry 806 for transmission by one or more of the one or more antennas 810. In various embodiments, the amplification through the transmit or receive signal paths may be done solely in the RF circuitry 806, solely in the FEM 808, or in both the RF circuitry 806 and the FEM 808.

[0124] In some embodiments, the FEM circuitry 808 may include a TX / RX switch to switch between transmit mode and receive mode operation. The FEM circuitry may include a receive signal path and a transmit signal path. The receive signal path of the FEM circuitry may include an LNA to amplify received RF signals and provide the amplified received RF signals as an output (e.g., to the RF circuitry 806). The transmit signal path of the FEM circuitry 808 may include a power amplifier (PA) to amplify input RF signals (e.g., provided by RF circuitry 806), and one or more filters to generate RF signals for subsequent transmission (e.g., by one or more of the one or more antennas 810).

[0125] In some embodiments, the PMC 812 may manage power provided to the baseband circuitry 804. In particular, the PMC 812 may control power-source selection, voltage scaling, battery charging, or DC-to-DC conversion. The PMC 812 may often be included when the device 800 is capable of being powered by a battery, for example, when the device is included in a UE. The PMC 812 may increase the power conversion efficiency while providing desirable implementation size and heat dissipation characteristics.

[0126] While Figure 8 shows the PMC 812 coupled only with the baseband circuitry 804. However, in other embodiments, the PMC 812 may be additionally or alternatively coupled with, and perform similar power management operations for, other components such as, but not limited to, application circuitry 802, RF circuitry 806, or FEM 808.

[0127] In some embodiments, the PMC 812 may control, or otherwise be part of, various power saving mechanisms of the device 800. For example, if the device 800 is in anRRC_Connected state, where it is still connected to the RAN node as it expects to receive traffic shortly, then it may enter a state known as Discontinuous Reception Mode (DRX) after a period of inactivity. During this state, the device 800 may power down for brief intervals of time and thus save power.

[0128] If there is no data traffic activity for an e8ended period of time, then the device 800 may transition off to an RRC_Idle state, where it disconnects from the network and does not perform operations such as channel quality feedback, handover, etc. The device 800 goes into a very low power state and it performs paging where again it periodically wakes up to listen to the network and then powers down again. The device 800 may not receive data in this state, in order to receive data, it can transition back to RRC_Connected state.

[0129] An additional power saving mode may allow a device to be unavailable to the network for periods longer than a paging interval (ranging from seconds to a few hours). During this time, the device is totally unreachable to the network and may power down completely. Any data sent during this time incurs a large delay and it is assumed the delay is acceptable.

[0130] Processors of the application circuitry 802 and processors of the baseband circuitry 804 may be used to execute elements of one or more instances of a protocol stack. For example, processors of the baseband circuitry 804, alone or in combination, may be used to execute Layer 3, Layer 2, or Layer 1 functionality, while processors of the application circuitry 804 may utilize data (e.g., packet data) received from these layers and further execute Layer 4 functionality (e.g., transmission communication protocol (TCP) and user datagram protocol (UDP) layers). As referred to herein, Layer 3 may comprise a radio resource control (RRC) layer, described in further detail below. As referred to herein, Layer 2 may comprise a medium access control (MAC) layer, a radio link control (RLC) layer, and a packet data convergence protocol (PDCP) layer, described in further detail below. As referred to herein, Layer 1 may comprise a physical (PHY) layer of a UE / RAN node, described in further detail below.

[0131] Figure 9 illustrates example interfaces of baseband circuitry in accordance with some embodiments. As discussed above, the baseband circuitry 804 of Figure 8 may comprise processors 804A-804E and a memory 804G utilized by said processors. Each of the processors 804A-804E may include a memory interface, 904A-904E, respectively, to send / receive data to / from the memory 804G.

[0132] The baseband circuitry 804 may further include one or more interfaces to communicatively couple to other circuitries / devices, such as a memory interface 912 (e.g., an interface to send / receive data to / from memory external to the baseband circuitry 804), an application circuitry interface 914 (e.g., an interface to send / receive data to / from the application circuitry 802 of Figure 8), an RF circuitry interface 916 (e.g., an interface to send / receive data to / from RF circuitry 806 of Figure 8), a wireless hardware connectivity interface 918 (e.g., an interface to send / receive data to / from Near Field Communication (NFC) components, Bluetooth® components (e.g., Bluetooth® Low Energy), Wi-Fi® components, and other communication components), and a power management interface 920 (e.g., an interface to send / receive power or control signals to / from the PMC 812. Figure 10: Control Plane Protocol Stack

[0133] Figure 10 is an illustration of a control plane protocol stack in accordance with some embodiments. In one embodiment, a control plane 1000 may be a communications protocol stack between one or more UEs such as, for example, UE 801 (or alternatively, the UE 802), and / or one or more RAN nodes 811 (or alternatively, the RAN node 812), and a mobility management entity (MME) 821.

[0134] The PHY layer 1001 may transmit or receive information used by the MAC layer 1002 over one or more air interfaces. The PHY layer 1001 may further perform link adaptation or adaptive modulation and coding (AMC), power control, cell search (e.g., for initial synchronization and handover purposes), and other measurements used by higher layers, such as the RRC layer 1005. The PHY layer 1001 may still further perform error detection on the transport channels, forward error correction (FEC) coding / decoding of the transport channels, modulation / demodulation of physical channels, interleaving, rate matching, mapping onto physical channels, and Multiple Input Multiple Output (MIMO) antenna processing.

[0135] The MAC layer 1002 may perform mapping between logical channels and transport channels, multiplexing of MAC service data units (SDUs) from one or more logical channels onto transport blocks (TB) to be delivered to PHY via transport channels, de- multiplexing MAC SDUs to one or more logical channels from transport blocks (TB) delivered from the PHY via transport channels, multiplexing MAC SDUs onto TBs, scheduling information reporting, error correction through hybrid automatic repeat request(HARQ), and logical channel prioritization.

[0136] The RLC layer 1003 may operate in a plurality of modes of operation, including: Transparent Mode (TM), Unacknowledged Mode (UM), and Acknowledged Mode (AM). The RLC layer 1003 may execute transfer of upper layer protocol data units (PDUs), error correction through automatic repeat request (ARQ) for AM data transfers, and concatenation, segmentation and reassembly of RLC SDUs for UM and AM data transfers. The RLC layer 1003 may also execute re-segmentation of RLC data PDUs for AM data transfers, reorder RLC data PDUs for UM and AM data transfers, detect duplicate data for UM and AM data transfers, discard RLC SDUs for UM and AM data transfers, detect protocol errors for AM data transfers, and perform RLC re-establishment.

[0137] The PDCP layer 1004 may execute header compression and decompression of IP data, maintain PDCP Sequence Numbers (SNs), perform in-sequence delivery of upper layer PDUs at re-establishment of lower layers, eliminate duplicates of lower layer SDUs at re- establishment of lower layers for radio bearers mapped on RLC AM, cipher and decipher control plane data, perform integrity protection and integrity verification of control plane data, control timer-based discard of data, and perform security operations (e.g., ciphering, deciphering, integrity protection, integrity verification, etc.).

[0138] The main services and functions of the RRC layer 1005 may include broadcast of system information (e.g., included in Master Information Blocks (MIBs) or System Information Blocks (SIBs) related to the non-access stratum (NAS)), broadcast of system information related to the access stratum (AS), paging, establishment, maintenance and release of an RRC connection between the UE and E-UTRAN (e.g., RRC connection paging, RRC connection establishment, RRC connection modification, and RRC connection release), establishment, configuration, maintenance and release of point to point Radio Bearers, security functions including key management, inter radio access technology (RAT) mobility, and measurement configuration for UE measurement reporting. Said MIBs and SIBs may comprise one or more information elements (IEs), which may each comprise individual data fields or data structures.

[0139] In one example, a UE (e.g., UE 106A-N) and a RAN node (e.g., base station 102) 811 may utilize a Uu interface (e.g., an LTE-Uu interface) to exchange control plane data via a protocol stack comprising the PHY layer 1001, the MAC layer 1002, the RLC layer 1003, the PDCP layer 1004, and the RRC layer 1005.

[0140] The non-access stratum (NAS) protocols 1006 form the highest stratum of the control plane between the UE (e.g., UE 106A-N) 801 and an MME 821. The NAS protocols 1006 support the mobility of the UE (e.g., UE 106A-N) 801and the session management procedures to establish and maintain IP connectivity between the UE (e.g., UE 106A-N) 801and a P-GW.

[0141] The S1 Application Protocol (S1-AP) layer 1015 may support the functions of the S1 interface and comprise Elementary Procedures (EPs). An EP is a unit of interaction between a RAN node (e.g., base station 102)811and the CN. The S1-AP layer 1015 services may comprise two groups: UE-associated services and non UE-associated services. These services perform functions including, but not limited to: E-UTRAN Radio Access Bearer (E- RAB) management, UE capability indication, mobility, NAS signaling transport, RAN Information Management (RIM), and configuration transfer.

[0142] The Stream Control Transmission Protocol (SCTP) layer (alternatively referred to as the SCTP / IP layer) 1014 may ensure reliable delivery of signaling messages between the RAN node (e.g., base station 102) 811 and a MME 821 based, in part, on the IP protocol, supported by the IP layer 1013. The L2 layer 1012 and the L1 layer 1011 may refer to communication links (e.g., wired or wireless) used by the RAN node (e.g., base station 102) and the MME to exchange information.

[0143] The RAN node (e.g., base station 102) 811and the MME 821may utilize an S1- MME interface to exchange control plane data via a protocol stack comprising the L1 layer 1011, the L2 layer 1012, the IP layer 1013, the SCTP layer 1014, and the S1-AP layer 1015. Figure 11: CSI Feedback

[0144] In 3GPP standards development, channel state information (CSI) feedback has been an important topic in almost every 3GPP standards release. Further, nearly all CSI codebook design has been focused on feedback based on a current CSI reference signal (CSI-RS) measurement. Figure 11 illustrates an example timing diagram 1100 of utilizing CSI feedback, according to some embodiments. As illustrated in Figure 11, a base station 102 (e.g., gNB) transmits a CSI measurement configuration to a UE 106 that provides instructions for the UE to measure CSI reference signals (CSI-RS) at 1110. A CSI configuration may include information about the types of CSI-RS, the time and / orfrequency to measure reference signals such as CSI-RS, and the like. The base station 102 (e.g., gNB) can transmit a CSI-RS to a UE 106 at 1120. UE 106 performs measurements on the CSI-RS and performs channel estimation at 1130. That is, for example, UE 106 may estimate a raw channel matrix. Based on the estimated raw channel matrix, UE 106 provides feedback to the BS 102 at 1140. As illustrated in Figure 11, this feedback may generally be referred to as a CSI report. A CSI report may include various types of feedback information (e.g., Channel Quality Information (CQI), Layer Indicator, etc.) and may further represent the estimated raw channel matrix in various ways. For example, a precoding matrix may be derived from the raw channel matrix. Further, a precoding matrix may be indexed according to codebooks at UE 106 and BS 102, (e.g., a Type I or Type II codebook) and a CSI report may include a precoding matrix index (PMI) (i.e., precoding codeword or precoding matrix indicator) from which BS 102 can derive a precoding matrix using the shared codebook. BS 102 may design a downlink transmission based on the feedback at 1150. That is, for example, BS 102 may select a channel for a downlink transmission based on the derived precoding matrix. BS 102 performs the downlink transmission according to the design at 1160.

[0145] It should be noted that CSI feedback may incur significant overhead. That is, for example, frequent signaling of CSI reports from multiple UEs may incur significant overhead in the communication bandwidth. Massive multiple-input and multiple output (MIMO), in which the gNB uses a large array of antennas, can considerably improve the system performance. However, the benefit of massive MIMO is based on the knowledge of downlink CSI. In frequency division duplexing (FDD) massive MIMO systems, the UE will feedback the downlink CSI-RS measurements to the BS through the uplink because of the lack of channel reciprocity. The feedback overhead can be substantial due to the high dimension of the CSI in massive MIMO systems. In addition, a large amount of power is used at the UE to provide the feedback.

[0146] Further, CSI feedback performance may be impacted by channel aging. That is, in an implemented system, channel characteristics are inherently time varying and the processing delays and UE mobility may cause a channel estimated from a CSI-RS to degrade by the time a downlink transmission occurs according to the downlink derived according to the CSI.Figures 12, 13, 14, and 15 AI / ML based CSI Feedback Compression and CSI Prediction

[0147] One way of reducing the amount of feedback and reducing power consumption at the UE is through the use of CSI compression and CSI prediction Artificial Intelligence (AI) / Machine Learning (ML) based models. As provided in 3rd Generation Partnership Project; Technical Specification Group Radio Access Network; Study on Artificial Intelligence (AI) / Machine Learning (ML) for NR air interface (Release 18), 3GPP TR 38.843 V18.0.0 (2023-12), the 3GPP standards development, (hereinafter 3GPP TR 38.843) AI / ML-based techniques for compressing CSI feedback and AI / ML-based techniques for channel prediction are currently being studied. It should be noted that AI / ML-based may refer to various AI and Machine Learning (ML) techniques, which may be referred to as AI / ML. Figure 12 illustrates an example of a two-sided AI / ML model for CSI compression, according to some embodiments. Figure 13 illustrates an example of a one-sided AI / ML model for CSI prediction, according to some embodiments. Figure 14 illustrates an example of an AI / ML model for joint CSI prediction and compression, according to some embodiments.

[0148] In the example illustrated in Figure 12, UE 106 receives a CSI-RS and CSI Measurement and Channel Estimator 1210 measures the CSI-RS and performs channel estimation, for example, as described above. For example, CSI Measurement and Channel Estimator 1210 may derive a raw channel matrix, a precoding matrix, and / or a PMI. AI / ML-Based Encoder 1220 may receive one or more of a raw channel matrix, a precoding matrix, and a PMI and generate a feedback bitstream according to AI / ML-based encoding techniques. For example, AI-Based Encoder 1220 may receive a raw channel matrix and compress the raw channel matrix according to several Neural Network (NN) layers, for example, one or more convolution layers, and generate a feedback bitstream. For example, AI / ML-Based Encoder 1220 may include an autoencoder. Figure 15 illustrates an example of a compressing a channel matrix according to convolution layers. In the example illustrated in Figure 15, an MxM channel / precoding matrix is downsampled by a factor of four by each convolution layer. For example, if M equal 1024, convolution layer 0 produces a 512x512 matrix, convolution layer 1 produces a 256x256 matrix, and convolution layer 2 produces a 128x128 matrix. It should be noted that a Neural Network may include hundreds or thousands of layers, and in some cases, may be referred to as a Deep Neural Network (DNN) of a Deep Network (DN). Further, it should be noted that, in addition to convolution layers, a AI / ML-Based Encoder may include several other types of layers, for example,fully connected (FC) layers, long short-term memory networks (LSTMs), variational autoencoder (VAE), generative adversarial networks (GAN), batch normalizations and attention mechanisms. Examples of AI / ML-Based Encoders for CSI include, but are not limited to CsiNet, ConvCsiNet, CsiNet+, Attention-CsiNet, CRNet, LSTM-Attention CsiNet, DS.NLCsiNet, DCGAN, PRVNet, CF-FCFNN, CLNet, ENet, DCRNet, CsiNet+DNN, MRFNet, ACCsiNet, DFECsiNet TransNet, CsiFormer, and CVLNet. The techniques described herein may be generally applicable to various AI / ML-Based models.

[0149] AI / ML-Based Decoder 1230 may perform reciprocal functions of AI / ML-Based Encoder 1220. That is, AI / ML-Based Decoder 1230 receives a feedback bitstream reconstructs the corresponding information that was encoded, e.g., a raw channel matrix. Downlink Transmission Designer 1240 uses this reconstructed information to design a downlink transmission for example, as described above. A two-sided AI / ML model for CSI compression attempts to compress CSI feedback information and thereby reduce the overhead of CSI feedback.

[0150] 3GPP TR 38.843 provides that for CSI compression using a two-sided model use case, considered AI / ML model training collaborations include: Type 1: Joint training of the two-sided model at a single side / entity, e.g., UE-sided or Network-sided; Type 2: Joint training of the two-sided model at network side and UE side, respectively; and Type 3: Separate training at network side and UE side, where the UE-side CSI generation part and the NW-side CSI reconstruction part are trained by UE side and network side, respectively.

[0151] In the example illustrated in Figure 13, UE 106 receives a CSI-RS and CSI Measurement and Channel Estimator 1210 measures the CSI-RS and performs channel estimation, for example, as described above. AI / ML Model 1310 receives information from CSI Measurement and Channel Estimator 1210. For example, AI / ML Model 1310 may receive previous CSI-RS measurements and / or previous estimated raw channel matrices. AI / ML Model 1310 performs channel prediction according to an CSI prediction AI / ML model based on the received data.

[0152] For example, AI / ML model 1310 may be a one-dimensional Long short-term memory (LSTM) AI / ML model for CSI prediction using a time domain. In some example, the AI / ML model 1310 may be used for predicting time domain correlation only (such as, for example, the LSTM). In one example the time-series CSI measurements may be fedinto the LSTM layer, which outputs a vector capturing temporal dependencies and this vector may be fed into a fully connected (FC) layer to generate the CSI prediction output. In other examples, AI / ML model 1310 may be a two-dimensional convolutional neural network (“CNN”) AI model for CSI prediction using a time domain and a frequency domain. For example, a 2D CNN can capture a batch of input data, where each sample can comprise a time series of CSI measurements across different frequency subcarriers. This input tensor may be passed through a series of 2D convolutional layers (e.g., neural network) and enable the model to identify patterns in the CSI that extend across both time steps and subcarriers. AI / ML model 1310 may learn and provide CSI predictions across the future time steps and subcarriers based on the measurements of the CSI-RS input into the AI / ML model as training. In other examples, AI / ML model 1310 may be a three- dimensional convolutional neural network (CNN) AI / ML model for CSI prediction using a time domain and a frequency domain and a spatial (antenna) domain. As the inputs pass through the 3D convolutional layers (e.g., neural network), AI / ML model 1310 may learn and predict the future CSI values across time, frequency, and antenna (spatial) domains. It should be noted that due to different designs, the model data collection categorization information can be different.

[0153] Further, the predicted channel may then be used as a CSI-Report to provide uplink feedback to BS 102. That is, for example, a PMI may be derived from a predicted channel matrix. Downlink Transmission Designer 1240 can use this uplink feedback to design a downlink transmission for example, as described above. That is, according to CSI prediction, a UE may fine-tune CSI feedback using AI / ML-based encoding techniques and the fine-tuned CSI feedback may result in a downlink design which is less susceptible to channel aging. Further, in some case, a CSI prediction can be used by the UE rather than a subsequent CSI-RS transmission. This can reduce CSI-RSs which are transmitted from the BS to the UE.

[0154] 3GPP TR 38.843 provides the following cases for CSI prediction: For model training, training data can be generated by UE. For UE-side model inference, input data is internally available at UE. For performance monitoring at the NW side, calculated performance metrics (if needed) or data needed for performance metric calculation (if needed) can be generated by UE and terminated at gNB. Further, 3GPP TR 38.843 specifies the following types of performance monitoring:Type 1: UE calculates the performance metric(s) UE reports performance monitoring output that facilitates functionality fallback decision at the network -Performance monitoring output details can be further defined -NW may configure threshold criterion to facilitate UE side performance monitoring (if needed). NW makes decision(s) of functionality fallback operation (fallback mechanism to legacy CSI reporting). Type 2: UE reports predicted CSI and / or the corresponding ground-truth NW calculates the performance metrics. NW makes decision(s) of functionality fallback operation (fallback mechanism to legacy CSI reporting). Type 3: UE calculates the performance metric(s) UE reports performance metric(s) to the NW NW makes decision(s) of functionality fallback operation (fallback mechanism to legacy CSI reporting).

[0155] In the example illustrated in Figure 14, a UE 106 and BS may use AI / ML model(s) for joint CSI prediction and compression. That is, the predicted channel may be used as input to AI / ML-Based Encoder 1220 and AI / ML-Based Encoder 1220 may generate a feedback bitstream according to AI-based encoding techniques.

[0156] In general, AI / ML models, including, for example, AI / ML-based models for CSI prediction and compression are trained with the scenario specific properties. It may be impractical to acquire a sufficient amount of training data for a target scenario for a particular deployment, due to the expense and time required to acquire training and collecting data. For example, training a single Deep Neural Network (DNN) may require thousands of samples andmultiple epochs. Users often encounter varying wireless transmission environments, which may necessitate the collection of new data and retraining of models. Thus, it may be impractical to train a new DNN for each user, due to the excessive time and data costs involved. As such, it is desirable to develop techniques that can be used to quickly adapt a AI / ML-based model to new environments using only a small amount of data after the AI / ML-based model is deployed.

[0157] Further, it should be noted that, by the time a model achieves optimal performance, the UE may have already moved into a new wireless channel environment, rendering the model ineffective. Thus, it may be desirable to obtain a model that performs adequately as quickly as possible, rather than trying to obtain absolute optimal performance. Updating models can be very costly, if training takes place without any of the prior training and if all the parameters of the NN are trained. In the case of AI / ML-based models for CSI use cases, there are a plethora of different AI / ML-based models, for example, as provided above, according to different configurations and scenarios. For example, setting for numerous parameters may be considered, such as, for example: UE Speed; Signal-to-Interference-plus-Noise Ratio (SINR); Outdoor / Indoor UE position; line of sight (LOS) / non-line-of-sight (NLOS); Propagation Model: Urban Macrocell (UMa) / Urban Microcell (Umi); Number of Antenna ports; Carrier frequency; and / or bandwidth.

[0158] This disclosure describes techniques that can be applied to quickly adapt a AI / ML- based model to an environment using only a small amount of data after the AI / ML-based model is deployed. In one example, according to techniques described herein, transfer learning may be applied, in which a pre-trained model trained with a large dataset is deployed and subsequently fine-tuned. In one example, according to techniques described herein, meta learning may be applied in which a pre-trained model is trained with an aggregate of small size datasets across different tasks and initialization parameters are optimized. It should be noted that in some examples, the techniques described herein are described in for AI / ML-based model for CSI use cases, the techniques described herein are generally applicable to AI / ML-based models which may be utilized for various use cases. For example, the techniques described herein may be used to adapt any type of AI / ML-based model deployed at a UE. Figures 16 and 17: Transfer Learning and Specific Layer Tuning

[0159] Machine learning and deep learning typically involve modeling, training, and testing within independent domains. In contrast, transfer learning aims to transferknowledge from source domains to target domains, enabling the target domains to achieve improved learning outcomes. The source domains may refer to the domain that has be used to train a neural network. The target domains may be a new domain for the neural network and may include, for example, new wireless channel environments. In general, the dataset in source domains is usually abundant, whereas the dataset in target domains tends to be limited. Obtaining a large number of samples in a target domain is not always feasible. Even when it is possible, the data and time costs required to train a neural network from scratch may be prohibitively high. Transfer learning allows for achieving good model performance by fine-tuning a pre-trained model with a small number of samples on similar tasks. According to the techniques herein, deep learning may be combined with transfer learning in order to address the issue of high training cost. For example, according to the techniques herein, in the case of AI / ML-based model CSI use cases, by applying transfer learning, channel-independent correlations can be extracted from the downlink CSI matrix, which can then be used to train DNNs for other channels.

[0160] As described above, it may be desirable to obtain a model that performs adequately as quickly as possible, rather than trying to obtain absolute optimal performance, as a UE may have already transitioned to a new wireless channel environment before optimal performance can be achieved. In one example, in order to obtain a model that performs adequately as quickly as possible, the number of sample sizes used for fine-tuning on model performance may be reduced, with the goal of minimizing the number of samples required in transfer learning to reduce training costs during the training process. Additionally, while fine-tuning the entire neural network generally yields better performance than fine-tuning only a part of the neural network, fine-tuning only certain layers of the network while keeping the remaining layers unchanged, may reduce the training costs. According to the techniques herein, training costs may be reduced by controlling the number of network parameters that are to be fine-tuned, while ensuring that the model performance meets the practical application requirements.

[0161] Figure 16 is a conceptual diagram illustrating applying transferring to an AI / ML based model according to the techniques herein. Figure 16 may correspond to using an AI / ML based model for CSI feedback applications. As illustrated in Figure 16, a model is pre-trained according to a set of source domain parameters and the model is fine tuned for each subsequent target domain. That is, for example, an AI / ML model may be trained for CSI compression and / or prediction based on a large set of training data. The AI / ML model may be deployed at a UE and fined tuned to each particular wireless channel environmentthe UE encounters. For example, the AI / ML model may be fine-tuned, as the UE moves to different wireless environments. Further, an AI / ML model trained for a particular task, for example, a type of CSI compression may be trained for a different task, for example, another type of CSI compression.

[0162] Figure 17 illustrates a block diagram of an example of a method 1700 to perform training or performance monitoring of an AI / ML-based model, according to some embodiments. The method shown in Figure 17 may be utilized as part of AI / ML-based model Life Cycle Management (LCM). The method shown in Figure 17 may be used in conjunction with any of the systems, methods, or devices shown in the Figures, among other devices. That is, Figure 17 provides an example of a procedure for transfer learning and model validation which may be performed, for example, by a UE. In various embodiments, some of the method elements shown may be performed concurrently, in a different order than shown, or may be omitted. Additional method elements may also be performed as desired. As shown, this method may operate as follows.

[0163] A UE 106, or other device, employs a pre-trained model for a nominal task, ^^(source task) and source domain ^^at 1702. For example, a UE 106 may employ a pre- trained model for CSI compression. The pre-trained model may be trained on a large set of channel / precoding matrices. The UE 106 performs model monitoring at 1704. For example, UE 106 may calculate the performance metric(s) and / or UE 106 may report performance metric(s) to the NW. Performance metrics (e.g., Key Performance Indicators (KPIs)) may be configured by the NW. UE 106 determines if performance has degraded at 1706. As illustrated in Figure 17, if model monitoring indicates good performance, i.e., performance has not degraded, the UE 1702 continues to employ the model for the task and monitor performance.

[0164] As illustrated in Figure 17, if model monitoring indicates performance has degraded, the UE 106 performs model validation across available models at 1708. That is, for example, the UE 106 may have a database of AI / ML-based models. For example, UE 106 may have a database of models for CSI prediction. In one example, the UE 106 may store the samples (i.e., input to the model) that indicated performance degradation. These samples may be referred to as a test domain, ^^^^^. UE 106 may perform model validation across its available models, or a subset thereof, with source data ^^^^^. UE 106 determines if a model is found, such that the performance improves at 1710. If a model is found suchthat the performance improves, as illustrated in Figure 17, UE 106 switches to that model at 1712 and employs the model at 1702. That is, the validated model becomes the pre- trained model.

[0165] As illustrated in Figure 17, if there are no models available that improve performance, UE fine-tunes the employed model at 1714. UE 106 may collect a limited number of training samples from a new environment, which may be referred to as target domain ^^,^^^^^. UE 106 performs transfer learning from current task ^^. In one example, UE 106 may perform transfer learning by fine-tuning the last layers, while keeping the parameters from other layers and the statistics from normalization layers unchanged. Fine- tuning may include for example, adjusting the values of kernels in a convolution layer. In a case where, a neural network includes dozens, hundreds, or thousands of layers, UE 106 may fine-tune a small percentage (e.g., less than 10%) of the layers, which are the final layers. For example, referring to Figure 15, if N is equal to 10, UE 106 may fine-tune layers 9 and 10. Thus, UE 106 adapts the pre-trained model for a nominal task, ^^to a model for a target task, ^^. It should be noted that by fine-tuning the final layers, the adaptation to target task, ^^, may be performed very efficiently.

[0166] After fine-tuning, UE 106 performs of a model validation for a target task, ^^at 1716. For example, UE 106 may perform a validation test by validating model ^^with ^^^^^. UE 106 determines if performance is improved at 1716. As illustrated in Figure 17, if performance is improved, model ^^is added to the available models at 1720 and UE 106 switches to that model at 1712 and employs the model at 1702. That is, the fine-tuned becomes the employed pre-trained model. As illustrated in Figure 17, if performance does not improve, UE 106 may fall back to a legacy mode at 1722. That is, for example, UE may fall back to a legacy mode and perform model validation according to the conditions configured by NW. Further, UE 106 may notify the NW about falling back to a legacy mode. In some case, NW may update a UE 106 database with another model. Figure 18: Meta Learning with Model Validation

[0167] As described above, in transfer learning a pre-trained model trained with a large dataset is deployed and subsequently fine-tuned. In practice, obtaining a large number of samples from a single wireless channel environment to pre-train a model can be challenging. Conversely, it is relatively easier to acquire a diverse dataset composed ofsamples from multiple wireless channel environments, where each environment contributes only a small number of samples. In meta learning, the network structure and parameter updating model may be fixed and the system focuses on learning an initialization that allows for rapid adaptation to new tasks with a small number of samples. That is, unlike transfer learning where a large number of samples may be used to train a DNN to obtain a pre-trained model, meta-learning learns a model initialization through multiple tasks, where each task provides only a small number of samples. After the Deep Learning (DL) model acquires initialization during the meta-training phase, it can achieve rapid convergence when retrained on the target task dataset. This dataset may be augmented from a small amount of collected data capturing the statistical features of the wireless channels. The goal of few-shot meta-learning is to train a model that can quickly adapt to a new task using only a few datapoints and training iterations. To accomplish this, the model or learner may be trained during a meta-learning phase on a set of tasks, such that the trained model can quickly adapt to new tasks using only a small number of examples or trials.

[0168] When training a model to solve one specific task, for example, to a particular CSI matrix compression, a loss may be defined as: ^ ^^^ ℒ^(^). The optimal ^ may befound by walking through the gradient: ^ ← ^ − ^^ℒ^(^). That is, a loss function ℒ^(^)for a task ^ may be minimize to optimize a model. Applying this to type of regression or a few-shot task (i.e., with a very small dataset), is known to perform poorly on, for example, neural networks, since there is simply too little data for too many parameters, leading to overfitting. That is, once a model has been optimized for a particular task, there are too few samples to allow the model to perform well for a different task. For example, if a model is optimized to compress a particular type of CSI matrix, the model may perform poorly when compressing a different type of CSI matrix. The key idea of meta learning is to mitigate this problem by learning not only from the data regarding exactly the particular task, but also from data of similar tasks. For example, in addition to training a model to optimally compression a particular type of CSI matrix, a model may be trained to compress multiple types of CSI matrices.

[0169] According the techniques herein, in one example, to incorporate meta learning, an assumption may be made that task ^ comes from some distribution of tasks ^(^) and data may be sampled from this distribution. It should be noted that rather than simply using the data from other tasks to find parameters that are optimal for all tasks, the option of fine-tuning a model (i.e., taking additional optimizer steps on data from the new task) may be applied. After applying meta learning, the pre-fine-tune-version of the model may converge for each new task. Thus, an optimization objective may be expressed as: ^ ^^^ ^^[ℒ^(^(!))] Where, ^ is an optimization algorithm that maps ! to a new parameter vector^(!).

[0170] Thus, the objective is learned with respect to a variety of tasks. This, together with the fact that it can further be regarded as the initialization of the optimizer, allows ! to be interpreted as above task-level and thus, acquires the status of a meta-parameter. Optimizing such a meta-parameter corresponds to meta-learning. That is, the goal of meta learning may be to optimize !.

[0171] In one example, ! may be optimized based on the following: (1) Initialize parameter vector !: (2) For iterations ^ =1, 2,⋯,^, where ^ is a number of tasks, do: (i) Sample a number of tasks ^^for ^(^). Collect samples for training and testing task ^^;(ii) For each task ^^, obtain ^^ = ^'(!) by minimizing ℒ^',^^^^^(!) ona few training samples. That is, perform the updates ^^ ← ! −^^ℒ^',^^^^^(!);(iii)^^, update ! by gradient descent such that it minimizes ℒ(!) = ∑ℒ^',^^^^(^^). Updating ! requires evaluating ^ the gradient oflosses on a set of test data. The gradient of the overall loss is obtained as follows: ^^ℒ(!) =∑ ^^ℒ^',^^^^(^^) ^ (3) Update ! via gradient descent, using a new learning rate : ! ← ! − ^^ℒ(!)(4) Go to (2) with updated !.

[0172] After an number of iterations of updating ! to an optimized initialization parameter !*, a pre-trained model which can be further fine-tuned for each individual task ^^is generated.

[0173] It should be noted that for a real deployment the pre-trained model may be fine- tuned with only few samples from task ^^, after the optimum initialization parameter !* has been obtained. In one example, fine tuning the model for task ^ includes performing the^updates ^ ← !∗ − ^ ℒ (!∗^^ ^). should be noted that when ! is not optimized many,^^^^^'to new tasks, when ! is optimized fewer samplesare convergence new tasks.

[0174] Figure 18 illustrates a block diagram of an example of a method 1800 to perform training or performance monitoring of an AI / ML-based model, according to some embodiments. The method shown in Figure 18 may be utilized as part of AI / ML-based model Life Cycle Management (LCM). The method shown in Figure 18 may be used in conjunction with any of the systems, methods, or devices shown in the Figures, among other devices. That is, Figure 18 provides an example of a procedure for meta learning and model validation which may be performed, for example, by a UE. In various embodiments, some of the method elements shown may be performed concurrently, in a different order than shown, or may be omitted. Additional method elements may also be performed as desired. As shown, this method may operate as follows.

[0175] A UE 106, or other device, selects a pre-trained model for a task, ^ at 1802.^The model may be pre-trained according to a meta learning procedure. For example, the model may be pre-trained by optimizing !, as described above. UE fine-tunes the pre- trained model and employs the fine-tuned model at 1804. That is, the UE 106 may collect a limited number of training samples from the environment and perform fine-tuning ∗according to: ^′ ← ! − ^ ℒ (!). The UE 106 performs model monitoring at 1806.^ ^ ^, ,^^^^^'For example, UE 106 the performance metric(s) and / or UE 106 may reportperformance metric(s) to the NW. Performance metrics (e.g., Key Performance Indicators (KPIs)) may be configured by the NW. UE 106 determines if performance has degraded at 1808. As illustrated in Figure 18, if model monitoring indicates good performance, i.e., perform has not degraded, the UE 1806 continues to employ the model for the task and monitor performance.

[0176] As illustrated in Figure 18, if model monitoring indicates performance has degraded, the UE 106 performs model validation across available models at 1810. That is, for example, the UE 106 may have a database of AI / ML-based models. For example, UE 106 may have a database of models for CSI prediction. In one example, the UE 106 may store the samples (i.e., input to the model) that indicated performance degradation. These samples may be referred to as a test domain, ^^^^^. UE 106 may perform model validation across its available models, or a subset thereof, with source data ^^^^^. UE 106 determines if a model is found, such that the performance improves at 1812. If a model is found such that the performance improves, as illustrated in Figure 18, UE 106 switches to that model at 1814 and selects the model at 1802. That is, the validated model becomes the pre-trained model.

[0177] As illustrated in Figure 18, if there no models available that improve performance, UE fine-tunes the employed model at 1816. UE 106 may collect a limited number of training samples from a new environment, which may be referred to as target domain ^^,^^^^^. In one example, UE 106 may perform fine-tuning by fine-tuning the last layers of a NN, while keeping the parameters from other layers and the statistics from normalization layers unchanged. Fine-tuning may include for example, adjusting the values of kernels in a convolution layer. In a case where, a neural network includes dozens, hundreds, or thousands of layers, UE 106 may fine a small percentage (e.g., less than 10%) of the layers, which are the final layers. In this manner, the adaptation to target task ^- canbe performed very efficiently. ^- ← ! − ^^ℒ^.,^^^^^(!) (Output is model for task j, ^-).

[0178] After fine-tuning,validation for a target task, ^- at 1818. For example, UE 106 may perform a validation test by validating model ^- with ^^^^^. UE 106 determines if performance is improved at 1820. As illustrated in Figure 18, if performance is improved, model ^- is added to the available models at 1822 and UE 106 switches to that model at 1814 and selects the model at 1802. That is, the fine-tuned becomes the selected pre-trained model. As illustrated in Figure 18, if performance does not improve, UE 106 may fall back to a legacy mode at 1824. That is, for example, UE may fall back to a legacy mode and perform model validation according to the conditions configured by NW. Further, UE 106 may notify the NW about falling back to a legacy mode. In some cases, NW may update a UE 106 database with another model. Further, although not explicitly illustrated in Figure 18, UE 106 may periodically update theparameter vector ! by collecting some test data ^^ℒ(!) = ∑^^ℒ^',^^^^(^^) and reinitiate ^ process1800 at 1802.

[0179] Embodiments described herein provide apparatus, systems, and methods for employing a pre-trained model for a task, determining performance of the pre-trained model has degraded based on a set of samples, and fine-tuning the pretrained model. Figure 19: Life Cycle Management (LCM) of an AI / ML-based model deployed in a wireless communications network

[0180] Figure 19 illustrates a block diagram of an example of a method 1900 for Life Cycle Management (LCM) of an AI / ML-based model deployed in a wireless communications network, according to some embodiments. The method shown in Figure 19 may be used in conjunction with any of the systems, methods, or devices shown in the Figures, among other devices. In various embodiments, some of the method elements shown may be performed concurrently, in a different order than shown, or may be omitted. Additional method elements may also be performed as desired. As shown, this method may operate as follows.

[0181] At 1910, a device in a wireless communications network, for example, a user equipment device (UE), such as UE 106, employs a pre-trained model for a task. For example, a UE may employ an AI / ML-based model for CSI compression and / or prediction. The AI / ML-based model may be pre-training according to a transfer learning or a meta learning process described above.

[0182] At 1920, the device may determine model performance has degraded. For example, a UE may determine model performance has degraded based on KPIs, for example, as described above.

[0183] At 1930, the device may determine that a model with improved performance is not available. For example, a device may have a database of models. In one example, the device may perform model validation across its available models with test data.

[0184] At 1940, the device may fine-tune the pre-trained model. In one example, the device may perform fine-tuning by fine-tuning the last layers of a neural network, while keeping the parameters from other layers and the statistics from normalization layers unchanged. Fine-tuning may include for example, adjusting the values of kernels in aconvolution layer.

[0185] In some examples, employing a pre-trained model for a task includes employing a pre-trained model that has been trained on source domain data for the task.

[0186] In some examples, employing a pre-trained model for a task includes employing a pre-trained model that has been trained by optimizing a parameter for a distribution of tasks.

[0187] In some examples, optimizing a parameter for a distribution of tasks includes minimizing a loss function for a task and updating the parameter based on the minimized a loss function for the task.

[0188] In some examples, fine-tuning the pretrained model includes fine-tuning the pretrained model a number of training samples from an environment.

[0189] In some examples, fine-tuning the pretrained model includes fine-tuning a subset of layers of a neural network.

[0190] In some examples, the subset of layers are final layers of the neural network.

[0191] In some examples, fine-tuning the pretrained model further includes keeping parameters from other layers and statistics from normalization layers unchanged.

[0192] In some examples, the fine-tuned model is validated using the set of samples used to determine performance has degraded.

[0193] In some examples, the fine-tuned model is stored to a database.

[0194] In some examples, a task is Channel State Information (CSI) compression.

[0195] In some examples, a task is Channel State Information (CSI) prediction.

[0196] It is well understood that the use of personally identifiable information should follow privacy policies and practices that are generally recognized as meeting or exceeding industry or governmental requirements for maintaining the privacy of users. In particular, personally identifiable information data should be managed and handled so as to minimize risks of unintentional or unauthorized access or use, and the nature of authorized use should be clearly indicated to users.

[0197] Embodiments of the present disclosure may be realized in any of various forms. For example, some embodiments may be realized as a computer-implementedmethod, a computer- readable memory medium, or a computer system. Other embodiments may be realized using one or more custom-designed hardware devices such as ASICs. Still other embodiments may be realized using one or more programmable hardware elements such as FPGAs.

[0198] In some embodiments, a non-transitory computer-readable memory medium may be configured so that it stores program instructions and / or data, where the program instructions, if executed by a computer system, cause the computer system to perform a method, e.g., any of the method embodiments described herein, or, any combination of the method embodiments described herein, or, any subset of any of the method embodiments described herein, or, any combination of such subsets.

[0199] In some embodiments, a device (e.g., a UE 106) may be configured to include a processor (or a set of processors) including one or more baseband processors and one or more application processors and a memory medium, where the memory medium stores program instructions, where the processor is configured to read and execute the program instructions from the memory medium, where the program instructions are executable to implement any of the various method embodiments described herein (or, any combination of the method embodiments described herein, or, any subset of any of the method embodiments described herein, or, any combination of such subsets). The device may be realized in any of various forms.

[0200] Any of the methods described herein for operating a user equipment (UE) may be the basis of a corresponding method for operating a base station, by interpreting each message / signal X received by the UE in the downlink as message / signal X transmitted by the base station, and each message / signal Y transmitted in the uplink by the UE as a message / signal Y received by the base station.

[0201] Although the embodiments above have been described in considerable detail, numerous variations and modifications will become apparent to those skilled in the art once the above disclosure is fully appreciated. It is intended that the following claims be interpreted to embrace all such variations and modifications.

Claims

CLAIMS What is claimed is:

1. An apparatus of a user equipment (UE) configured to perform training or performance monitoring for an artificial intelligence (AI) / machine learning (ML) model, the apparatus comprising: one or more processors configured to: employ a pre-trained model for a task; determine performance of the pre-trained model has degraded based on a set of samples; and fine-tune the pretrained model; and a memory coupled to the one or more processors and configured to store samples.

2. The apparatus of claim 1, wherein employing a pre-trained model for a task includes employing a pre-trained model that has been trained on source domain data for the task.

3. The apparatus of claim 1, wherein employing a pre-trained model for a task includes employing a pre-trained model that has been trained by optimizing a parameter for a distribution of tasks.

4. The apparatus of claim 3, wherein optimizing a parameter for a distribution of tasks includes minimizing a loss function for a task and updating the parameter based on the minimized a loss function for the task.

5. The apparatus of any of claims 1-4, wherein fine-tuning the pretrained model includes fine-tuning the pretrained model a number of training samples from an environment.

6. The apparatus of any of claims 1-5, wherein fine-tuning the pretrained model includes fine-tuning a subset of layers of a neural network.

7. The apparatus of claim 6, wherein the subset of layers are final layers of the neural network.

8. The apparatus of claim 6, wherein fine-tuning the pretrained model further includes keeping parameters from other layers and statistics from normalization layers unchanged.

9. The apparatus of any of claims 1-8, wherein the one or more processors are further configured to validate the fine-tuned model using the set of samples used to determine performance has degraded.

10. The apparatus of any of claims 1-9, wherein the one or more processors are further configured to store the fine-tuned model to a database.

11. An apparatus of a user equipment (UE) configured to perform training or performance monitoring for an artificial intelligence (AI) / machine learning (ML) model, the apparatus comprising: one or more processors configured to: employ a pre-trained model for a task that has been trained on source domain data for the task; determine performance of the pre-trained model has degraded based on a set of samples; and fine-tune the pretrained model; and a memory coupled to the one or more processors and configured to store samples.

12. An apparatus of a user equipment (UE) configured to perform training or performance monitoring for an artificial intelligence (AI) / machine learning (ML) model, the apparatus comprising: one or more processors configured to: employ a pre-trained model for a nominal task that has been trained on a source domain; perform performance modeling on the pre-trained model;determine performance of the pre-trained model has degraded based on a test domain; perform model validation across available models using the test domain; determine none of the available models improve performance; fine-tune a final set of layers of a neural network of the pre-trained model; and validating the fine-tuned pre-trained model to a target task; and a memory coupled to the one or more processors and configured to store the fine-tuned pre-trained model.

13. The apparatus of any of claims 11-12, wherein fine-tuning the pretrained model further includes keeping parameters from other layers and statistics from normalization layers unchanged.

14. The apparatus of any of claims 12-13, wherein validating the fine-tuned pre- trained model to a target task includes validating the fine-tuned pre-trained model using the test domain.

15. The apparatus of any of claims 12-13, wherein the nominal task is Channel State Information (CSI) compression.

16. The apparatus of any of claims 12-13, wherein the nominal task is Channel State Information (CSI) prediction.

17. A method to perform training or performance monitoring for an artificial intelligence (AI) / machine learning (ML) model, the method comprising: employing a pre-trained model for a task; determining performance of the pre-trained model has degraded based on a set of samples; and fine-tuning the pretrained model.

18. The method of claim 17, wherein employing a pre-trained model for a taskincludes employing a pre-trained model that has been trained on source domain data for the task.

19. The method of claim 17, wherein employing a pre-trained model for a task includes employing a pre-trained model that has been trained by optimizing a parameter for a distribution of tasks.

20. The method of claim 19, wherein optimizing a parameter for a distribution of tasks includes minimizing a loss function for a task and updating the parameter based on the minimized a loss function for the task.

21. The method of claim 19, wherein optimizing a parameter for a distribution of tasks comprises minimizing a loss function across all tasks and updating the parameter based on the minimized a loss function evaluated on testing datasets for all tasks.

22. The method of any of claim 17-21, wherein fine-tuning the pretrained model includes fine-tuning the pretrained model using a number of training samples from an environment.

23. The method of any of claim 17-22, wherein fine-tuning the pretrained model includes fine-tuning a subset of layers of a neural network.

24. The method of claim 23, wherein the subset of layers are final layers of the neural network.

25. The method of claim 23, wherein fine-tuning the pretrained model further includes keeping parameters from other layers and statistics from normalization layers unchanged.

26. The method of any of claim 17-25, further comprising validating the fine-tuned model using the set of samples used to determine performance has degraded.

27. The method of any of claim 17-26, further comprising storing the fine-tuned model to a database.

28. A method to perform training or performance monitoring for an artificial intelligence (AI) / machine learning (ML) model, the method comprising: employing a pre-trained model for a nominal task that has been trained on a source domain; performing performance modeling on the pre-trained model; determining performance of the pre-trained model has degraded based on a test domain; performing model validation across available models using the test domain; determining none of the available models improve performance; fine-tuning a final set of layers of a neural network of the pre-trained model; validating the fine-tuned pre-trained model to a target task; and storing the fine-tuned pre-trained model.

29. The method of claim 28, wherein fine-tuning the pretrained model further includes keeping parameters from other layers and statistics from normalization layers unchanged.

30. The method of any of claims 28-29, wherein validating the fine-tuned pre-trained model to a target task includes validating the fine-tuned pre-trained model using the test domain.

31. The method of any of claims 28-30, wherein the nominal task is Channel State Information (CSI) compression.

32. The method of any of claims 28-30, wherein the nominal task is Channel State Information (CSI) prediction.

33. A baseband processor configured to cause a user equipment (UE) to perform any of the methods of claims 17-32.

34. A baseband processor configured to cause a base station to perform one or more of the methods of claims 17-32.

35. An apparatus configured to cause a user equipment (UE), having one or more processors coupled to a memory, to perform any of the methods of claims 17-32.

36. An apparatus configured to cause base station, having one or more processors coupled to a memory, to perform any of the methods of claims 17-32.

37. A computer program product, comprising computer instructions which, when executed by one or more processors, perform any of the operations described herein.

Images