Capability reporting for multi-model artificial intelligence / machine learning user device features

Abstract

The present invention provides an apparatus, method, computer program, computer program product, and computer-readable medium for capability reporting for a multi-model AI / ML UE feature. The method includes generating user equipment capability information, the generating user equipment capability information including identifying at least one machine learning model available at the user equipment for a given scenario, assigning a unique identification to each of the at least one machine learning model, and associating the at least one machine learning model having the unique identification with at least one of a parameter list, a data set, and a registration ID, and reporting the generated user equipment capability information to a network entity.

Description

[Technical Field]

[0001] Various example embodiments relate to apparatuses, methods, systems, computer programs, computer program products, and computer-readable media for capability reporting for multi-model AI / ML UE features.

[0002] (abbreviation) 5G - Fifth Generation gNB - 5G / NR base station NR - New Radio RAN - Radio Access Network AI - Artificial Intelligence ML - Machine Learning UE - User Equipment UL - Uplink DL - Downlink DCI - Downlink Control Information MAC CE - Medium Access Control Control Element MIMO - Multiple Input and Multiple Output NN - Neural Network CSI - Channel State Information SSB - Synchronous Signal Block RS - Reference signal TRP - Transmission Point Tx - Transmitter Rx - Receiver RB - Residual Block CNN - Convolutional Neural Network [Background technology]

[0003] Certain aspects of the present invention relate to a Study Item (SI) on Artificial Intelligence (AI) / Machine Learning (ML) for the Rel-18 New Radio (NR) Air Interface (see 3GPP RP-213599).

[0004] The SI aims to explore the benefits of augmenting the air interface with features that enable support for AI / ML-based algorithms for enhanced performance and / or reduced complexity / overhead. The goal of these considerations is to build a foundation for future air interface use cases that leverage AI / ML techniques. The initial set of use cases to be covered includes CSI feedback enhancements (e.g., overhead reduction, improved accuracy, prediction), beam management (e.g., beam prediction in time and / or spatial domain versus overhead and latency reduction, improved beam selection accuracy), and positioning accuracy enhancement. For those use cases, the benefits should be evaluated (utilizing the developed methodology and defined KPIs), and the potential impact on the specification should be evaluated, including PHY layer aspects and protocol aspects.

[0005] One of the main expected consequences of such considerations is that the AI / ML approaches for the selected sub-use cases need to be sufficiently diverse to support the different requirements at the gNB-UE cooperation level.

[0006] It should be noted that additional use cases may be addressed during the "AI / ML for Air Interface" Work Item (WI) phase. Starting with Release 18, companies are likely to propose a wide variety of use cases and applications for ML in gNBs and UEs. The goal is to explore the benefits of augmenting the air interface with features that enable improved support for AI / ML-based algorithms for enhanced performance and / or reduced complexity / overhead. Enhanced performance here depends on the considered use case and could be, for example, improved throughput, robustness, accuracy, or reliability. The goal is for enough use cases to be considered to enable the identification of a common AI / ML framework, including functional requirements for AI / ML architectures, that can be used in subsequent projects. Research should also identify areas where AI / ML can improve the performance of air interface functions. The impact of specifications will be evaluated to improve the overall understanding of what is needed to enable AI / ML techniques for the air interface.

[0007] Generalizing AI / ML models to cover different scenarios / configurations is seen as a significant challenge that RAN1 should investigate, and there are different approaches to solving it. It is highly unlikely that a single ML model will ultimately be necessary and sufficient for each and every deployment scenario. For example, developing a "universal" ML model for beam prediction in the spatial domain that performs well across different scenarios / configurations / setups / parameters can be challenging, as its performance must be validated against data for each and every deployment scenario. Also, having one universal model can result in higher training / inference complexity, and performance issues may arise if the model cannot perform well in all scenarios. Summary of the Invention

[0008] Various example embodiments seek to address at least some of the issues and / or problems and shortcomings noted above.

[0009] It is an object of various example embodiments to provide an apparatus, method, system, computer program, computer program product, and computer readable medium for capability reporting for multi-model AI / ML UE features.

[0010] According to one aspect of various example embodiments, there is provided a method for use in a user equipment, the method comprising: generating user equipment capability information, Identifying at least one machine learning model available on a user device for a given scenario; assigning a unique identification to each of the at least one machine learning model; Associating at least one machine learning model having a unique identification with at least one of a parameter list, a dataset, and a registration ID; generating user equipment capability information, including: reporting the generated user equipment capability information to a network entity; Includes.

[0011] According to another aspect of various example embodiments, there is provided an apparatus for use in a user equipment, the apparatus comprising: A means for generating user equipment capability information, comprising: means for identifying at least one machine learning model available on the user equipment for a given scenario; means for assigning a unique identification to each of the at least one machine learning model; means for associating at least one machine learning model having a unique identification with at least one of a parameter list, a dataset, and a registration ID; means for generating user equipment capability information, means for reporting the generated user equipment capability information to a network entity; Equipped with.

[0012] According to another aspect of the present invention there is provided a computer program product comprising code means adapted to produce the steps of any of the methods as described above when loaded into the memory of a computer.

[0013] According to a further aspect of the present invention there is provided a computer program product as defined above, the computer program product comprising a computer readable medium having software code portions stored thereon.

[0014] According to a further aspect of the present invention there is provided a computer program product as defined above, wherein the program is directly loadable into an internal memory of a processing device.

[0015] According to one aspect of various example embodiments, there is provided a computer-readable medium having stored thereon a computer program as detailed above.

[0016] According to an exemplary aspect, a computer program product is provided comprising computer-executable computer program code configured to, when the program is run on a computer (e.g., a computer of an apparatus according to any one of the aforementioned exemplary apparatus-related aspects of the present disclosure), cause the computer to perform a method according to any one of the aforementioned exemplary method-related aspects of the present disclosure.

[0017] Such a computer program product may comprise (or embody) a (tangible) computer-readable (storage) medium or the like, on which computer-executable computer program code is stored and / or from which the program may be directly loadable into the internal memory of the computer or its processor.

[0018] Further aspects and features of the present invention are set out in the dependent claims.

[0019] These and other objects, features, details and advantages will become more fully apparent from the following detailed description of various aspects / embodiments, which is to be read in connection with the accompanying drawings. [Brief explanation of the drawings]

[0020] [Figure 1] FIG. 1 is a sequence diagram illustrating an example of a signaling procedure according to certain aspects of the present invention. [Figure 2] 1 is a flowchart illustrating an example method according to certain aspects of the present invention. [Figure 3] FIG. 1 is a block diagram illustrating another example of an apparatus according to certain aspects of the present invention. [Figure 4] FIG. 1 is a block diagram illustrating another example of an apparatus according to certain aspects of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0021] The present disclosure is described herein with reference to certain non-limiting examples and what are presently considered to be the contemplated embodiments. Those skilled in the art will understand that the present disclosure is in no way limited to these examples, but may be more broadly applicable.

[0022] It should be noted that the following description of the present disclosure and its embodiments primarily refers to specifications used as non-limiting examples for certain exemplary network configurations and deployments. That is, the present disclosure and its embodiments are described primarily in relation to 3GPP specifications, used as non-limiting examples for certain exemplary network configurations and deployments. Accordingly, the description of the exemplary embodiments presented herein will specifically refer to terminology directly related thereto. Such terminology is used only in the context of the presented non-limiting examples and, of course, does not limit the present disclosure in any way. Rather, any other communications or communication-related system deployments, etc., may also be utilized as long as they conform to the features described herein.

[0023] Hereinafter, various embodiments and implementations of the present disclosure, as well as aspects or embodiments thereof, will be described with some variations and / or alternatives. It should be noted that, generally, all of the variations and / or alternatives described may be presented alone or in any conceivable combination (including combinations of individual features of the various variations and / or alternatives), subject to certain needs and constraints.

[0024] According to some aspects of the present invention, it is proposed to use multiple ML models for the same scenario (e.g., CSI compression, beam prediction, positioning, etc.), and the models can be changed depending on the situation / scenario / configuration. A scenario refers to at least one of a feature / sub-use case / use case. The issues mentioned above can be avoided by using multiple ML models for the same scenario. Therefore, for some scenarios, it must be specified how multiple ML models can be defined in the NR framework.

[0025] If for a given scenario more than one model can be used / selected for model inference at the UE side, the UE capability reporting (UE feature) will take into account:

[0026] The UE indicates the number of ML models (e.g., X=4, 6, 8) that the UE can support for the same functionality (e.g., CSI compression with a two-sided model, beam prediction in the spatial domain with a one-sided model, positioning estimated in NLOS scenarios with a one-sided model). The UE displays the ML model ID (a number that uniquely identifies different models that support the same functionality, e.g., for X=4, ID=1, 2, 3, 4) and displays the association between the ML model ID and the parameter list and / or dataset and / or registration ID; The parameter list defines radio conditions (deployment type, applicable scenarios, gNB / UE antenna configuration, scattering parameters, etc.), and / or ML-specific details (reportable quantities, required / associated measurement configurations, required / associated auxiliary information, ML model inputs / outputs and dimensionality), and / or limitations / conditions (inference delay, required warm-up time, fine-tuning requirements). o The Dataset refers to a version of the Dataset, which is accessible to both the UE and the network through other means (operator controlled server, proprietary cloud) and the network understands the aspects related to the model based on the associated Dataset of the Model ID. ○ A Registration ID refers to a unique version of an ML model with a unique identifier. Here, the trained model can reside in an operator-controlled or proprietary server, and the network may not have full access to the model. Nevertheless, details / parameters associated with the model can be made known to the network by referencing the Registration ID.

[0027] The UE indicates whether the model can be switched from one to another. If the switch is dynamic, the UE may also indicate any associated delay considerations for switching from one model to another. o In one variation, model switching between some model combinations may not be supported, while switching between some other combinations may be supported. The UE displays whether the model can be disabled and the possibility of fallback to a parametric model. o If fallback is supported, any associated delay considerations for switching and the parameter list associated with the parametric model (e.g., for CSI compression, Type I or Type II codebook-based reporting can be considered as a parametric model) may also be displayed by the UE. The UE indicates whether more than one model may be activated at a given time. ○ If yes, the UE further reports the number of models supported in parallel, which model IDs can be supported in parallel, and any associated considerations / restrictions for applying parallel operation of ML models. o In one example, for inter-cell beam prediction in the spatial domain, the beam prediction can use multiple models in parallel, where each model can be a cell-specific model. For each of these cell-specific models, the UE measures the beam corresponding to that cell and uses the measurements at the input of the model, and the model output results in the best predicted beam for the same cell.

[0028] Based on the received UE capability information, the network configures ML model parameters, decides / supports model switching, or considers activation of more than one model at a given time for a given feature-related support for the UE.

[0029] In the following, a specific example of the above-mentioned aspects is described by considering the case of beam management (spatial domain beam prediction) as an example.

[0030] Specifically, according to a specific example: The UE indicates the number of ML models (e.g., X=4, 6, 8) that the UE can support for beam prediction in the spatial domain. · The UE displays the ML model ID (ID=1, 2, 3, 4 for X=4) and displays the association between the ML model ID and the parameter list. ○ Model ID1 - Frequency Range 2 (FR2), Subcarrier Spacing (SCS) 120KHz, High Density Urban Scenario, Base Station (BS) Antenna Configuration (One Panel: (M, N, P, Mg, Ng) = (4, 8, 2, 1, 1), (dV, dH) = (0.5, 0.5)λ), Set A Dimensions (64 beams, Layout M1), Set B Dimensions (4 or 8 or 16 beams, Layout N1), Auxiliary Information (NULL), Reportable Quantities (Beam ID, L1-RSRP), etc. ○ Model ID2 - FR2, SCS 60KHz, indoor hotspot scenario, BS antenna configuration (one panel: (M, N, P, Mg, Ng) = (2, 2, 2, 1, 1), (dV, dH) = (0.5, 0.5)λ), set A dimensions (32 beams, layout M2), set B dimensions (8 or 16 beams, layout N2), auxiliary information (beam angles), reportable quantities (beam IDs), etc. The UE indicates that the model can be switched from one to another. ○Model switching is dynamic: multiple combinations (combination 1, 2, ...) and associated parameters for switching can be provided. The UE indicates that the model can be disabled and can fall back to a parametric model. o The associated delay considerations for the switchover and the parameter list associated with the parametric model may also be displayed by the UE. The UE indicates that only one model can be active at a given time.

[0031] Based on the received UE capability information, the network configures ML model related parameters, decides / supports model switching, or considers activation of more than one model at a given time for a given supported feature for the UE.

[0032] The above multiple ML models for the same scenario (e.g., use case feature group) can be displayed in the form of a table below.

[0033] [Table 1] Table 1: Multiple Model ID Coexistence and Switching

[0034] Table 1 discusses how a UE reports coexistence and switching when using multiple ML models for the same scenario. In the above example, the UE supports four such ML models with ML model IDs 1 through 4 for different parameter lists (optionally providing a registration ID (global or vendor-specific ID) and a dataset ID used for training). Combinations refer to potential combinations of ML models that are allowed to operate together. For example, combination 1 indicates that ML models 1 through 3 can operate together when configured by the network, e.g., in a particular band combination (in combination 1, model ID 4 cannot be configured for the UE). Furthermore, Table 2 below describes how combinations should be interpreted by the network for configuration purposes.

[0035] [Table 2] Table 2: Description of each combination

[0036] FIG. 1 is a sequence diagram illustrating an example of a signaling procedure according to some embodiments of the present invention.

[0037] In step S11 of Figure 1, the network initiates the fetching of combinations of ML model IDs for the same scenario (e.g., use case), i.e., the network requests the UE to provide information about the ML models that are available at the UE and can be combined for a particular scenario.

[0038] The network therefore sends a corresponding request to the user equipment in step S12, which indicates, among other things, the available ML models and scenarios for which combinations should be provided. Optionally, this may require, for example, a registration ID to be provided by the user equipment.

[0039] In step S13, the UE identifies one or more machine learning models available to the user equipment for the given scenario included in the request received in step S12. The user equipment then assigns a unique identification to each of the one or more machine learning models. The user equipment further associates the one or more machine learning models with the unique identification with at least one of a parameter list, a dataset, and optionally a registration ID. That is, the user equipment generates user equipment capability information and compiles a list of ML model ID combinations.

[0040] Then, in step S14, the user equipment reports the generated user equipment capability information, including the available ML models and combinations, to the network.

[0041] The network stores the available ML models and combinations, which are represented by ML model IDs and constitute combinations of ML model IDs.

[0042] This configuration is then sent to the user equipment in step S16.

[0043] In step S17, the user equipment confirms receipt of the configured ML model ID combination and starts operating according to the configured ML model ID combination received from the network.

[0044] A more general description of an exemplary version of the present invention is provided below with respect to FIGS.

[0045] FIG. 2 is a flow chart illustrating an example of a method according to some example versions of the present invention.

[0046] According to an example version of the invention, a method can be implemented in, or can be part of, a user equipment, etc. The method includes, in step S21, generating user equipment capability information. Generating the user equipment capability information in step S21 includes identifying at least one machine learning model available on the user equipment for a given scenario, assigning a unique identification to each of the at least one machine learning model, and associating the at least one machine learning model with the unique identification with at least one of a parameter list, a dataset, and a registration ID.

[0047] Further, the method includes, in step S22, reporting the generated user equipment capability information to a network entity. The network entity may be a base station such as a gNB.

[0048] According to some example versions of the invention, the method further includes receiving a request from the network entity indicating a predetermined scenario for which a machine learning model is to be identified.

[0049] According to some example versions of the invention, the method further includes receiving, from the network entity, a configuration for a machine learning model to be used for the given scenario based on the reported user equipment capability information, and operating in accordance with the received configuration.

[0050] According to some example versions of the present invention, the user equipment capability information includes an indication of whether an identified machine learning model can be switched from one to another when applied to a given scenario.

[0051] According to some example versions of the present invention, the user equipment capability information includes an indication of whether two or more of the identified machine learning models may be activated at a given time when applied to a given scenario.

[0052] According to some example versions of the present invention, the user equipment capability information includes an indication of whether an identified machine learning model can be disabled in order to switch to a parametric model when applied to a given scenario.

[0053] According to some example versions of the invention, the parameter list defines at least one of: radio conditions inducing at least one of deployment type, applicable scenarios, base station / user equipment antenna configurations, and scattering parameters; machine learning specific details including at least one reportable quantity, required measurement configuration, required auxiliary information, inputs / outputs and dimensions of the machine learning model; and constraints / conditions including at least one inference delay, required warm-up time, and fine-tuning requirements.

[0054] According to some example versions of the present invention, a dataset refers to a version of a dataset that is accessible to both user equipment and a network through an operator-controlled server and / or other means, including a proprietary cloud, and the dataset specifies aspects related to the model.

[0055] According to some example versions of the present invention, a registration ID refers to a unique version of a machine learning model with a unique identifier.

[0056] FIG. 3 is a block diagram illustrating another example of an apparatus according to some example versions of the present invention.

[0057] 3 shows a block circuit diagram illustrating the configuration of an apparatus 30, which is configured to implement the various aspects of the present invention described above. It should be noted that the apparatus 30 shown in FIG. 3 may include several additional elements or functions in addition to those described herein below, which are not essential for understanding the present invention and are therefore omitted herein for the sake of brevity. Furthermore, the apparatus may also be another device or the like having similar functionality, such as a chipset, chip, module, etc., which may be part of the apparatus or attached to the apparatus as a separate element.

[0058] The device 30 may comprise a processing function or processor 31, such as a CPU or the like, that executes instructions provided by a program or the like. The processor 31 may comprise one or more processing sections dedicated to specific processes as described below, or the processes may run within a single processor. The sections for performing such specific processes may also be provided in separate elements, or in one or more processors or processing sections, e.g., in one physical processor such as a CPU, or in several physical entities. Reference symbol 32 denotes a transceiver or input / output (I / O) unit (interface) connected to the processor 31. The I / O unit 32 may be used to communicate with one or more other network elements, entities, terminals, or the like. The I / O unit 32 may be a combined unit with communication equipment toward several network elements, or may comprise a distributed structure with multiple different interfaces for different network elements. The device 30 further comprises at least one memory 33, e.g., for storing data and programs to be executed by the processor 31 and / or usable as a working memory for the processor 31.

[0059] The processor 31 is configured to perform the processing associated with the above-described aspects.

[0060] In particular, the apparatus 30 may be implemented in or part of user equipment and may be configured to process as described in relation to FIG.

[0061] Thus, according to some example versions of the present invention, there is provided an apparatus 30 for use in user equipment, comprising at least one processor 31 and at least one memory 33 for storing instructions to be executed by the processor 31, wherein the at least one memory 33 and the instructions are configured by the at least one processor 31 to cause the apparatus 30 to generate at least user equipment capability information, the generation including identifying at least one machine learning model available in the user equipment for a given scenario, assigning a unique identification to each of the at least one machine learning model, and associating the at least one machine learning model with the unique identification with at least one of a parameter list, a dataset, and a registration ID, and reporting the generated user equipment capability information to a network entity.

[0062] Furthermore, the present invention may be implemented by an apparatus for user equipment, as shown in FIG. 4, comprising means for performing the above-mentioned processing.

[0063] That is, according to some example versions of the present invention, as shown in Figure 4, an apparatus for use in user equipment comprises means for generating user equipment capability information 41. The generating means 41 includes means for identifying 411 at least one machine learning model available in the user equipment for a given scenario, means for assigning 412 a unique identification to each of the at least one machine learning model, and means for associating 413 the at least one machine learning model having the unique identification with at least one of a parameter list, a dataset, and a registration ID. Further, the apparatus comprises means 42 for reporting the generated user equipment capability information to a network entity.

[0064] Further, according to some example versions of the present invention, there is provided a computer program comprising instructions that, when executed by an apparatus for use with user equipment, cause the apparatus to generate user equipment capability information, the generation including identifying at least one machine learning model available at the user equipment for a given scenario, assigning a unique identification to each of the at least one machine learning model, and associating the at least one machine learning model with the unique identification with at least one of a parameter list, a dataset, and a registration ID; and reporting the generated user equipment capability information to a network entity.

[0065] The computer program product may comprise code means adapted to produce the steps of any of the methods as described above when loaded into the memory of a computer.

[0066] According to some example versions of the present invention, there is provided a computer program product as defined above, the computer program product comprising a computer readable medium having software code portions stored thereon.

[0067] According to some example versions of the present invention, a computer program product as defined above is provided, the program being directly loadable into the internal memory of a processing device / apparatus.

[0068] According to some example versions of the present invention, there is provided a computer-readable medium storing a computer program as detailed above.

[0069] According to some example versions of the present invention, a computer program product is provided comprising computer-executable computer program code configured to, when the program is run on a computer (e.g., a computer of an apparatus according to any one of the aforementioned exemplary apparatus-related aspects of the present disclosure), cause the computer to perform a method according to any one of the aforementioned exemplary method-related aspects of the present disclosure.

[0070] Such a computer program product may comprise (or embody) a (tangible) computer-readable (storage) medium or the like, on which computer-executable computer program code is stored and / or from which the program may be directly loadable into the internal memory of the computer or its processor.

[0071] Furthermore, the present invention may be embodied by an apparatus for use in user equipment, comprising respective circuitry for performing the above-described processes.

[0072] That is, according to some example versions of the present invention, an apparatus for use in user equipment is provided, including a generating circuit for generating user equipment capability information, the generating circuit including an identifying circuit for identifying at least one machine learning model available in the user equipment for a given scenario, an assigning circuit for assigning a unique identification to each of the at least one machine learning model, an associating circuit for associating the at least one machine learning model having the unique identification with at least one of a parameter list, a dataset, and a registration ID, and a reporting circuit for reporting the generated user equipment capability information to a network entity.

[0073] For further details regarding the functionality of the apparatus and computer program, reference is made to the above description of the method according to some example versions of the invention, as described in relation to FIG.

[0074] In the above exemplary description of the device, only units / means relevant for understanding the principles of the present invention are described using functional blocks. The device may include additional units / means necessary for each operation. However, the description of these units / means is omitted in this specification. The arrangement of the functional blocks of the device should not be construed as limiting the present invention, and a function can be performed by one block or further divided into sub-blocks.

[0075] In the above description, when an apparatus (or some other means) is described as being configured to perform a certain function, this should be interpreted as being equivalent to stating that a (i.e., at least one) processor or corresponding circuitry, potentially in conjunction with computer program code stored in the memory of the respective apparatus, is configured to cause the apparatus to perform at least the described function. Also, such functions should be interpreted as being equivalently implementable by circuits or means specifically configured to perform the respective functions (i.e., the phrase "a unit configured to" should be interpreted as being equivalent to phrases such as "means for").

[0076] As used in this application, the term "circuitry" may refer to one or more, or all, of the following: (a) hardware-only circuit implementations (e.g., implementations using only analog and / or digital circuitry); and (b) A combination of hardware circuitry and software, such as (where applicable): (i) a combination of analog and / or digital hardware circuitry and software / firmware; and (ii) any portion of a hardware processor (including a digital signal processor) with software, software, and memory that work together to cause a device, such as a mobile phone or server, to perform various functions; and (c) A hardware circuit and / or processor, such as a microprocessor, or a portion of a microprocessor, that requires software (e.g., firmware) for operation, but may be absent when software is not necessary for operation.

[0077] This definition of circuit applies to all uses of the term in this application, including within any patent claims. As a further example, the term circuit, as used in this application, also covers merely a hardware circuit or processor (or processors), or implementations of portions of a hardware circuit or processor, and their accompanying software and / or firmware. The term circuit also covers, for example, and where applicable to particular patent claim elements, a baseband integrated circuit or a processor integrated circuit for a mobile device, or similar integrated circuit in a server, cellular network device, or other computing or network device.

[0078] For the purposes of the present invention as set forth herein above, the following should be noted.

[0079] - Method steps that are likely to be implemented as software code portions and run using a processor in an apparatus (as an example of a device, apparatus, and / or modules thereof, or as an example of an entity comprising an apparatus and / or modules thereof) can be specified using any known or future-developed programming language, as long as the functionality defined by the method steps is maintained without being affected by the software code.

[0080] In general, any method step is suitable to be implemented as software or by hardware without changing the concept of the aspect / embodiment and without modifying it in terms of the functionality implemented.

[0081] - the method steps and / or devices, units or means likely to be implemented as hardware components in the apparatus defined above, or any modules thereof (e.g. a device performing the functions of an apparatus according to the above mentioned aspects / embodiments), are hardware independent and can be implemented using any known or future developed hardware technology, or any hybrid thereof, such as MOS (Metal Oxide Semiconductor), CMOS (Complementary Metal-Oxide Semiconductor), BiMOS (Bipolar Metal-Oxide Semiconductor), BiCMOS (Bipolar Complementary Metal-Oxide Semiconductor), ECL (Emitter Coupled Logic), TTL (Transistor-Transistor Logic), etc., such as using ASIC (Application Specific Integrated Circuit) components, FPGA (Field Programmable Gate Array) components, CPLD (Combined Programmable Logic Device) components, APU (Accelerated Processor Unit), GPU (Graphics Processor Unit) or DSP (Digital Signal Processor) components.

[0082] - Devices, units or means (e.g. apparatus as defined above, or any of their respective units / means) may be implemented as separate devices, units or means, but this does not exclude them being implemented in a distributed manner throughout a system, as long as the functionality of the devices, units or means is maintained.

[0083] - an apparatus may be represented by a semiconductor chip, a chipset or a (hardware) module comprising such a chip or chipset, but this does not exclude the possibility that the functionality of the apparatus or module may be implemented as software in a (software) module, such as a computer program or computer program product with executable software code portions for execution / running on a processor, instead of being implemented in hardware.

[0084] A device may be considered as an apparatus or as an assembly of two or more apparatuses, whether functionally associated with each other or functionally independent of each other but within the same device housing, for example.

[0085] In general, it should be noted that each functional block or element according to the above-described aspects can be implemented by any known means, either in hardware and / or software, when adapted only to perform the described function of the respective part. The described method steps can be realized in separate functional blocks or by separate devices, or one or more of the method steps can be realized in a single functional block or by a single device.

[0086] In general, any method step is suitable to be implemented as software or by hardware without changing the concept of the present invention. Devices and means may be implemented as individual devices, but this does not exclude them being implemented in a distributed manner throughout a system, as long as the functionality of the devices is maintained. Such and similar principles should be considered to be known to those skilled in the art.

[0087] Software in the sense of this description comprises software code and thus code means or portions, or computer programs, or computer program products, for performing the respective functions, as well as software (or computer programs, or computer program products) embodied on tangible media such as computer readable (storage) media on which the respective data structures or code means / portions are stored, or potentially embodied in signals or chips during their processing.

[0088] It should be noted that the above-described aspects / embodiments and general and specific examples are presented for illustrative purposes only and are not intended to limit the present invention in any way. Rather, all variations and modifications that fall within the scope of the appended claims are intended to be covered.

Claims

1. 1. A method for use in a user equipment, comprising: generating user equipment capability information, Identifying at least one machine learning model available on the user equipment for a given scenario; assigning a unique identification to each of the at least one machine learning model; associating the at least one machine learning model having the unique identification with at least one of a parameter list, a dataset, and a registration ID; generating user equipment capability information, including: reporting the generated user equipment capability information to a network entity; A method comprising:

2. receiving a request from the network entity indicating the predetermined scenario for which the machine learning model is to be identified; The method of claim 1 further comprising:

3. receiving, from the network entity, a configuration for the machine learning model to be used for the given scenario based on the reported user equipment capability information; operating in accordance with said received configuration; and The method of claim 1 further comprising:

4. 2. The method of claim 1, wherein the user equipment capability information includes an indication of whether the identified machine learning model can be switched from one to another when applied to the given scenario.

5. 2. The method of claim 1 , wherein the user equipment capability information includes an indication of whether two or more of the identified machine learning models, when applied to the given scenario, may be activated at a given time.

6. 2. The method of claim 1 , wherein the user equipment capability information includes an indication of whether the identified machine learning model, when applied to the given scenario, can be disabled in favor of a parametric model.

7. The parameter list: Radio conditions that induce at least one of the following: deployment type, applicable scenario, base station / user equipment antenna configuration, and scattering parameters; Machine learning specific details including at least one reportable quantity, required measurement configurations, required auxiliary information, inputs / outputs and dimensions of said machine learning model; with limitations / conditions, including at least one inference delay, required warm-up time, and fine-tuning requirements. The method of claim 1 , further comprising defining at least one.

8. 10. The method of claim 1, wherein the dataset refers to a version of the dataset accessible to both the user equipment and a network through an operator-controlled server and / or other means, including a proprietary cloud, and the dataset specifies aspects associated with a model.

9. The method of any one of claims 1 to 8, wherein the registration ID refers to a unique version of the machine learning model having a unique identifier.

10. A means for generating user equipment capability information, comprising: means for identifying at least one machine learning model available on the user equipment for a given scenario; means for assigning a unique identification to each of the at least one machine learning model; means for associating the at least one machine learning model having the unique identification with at least one of a parameter list, a dataset, and a registration ID; means for generating user equipment capability information, means for reporting the generated user equipment capability information to a network entity; An apparatus comprising:

11. 11. The apparatus of claim 10, comprising: means for receiving a request from the network entity indicating the predetermined scenario for which the machine learning model is to be identified.

12. receiving, from the network entity, a configuration for the machine learning model to be used for the given scenario based on the reported user equipment capability information; operating in accordance with said received configuration; and 11. The apparatus of claim 10, comprising means for performing:

13. 11. The apparatus of claim 10, wherein the user equipment capability information includes an indication of whether the identified machine learning model can be switched from one to another when applied to the given scenario.

14. 11. The apparatus of claim 10, wherein the user equipment capability information includes an indication of whether two or more of the identified machine learning models, when applied to the given scenario, may be activated at a given time.

15. 11. The apparatus of claim 10, wherein the user equipment capability information includes an indication of whether the identified machine learning model can be disabled in favor of a parametric model when applied to the given scenario.

16. The parameter list: Radio conditions that induce at least one of the following: deployment type, applicable scenario, base station / user equipment antenna configuration, and scattering parameters; Machine learning specific details including at least one reportable quantity, required measurement configurations, required auxiliary information, inputs / outputs and dimensions of said machine learning model; with limitations / conditions, including at least one inference delay, required warm-up time, and fine-tuning requirements. The apparatus of claim 10, further comprising:

17. 11. The apparatus of claim 10, wherein the dataset refers to a version of the dataset accessible to both the user equipment and a network through an operator-controlled server and / or other means including a proprietary cloud, and the dataset specifies aspects associated with a model.

18. The apparatus of any one of claims 10 to 17, wherein the registration ID references a unique version of the machine learning model having a unique identifier.

19. A computer program comprising instructions which, when executed by an apparatus, cause said apparatus to perform the method of any one of claims 1 to 8.

Images