Applying conditions for input parameters for UE-side prediction using ML model

Abstract

The second network node receives, from the first network node, a configuration for performing prediction by the ML model, the configuration including input conditions related to input parameters or values for the ML model. The second network node determines whether input parameters of the ML model satisfy input condition (s). The second network node, in response to the determination, performs one or both of the prediction using the ML model, or reports information to the first network node based on whether the prediction is performed. The first network node sends the configuration to the second network node, and one or both of receiving an output of a prediction performed on the second network node using the ML model or reporting information from the second network node based on whether the prediction is performed.

Description

Technical Field

[0001] Examples of embodiments described herein generally relate to using machine learning (ML) models for beam prediction by user equipment (UE) in a cellular network, and more specifically, to prediction by the UE applying conditions to input parameters of the ML model. Background Technology

[0002] Machine learning (ML) models are deployed in numerous locations within wireless networks, such as cellular networks. These models can be trained by the network or other elements and then used by the network or by a user equipment (UE), a device connected to the network. For example, a UE can perform measurements on the beams the network uses for communication. Those measurements on the beams (and potentially other input parameters) can be fed into an ML model that has been trained accordingly to output an indication of the specific beams the network should use for communication from the network to the UE. ML models can be trained for many other communication parameters. Summary of the Invention

[0003] This section is intended to include examples and is not intended to be restrictive.

[0004] In an exemplary embodiment, a method is disclosed, comprising: receiving from a first network node by a second network node a configuration for performing a prediction by a machine learning model, the configuration including one or more input conditions relating to input parameters or values ​​of input parameters for the machine learning model; determining by the second network node whether the input parameters for the machine learning model satisfy one or more input conditions; and performing one or both of the following by the second network node in response to the determination: using the prediction of the machine learning model, or reporting information to the first network node based on whether the prediction has been performed.

[0005] Additional exemplary embodiments include a computer program that includes instructions for performing the methods of the preceding paragraph when the computer program is run on the device. The computer program according to this paragraph is a computer program product including a computer-readable medium carrying instructions embodied therein for use with the device. Another example is a computer program according to this paragraph in which the program can be directly loaded into the internal memory of the device.

[0006] An exemplary device includes one or more processors and one or more memories storing instructions that, when executed by the one or more processors, cause the device to perform at least the following: receiving from a first network node by a second network node a configuration for performing a prediction by a machine learning model, the configuration including one or more input conditions related to input parameters or values ​​of the input parameters for the machine learning model; determining by the second network node whether the input parameters for the machine learning model satisfy one or more input conditions; and performing one or both of the following by the second network node in response to the determination: using the prediction of the machine learning model, or reporting information to the first network node based on whether the prediction has been performed.

[0007] An exemplary computer program product includes a computer-readable storage medium carrying instructions that, when executed by a device, cause the device to perform at least the following operations: receiving from a first network node by a second network node a configuration for performing a prediction by a machine learning model, the configuration including one or more input conditions related to input parameters for the machine learning model or the values ​​of the input parameters; determining by the second network node whether the input parameters for the machine learning model satisfy one or more input conditions; and performing one or both of the following by the second network node in response to the determination: using the prediction of the machine learning model, or reporting information to the first network node based on whether the prediction has been performed.

[0008] In another exemplary embodiment, an apparatus includes components for performing the following operations: receiving from a first network node by a second network node a configuration for performing a prediction by a machine learning model, the configuration including one or more input conditions related to input parameters for the machine learning model or the values ​​of the input parameters; determining by the second network node whether the input parameters for the machine learning model satisfy one or more input conditions; and performing one or both of the following by the second network node in response to the determination: using the prediction of the machine learning model, or reporting information to the first network node based on whether the prediction has been performed.

[0009] In an exemplary embodiment, a method is disclosed, comprising: sending from a first network node to a second network node a configuration for performing a prediction by a machine learning model at the second network node, the configuration including one or more input conditions relating to input parameters or values ​​of input parameters for the machine learning model; and receiving from the second network node, from the first network node, one or both of the output of a prediction performed on the second network node using the machine learning model or information reported from the second network node based on whether the prediction was performed.

[0010] Additional exemplary embodiments include a computer program that includes instructions for performing the methods of the preceding paragraph when the computer program is run on the device. The computer program according to this paragraph is a computer program product including a computer-readable medium carrying instructions embodied therein for use with the device. Another example is a computer program according to this paragraph in which the program can be directly loaded into the internal memory of the device.

[0011] An exemplary device includes one or more processors and one or more memories storing instructions that, when executed by the one or more processors, cause the device to perform at least: sending from a first network node to a second network node a configuration for performing a prediction by a machine learning model at the second network node, the configuration including one or more input conditions related to input parameters or values ​​of input parameters for the machine learning model; and receiving from the second network node, from the first network node, one or both of the output of a prediction performed on the second network node using the machine learning model or reporting information from the second network node based on whether the prediction was performed.

[0012] An exemplary computer program product includes a computer-readable storage medium carrying instructions that, when executed by a device, cause the device to perform at least the following operations: sending from a first network node to a second network node a configuration for performing a prediction by a machine learning model at the second network node, the configuration including one or more input conditions related to input parameters for the machine learning model or the values ​​of the input parameters; and receiving from the second network node one or both of the output of a prediction performed on the second network node using the machine learning model or information reported from the second network node based on whether the prediction was performed.

[0013] In another exemplary embodiment, an apparatus includes components for performing the following operations: sending from a first network node to a second network node a configuration for performing a prediction by a machine learning model at the second network node, the configuration including one or more input conditions related to input parameters or values ​​of input parameters for the machine learning model; and receiving from the second network node, from the first network node, one or both, the output of a prediction performed at the second network node using the machine learning model or information reported from the second network node based on whether the prediction was performed. Attached Figure Description

[0014] In the attached diagram: Figure 1 This is a flowchart of a method in which conditions are applied to input data for UE-side prediction using an ML model; Figure 2It is a block diagram showing the ML model, its input parameters, and outputs, and also shows examples of input conditions (multiple) related to the input parameters or the values ​​of the input parameters for the ML model; Figure 3 These are signaling diagrams and flowcharts illustrating the conditions for input parameters used in ML model-based UE-side prediction; and Figure 4 This is a block diagram of a possible, non-limiting, exemplary system in which exemplary embodiments can be practiced. Detailed Implementation

[0015] Abbreviations that can be found in the specification and / or drawings are defined below at the end of the detailed description.

[0016] The word "exemplary" is used herein to mean "serving as an example, instance, or illustration." Any embodiment described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments. All embodiments described in this detailed description are exemplary embodiments provided to enable those skilled in the art to make or use the invention, and are not intended to limit the scope of the invention as defined by the claims.

[0017] When used with more than one reference numeral, word, or abbreviation in this specification, and as is generally used in this specification, " / " may be interpreted as "or," "and," or "both." As used herein, "at least one of the following: " and "at least one of " and similar wording, wherein the list of two or more elements is connected by "and" or "or," means at least any one of these elements, or at least any two or more of these elements, or at least all of these elements.

[0018] As used herein, the singular forms “a,” “an,” and “the” are also intended to include the plural forms, unless the context clearly indicates otherwise. It will be further understood that the terms “comprising,” “having,” and / or “including” as used herein specify the presence of stated features, elements, and / or components, but do not exclude the presence or addition of one or more other features, elements, components, and / or combinations thereof.

[0019] Any flowcharts in this article (such as...) Figure 1 or 2) or signaling diagram (such as Figure 3 A block diagram is considered a logic flowchart and illustrates the operation of an exemplary method according to an exemplary embodiment, the execution result of computer program instructions embodied in a computer-readable storage medium, the function executed by logic implemented in hardware, and / or interconnecting components for executing the function. Figure 4The diagram also illustrates the operation of an exemplary method according to an exemplary embodiment, the execution result of computer program instructions embodied in a computer-readable storage medium, the function executed by logic implemented in hardware, and / or interconnecting components for performing the function.

[0020] The example in this article relates to a 3GPP research project for AI / ML technologies in 3GPP, see Qualcomm (Host), “Research on Artificial Intelligence (AI) / Machine Learning (ML) for the NR Air Interface” RP-213599, 3GPP TSG RAN Meeting #94e, Electronic Conference, December 6-17, 2021. This research project has the following objectives, which are enclosed between open and close quotation marks: 4. Objectives 4.1 Target of SI or core WI or test WI For each target use case, the 3GPP framework for AI / ML for the air interface was studied in terms of performance, complexity and potential specification impact.

[0021] Use cases of interest: The initial set of use cases includes: CSI feedback enhancements include, for example, reduced overhead, improved accuracy, and improved prediction [RAN1]. Beam management, such as temporal beam prediction and / or spatial beam prediction, to reduce overhead and latency, and improve beam selection accuracy [RAN1]. Enhanced localization accuracy for different scenarios, including those with heavy NLOS conditions [RAN1] Before RAN#98, representative sub-use cases were identified for each use case for characterization and baseline performance evaluation. The AI / ML approach for the selected sub-use cases needs to be diverse enough to support various requirements for g NB-UE collaboration levels. Note: The use cases selected for this study are intended solely to develop a framework for applying AI / ML to the air interface for these and other use cases. The selection itself is not intended to provide any indication of the prospects for any future standardization projects.

[0022] AI / ML models, terminology, and descriptions are used to identify common and specific characteristics of the framework survey: Characterizing the definition stages and associated complexities of AI / ML related algorithms: Model generation, including model training (including input / output, preprocessing / postprocessing, online / offline (if applicable)), model validation, and model testing (if applicable). Inference operations, such as input / output, preprocessing / postprocessing (if applicable). Identify the various collaboration levels between the UE and gNB associated with the selected use case, for example... No collaboration: Based solely on the implemented AI / ML algorithm, with no information exchange [for comparison purposes]. Various levels of UE / gNB collaboration aimed at individual or joint ML operations.

[0023] Representation of AI / ML model lifecycle management: e.g., model training, model deployment, model inference, model monitoring, and model updates. Datasets used for training, validation, testing, and inference. Use common symbols and terms to identify AI / ML related functions, processes, and interfaces. Note: Consider the work to be done on FS_NR_ENDC_data_collect when appropriate. For the use cases under consideration: 1) Evaluate the performance benefits of AI / ML-based algorithms for agreed-upon use cases in the final representative set: A methodology based on statistical models (from TR 38.901 and TR 38.857 [Location]) is used for link and system-level simulation.

[0024] The 3GPP evaluation methodology should be expanded as needed to better suit AI / ML-based technologies.

[0025] Whether optional field data are needed to further evaluate performance and robustness in real-world environments should be discussed as part of this study.

[0026] For the selected use cases, there should be common assumptions when building the datasets for training, validation, and testing.

[0027] Considering sufficient model training strategies, collaboration levels, and associated impacts Consider the agreed-upon (multiple) base AI models for calibration. The descriptions of the AI ​​models used for evaluation and the training methodologies should be reported for informational and cross-checking purposes. KPIs: Determine common KPIs and corresponding requirements for AI / ML operations. Determine use case-specific KPIs and baselines for the selected use cases.

[0028] The performance, inference latency, and computational complexity of AI / ML-based algorithms should be compared with the baseline of existing technologies. The overhead, power consumption (including computation), memory storage and hardware requirements (including for a given processing latency), and generalization capabilities associated with enabling the corresponding AI / ML scheme should be considered.

[0029] 2) Assess the potential normative impact, particularly on the usage of conventions in the final representative set and on the public framework: Regarding the PHY layer, for example (RAN1) Consider aspects related to potential specifications for AI model lifecycle management and the construction of datasets for training, validation, and testing for selected use cases. Use case and collaboration level-specific specification impacts, such as new signaling, components for training and validation data assistance, auxiliary information, measurement, and feedback. Regarding protocols, for example, (RAN2)-RAN2 only begins work after sufficient progress has been made in use case studies in RAN1. The RAN1 input considers aspects related to, for example, capability indicators, configuration and control processes (training / inference), and the management of data and AI / ML models. Collaboration level-specific specification impact for each use case Regarding interoperability and testability, for example, (RAN4)-RAN4 only began working after sufficient progress had been made in use case studies in RAN1 and RAN2. Requirements and testing framework for validating AI / ML-based performance enhancements and ensuring that AI / ML-enabled UEs and gNBs meet or exceed existing minimum requirements where applicable. Consider the needs and impacts of defining AI / ML processing capabilities. Note 1: We do not expect to specify a particular AI / ML model; that is left to the implementation. User data privacy must be preserved.

[0030] Note 2: Research on AI / ML for the air interface is based on the current RAN architecture and should not introduce new interfaces. In addition to RP-213599, there are several protocols regarding the 3GPP status regarding AI / ML, including the following, and can be found in, for example, Nokia, Nokia Shangai Bell, “Other Aspects of ML for Beam Management”, 3GPP TSG RAN WG1#113, Incheon, Korea, May 22-26, 2023, and R1-2304684.

[0031] protocol For BM case 1 with UE-side AI / ML models, we investigate the potential canonical impacts of L1 signaling reporting the following information about AI / ML model inference to the NW: Multiple beams based on the output of AI / ML model inference FFS: Predicted L1-RSRP corresponding to (multiple) beams FFS: Other Information protocol For BM case 2 with UE-side AI / ML models, we investigate the potential canonical impacts of L1 signaling reporting the following information about AI / ML model inference to the NW: Beams of N future time instances based on AI / ML model inference output FFS: The value of N FFS: Predicted L1-RSRP corresponding to (multiple) beams Information regarding the timestamps corresponding to the reported beam(s). FFS: Explicit or Implicit FFS: Other Information protocol For BM Case 1 and BM Case 2 with UE-side AI / ML models, the following alternatives for model monitoring with potential downward selection are investigated: Alternative Solution 1. UE-side model monitoring UE monitoring performance metrics The UE makes (multiple) decisions regarding model selection, activation, deactivation, switching, and rollback. Alternative Option 2. NW-side Model Monitoring NW monitors (multiple) performance metrics NW makes (multiple) decisions regarding model selection, activation, deactivation, switching, and rollback. Alternative Option 3. Hybrid Model Monitoring U monitors (multiple) performance metrics NW makes (multiple) decisions regarding model selection, activation, deactivation, switching, and rollback. protocol Regarding NW-side model monitoring of network-side AI / ML models for BM case 1 and BM case 2, the necessity and potential canonical impacts from the following aspects were investigated: UE report based on beam(s) measurements of a set of beams indicated by the gNB. Signaling, such as RRC-based and L1-based signaling Note: Performance and UE complexity, power consumption should be taken into account. As these protocols demonstrate, there is active research in the AI / ML field. In this field, UEs can be configured to perform ML-based predictions using network-deployed ML models. Since the model is trained by the functions of the wireless communication network (e.g., the model may logically interface with RAN functions) or may be part of RAN functions, the required input parameters can be configured by the network (e.g., the input parameters used to perform predictions, such as beam index / L1-RSRP, location, etc.).

[0032] Then, the UE can be provided with an ML model (e.g., by a RAN or ML function associated with the RAN / core network) and configured to perform predictions and provide the prediction output (which may also be referred to as the result) to the network.

[0033] However, even though the NW has been configured with input parameters, there may be other limitations associated with the provided ML model and input parameters. For example, incorrect or invalid input parameters may lead to inaccurate or invalid predictions, which could cause the network to perform actions (such as control / data beam configuration) based on invalid predictions, and the results may be suboptimal, or even lead to link failures.

[0034] The embodiments described in this paper address these and other issues. As an example, a UE that uses an ML model to perform predictions (such as beam prediction) can perform predictions without using an ML model, for example, based on whether certain conditions(s) are not met. This is in Figure 1 As described in the text, Figure 1 This is an overview used to provide examples for this article.

[0035] refer to Figure 1 The figure is a flowchart of a method in which conditions are applied to input data for UE-side prediction using an ML model. In box 110, the UE obtains an ML model from the network, or accesses an already obtained ML model, which has input parameters and outputs one or more communication parameters. In this paper, an ML model is a model that has been trained to identify certain types of patterns for input parameters and create corresponding outputs for one or more communication parameters. One such example is for beam selection based on input parameters. The model has been applied to a set of data during training, and an algorithm has been implemented to infer and learn from this data to produce outputs. While this paper primarily uses the term "ML model," AI models or other models trained to identify patterns and produce outputs are considered to fall under the term "ML model." It should also be noted that the term "output" can encompass the following (in which beam selection is used as an example): a single output, such as an indication of the beam to be selected (as a communication parameter); multiple outputs, such as an indication of the beam to be selected (as a communication parameter) and an indication of the probability of selecting the correct beam, or an indication of multiple beams that can be selected (as multiple communication parameters) and the corresponding probability that these beams are the best beams to be selected. As an example, an ML model may refer to at least one or more of the following: (but not limited to) (multiple) neural networks, (multiple) convolutional neural networks, (multiple) recurrent neural networks, deep reinforcement learning, transformers, convolutional-LSTM (where LSTM stands for Long Short-Term Memory), autoencoder-decoder LSTM, etc.

[0036] In box 120, the UE receives (e.g., additional) configuration for performing ML-based prediction. This configuration includes one or more input conditions related to input parameters or values ​​of input parameters for the ML model. In box 130, the UE determines whether the input parameters(s) satisfy or do not satisfy the input conditions(s) used for the current prediction. This is referred to as prediction because the model is based on and previously trained on the input parameters, and the output is predicted based on those inputs, such that the output is predicted based on those previously checked input parameters. In other words, the ML model correctly predicts its output based on its previous training.

[0037] In response to the input parameters satisfying (multiple) input conditions (reference numeral 140), blocks 160 and 170 are executed. In block 160, the UE uses an ML model to perform a prediction. In block 170, the UE reports the prediction output to the network.

[0038] In response to an input parameter not satisfying (multiple) input conditions (reference numeral 150), there are two possible alternative paths (alternative schemes) 151 (alternative scheme 1, Alt. 1) and 152 (alternative scheme 2, Alt. 2). In block 180, for alternative scheme 1 151, the UE does not perform prediction using the ML model. Instead, the UE reports at least one indication that the criterion was not met and / or information about the unmet criterion. This reporting occurs in block 190. In one embodiment, for block 190, the UE can be configured (e.g., in a separate reporting instance) to report to the network which input parameter condition was not met. In this case, the UE can be configured to exclude predictions to the network (information predicted as output).

[0039] Alternatively, in Alternative 2 152, the UE is configured to perform a prediction (box 185) and report (box 190) the predicted value with an associated indication that the input parameter conditions were not met. In one example, the UE is configured to still report (box 186), but also indicate (box 190) that one or more input conditions are invalid. Another example is that the UE performs a prediction (box 185), but then performs a report in box 190 without reporting the prediction output. See reference numeral 187. This would, for example, allow the network to query the UE for the prediction later if the network determines that more information about the prediction result is useful. Alternative 2 152 allows the network to still receive (or query later) the report, and also receive feedback on the report based on the unmet conditions(s). For example, this could be used for additional training of ML models or for other purposes.

[0040] Box 195 indicates possible techniques for reporting. That is, in one embodiment, any reporting can be performed using one of the following: UCI (Uplink Control Information) on PUSCH / PUCCH, MAC CE, or RRC. As a non-limiting example, PUSCH / PUCCH / MAC CE can be used to report predictions performed using valid input parameters. In another non-limiting example, if invalid inputs are used and a prediction is reported, it can be reported using RRC or RRC and PUSCH / PUCCH / MAC CE. RRC-level reporting can be used to report invalid parameters or parameters or conditions deemed invalid for prediction.

[0041] Now that an overview has been provided, more details are coming up.

[0042] refer to Figure 2 This figure is a block diagram showing an ML model, its input parameters, and its output, and also shows examples of (multiple) input conditions related to the input parameters or the values of the input parameters for the ML model. The ML model 200 (which can be a neural network) has a certain number of input parameters 210, in this case input parameters 210-1 to 210-N, where N>1. The ML model 200 has been trained to produce an output 220 based on the input parameters 210. The output 220 can include the following as examples: 220-1, communication parameters (CP), such as an indication of a beam; 220-2, CP and corresponding probabilities (e.g., the number of probabilities indicating that this is the correct beam to be selected); 220-3, multiple CPs, e.g., with implicit probabilities, such as beam 1 and beam 2 in that order, where the probability that beam 1 is correct is higher than that of beam 2; or 220-4, multiple CPs and corresponding probabilities.

[0043] Note that the terms "input parameter" and "input condition" are only used to describe the corresponding elements used herein, and other terms may be used. For example, "input parameter" can be shortened to "input" or any other term indicating an input to the ML model 200. As a non-limiting example, "input condition" can be referred to as an input criterion or rule, and corresponds to any element that sets a condition for an input parameter to determine whether a prediction should be performed based on the input parameter that satisfies the condition.

[0044] The input condition 230 is related to the input parameter 210 or the value of the input parameter 210 for the ML model 200. Multiple examples of the input condition 230 are shown in Figure 2 and described below.

[0045] In one embodiment, the UE is configured with at least one requirement as an input condition 230-1 for the (multiple) input parameters 210 for ML-based prediction for the UE. If the requirement as the input condition 230-1 is not satisfied for the input parameter 210 (as described in Figure 1 ), then in the case indicated by the configuration (see block 120), the UE is not allowed to use the input parameter 210 for prediction for the ML model. Also see the reference numerals 150 and alternative 1 151 in Figure 1 .

[0046] The input parameter and the input condition 230-2 are also considered. In this example, there are N input parameters, and the UE has a requirement (as the input condition 230-2) for one of the following for the current prediction: a) All N input parameters for the current prediction, or b) At least M < N input parameters; or c) A specific parameter.

[0047] In other words, if the UE does not meet the applicable requirements (a), (b), or (c) for the current prediction, the UE is not required to perform the prediction (or is not allowed to perform the prediction that will be reported) (see Figure 1 In alternative schemes 1, 151 and 180, the UE is configured not to report predictions to the network, or is configured to report values ​​or indicators where predictions were not performed, or the reported predictions were performed using invalid input values, or not all(multiple) requirements for the input conditions are met (see example...). Figure 1 (See box 190). When the UE cannot perform prediction (or does not need to perform prediction), the UE can be configured to indicate such information to the network.

[0048] In one embodiment, the condition used to perform beam prediction for a particular model can be cell quality or beam quality. See also Figure 3 Input condition 230-3. Under this embodiment, the following examples are possible. Quality can refer to, for example, RSRP and / or SINR (signal-to-interference-plus-noise ratio) / RSRQ (reference signal reception quality).

[0049] a) The model may require or be trained to predict cell quality associated with the UE that is measured / observed and is above a threshold level (see Input Condition 230-3a).

[0050] (b) At least X (e.g., N) beam quality values ​​must be higher than the threshold used as input to the model. See Input Conditions 230-3b. In the example, X ≤ N. The threshold can be configured by the network or associated with ML model 200.

[0051] c) As an example, the prediction can be, for example, L1-RSRP, or the beam ID(s) / spatial domain(s) that the UE may not be able to receive for a measurement sample (e.g., beam quality or cell quality) or beam prediction), for example, at least one measurement sample (for cell quality or beam quality for beam prediction) has a quality lower than the input threshold configured for the model. See Input Conditions 230-3c. More specifically, the prediction can be the L1-RSRP or (multiple) beam IDs or (multiple) top K beam IDs indicating beam quality. When the NW uses the predicted L1-RSRP / (multiple) beam IDs in the system throughput calculation, the NW can obtain the cell quality.

[0052] d) As an example, the UE cannot obtain at least one or more of the required measurements due to the reception of other signals and channels (e.g., system information). In this example (see Input Condition 230-3d), the UE must obtain all the required measurements (as input parameters) in order to satisfy the input parameters (see... Figure 1(See attached figure 140).

[0053] e) As another example, the UE cannot feed measurements for one or more (or all) of the required Rx panels (or beams). For example, the model might need to feed measurements (A <= B) of at least A antenna panels (or beams) out of B antenna panels (or beams) at the UE to the network. This is shown as input condition 230-3e: Requires measurements for at least A antenna panels out of B antenna panels at the UE (as an input parameter). The value of A can be configured by the network or defined in the specification. B can be the UE capability used to perform the measurements. The value of B can be provided to the network as a UE capability. In some examples, the value of B can be configured by the network and / or associated with the ML model, such that at least B panels (or beams) must be used to obtain the measurements (which are fed to the network).

[0054] f) As an example, the UE performs other tasks and cannot perform measurements; these other tasks could be, for example, random access. This can be handled under input condition 230-3d.

[0055] In a specific example, the input parameter could be location (e.g., positioning) information for the UE, and a specific threshold could exist for this. See Input Condition 230-4. The specific threshold could be as follows.

[0056] a) If the UE determines that there is an uncertainty associated with the location information used as input for the model that is greater than the uncertainty threshold, then see input condition 230-4a: i) For example, location can be satellite-based, and there may be a minimum number of satellites visible / used for location information, see input condition 230-4ai (i.e., if the number is less than the minimum number of satellites, the condition is not met because the uncertainty is higher with a smaller number of satellites); or ii) If the location uncertainty is greater than Z meters (or higher than the value Z), see input condition 230-4aii (i.e., if the UE is greater than Z meters, the condition is not met because the uncertainty increases with the number of meters). This can be indicated by the (satellite) positioning software, or the uncertainty information can be determined by the UE or provided by a positioning entity (such as LMF, location management function). Positioning accuracy can be enhanced, for example, using additional positioning enhancement techniques such as wireless access points (Wi-Fi or 3GPP positioning methods), and therefore in some examples, the residual location uncertainty may not be a value given by satellite positioning, but a combination of multiple techniques. In some examples, the location uncertainty threshold (value Z) can be a configurable value (e.g., configured by the network using RRC). In another example, the location uncertainty threshold can be a value associated with an ML model. Example values ​​could be 1 meter, 5 meters, 10 meters, etc. In one example, the UE can be configured to determine that input conditions for location-specific input parameters are valid when the location uncertainty is less than (or equal to) the value Z. In some cases, the UE can be configured to perform prediction when the location uncertainty is greater than the Z value. When a UE reports a prediction with an uncertainty greater than Z, the UE can be configured to instruct the reported result to be executed. In some examples, the UE can be configured with a location uncertainty threshold (e.g., configured by the network and / or associated with an ML model), and when the location uncertainty is above the threshold, the UE can consider location-specific input conditions invalid. In some cases, the UE is not required to predict and / or report prediction results. Alternatively, when the location uncertainty is below the threshold, the UE is expected to execute the prediction or consider location-specific input conditions to be valid.

[0057] refer to Figure 3 This diagram is a signaling diagram and flowchart applying conditions to input parameters for UE-side prediction using an ML model. In this diagram, base station 70 (which may also be a first network node) performs signaling with UE 10 (which may be a second network node). In block 310, base station 70 sends ML model 200 to the UE. As an example, what is sent could be an instruction 315 defining ML model 200. In block 320, base station 70 sends (e.g., additional) configurations for performing ML-based predictions (e.g., multiple input conditions to the UE), which are shown as (e.g., additional) configuration 325.

[0058] UE 10 in box 321 uses an ML model to apply conditions to the input parameters, such as Figure 1 and 2 As described in [the document]. Boxes 326 and 340 indicate the possible reports that the UE can perform. In box 326, UE 10 reports the predicted output (see also [the document]). Figure 1(Box 170) and the signal indication 330 of the predicted output. The base station 70 receives the indication and, in box 335, performs (a plurality of) actions based on the indicated predicted output, such as (box 336) using the beam indicated for the UE by the predicted output 330.

[0059] As another possibility for reporting, UE 10 in box 340 follows... Figure 1 Box 190 executes a report. Report 345 may include an indication that conditions(s) are not met and / or information about the unmet conditions(s). Base station 70 receives report 345 and, in box 350, performs (multiple) actions based on the report's (multiple) input parameters not meeting (multiple) input conditions(s). Box 360 indicates possible actions, such as further training ML model 200 or deploying alternative ML model 200 for the UE based on the report.

[0060] Go to Figure 4 The figure illustrates a block diagram of a possible, non-limiting example of a cellular network 1 connected to a user equipment (UE) 10. Figure 4 The cellular network shows several network elements: base station 70; and core network 90.

[0061] exist Figure 4 In this embodiment, User Equipment (UE) 10 (e.g., a second network node) wirelessly communicates with base station 70 (e.g., a first network node) of cellular network 1 via radio link 11. UE 10 is a wireless communication device, such as a mobile device, configured to access the cellular network. UE 10 is shown having one or more antennas 28. The ellipsis 2 indicates that multiple UEs 10 may exist that wirelessly communicate with base station 70 via radio link. UE 10 includes one or more processors 13, one or more memories 15, and other circuitry 16. The other circuitry 16 includes one or more receivers (Rx) 17 and one or more transmitters ((multiple) Tx) 18. Procedure 12 is used to cause UE 10 to perform the operations described herein. Procedure 12 includes, can access, or otherwise causes ML model 200 to perform predictions based on input parameters 210. For UE 10, the other circuitry 16 may include circuitry such as user interface elements (not shown) for a display-like device.

[0062] Base station 70 (e.g., a second network node and a network element of cellular network 1) provides UE 10 with access to cellular network 1 and data network 91 via core network 90 (e.g., via the User Level Function (UPF) of core network 90). That is, base station 70 is an access node providing access to network 1 for UE 10. Base station 70 is shown having one or more antennas 58. Typically, base station 70 may be referred to as RAN node 70, although many will alternatively refer to the base station as gNB (gNode B, base station for NR, New Radio). However, many other examples of RAN nodes exist, including eNB (evolved Node B) or TRP (transmitter-receiver point). Base station 70 includes one or more processors 73, one or more memories 75, and other circuitry 76. The other circuitry 76 includes one or more receivers (Rx) 77 and one or more transmitters (Tx) 78. Procedure 72 is used to cause base station 70 to perform the operations described herein. As an example, program 72 can also access a version of ML model 200, which can send ML model 200 to UE 10 or update it through additional training.

[0063] Note that the examples in this document are not limited to a specific wireless technology, although the examples use cellular technology. That is, base station 70 can alternatively be implemented via other wireless technologies, such as Wi-Fi (a wireless network protocol used by devices to communicate without a direct cable connection). In the case of Wi-Fi, link 11 can be characterized as a wireless link. For Wi-Fi, base station 70 is generally referred to as an access point, and UE 10 can be referred to as a wireless (e.g., mobile) device, and network 1 is a wireless network. As an example, core network 90 can be an Internet service provider or a wireless wide area network, and data network 91 can be the Internet.

[0064] Two or more base stations 70 communicate using, for example, multiple links 79. The multiple links 79 may be wired or wireless or both, and may implement, for example, an Xn interface for 5G (fifth generation), an X2 interface for LTE (long-term evolution), or other suitable interfaces for other standards.

[0065] Cellular network 1 may include a core network 90, as shown in the third illustration, which may include core network functions and provide connectivity to data network 91 (such as telephone networks and / or data communication networks (e.g., the Internet)) via one or more links 81. Core network 90 includes one or more processors 93, one or more memories 95, and other circuitry 96. The other circuitry 96 includes one or more receivers (Rx) 97 and one or more transmitters (Tx) 98. Procedure 92 is used to cause core network 90 to perform the operations described herein.

[0066] The core network 90 can be a 5GC (5G core network). The core network 90 can implement or include multiple network functions (NFs) 99, and program 92 can include one or more NFs 99. The 5G core network can use hardware such as memory and processors, as well as a virtualization layer. It can be a single standalone computing system, a distributed computing system, or a cloud computing system. The NFs 99 of the core network (as network elements) can be containers or virtual machines running on the hardware of the computing system(s) constituting the core network 90.

[0067] Core network functions for 5G can include access and mobility management functions provided by network functions such as Access and Mobility Management Functions (AMF) and session management functions provided by network functions such as Session Management Functions (SMF). Core network functions for access and mobility management in LTE (Long Term Evolution) networks can be provided by MME (Mobility Management Entity) and / or SGW (Serving Gateway) functions that route data to the data network. Many others are possible, such as those provided by… Figure 4 Examples shown include: AMF; SMF; MME; SGW; GMLC (Gateway Mobility Location Center); LMF (Location Management Function); UDM (Unified Data Management) / UDR (Unified Data Repository); NRF (Network Repository Function); and / or E-SMLC (Evolved Serving Mobility Location Center). These are merely exemplary core network functionalities that can be provided by the core network 90, and note that both 5G and LTE core network functionalities can be provided by the core network 90. ​​RAN node 70 is coupled to the core network 90 via backhaul link 31. RAN node 70 and core network 90 may include an NG (Next Generation) interface for 5G, or an S1 interface for LTE, or other suitable interfaces for other radio access technologies used for communication via backhaul link 31.

[0068] Any methods presented in this paper are applicable to any (wireless) technology using ML, as well as any technology such as 6G and higher.

[0069] In data network 91, there is a computer-readable medium 94. The computer-readable medium 94 contains instructions that, when downloaded and installed into the memory 15, 75, or 95 of the corresponding UE 10, base station 70, and / or (multiple) core network elements 90 and executed by (multiple) processors 13, 73, or 93, cause the corresponding device to perform the corresponding actions described herein. The computer-readable medium 94 may be implemented in other forms, such as via a compressed disk or memory stick.

[0070] Programs 12, 72, and 92 contain instructions stored by one or more corresponding memories 15, 75, or 95. When executed by one or more corresponding processors 13, 73, or 93, these instructions cause the corresponding devices 10, 70, or 90 to perform the operations described herein. The computer-readable memories 15, 75, or 95 can be of any type suitable to the local technical environment and can be implemented using any suitable data storage technology, such as semiconductor-based memory devices, flash memory, firmware, magnetic storage devices and systems, optical storage devices and systems, fixed memory, and removable memory. The computer-readable memories 15, 75, and 95 can be components for performing storage functions. As a non-limiting example, one or more processors 13, 73, and 93 can be of any type suitable to the local technical environment and can include one or more of general-purpose computers, special-purpose computers, microprocessors, digital signal processors (DSPs), and processors based on multi-core processor architectures. Processors 13, 73, and 93 can be components for causing their respective devices to perform various functions, such as those described herein. Specifically, the components used to perform the functions described herein may include one or more processors 13, 73 and 93 and one or more corresponding memories 15, 75 or 95, which individually store instructions that, when executed by the corresponding processor, cause the corresponding means to be executed.

[0071] Receivers 17, 77, and 97, and transmitters 18, 78, and 98 can implement wired or wireless interfaces. Receivers and transmitters can be grouped together as transceivers.

[0072] Cellular network 1 can achieve network virtualization, which is the process of combining hardware and software network resources and network functions into a single software-based management entity (virtual network). Network virtualization involves platform virtualization, often combined with resource virtualization. Network virtualization is categorized as external, combining many networks or parts of networks into virtual units or internal, thereby providing network-like functionality to a software container on a single system. Note that the virtualized entities created by network virtualization (such as network function 99) are still implemented to some extent using hardware such as processors 73 and / or 93 and memory 75 and / or 95, and such virtualized entities also produce technical effects.

[0073] Typically, various embodiments of user equipment 10 may include, but are not limited to, cellular phones (such as smartphones, mobile phones, cell phones, Voice over Internet Protocol (VoIP) phones and / or wireless local loop phones), tablet computers, portable computers, vehicles or in-vehicle devices for wireless V2X (vehicle-to-everything) communication, image capture devices such as digital cameras, gaming devices, music storage and recycle bins, internet-connected appliances (including Internet of Things, IoT devices), IoT devices with sensors and / or actuators for applications such as automation, and portable units or terminals combining such functionalities, laptop embedded devices (LEE), laptop devices (LME), Universal Serial Bus (USB) dongles, smart devices, wireless customer premises equipment (CPE), Internet of Things (IoT) devices, watches or other wearable devices, head-mounted displays (HMDs), vehicles, drones, medical devices and applications (e.g., remote surgery), industrial devices and applications (e.g., robots and / or other wireless devices operating in industrial and / or automated processing chain environments), consumer electronics devices, devices operating on commercial and / or industrial wireless networks, etc. In other words, UE 10 can be any terminal device capable of wireless communication. By way of example and not limitation, UE can also be referred to as a communication device, terminal device (MT), subscriber station (SS), portable subscriber station, mobile station (MS), or access terminal (AT).

[0074] Without limiting the scope, interpretation, or application of the claims that follow in any way, the technical effects and / or advantages of one or more example embodiments disclosed herein are that the prediction quality can be trusted by the network (e.g., for beam management decisions) when the UE performs ML-based predictions in a communication network. Another technical effect and / or advantage of one or more example embodiments disclosed herein is that the technology provides the network with further input to determine why predictions are unsuccessful; for example, it can be used to further train the model or deploy alternative models for the UE based on reports.

[0075] Here are some additional examples.

[0076] Example 1. A method comprising: receiving from a first network node by a second network node a configuration for performing a prediction by a machine learning model, the configuration including one or more input conditions relating to input parameters or values ​​of input parameters for the machine learning model; determining by the second network node whether the input parameters for the machine learning model satisfy one or more input conditions; and performing one or both of the following by the second network node in response to the determination: using the prediction of the machine learning model, or reporting information to the first network node based on whether the prediction has been performed.

[0077] Example 2. The method according to Example 1, wherein: determining comprises determining that input parameters for a machine learning model satisfy one or more input conditions; and performing comprises: based on determining that input parameters for a machine learning model satisfy one or more input conditions, using the machine learning model to perform a prediction; and based on performing the prediction, reporting an indication of the output of the machine learning model.

[0078] Example 3. The method according to Example 1, wherein: determining comprises determining that input parameters for a machine learning model do not satisfy one or more input conditions; and performing comprises: based on determining that input parameters for a machine learning model do not satisfy one or more input conditions, not using the machine learning model to perform a prediction; and reporting at least one indication that one or both of the one or more input conditions are not satisfied or information about the one or more input conditions that are not satisfied.

[0079] Example 4. The method according to Example 1, wherein: determining comprises determining that input parameters for a machine learning model do not satisfy one or more input conditions; and performing comprises: based on determining that input parameters for a machine learning model do not satisfy one or more input conditions, using the machine learning model to perform a prediction; based on performing the prediction, reporting an indication of the output of the machine learning model; and reporting at least one indication that one or both of the one or more input conditions are not satisfied or information about the one or more input conditions that are not satisfied.

[0080] Example 5. The method according to any one of Examples 1 to 4, wherein one or more input conditions comprise at least one requirement for input parameters for machine learning-based prediction by a second network node, and determining comprises determining whether the input parameters for the machine learning model satisfy the at least one requirement.

[0081] Example 6. The method according to any one of Examples 1 to 4, wherein: there are N input parameters, and the second network node has one of the following requirements as input conditions for the current prediction: a) all N input parameters for the current prediction, or b) at least M input parameters, where M < N; or c) a specific input parameter; and determining comprises determining whether the input parameters for the machine learning model satisfy the requirement.

[0082] Example 7. The method according to any one of Examples 1 to 4, wherein one or more input conditions comprise one or both of cell quality or beam quality, and determining comprises determining whether the input parameters for the machine learning model satisfy one or both of cell quality or beam quality.

[0083] Example 8. The method according to Example 7, wherein if the selected quality in cell quality or beam quality is higher than a threshold level value, the selected quality is satisfied.

[0084] Example 9. According to the method of Example 7, at least X of the N beam quality values ​​used as input parameters must be higher than a threshold for the machine learning model, so that the beam quality as an input condition is satisfied.

[0085] Example 10. According to the method of Example 7, at least one measurement sample for cell quality or beam quality has a quality lower than the input threshold configured for the machine learning model, such that the beam quality as an input condition is not met.

[0086] Example 11. According to the method of Example 7, the second network node must obtain all the required measurements as input parameters so that one or both of cell quality or beam quality are satisfied.

[0087] Example 12. According to the method of Example 7, the second network node is required to measure at least A antenna panels out of B antenna panels at the second network node as input parameters, where A <= B, otherwise the input parameters satisfy or do not satisfy one or both of cell quality or beam quality.

[0088] Example 13. According to any one of Examples 1 to 4, wherein one or more input conditions include location information input for the second network node.

[0089] Example 14. According to the method of Example 13, the uncertainty associated with the location information used as input parameters for a machine learning model must be greater than an uncertainty threshold so that the input parameters for the machine learning model do not satisfy uncertainty as an input condition.

[0090] Example 15. According to the method of Example 14, wherein the uncertainty includes a limitation on the minimum number of visible or used satellites for location information. Example 16. According to the method of Example 14, wherein the uncertainty includes location uncertainty greater than Z meters.

[0091] Example 17. A method comprising: sending from a first network node to a second network node a configuration for performing a prediction by a machine learning model at the second network node, the configuration including one or more input conditions relating to input parameters or values ​​of input parameters for the machine learning model; and receiving from the second network node, from the first network node, one or both, output of a prediction performed at the second network node using the machine learning model or information reported from the second network node based on whether the prediction was performed.

[0092] Example 18. A method according to any one of Examples 1 to 17, wherein the first network node includes an access node that provides access to the wireless network by the second network node, and the second network node includes a wireless device capable of connecting to the wireless network via the access node.

[0093] Example 19. A method according to any one of Examples 1 to 17, wherein the first network node includes a base station providing access to a cellular network by a second network node, and the second network node includes user equipment capable of connecting to the cellular network via the base station.

[0094] Example 20. A computer program including instructions for performing a method of any one of Examples 1 to 19 when the computer program is running on a device.

[0095] Example 21. A computer program according to Example 20, wherein the computer program is a computer program product comprising a computer-readable medium carrying instructions embodied therein for use with a device.

[0096] Example 22. A computer program according to Example 20, wherein the computer program can be directly loaded into the internal memory of the device.

[0097] Example 23. An apparatus comprising components for performing: receiving from a first network node by a second network node a configuration for performing a prediction by a machine learning model, the configuration including one or more input conditions relating to input parameters or values ​​of input parameters for the machine learning model; determining by the second network node whether the input parameters for the machine learning model satisfy one or more input conditions; and performing one or both of the following by the second network node in response to the determination: using the prediction of the machine learning model, or reporting information to the first network node based on whether the prediction has been performed.

[0098] Example 24. The apparatus according to Example 23, wherein: determining includes determining that the input parameters for the machine learning model satisfy one or more input conditions; and performing includes: using the machine learning model to perform a prediction based on determining that the input parameters for the machine learning model satisfy one or more input conditions; and an instruction to report the output of the machine learning model based on performing the prediction.

[0099] Example 25. The apparatus according to Example 23, wherein: determining includes determining that the input parameters for the machine learning model do not satisfy one or more input conditions; and performing includes: not using the machine learning model to perform a prediction based on determining that the input parameters for the machine learning model do not satisfy one or more input conditions; and reporting at least one indication indicating that one or both of the one or more input conditions are not satisfied, or information about the one or more input conditions that are not satisfied.

[0100] Example 26. The apparatus according to Example 23, wherein: determining comprises determining that an input parameter for a machine learning model does not meet one or more input conditions; and performing comprises: based on determining that an input parameter for a machine learning model does not meet one or more input conditions, using the machine learning model to perform a prediction; reporting an indication of the output of the machine learning model based on performing the prediction; and reporting at least one indication indicating that one or both of the one or more input conditions are not met or information about the one or more input conditions that are not met.

[0101] Example 27. The apparatus according to any one of Examples 23 to 26, wherein the one or more input conditions comprise at least one requirement for an input parameter for machine learning-based prediction by a second network node, and determining comprises determining whether the input parameter for the machine learning model meets the at least one requirement.

[0102] Example 28. The apparatus according to any one of Examples 23 to 26, wherein: there are N input parameters, and the second network node has one of the following requirements as input conditions for the current prediction: a) all N input parameters for the current prediction, or b) at least M input parameters, where M < N; or c) a specific input parameter; and determining comprises determining whether the input parameter for the machine learning model meets the requirement.

[0103] Example 29. The apparatus according to any one of Examples 23 to 26, wherein the one or more input conditions comprise one or both of cell quality or beam quality, and determining comprises determining whether the input parameter for the machine learning model meets one or both of cell quality or beam quality.

[0104] Example 30. The apparatus according to Example 29, wherein if the selected quality in cell quality or beam quality is higher than a threshold level value, the selected quality is met.

[0105] Example 31. The apparatus according to Example 29, wherein at least X of the N beam quality values used as input parameters must be higher than a threshold for the machine learning model for the beam quality as an input condition to be met.

[0106] Example 32. The apparatus according to Example 29, wherein at least one measurement sample for cell quality or beam quality has a quality lower than an input threshold configured for the machine learning model for the beam quality as an input condition not to be met.

[0107] Example 33. The apparatus according to Example 29, wherein the second network node must obtain all required measurements as input parameters for one or both of cell quality or beam quality to be met.

[0108] Example 34. The apparatus according to Example 29, wherein the second network node is required to measure at least A of the B antenna panels at the second network node as input parameters, where A <= B, otherwise the input parameters satisfy or do not satisfy one or both of cell quality or beam quality.

[0109] Example 35. An apparatus according to any one of Examples 23 to 26, wherein one or more input conditions include location information input for a second network node.

[0110] Example 36. An apparatus according to Example 35, wherein the uncertainty associated with location information used as input parameters for a machine learning model must be greater than an uncertainty threshold such that the input parameters for the machine learning model do not satisfy uncertainty as an input condition.

[0111] Example 37. The apparatus of Example 36, wherein the uncertainty includes a limitation on the minimum number of visible or used satellites for location information; Example 38. The apparatus of Example 36, wherein the uncertainty includes a location uncertainty greater than Z meters.

[0112] Example 39. An apparatus comprising components for performing the following operations: sending from a first network node to a second network node a configuration for performing a prediction by a machine learning model at the second network node, the configuration including one or more input conditions relating to input parameters or values ​​of input parameters for the machine learning model; and receiving from the second network node, from the first network node, one or both, the output of a prediction performed at the second network node using the machine learning model or information reported from the second network node based on whether the prediction was performed.

[0113] Example 40. An apparatus according to any one of Examples 23 to 39, wherein the first network node includes an access node that provides access to the wireless network by the second network node, and the second network node includes a wireless device capable of connecting to the wireless network via the access node.

[0114] Example 41. An apparatus according to any one of Examples 23 to 39, wherein the first network node includes a base station providing access to a cellular network by a second network node, and the second network node includes user equipment capable of connecting to the cellular network via the base station.

[0115] Example 42. An apparatus of any of the foregoing apparatus examples, wherein the apparatus comprises: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to be executed.

[0116] Example 43. An apparatus comprising: one or more processors; and one or more memories storing instructions that, when executed by the one or more processors, cause the apparatus to perform at least: receiving, by a second network node, a configuration for performing a prediction by a machine learning model from a first network node, the configuration including one or more input conditions relating to input parameters for the machine learning model or the values ​​of the input parameters; determining, by the second network node, whether the input parameters for the machine learning model satisfy one or more input conditions; and, in response to the determination, performing one or both of the following: using the prediction of the machine learning model, or reporting information to the first network node based on whether the prediction has been performed.

[0117] Example 44. The apparatus according to Example 43, wherein: determining includes determining that the input parameters for the machine learning model satisfy one or more input conditions; and performing includes: using the machine learning model to perform a prediction based on determining that the input parameters for the machine learning model satisfy one or more input conditions; and an instruction to report the output of the machine learning model based on performing the prediction.

[0118] Example 45. The apparatus according to Example 43, wherein: determining includes determining that the input parameters for the machine learning model do not satisfy one or more input conditions; and performing includes: not using the machine learning model to perform a prediction based on determining that the input parameters for the machine learning model do not satisfy one or more input conditions; and reporting at least one indication indicating that one or both of the one or more input conditions are not satisfied, or information about the one or more input conditions that are not satisfied.

[0119] Example 46. The apparatus according to Example 43, wherein: determining includes determining that the input parameters for the machine learning model do not satisfy one or more input conditions; and performing includes: using the machine learning model to perform a prediction based on determining that the input parameters for the machine learning model do not satisfy one or more input conditions; reporting an indication of the output of the machine learning model based on performing the prediction; and reporting at least one indication indicating that one or both of the one or more input conditions are not satisfied, or information about the one or more input conditions that are not satisfied.

[0120] Example 47. An apparatus according to any one of Examples 43 to 46, wherein one or more input conditions include at least one requirement for input parameters for machine learning-based predictions by a second network node, and determining includes determining whether the input parameters for the machine learning model satisfy the at least one requirement.

[0121] Example 48. The apparatus according to any one of Examples 43 to 46, wherein: there are N input parameters, and the second network node has one of the following requirements as input conditions for the current prediction: a) for all N input parameters for the current prediction, or b) at least M input parameters, where M < N; or c) specific input parameters; and the determination includes determining whether the input parameters for the machine learning model meet the requirements.

[0122] Example 49. The apparatus according to any one of Examples 43 to 46, wherein one or more input conditions include one or both of cell quality or beam quality, and the determination includes determining whether the input parameters for the machine learning model meet one or both of cell quality or beam quality.

[0123] Example 50. The apparatus according to Example 49, wherein if the selected quality in cell quality or beam quality is higher than the threshold level value, the selected quality is met.

[0124] Example 51. The apparatus according to Example 49, wherein at least X of the N beam quality values used as input parameters must be higher than the threshold for the machine learning model so that the beam quality as an input condition is met.

[0125] Example 52. The apparatus according to Example 49, wherein at least one measurement sample for cell quality or beam quality has a quality lower than the input threshold configured for the machine learning model so that the beam quality as an input condition is not met.

[0126] Example 53. The apparatus according to Example 49, wherein the second network node must obtain all required measurements as input parameters so that one or both of cell quality or beam quality are met.

[0127] Example 54. The apparatus according to Example 49, wherein the second network node is required to measure at least A of the B antenna panels at the second network node as input parameters, where A <= B, otherwise the input parameters meet or do not meet one or both of cell quality or beam quality.

[0128] Example 55. The apparatus according to any one of Examples 43 to 46, wherein one or more input conditions include a position information input for the second network node.

[0129] Example 56. The apparatus according to Example 55, wherein the uncertainty associated with the position information used as an input parameter for the machine learning model must be greater than the uncertainty threshold so that the input parameter for the machine learning model does not meet the uncertainty as an input condition.

[0130] Example 57. The apparatus of Example 56, wherein the uncertainty includes a limitation on the minimum number of visible or used satellites for location information; Example 58. The apparatus of Example 56, wherein the uncertainty includes a location uncertainty greater than Z meters.

[0131] Example 59. An apparatus comprising: one or more processors; and one or more memories storing instructions that, when executed by the one or more processors, cause the apparatus to perform at least: sending from a first network node to a second network node a configuration for performing a prediction by a machine learning model at the second network node, the configuration including one or more input conditions relating to input parameters or values ​​of the input parameters for the machine learning model; and receiving from the second network node, from the first network node, one or both, an output of a prediction performed at the second network node using the machine learning model or information reported from the second network node based on whether the prediction was performed.

[0132] Example 60. An apparatus according to any one of Examples 43 to 59, wherein the first network node includes an access node that provides access to the wireless network by the second network node, and the second network node includes a wireless device capable of connecting to the wireless network via the access node.

[0133] Example 61. An apparatus according to any one of Examples 43 to 59, wherein the first network node includes a base station providing access to a cellular network by a second network node, and the second network node includes user equipment capable of connecting to the cellular network via the base station.

[0134] As used in this application, the term "circuit" may refer to one or more or all of the following: (a) Hardware circuit implementation only (e.g., implemented with purely analog and / or digital circuits), and (b) Combinations of hardware circuitry and software, such as (if applicable): (i) combinations of (multiple) analog and / or digital hardware circuitry and software / firmware, and (ii) any part of a hardware processor having software (including (multiple) digital signal processors, software, and (multiple) memories, which work together to enable a device (such as a mobile phone or server) to perform various functions) and (c) The operation requires software (e.g., firmware) for the operation of (multiple) hardware circuits and / or (multiple) processors, such as (multiple) microprocessors or parts thereof, but the software may be absent when the operation does not require the software.

[0135] This definition of "circuit" applies to all uses of the term in this application. As a further example, as used in this application, the term "circuit" also covers only hardware circuitry or processors (or processors), or portions of hardware circuitry or servers and their accompanying software and / or firmware implementations. For example, where applicable to certain claim elements, the term "circuit" also covers baseband integrated circuits or processor integrated circuits for a UE, or similar integrated circuits in servers, cellular network devices, or other computing or network devices.

[0136] The embodiments described herein can be implemented as software (executed by one or more processors), hardware (e.g., application-specific integrated circuits), or a combination of software and hardware. In example embodiments, software (e.g., application logic, instruction set) is maintained on any of a variety of conventional computer-readable media. In the context of this document, "computer-readable media" can be any medium or component that can contain, store, communicate, propagate, or transmit instructions for use by or in connection with an instruction execution system, apparatus, or device (such as a computer), wherein an example of a computer is, for example, in... Figure 4 The computer-readable medium may include computer-readable storage media (e.g., memories 15, 75, and 95 or other devices), which may be any medium or component that can contain, store, and / or transmit instructions for use by or in connection with an instruction execution system, apparatus, or device (such as a computer). Computer-readable storage media do not include propagating signals and can therefore be considered non-transitory. As used herein, the term "non-transitory" is a limitation of the medium itself (i.e., tangible, not signaling), not a limitation of the persistence of data storage (e.g., RAM, random access memory, as opposed to ROM, read-only memory).

[0137] If necessary, the different functions discussed in this article can be executed in different orders and / or simultaneously with each other. Furthermore, if necessary, one or more of the above functions can be optional or can be combined.

[0138] Although various aspects of the invention are set forth in the independent claims, other aspects of the invention include other combinations of features from the described embodiments and / or dependent claims with features of the independent claims, and not just combinations expressly set forth in the claims.

[0139] This document also notes that while exemplary embodiments of the invention have been described above, these descriptions should not be construed as limiting. Rather, various changes and modifications may be made without departing from the scope of the invention as defined in the appended claims.

[0140] The following abbreviations, which can be found in the instruction manual and / or accompanying drawings, are defined as follows:

Claims

1. A method comprising: The second network node receives from the first network node a configuration for performing predictions by a machine learning model, the configuration including one or more input conditions related to input parameters for the machine learning model or the values ​​of the input parameters; The second network node determines whether the input parameters for the machine learning model satisfy one or more input conditions; as well as In response to the determination, the second network node performs one or both of the following: using the prediction of the machine learning model, or reporting information to the first network node based on whether the prediction has been performed.

2. An apparatus comprising components for performing the following: The second network node receives from the first network node a configuration for performing predictions by a machine learning model, the configuration including one or more input conditions related to input parameters for the machine learning model or the values ​​of the input parameters; The second network node determines whether the input parameters for the machine learning model satisfy one or more input conditions; as well as In response to the determination, the second network node performs one or both of the following: using the prediction of the machine learning model, or reporting information to the first network node based on whether the prediction has been performed.

3. The apparatus according to claim 2, wherein: The determination includes determining that the input parameters for the machine learning model satisfy one or more input conditions; and The execution includes: The prediction is performed using the machine learning model based on the determination that the input parameters for the machine learning model satisfy one or more input conditions; as well as Instructions to report the output of the machine learning model based on the execution of the prediction.

4. The apparatus according to claim 2, wherein: The determination includes determining that the input parameters for the machine learning model do not satisfy one or more input conditions; and The execution includes: Based on the determination that the input parameters for the machine learning model do not satisfy one or more input conditions, the machine learning model is not used to perform the prediction; as well as The report indicates at least one indication or information regarding the non-metd one or more input conditions.

5. The apparatus according to claim 2, wherein: The determination includes determining that the input parameters for the machine learning model do not satisfy one or more input conditions; and The execution includes: The prediction is performed using the machine learning model based on the determination that the input parameters for the machine learning model do not satisfy one or more input conditions; Instructions to report the output of the machine learning model based on the execution of the prediction; and The report indicates at least one indication or information regarding the non-metd one or more input conditions.

6. The apparatus according to any one of claims 2 to 5, wherein the one or more input conditions include at least one requirement for the input parameter for machine learning-based prediction by the second network node, and the determination includes determining whether the input parameter for the machine learning model meets the at least one requirement.

7. The apparatus according to any one of claims 2 to 5, wherein: there are N input parameters, and the second network node has one of the following requirements as input conditions for the current prediction: a) all N input parameters for the current prediction, or b) at least M input parameters, where M < N; or c) a specific input parameter; and the determination includes determining whether the input parameter for the machine learning model meets the requirement.

8. The apparatus according to any one of claims 2 to 5, wherein the one or more input conditions include one or both of cell quality or beam quality, and the determination includes determining whether the input parameter for the machine learning model meets the one or both of the cell quality or the beam quality, and wherein: if the selected quality in the cell quality or the beam quality is higher than the threshold level value, the selected quality is met; or at least X of the N beam quality values used as input parameters must be higher than the threshold for the machine learning model to meet the beam quality as an input condition; or at least one measurement sample for the cell quality or the beam quality has a quality lower than the input threshold configured for the machine learning model so that the beam quality as an input condition is not met; or the second network node must obtain all required measurements as input parameters to meet one or both of the cell quality or the beam quality; or the second network node is required to measure at least A of the B antenna panels at the second network node as input parameters, where A <= B, otherwise the input parameter meets or does not meet one or both of the cell quality or the beam quality.

9. The apparatus according to any one of claims 2 to 5, wherein the one or more input conditions include a position information input for the second network node.

10. The apparatus according to claim 9, wherein the uncertainty associated with the position information used as an input parameter for the machine learning model must be greater than the uncertainty threshold for the input parameter for the machine learning model not to meet the uncertainty as an input condition.

11. The apparatus according to claim 10, wherein the uncertainty includes a limit on the minimum number of satellites visible or used for the position information.

12. The apparatus according to claim 10, wherein the uncertainty includes a position uncertainty greater than Z meters.

13. An apparatus, comprising components for performing the following: Sending configuration from a first network node to a second network node for performing predictions by a machine learning model at the second network node, the configuration including one or more input conditions related to input parameters of the machine learning model or the values ​​of the input parameters; and The first network node receives from the second network node either the output of a prediction performed on the second network node using the machine learning model, or information reported from the second network node based on whether the prediction was performed, or both.

14. The apparatus of any one of claims 2 to 13, wherein the first network node includes an access node that provides access to the wireless network by the second network node, and the second network node includes a wireless device capable of connecting to the wireless network via the access node.

15. The apparatus of any one of claims 2 to 13, wherein the first network node includes a base station providing access to a cellular network by the second network node, and the second network node includes a user equipment capable of connecting to the cellular network via the base station.

Images