Prediction functionality

The implementation of a recovery mode for AI/ML-based beam prediction in mobile communication systems addresses the challenge of temporary performance degradation by enabling selective activation/deactivation based on performance thresholds, ensuring reliable operation and reducing unnecessary deactivation.

WO2026233760A1PCT designated stage Publication Date: 2026-11-12NOKIA TECHNOLOGIES OY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
PCT/EP2026/061914
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2025-05-09
Filing Date
2026-04-24
Publication Date
2026-11-12

AI Technical Summary

Technical Problem

Existing mobile communication systems lack effective procedures for managing performance degradation and failure recovery of artificial intelligence/machine learning (AI/ML) based beam prediction functionalities, particularly in temporary circumstances such as signal blockages, leading to unnecessary deactivation of these functionalities.

Method used

Implementing a recovery mode for AI/ML-based prediction functionalities in user equipment (UE) and network nodes, utilizing failure detection and recovery procedures based on performance thresholds and time periods to differentiate between temporary and permanent performance degradation, allowing for selective activation or deactivation of these functionalities.

Benefits of technology

Enables efficient and adaptive management of AI/ML-based beam prediction systems, preventing unnecessary deactivation during temporary performance issues and ensuring reliable operation by distinguishing between transient and permanent failures.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure EP2026061914_12112026_PF_FP_ABST
    Figure EP2026061914_12112026_PF_FP_ABST
Patent Text Reader

Abstract

This specification describes a first apparatus comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the first apparatus to: determine a recovery time period for a prediction functionality, wherein the prediction functionality is implemented by one or more artificial intelligence or machine learning (AI / ML) models, wherein the recovery time period comprises a recovery evaluation time period during which the first apparatus performs performance evaluation of the prediction functionality; determine whether or not a recovery condition is satisfied during the recovery evaluation time period; and transmit, to a second apparatus, an indication indicating whether or not the prediction functionality is applicable within the recovery time period.
Need to check novelty before this filing date? Find Prior Art

Description

iPREDICTION FUNCTIONALITYCROSS-REFERENCE TO RELATED APPLICATION

[0001] This application claims the benefit of EP application No. 25175348.9, filed May 09, 2025. The content of which are hereby incorporated by reference in their entirety.FIELD

[0002] Example embodiments relate to user equipment, network nodes, methods and / or computer programs for use in mobile communication systems. More particularly, the example embodiments relate to prediction functionalities, such as artificial intelligence or machine learning models (AIZML)for use in mobile communication systems.BACKGROUND

[0003] Prediction functionalities may be used for various functions in mobile communication systems. AI / ML based beam management is one of the key topics in 3GPP Rel-19 work item on AI / ML in air interface. Two main use-cases are spatial domain beam prediction (BM-Case1) and time domain beam prediction (BM-Case2). The scope of spatial beam prediction (BM-Case1 ) is to predict the best Tx (or pair of Tx and Rx) beams from a set of transmitted BS Tx beams (referred as SetA) out of smaller sub-set of transmitted beams (SetB). Time-domain beam prediction (BM-Case2) aims on predicting the best beam to be use in the next time instant or instants.

[0004] There remains a need for improvement in procedure for using prediction functionalities.SUMMARY

[0005] According to some aspects, there is provided the subject matter of the independent claims. Some further aspects are defined in the dependent claims.BRIEF DESCRIPTION OF DRAWINGS

[0006] Some example embodiments will now be described with reference to the accompanying drawings.

[0007] FIGs. 1 and 2 are block diagrams of systems in accordance with example embodiments;

[0008] FIGs. 3 is a block diagrams of an example scenarios;

[0009] FIG. 4 is a block diagram of example scenarios in accordance with example embodiments;

[0010] FIGs. 5 to 6 are flowcharts of algorithms in accordance with example embodiment;

[0011] FIGs. 7 to 8 are flowcharts of algorithms in accordance with example embodiment;

[0012] FIG. 9 is a block diagram of an example scenario in accordance with an example embodiments;

[0013] FIG.10 is a schematic diagram of a system that may be used to implement one or more of the example embodiments;

[0014] FIG. 11 shows tangible media for storing computer-readable code which when run by a computer may perform methods according to example embodiments described herein.DETAILED DESCRIPTION

[0015] The scope of protection sought for various embodiments of the invention is set out by the independent claims. The embodiments and features, if any, described in this specification that do not fall under the scope of the independent claims are to be interpreted as examples useful for understanding various embodiments of the invention.

[0016] FIG. 1 is a block diagram of an example system, indicated generally by the reference numeral 10, in accordance with an example embodiment.

[0017] The system 10 comprises a first apparatus 11 , such as a user equipment (UE), and a second apparatus 12, such as a network node or base station (gNB). The first apparatus may be in a mobile communications network associated with the second apparatus.

[0018] The UE 11 may comprise a prediction functionality 13, where the prediction functionality 13 may be implemented by one or more artificial intelligence or machine learning (AI / ML) models. The term ‘prediction functionality’ may be interchangeably referred to as ‘AI / ML feature / capability / functionality / model / configuration’ herein. A first apparatus may be interchangeably referred to as a UE herein. The second apparatus may be interchangeably referred to as a base station (gNB), or a network node (NW) herein.

[0019] A prediction model may be a machine learning (ML) model. The ML model may be used within a wireless network to perform (or assist with performing) one or more tasks. In general, one or more nodes (e.g., BS, gNB, eNB, RAN node, user node, UE, user device, relay node, or other wireless node) within a wireless network may use or employ a ML model, e.g., such as, for example a neural network model (e.g., which may be referred to as a neural network, an artificial intelligence (Al) neural network, an Al neural network model, an Al model, a machine learning (ML) model or algorithm, a model, or other term) to perform, or assist in performing, one or more ML-enabled tasks. Other types of models may also be used. A ML-enabled task may include tasks that may be performed (or assisted in performing) by a ML model, or a task for which a ML model has been trained to perform or assist in performing).

[0020] ML-based algorithms or ML models may be used to perform and / or assist with performing a variety of wireless and / or radio resource management (RRM) and / or RAN related functions or tasks to improve network performance, such as, e.g., in the UE for beam prediction (e.g., predicting a best beam or bestbeam pair based on measured reference signals), antenna panel or beam control, RRM (radio resource measurement) measurements and feedback (channel state information (CSI) feedback), link monitoring, Transmit Power Control (TPC), etc. In some cases, ML models may be used to improve performance of a wireless network in one or more aspects or as measured by one or more performance indicators or performance criteria.

[0021] Models (e.g., neural networks or ML models) may be or may include, for example, computational models used in machine learning made up of nodes organized in layers. The nodes are also referred to as artificial neurons, or simply neurons, and perform a function on provided input to produce some output value. A neural network or ML model may typically require a training period to learn the parameters, i.e., weights, used to map the input to a desired output. The mapping may occur via the function that is learned from a given data for the problem in question. Thus, the weights are weights for the mapping function of the neural network. Each neural network model or ML model may be trained for a particular task.

[0022] To provide the output given the input, the ML functionality of a neural network model or ML model should be trained, which may involve learning the proper value for a large number of parameters (e.g., weights and / or biases) for the mapping function (or of the ML functionality of the ML model). For example, the parameters may be used to weight and / or adjust terms in the mapping function. This training may be an iterative process, with the values of the weights and / or biases being tweaked over many (e.g., tens, hundreds and / or thousands) of rounds of training episodes or training iterations until arriving at the optimal, or most accurate, values (or weights and / or biases). In the context of neural networks (neural network models) or ML models, the parameters may be initialized, often with random values, and a training optimizer iteratively updates the parameters (e.g., weights) of the neural network to minimize error in the mapping function. In other words, during each round, or step, of iterative training the network updates the values of the parameters so that the values of the parameters eventually converge to the optimal values.

[0023] ML models may be trained in either a supervised or unsupervised manner, as examples. In supervised learning, training examples are provided to the ML model or other machine learning algorithm. A training example includes the inputs and a desired or previously observed output. Training examples are also referred to as labeled data because the input is labeled with the desired or observed output. In the case of a neural network (which may be a specific case of ML model), the network (or ML model) learns the values for the weights used in the mapping function or ML functionality of the ML model that most often result in the desired output when given the training inputs. In unsupervised training, the ML model learns to identify a structure or pattern in the provided input. In other words, the model identifies implicit relationships in the data. Unsupervised learning is used in many machine learning problems and typically requires a large set of unlabeled data.

[0024] According to an example embodiment, a ML model may be classified into (or may include) two broad categories (supervised and unsupervised), depending on whether there is a learning “signal” or “feedback” available to a model. Thus, for example, within the field of machine learning, there may be two main types of learning or training of a model: supervised, and unsupervised. The main difference between the two types is that supervised learning is done using known or prior knowledge of what the output values for certain samples of data should be. Therefore, a goal of supervised learning may be to learn a function that, given a sample of data and desired outputs, best approximates the relationship between input and output observable in the data. Unsupervised learning, on the other hand, does not have labeled outputs, so its goal is to infer the natural structure present within a set of data points.

[0025] In an example, a prediction may be based on a prediction model. The prediction model or the ML based algorithm may include a model or an algorithm based an artificial intelligence and / or machine learning (AI / ML).

[0026] In an example, artificial intelligence and / or machine learning (AI / ML) techniques may be implemented to improve the performance of wireless communication systems. The implementation of the AI / ML (techniques) may include implementation of mechanisms at the network side and / or the UE side. For example, the AI / ML techniques may enhance data collection for NR and dual connectivity scenarios. The AI / ML model, herein, is exchangeable with prediction model, Al and ML model, Al or ML model, Al model, or ML model.

[0027] FIG. 2 is a diagram illustrating functional framework for radio access network intelligence based on AI / ML. In an example, data collection may be a function that provides input data to model training and model inference functions. AI / ML algorithm specific data preparation (e.g., data pre-processing and cleaning, formatting, and transformation) may not be carried out in the data collection function. In an example, input data may include measurements from UEs or different network entities, feedback from actor, output from an AI / ML model, and / or the like. In an example, training data may include the data needed as input for the AI / ML model training function.

[0028] In an example, inference data may include the data needed as input for the AI / ML model inference function. In an example, model training may be a function that performs the AI / ML model training, validation, and testing which may generate model performance metrics as part of the model testing procedure. The model training function may perform data preparation (e.g., data pre-processing and cleaning, formatting, and transformation) based on training data delivered by a data collection function. In an example, model deployment / update may be employed to initially deploy a trained, validated, and tested AI / ML model to the model inference function or to deliver / provide an updated model to the model inference function. In an example, the model inference may be a function that provides AI / ML model inference output(e.g. , predictions or decisions). In an example, the model inference function may provide model performance feedback to model training function. The model inference function may perform data preparation (e.g., data pre-processing and cleaning, formatting, and transformation) based on inference data delivered / provided by a data collection function. In an example, the output may include the inference output of the AI / ML model produced by a model inference function. In an example, the actor may include a function that receives the output from the model inference function and triggers or performs corresponding actions. The actor may trigger actions directed to other entities (of the AI / ML model, network, and / or the like) or to itself. In an example, the feedback may include information that may be needed to derive training data, inference data or to monitor the performance of the AI / ML model and its impact to the network through updating of key performance indicators (KPIs) and performance counters.

[0029] The techniques described herein may be applied to a wide variety of wireless technologies or wireless networks, such as 5G (New Radio (NR)), cmWave, and / or mmWave band networks, loT, MTC, eMTC, eMBB, URLLC, 6G, etc., or any other wireless network or wireless technology. These example networks, technologies or data service types are provided only as illustrative examples.

[0030] 3GPP Rel-19 Wl (Work Item) on AI / ML for NR Air Interface [RP-243244] is based on the study of AI / ML techniques in NR air interface from Rel-18 FS_NR_AIML_Air with the outcomes captured in [TR 38.843],

[0031] The terms ‘prediction functionality’, ‘AI / ML features’, ‘AI / ML models’, ‘AI / ML configurations’, ‘AI / ML functionality’, have been used interchangeably throughout the description, and are intended to have the same broad scope as the term ‘prediction functionality’.

[0032] FIG. 3 is a block diagram of an example scenario (representing signals in time and / or frequency domain), indicated generally by the reference numeral 300. The scenario 300 shows an illustration of measurements and reporting without the use of a prediction functionality.

[0033] Measurements 301 may represent reference signal (RS) measurements. For example, when a UE identifies new reference signals, the UE may perform measurements of the received RS, for example, with a RS periodicity 304. A measurement period 303 may be a reporting delay, that may be caused due to the need for the UE to sweep beams (e.g. Tx beams) transmitted by a gNB with multiple beams (Rx beams) received by the UE.

[0034] Signals 302 may represent reporting of measurement results, where the UE may report best measured beams to the gNB. The measurement reporting may have a reporting periodicity 305.

[0035] The scenario 300 relate to beam management (BM) (e.g. for L1-RSRP measurements in FR2) as an illustrative use-case. However, the similar principles may be applicable for the other use-cases as well, e.g., for CSI prediction.

[0036] Some requirements for the reporting delay are defined in TS 38.133 (Clause 9.5.4 for L1 -RSRP in FR2, Table 9.5.4.1-2 for SSB-based measurements in frequency 2 (FR2)), as follows:• For non-DRX measurements, UE shall be capable of performing L1-RSRP measurements based on the configured SSB resource for L1-RSRP computation, and the UE physical layer shall be capable of reporting L1-RSRP measured over the measurement period ofT_L 1 RSRP_Measurement_Period_SSB• T_L1-RSRP_Measurement_Period_SSB = max(TReport, ceil(M*P*N)*TSSB• M=1 if higher layer parameter timeRestrictionForChannelMeasurement is configured, and M=3 otherwise• N=8 for FR2• The value of P depends on the measurement gap configuration.• TReport is configured periodicity for reporting.

[0037] Similarly, CSI-RS measurement reporting delay is defined TS 38.133 (Clause 9.5.4.2, Table 9.5.4.2-2), such that the value of N for CSI-RS measurement reporting may be different (smaller in most cases) than 8.

[0038] Although beam failure recovery requirements exist in RAN4, there is no requirements yet defined in RAN4 for AI / ML model / functionality / capability failure recovery. Performance degradation, such as beam failure, may lead to failure of a prediction functionality (AI / ML model / functionality / capability) failure.However, legacy techniques may require disabling the prediction functionality regardless of whether or not the performance degradation and / or prediction functionality failure is temporary. For example, if the performance degradation and AI / ML functionality failure is due to temporary circumstances, such as a temporary blockage obstructing signals to the UE, it may not be desirable to indefinitely disable or deactivate the AI / ML functionality since the blockage can move away and the signals to the UE may be resumed. Alternatively, in a scenario where the circumstances / conditions may have changed for indefinite period, it may be desirable to disable or deactivate the AI / ML functionality until some other event is triggered such as, UE switch off or site change, or the like.

[0039] Example embodiments described herein provide procedures for determining whether or not the performance degradation and / or prediction functionality failure is temporary, and failure recovery procedures for prediction functionalities.

[0040] FIG. 4 is a block diagram of an example scenario (representing signals in time and / or frequency domain), indicated generally by the reference numeral 400, in accordance with example embodiments. Scenario 400 provides an illustration of recovery procedure for a prediction functionality. Illustrated is reference signals or measurements 401 in a spatial domain, and corresponding monitoring metrics 402corresponding to a prediction functionality. When the prediction functionality is being used or is activated, the RS / measurements 401 may comprise measurement RS set for prediction (Set B) 409 (a, b, c, d) and / or monitoring RS set (e.g. Set A- SetB) 410 (a, b, c), where a monitoring periodicity is indicated by 403, and a measurement periodicity is indicated by 404. The elements of FIG. 4 are described further below in conjunction with FIG. 5.

[0041] FIG. 5 is a flowchart of an algorithm, indicated generally by the reference numeral 500, in accordance with an example embodiment. The operations of the algorithm 500 may be performed at a first apparatus, such as a user equipment (UE). FIG. 5 is described in conjunction with FIG. 4 for better understanding.

[0042] At operation 501 , the first apparatus may receive a first configuration from a second apparatus, wherein the first configuration relates to a recovery mode for a prediction functionality. The prediction functionality may be implemented by one or more artificial intelligence or machine learning (AI / ML) models (e.g. AI / ML feature / capability / functionality / model / configuration).

[0043] Next, at operation 502, the UE may enable the recovery mode by performing failure recovery based on the first configuration. The first configuration may comprise at least one of:• a failure detection time period, such as the failure evaluation period 414;• a recovery time period, such as recovery period (T R) 412;• a recovery evaluation time period, such as recovery evaluation period (TR_EV3I) 415;• a first performance threshold of failure recovery for the prediction functionality, such as threshold (Thout) 405;• a second performance threshold of failure recovery for the prediction functionality, such as threshold (Thin) 413; or• one or more reference signal resources, such as RS 411 , for performing measurements in the recovery mode.

[0044] In some examples, the first apparatus may perform failure recovery based on the first configuration by first detecting a failure in the prediction functionality during the failure detection time period. For example, detecting a failure may comprise performing, during the failure detection (evaluation) time period 414, an evaluation of the prediction functionality based on a comparison of a performance of the prediction functionality with the first performance threshold. For example, the first apparatus may determine that the failure has occurred in the prediction functionality if the performance of the prediction functionality is determined to be lower than the first performance threshold (405). For example, the monitoring metric 406a is higher than the first performance threshold 405, but the monitoring metric later deteriorates, as shown by monitoring metrics 406b and 406c. As monitoring metrics 406b and 406c are lower than the firstperformance threshold 405, the UE may determine that a failure has occurred in the prediction functionality, for example, at the time instance 407. In some examples, the failure detection period may start when a monitoring metric lower than the first performance threshold 405 is first detected. In one example, if the monitoring metrics became higher than the first performance threshold 405 before the expiry of the failure detection period, prediction functionality may not be considered to have failed, and may be continued to be used for predictions.

[0045] In some examples, the recovery period (TR) 412 may start after the failure at time instance 408. The operation for failure recovery may further comprise performing failure recovery of the prediction functionality during the recovery time period (TR) 412. In some examples, the recovery time period (TR) 412 comprises a recovery evaluation time period (TR_EV3I) 415. For example, the first apparatus may perform failure recovery based on the first configuration by performing, during the evaluation time period (TR_EV3I) 415, an evaluation of the prediction functionality based on a comparison of a performance of the prediction functionality with the second performance threshold (Thin) 413. For example, the monitoring metric 406d is lower than the second performance threshold 413, but the monitoring metric later improves, as shown by monitoring metrics 406e, 406f, and 406g. As monitoring metrics 406e, 406f, and 406g are higher than the second performance threshold (Thin) 413, the UE may determine that the failure of prediction functionality is no longer applicable, and the performance of the prediction functionality has improved, and therefore the prediction functionality is recovered. The UE may then enable an active mode for the prediction functionality if, within the recovery time period, the performance of the prediction functionality is determined to be equal to or higher than the second performance threshold. In some examples, the evaluation time period (TR Evai) 415 may start when a monitoring metric higher than the second performance threshold 413 is first detected. In one example, if the monitoring metrics became lower than the second performance threshold 413 before the expiry of the evaluation time period (TR_EV3I) 415, the prediction functionality may not be considered to have recovered.

[0046] Alternatively, the UE may deactivate (e.g. indefinitely, or at least until an event or trigger) the prediction functionality if, within the recovery time period (TR) 412, the performance of the prediction functionality is determined to be lower than the second performance threshold 413.

[0047] In some examples, the first performance threshold 405 may be higher than the second performance threshold 413. As such, the first performance threshold 405 for entering a recovery mode is lower than the second performance threshold 413 for re-activating the prediction functionality.

[0048] In some examples, the first apparatus may determine the performance of the prediction functionality by performing recovery verification using the one or more reference signal resource (412) for performing measurements in the recovery mode. For example, the one or more reference signalresources for performing measurements in the recovery mode may comprise at least one of: reference signals for performing inference measurements; or measurements, suitable for recovery verification, provided by one or more active functionalities associated with the first apparatus.

[0049] In some examples, when the prediction functionality is in recovery period 412, measurements may be performed and / or reported based on full RS set (411) rather than predicted measurements. Once the prediction functionality is considered to be recovered at time instance 408, the RS / measurements may again be based on monitoring RS set 409 and / or measurement RS set 410.

[0050] In some examples, the first apparatus may perform beam predictions or beam inference using the prediction functionality. The beam predictions may comprise predicted at least one reference signal measurement. The first apparatus may then obtain at least one reference signal measurement. The first apparatus may then compare the beam predictions or beam inference with the at least one reference signal measurements. The performance of the prediction functionality may be determined based on said comparison.

[0051] In some examples, the first configuration is received via a radio resource control message from the second apparatus and the radio resource control message relates to an initial configuration for the first apparatus.

[0052] In some examples, the first apparatus may send a first indication to the second apparatus, indicating that the prediction functionality is entered into an active mode if failure recovery is successful, wherein in the active mode, the prediction functionality is applicable. The first indication may comprise one or more of a user equipment (UE) assistance information (UAI) or a RRC reconfiguration complete message.

[0053] In some examples, the prediction functionality is suitable at least for beam prediction. As such, the first apparatus may pause sending beam prediction results to the second apparatus while the first apparatus is in the recovery mode for the prediction functionality (e.g. during recovery period 412). The first apparatus may resume sending beam prediction results to the second apparatus while the first apparatus is in the active mode for the prediction functionality. As such, for example, during the recovery period 412, the UE may use the RS 411 for obtaining beam prediction results, but these beam prediction results may not be sent to the second apparatus. These beam prediction results obtained during the recovery period 412 may only be used for recovery verification (e.g. verifying whether the prediction functionality has recovered).

[0054] FIG. 6 is a flowchart of an algorithm, indicated generally by the reference numeral 600, in accordance with an example embodiment. The operations of the algorithm 600 may be performed at aiosecond apparatus, such as a base station or network node (gNb / NW). FIG. 6 is described in conjunction with FIG. 4 for better understanding.

[0055] At operation 601 , the second apparatus may send a first configuration to a first apparatus, wherein the first configuration relates to a recovery mode for a prediction functionality, the prediction functionality is implemented by one or more artificial intelligence or machine learning (AI / ML) models (e.g. AI / ML feature / capability / functionality / model / configuration).

[0056] Next, at operation 602, the second apparatus may receive a first indication from the first apparatus indicating the applicability of the prediction functionality.

[0057] In some examples, the first configuration comprises at least one of:• a failure detection time period, such as the failure evaluation period 414;• a recovery time period, such as recovery period (T R) 412;• a recovery evaluation time period, such as recovery evaluation period (TR_EV3I) 415;• a first performance threshold of failure recovery for the prediction functionality, such as threshold (Thout) 405;• a second performance threshold of failure recovery for the prediction functionality, such as threshold (Thin) 413; or• one or more reference signal resources, such as RS 411 , for performing measurements in the recovery mode.

[0058] In some examples, the one or more reference signal resources (e.g. RS 411) for performing measurements in the recovery mode may comprise at least one of: reference signals for performing inference measurements; or measurements, suitable for recovery verification, provided by one or more active functionalities associated with the first apparatus.

[0059] In some examples, the first indication comprises one or more of a user equipment (UE) assistance information (UAI) or a RRC reconfiguration complete message.

[0060] In some examples, the first configuration is sent, to the first apparatus, via a radio resource control message, and the radio resource control message relates to an initial configuration for the first apparatus.

[0061] FIG. 7 is a flowchart of an algorithm, indicated generally by the reference numeral 700, in accordance with an example embodiment. The operations of the algorithm 700 may be performed at a first apparatus, such as a UE.

[0062] At operation 701 , the first apparatus may determine a recovery time period (e.g. recovery time period 412) for a prediction functionality, wherein the prediction functionality is implemented by one or more artificial intelligence or machine learning (AI / ML) models. The recovery time period (412) may comprise anrecovery evaluation time period (415) during which the first apparatus performs performance evaluation of the prediction functionality.

[0063] Next, at operation 702, the UE may determine whether or not a recovery condition is satisfied during the recovery evaluation time period (415).

[0064] At operation 703, the UE may transmit, to a second apparatus (e.g. gNB), an indication indicating whether or not the prediction functionality is applicable within the recovery time period (412).

[0065] In some examples, the recovery time period is determined based on the recovery evaluation time period and a first processing time. For example, the recovery time period is based on an equation: T = K x Ceil((T1+D) / K), where T is the recovery time period, T1 is the recovery evaluation time period, K is a scaling factor to T1 , and D is a first processing time.

[0066] In some examples, the UE shall be capable of transmitting re-applicability indication to the gNB within a period of T (recovery time period). For example, the UE shall be able to evaluate whether the prediction functionality (AI / ML-enabled feature) inference / prediction quality estimated over the recovery time period becomes better than the second performance threshold. The UE may determine the monitoring metric (performance metric), by comparing the predictions / inferences with the measurement of configured monitoring RS resources during this duration.

[0067] In some examples, the first apparatus may receive the first configuration from the second apparatus, wherein the first configuration relates to a recovery mode for the prediction functionality.

[0068] In some examples, the first configuration comprises at least one of:• a failure detection time period, such as the failure evaluation period 414;• a recovery time period, such as recovery period (T R) 412;• a recovery evaluation time period, such as recovery evaluation period (TR_EV3I) 415;• a first performance threshold of failure recovery for the prediction functionality, such as threshold (Thout) 405;• a second performance threshold of failure recovery for the prediction functionality, such as threshold (Thin) 413; or• one or more reference signal resources, such as RS 411 , for performing measurements in the recovery mode.

[0069] In some examples, the first apparatus may perform failure recovery during the recovery time period (412) based on the first configuration. For example, the first apparatus may perform failure recovery by detecting a failure has occurred in the first model during the failure detection time period, and then by performing failure recovery of the first model during the recovery time period.

[0070] In some examples, determining whether the recovery condition is satisfied comprises performing, during a failure detection time period (414), an evaluation of the prediction functionality based on a comparison of a performance of the prediction functionality with a first performance threshold (405). For example, the first apparatus may determine that the failure has occurred in the prediction functionality if the performance of the prediction functionality is determined to be lower than the first performance threshold (405).

[0071] In some examples, the first apparatus may perform failure recovery based on the first configuration by performing, during the evaluation time period (TR_EVai) 415, an evaluation of the prediction functionality based on a comparison of a performance of the prediction functionality with the second performance threshold (Thin) 413.

[0072] The UE may then enable an active mode for the prediction functionality if, within the recovery time period, the performance of the prediction functionality is determined to be equal to or higher than the second performance threshold.

[0073] Alternatively, the UE may deactivate (e.g. indefinitely, or at least until an event or trigger) the prediction functionality if, within the recovery time period (TR) 412, the performance of the prediction functionality is determined to be lower than the second performance threshold 413.

[0074] In some examples, the first apparatus may determine the performance of the prediction functionality by performing recovery verification using the one or more reference signal resource (412) for performing measurements in the recovery mode. For example, the one or more reference signal resources for performing measurements in the recovery mode may comprise at least one of: reference signals for performing inference measurements; or measurements, suitable for recovery verification, provided by one or more active functionalities associated with the first apparatus.

[0075] FIG. 8 is a flowchart of an algorithm, indicated generally by the reference numeral 800, in accordance with an example embodiment. The operations of the algorithm 800 may be performed at a second apparatus, such as a gNB (network node, NW).

[0076] At operation 801 , the second apparatus may send, to a first apparatus, a first configuration, wherein the first configuration relates to a recovery mode for prediction functionality, wherein the prediction functionality is implemented by one or more artificial intelligence or machine learning (AI / ML) models.

[0077] Next, at operation 802, the second apparatus may receive, from the first apparatus, an indication indicating whether or not the prediction functionality is applicable within the recovery time period. The indication may, for example, be carried in a UE assistance information (UAI) or a RRC reconfiguration complete message

[0078] In some examples, the first configuration comprises at least one of:• a failure detection time period, such as the failure evaluation period 414;• a recovery time period, such as recovery period (T R) 412;• a recovery evaluation time period, such as recovery evaluation period (TR_EV3I) 415;• a first performance threshold of failure recovery for the prediction functionality, such as threshold (Thout) 405;• a second performance threshold of failure recovery for the prediction functionality, such as threshold (Thin) 413; or• one or more reference signal resources, such as RS 411 , for performing measurements in the recovery mode; the one or more reference signal resources for performing measurements in the recovery mode comprises at least one of: reference signals for performing inference measurements; or measurements, suitable for recovery verification, provided by one or more active functionalities associated with the first apparatus.

[0079] FIG. 9 is a message flow indicating a procedure for messages transmitted between a first apparatus (UE 91) and a second apparatus (gNB 92), according to example embodiments.

[0080] The gNB 92 may send a first configuration (RRCReconfiguration) to the UE 91 at operatiom 901. The first configuration may comprise a full CSI-ReportConfig(l) for beam prediction, comprising a recovery mode configuration (AIMLFailureRecoveryConfig). In one example, the MAC entity for the UE may be configured by RRC per AI / ML Fu nctionality / model / capability with an AI / ML failure recovery procedure. The AIMLFailureRecoveryConfig may be reconfigured by upper layers during RRC Reconfiguration. When a prediction functionality (AI / ML) is deactivated, the UE may perform AI / ML failure recovery procedure. The recovery mode configuration may comprise one or more of:• AIMLFailureDetectionTimer for the AI / ML failure detection per AI / ML fu nctionality / model / capability (e.g. failure detection period 414)• AIMLFailureRecoveryTimer for the AI / ML failure recovery procedure per AI / ML functionality / model / capability (e.g. recovery time period 412)• performance-ThresholdAIMLFR: a performance threshold for the AIML failure recovery (e.g. first performance threshold 405 and / or second performance threshold 413).

[0081] The UE 91 may send an indication 902 (RRCReconfigurationComplete) indicating that the configuration for recovery mode is complete on the UE side. The indication may indicate identity of the CSI report configuration, such as CSI-ReportConfigld=1 (i.e. ID of the CSI beam report is (1)). A value of one or more fields of the CSI-ReportConfig may indicate applicability of the prediction model and / or the recovery mode configuration.

[0082] Considering that the prediction functionality is applicable to the UE in general, the UE may activate (903) the CS I -Report Confi g (1 ), and may transmit (904) a CSI beam report (1 ) to the gNB. If a failure is detected, the UE may determine (905) that the CSI beam report (1) is inapplicable. Once it is determined that the CSI beam Report (1) is inapplicable, the UE may send (906) a UAI (UE assistance information) to indicate to the gNB that the CSI beam report (1) is inapplicable. Due to the failure of the prediction functionality, the UE may experience ambiguities in inference reporting (907).

[0083] Next, the gNB may respond to the UAI by sending (908) an RRC reconfiguration message. The RRC reconfiguration message may optionally comprise recovery mode configuration for CSI-ReportConfig(l). RRCreconfiguration message sent at operation 908 may be used by the gNB for reconfiguring the UE for an alternative functionality (AI / ML or non AI / ML) (e.g. due to the failure of a primary prediction functionality) and optionally for configuring failure recovery for failed AI / ML functionality, if needed. Alternatively, if the recovery mode configuration is comprised in the initial RRCReconfiguration (901), the recovery mode configuration may not need to be included in the RRC reconfiguration message of operation 908.

[0084] The UE 91 may then enable (909) the prediction functionality (CSI-ReportConfig(1 )) in recovery mode. If the performance of the prediction functionality recovers within the recovertime period, the CSI beam Report (1) may become applicable again, as determined by the UE in operation 910. If the CSI beam Report (1) is applicable again, the UE may send another UAL 911 to the gNB indicating that that the CSI beam report (1) is again applicable (CSI-ReportConfigld=1 : (re)applicability).

[0085] The gNB 92 may then send a RRCReconfiguraiton (912) comprising a full CSI -ReportCo nfi g (1 ) for beam prediction. The UE may then acknowledge the configuration by sending message 913 comprising indication RRCReconfigurationComplete (CSI-ReportConfigld= 1 applicability).

[0086] In case the prediction functionality is not recovered before the recovery time period expires, the UE 91 may deactivate (914) the prediction functionality (CSI-ReportConfig(1 )) definitely (or at least until an event or trigger occurs)Example Apparatus

[0087] FIG. 10 shows an apparatus according to some example embodiments, which may comprise the user terminal 100. The user equipment may be, for example a smartphone, or other terminal device. The apparatus may be configured to perform the operations described herein, for example operations described with reference to any disclosed process. The apparatus comprises at least one processor 102 and at least one memory 104 directly or closely connected to the processor. The memory 104 includes at least one random access memory (RAM) and at least one read-only memory (ROM). Computer program code(software) 105 is stored in the ROM. The apparatus may be connected to a transmitter (TX) and a receiver (RX). The apparatus may, optionally, be connected with a user interface (Ul) for instructing the apparatus and / or for outputting data. The at least one processor 102, with the at least one memory 104 and the computer program code 105 are arranged to cause the apparatus to at least perform at least the method according to any preceding process, for example as disclosed in relation to the flow diagrams of FIGS 5, 6, 7, 8 and message flow of FIG. 9 and related features thereof.

[0088] FIG. 11 shows a non-transitory media 110 according to some embodiments. The non-transitory media 110 is a computer readable storage medium. It may be e.g. a CD, a DVD, a USB stick, a blue ray disk, etc. The non-transitory media 110 stores computer program code, causing an apparatus to perform the method of any preceding process for example as disclosed in relation to the flow diagrams and related features thereof.

[0089] Names of network elements, protocols, and methods are based on current standards. In other versions or other technologies, the names of these network elements and / or protocols and / or methods may be different, as long as they provide a corresponding functionality. For example, embodiments may be deployed in 2G / 3G / 4G / 5G networks and further generations of 3GPP but also in non-3GPP radio networks such as WiFi.

[0090] A memory may be volatile or non-volatile. It may be e.g. a RAM, a SRAM, a flash memory, a FPGA block ram, a DCD, a CD, a USB stick, and a blue ray disk.

[0091] If not otherwise stated or otherwise made clear from the context, the statement that two entities are different means that they perform different functions. It does not necessarily mean that they are based on different hardware. That is, each of the entities described in the present description may be based on a different hardware, or some or all of the entities may be based on the same hardware. It does not necessarily mean that they are based on different software. That is, each of the entities described in the present description may be based on different software, or some or all of the entities may be based on the same software. Each of the entities described in the present description may be embodied in the cloud.

[0092] Implementations of any of the above described blocks, apparatuses, systems, techniques or methods include, as non-limiting examples, implementations as hardware, software, firmware, special purpose circuits or logic, general purpose hardware or controller or other computing devices, or some combination thereof. Some embodiments may be implemented in the cloud.

[0093] It is to be understood that what is described above is what is presently considered the preferred embodiments. However, it should be noted that the description of the preferred embodiments is given by way of example only and that various modifications may be made without departing from the scope as defined by the appended claims.Statements of Invention

[0094] In a first aspect, the example embodiments provide a first apparatus comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the first apparatus to: receive a first configuration from a second apparatus, wherein the first configuration relates to a recovery mode for a prediction functionality, wherein the prediction functionality is implemented by one or more artificial intelligence or machine learning (AI / ML) models; enable the recovery mode by performing failure recovery based on the first configuration, wherein the first configuration comprises at least one of: a failure detection time period; a recovery time period; a recovery evaluation time period; a first performance threshold of failure recovery for the prediction functionality; a second performance threshold of failure recovery for the prediction functionality; or one or more reference signal resources for performing measurements in the recovery mode.

[0095] In some examples, detecting a failure comprises: during the failure detection time period, performing an evaluation of the prediction functionality based on a comparison of a performance of the prediction functionality with the first performance threshold.

[0096] In some examples, the instructions further cause the first apparatus to determine that the failure has occurred in the prediction functionality if the performance of the prediction functionality is determined to be lower than the first performance threshold.

[0097] In some examples, the recovery time period comprises a recovery evaluation time period, wherein performing failure recovery based on the first configuration comprises: during the evaluation time period, performing an evaluation of the prediction functionality based on a comparison of a performance of the prediction functionality with the second performance threshold.

[0098] In some examples, the instructions further cause the first apparatus to: if, within the recovery time period, the performance of the prediction functionality is determined to be equal to or higher than the second performance threshold, enable an active mode for the prediction functionality; or if, within the recovery time period, the performance of the prediction functionality is determined to be lower than the second performance threshold, deactivate the prediction functionality. In some examples, the second performance threshold is higher than the first performance threshold.

[0099] In some examples, the instructions further cause the first apparatus to determine the performance of the prediction functionality by performing recovery verification using the one or more reference signal resource for performing measurements in the recovery mode.

[0100] In some examples, the one or more reference signal resources for performing measurements in the recovery mode comprises at least one of: reference signals for performing inference measurements;or measurements, suitable for recovery verification, provided by one or more active functionalities associated with the first apparatus.

[0101] In some examples, determining the performance comprises: performing beam predictions or beam inference using the prediction functionality, wherein the beam predictions comprise predicted at least one reference signal measurement; obtaining at least one reference signal measurement; and comparing the beam predictions or beam inference with the at least one reference signal measurements; wherein the performance is determined based on said comparison

[0102] In some examples, the first configuration is received via a radio resource control message from the second apparatus and the radio resource control message relates to an initial configuration for the first apparatus.

[0103] In some examples, the instructions further cause the first apparatus to: send a first indication to the second apparatus, indicating that the prediction functionality is entered into an active mode if failure recovery is successful, wherein in the active mode, the prediction functionality is applicable.

[0104] In some examples, the first indication comprises one or more of a user equipment (UE) assistance information or a RRC reconfiguration complete message.

[0105] In some examples, the prediction functionality is suitable at least for beam prediction, wherein the instructions further cause the first apparatus to: pause sending beam prediction results to the second apparatus while the first apparatus is in the recovery mode for the prediction functionality; or resume sending beam prediction results to the second apparatus while the first apparatus is in the active mode for the prediction functionality.

[0106] In a second aspect, this specification describes a method performed at a first apparatus, comprising: receiving a first configuration from a second apparatus, wherein the first configuration relates to a recovery mode for a prediction functionality, wherein the prediction functionality is implemented by one or more artificial intelligence or machine learning (AI / ML) models; enabling the recovery mode by performing failure recovery based on the first configuration, wherein the first configuration comprises at least one of: a failure detection time period; a recovery time period; a recovery evaluation time period; a first performance threshold of failure recovery for the prediction functionality; a second performance threshold of failure recovery for the prediction functionality; or one or more reference signal resources for performing measurements in the recovery mode.

[0107] In some examples, detecting a failure comprises: during the failure detection time period, performing an evaluation of the prediction functionality based on a comparison of a performance of the prediction functionality with the first performance threshold.

[0108] In some examples, the method further comprises determining that the failure has occurred in the prediction functionality if the performance of the prediction functionality is determined to be lower than the first performance threshold.

[0109] In some examples, the recovery time period comprises a recovery evaluation time period, wherein performing failure recovery based on the first configuration comprises: during the evaluation time period, performing an evaluation of the prediction functionality based on a comparison of a performance of the prediction functionality with the second performance threshold.

[0110] In some examples, the method further comprises: if, within the recovery time period, the performance of the prediction functionality is determined to be equal to or higher than the second performance threshold, enabling an active mode for the prediction functionality; or if, within the recovery time period, the performance of the prediction functionality is determined to be lower than the second performance threshold, deactivating the prediction functionality.

[0111] In some examples, the second performance threshold is higher than the first performance threshold.

[0112] In some examples, the method further comprises determining the performance of the prediction functionality by performing recovery verification using the one or more reference signal resource for performing measurements in the recovery mode.

[0113] In some examples, the one or more reference signal resources for performing measurements in the recovery mode comprises at least one of: reference signals for performing inference measurements; or measurements, suitable for recovery verification, provided by one or more active functionalities associated with the first apparatus.

[0114] In some examples, determining the performance comprises: performing beam predictions or beam inference using the prediction functionality, wherein the beam predictions comprise predicted at least one reference signal measurement; obtaining at least one reference signal measurement; and comparing the beam predictions or beam inference with the at least one reference signal measurements; wherein the performance is determined based on said comparison

[0115] In some examples, the first configuration is received via a radio resource control message from the second apparatus and the radio resource control message relates to an initial configuration for the first apparatus.

[0116] In some examples, the method further comprises: sending a first indication to the second apparatus, indicating that the prediction functionality is entered into an active mode if failure recovery is successful, wherein in the active mode, the prediction functionality is applicable.

[0117] In some examples, the first indication comprises one or more of a user equipment (UE) assistance information or a RRC reconfiguration complete message.

[0118] In some examples, the prediction functionality is suitable at least for beam prediction, wherein the instructions further cause the first apparatus to: pause sending beam prediction results to the second apparatus while the first apparatus is in the recovery mode for the prediction functionality; or resume sending beam prediction results to the second apparatus while the first apparatus is in the active mode for the prediction functionality.

[0119] In a third aspect, there is provided a computer program product comprising a set of instructions which, when executed on an apparatus, is configured to cause the apparatus to carry out the method of any preceding method definition.

[0120] According to a fourth aspect, there is provided a non-transitory computer readable medium comprising program instructions stored thereon for performing a method, comprising: receiving a first configuration from a second apparatus, wherein the first configuration relates to a recovery mode for a prediction functionality, wherein the prediction functionality is implemented by one or more artificial intelligence or machine learning (AI / ML) models; enabling the recovery mode by performing failure recovery based on the first configuration, wherein the first configuration comprises at least one of: a failure detection time period; a recovery time period; a recovery evaluation time period; a first performance threshold of failure recovery for the prediction functionality; a second performance threshold of failure recovery for the prediction functionality; or one or more reference signal resources for performing measurements in the recovery mode.

[0121] The program instructions of the fourth aspect may also perform operations according to any preceding method definition of the second aspect.

[0122] According to a fifth aspect, there is provided an apparatus comprising: means for receiving a first configuration from a second apparatus, wherein the first configuration relates to a recovery mode for a prediction functionality, wherein the prediction functionality is implemented by one or more artificial intelligence or machine learning (AI / ML) models; means for enabling the recovery mode by performing failure recovery based on the first configuration, wherein the first configuration comprises at least one of: a failure detection time period; a recovery time period; a recovery evaluation time period; a first performance threshold of failure recovery for the prediction functionality; a second performance threshold of failure recovery for the prediction functionality; or one or more reference signal resources for performing measurements in the recovery mode.

[0123] According to a sixth aspect, this specification describes an apparatus configured to perform any method as described with reference to the second aspect.

[0124] According to a seventh aspect, this specification describes an apparatus comprising: at least one processor; and at least one memory including computer program code which, when executed by the at least one processor, causes the apparatus to perform operations according to any preceding method definition of the second aspect.

[0125] According to an eighth aspect, this specification describes an apparatus comprising: a first module configured to receive a first configuration from a second apparatus, wherein the first configuration relates to a recovery mode for a prediction functionality, wherein the prediction functionality is implemented by one or more artificial intelligence or machine learning (AI / ML) models; and a second module configured to enable the recovery mode by performing failure recovery based on the first configuration, wherein the first configuration comprises at least one of: a failure detection time period; a recovery time period; a recovery evaluation time period; a first performance threshold of failure recovery for the prediction functionality; a second performance threshold of failure recovery for the prediction functionality; or one or more reference signal resources for performing measurements in the recovery mode.

[0126] According to a ninth aspect, this specification describes a second apparatus comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the second apparatus to: send a first configuration to a first apparatus, wherein the first configuration relates to a recovery mode for a prediction functionality, wherein the prediction functionality is implemented by one or more artificial intelligence or machine learning (AI / ML) models; receive a first indication from the first apparatus indicating the applicability of the prediction functionality; wherein the first configuration comprises at least one of: a failure detection time period; a recovery time period; a recovery evaluation time period; a first performance threshold of failure recovery for the prediction functionality; a second performance threshold of failure recovery for the prediction functionality; or one or more reference signal resources for performing measurements in the recovery mode.

[0127] In some examples, the one or more reference signal resources for performing measurements in the recovery mode comprises at least one of: reference signals for performing inference measurements; or measurements, suitable for recovery verification, provided by one or more active functionalities associated with the first apparatus

[0128] In some examples, the first indication comprises one or more of a user equipment (UE) assistance information or a RRC reconfiguration complete message.

[0129] In some examples, the first configuration is sent, to the first apparatus, via a radio resource control message, and the radio resource control message relates to an initial configuration for the first apparatus.

[0130] According to a tenth aspect, this specification describes a method, performed at a second apparatus, comprising: sending a first configuration to a first apparatus, wherein the first configurationrelates to a recovery mode for a prediction functionality, wherein the prediction functionality is implemented by one or more artificial intelligence or machine learning (AI / ML) models; receiving a first indication from the first apparatus indicating the applicability of the prediction functionality; wherein the first configuration comprises at least one of: a failure detection time period; a recovery time period; a recovery evaluation time period; a first performance threshold of failure recovery for the prediction functionality; a second performance threshold of failure recovery for the prediction functionality; or one or more reference signal resources for performing measurements in the recovery mode.

[0131] In some examples, the one or more reference signal resources for performing measurements in the recovery mode comprises at least one of: reference signals for performing inference measurements; or measurements, suitable for recovery verification, provided by one or more active functionalities associated with the first apparatus

[0132] In some examples, the first indication comprises one or more of a user equipment (UE) assistance information or a RRC reconfiguration complete message.

[0133] In some examples, the first configuration is sent, to the first apparatus, via a radio resource control message, and the radio resource control message relates to an initial configuration for the first apparatus.

[0134] According to an eleventh aspect, there is provided a computer program product comprising a set of instructions which, when executed on an apparatus, is configured to cause the apparatus to carry out the method of any preceding method definition according to the tenth aspect.

[0135] According to a twelfth aspect, there is provided a non-transitory computer readable medium comprising program instructions stored thereon for performing a method, comprising: sending a first configuration to a first apparatus, wherein the first configuration relates to a recovery mode for a prediction functionality, wherein the prediction functionality is implemented by one or more artificial intelligence or machine learning (AI / ML) models; receiving a first indication from the first apparatus indicating the applicability of the prediction functionality; wherein the first configuration comprises at least one of: a failure detection time period; a recovery time period; a recovery evaluation time period; a first performance threshold of failure recovery for the prediction functionality; a second performance threshold of failure recovery for the prediction functionality; or one or more reference signal resources for performing measurements in the recovery mode.

[0136] The program instructions of the twelfth aspect may also perform operations according to any preceding method definition of the tenth aspect.

[0137] According to a thirteenth aspect, there is provided an apparatus comprising: means for sending a first configuration to a first apparatus, wherein the first configuration relates to a recovery mode for a prediction functionality, wherein the prediction functionality is implemented by one or more artificialintelligence or machine learning (AI / ML) models; means for receiving a first indication from the first apparatus indicating the applicability of the prediction functionality; wherein the first configuration comprises at least one of: a failure detection time period; a recovery time period; a recovery evaluation time period; a first performance threshold of failure recovery for the prediction functionality; a second performance threshold of failure recovery for the prediction functionality; or one or more reference signal resources for performing measurements in the recovery mode.

[0138] According to a fourteenth aspect, this specification describes an apparatus configured to perform any method as described with reference to the tenth aspect.

[0139] According to a fifteenth aspect, this specification describes an apparatus comprising: at least one processor; and at least one memory including computer program code which, when executed by the at least one processor, causes the apparatus to perform operations according to any preceding method definition of the tenth aspect.

[0140] According to a sixteenth aspect, this specification describes an apparatus comprising: a first module configured to send a first configuration to a first apparatus, wherein the first configuration relates to a recovery mode for a prediction functionality, wherein the prediction functionality is implemented by one or more artificial intelligence or machine learning (AI / ML) models; a second module configured to receive a first indication from the first apparatus indicating the applicability of the prediction functionality; wherein the first configuration comprises at least one of: a failure detection time period; a recovery time period; a recovery evaluation time period; a first performance threshold of failure recovery for the prediction functionality; a second performance threshold of failure recovery for the prediction functionality; or one or more reference signal resources for performing measurements in the recovery mode.

[0141] According to a seventeenth aspect, the example embodiments provide a first apparatus comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the first apparatus to: determine a recovery time period for a prediction functionality, wherein the prediction functionality is implemented by one or more artificial intelligence or machine learning (AI / ML) models, wherein the recovery time period comprises a recovery evaluation time period during which the first apparatus performs performance evaluation of the prediction functionality; determine whether or not a recovery condition is satisfied during the recovery evaluation time period; and transmit, to a second apparatus, an indication indicating whether or not the prediction functionality is applicable within the recovery time period.

[0142] In some examples, the recovery time period is determined based on the recovery evaluation time period and a first processing time.

[0143] In some examples, the recovery time period is based on an equation: T = K x Cei l((T1 +D) IK), where T 1 is the recovery evaluation time period, K is a scaling factor to T1 , and D is a first processing time.

[0144] In some examples, the instructions further cause the first apparatus to receive a first configuration from the second apparatus, wherein the first configuration relates to a recovery mode for the prediction functionality.

[0145] In some examples, wherein the first configuration comprises at least one of: a failure detection time period; a recovery time period; a recovery evaluation time period; a first performance threshold of failure recovery for the prediction functionality; a second performance threshold of failure recovery for the prediction functionality; or one or more reference signal resources for performing measurements in the recovery mode.

[0146] In some examples, the instructions further cause the first apparatus to perform failure recovery during the recovery time period based on the first configuration, wherein performing failure recovery comprises: detect a failure has occurred in the first model during the failure detection time period; performing failure recovery of the first model during the recovery time period.

[0147] In some examples, determining whether the recovery condition is satisfied comprises: during a failure detection time period, performing an evaluation of the prediction functionality based on a comparison of a performance of the prediction functionality with a first performance threshold.

[0148] In some examples, performing failure recovery based on the first configuration comprises: during the recovery evaluation time period, performing an evaluation of the prediction functionality based on a comparison of a performance of the prediction functionality with the second performance threshold.

[0149] In some examples, the instructions further cause the first apparatus to: if, within the recovery time period, the performance of the prediction functionality is determined to be equal to or higher than the second performance threshold, enable an active mode for the prediction functionality; or if, within the recovery time period, the performance of the prediction functionality is determined to be lower than the second performance threshold, deactivate the prediction functionality.

[0150] In some examples, the instructions further cause the first apparatus to determine the performance of the predicted functionality by performing recovery verification using the one or more reference signal resource for performing measurements in the recovery mode.

[0151] In some examples, the one or more reference signal resources for performing measurements in the recovery mode comprises at least one of: reference signals for performing inference measurements; or measurements, suitable for recovery verification, provided by one or more active functionalities associated with the first apparatus.

[0152] In some examples, the instructions further cause the first apparatus to, during the recovery time period, determine performance of the prediction functionality, by performing: performing beam predictions or beam inference using the prediction functionality in an inference stage, wherein the beam predictions comprise predicted at least one reference signal measurements; obtaining first at least one reference signal measurements; and comparing the beam predictions or beam inference with the first at least one reference signal measurements; wherein the performance of the prediction functionality is determined based on said comparison.

[0153] In some examples, the first configuration is received via a radio resource control message from the second apparatus and the radio resource control message relates to an initial configuration for the first apparatus.

[0154] In some examples, the indication further indicates that the prediction functionality is entered into an active mode if failure recovery is successful, wherein in the active mode, the prediction functionality is applicable.

[0155] In some examples, the indication is carried in a UE assistance information or a RRC reconfiguration complete message.

[0156] According to an eighteenth aspect, this specification describes a method performed at a first apparatus, comprising: determining a recovery time period for a prediction functionality, wherein the prediction functionality is implemented by one or more artificial intelligence or machine learning (AI / ML) models, wherein the recovery time period comprises a recovery evaluation time period during which the first apparatus performs performance evaluation of the prediction functionality; determining whether or not a recovery condition is satisfied during the recovery evaluation time period; and transmitting, to a second apparatus, an indication indicating whether or not the prediction functionality is applicable within the recovery time period.

[0157] In some examples, the recovery time period is determined based on the recovery evaluation time period and a first processing time.

[0158] In some examples, the recovery time period is based on an equation: T = K x Ceil((T1+D) / K), where T 1 is the recovery evaluation time period, K is a scaling factor to T1 , and D is a first processing time.

[0159] In some examples, the method further comprises receiving a first configuration from the second apparatus, wherein the first configuration relates to a recovery mode for the prediction functionality.

[0160] In some examples, wherein the first configuration comprises at least one of: a failure detection time period; a recovery time period; a recovery evaluation time period; a first performance threshold of failure recovery for the prediction functionality; a second performance threshold of failure recovery for theprediction functionality; or one or more reference signal resources for performing measurements in the recovery mode.

[0161] In some examples, the method further comprises performing failure recovery during the recovery time period based on the first configuration, wherein performing failure recovery comprises: detect a failure has occurred in the first model during the failure detection time period; performing failure recovery of the first model during the recovery time period.

[0162] In some examples, determining whether the recovery condition is satisfied comprises: during a failure detection time period, performing an evaluation of the prediction functionality based on a comparison of a performance of the prediction functionality with a first performance threshold.

[0163] In some examples, performing failure recovery based on the first configuration comprises: during the recovery evaluation time period, performing an evaluation of the prediction functionality based on a comparison of a performance of the prediction functionality with the second performance threshold.

[0164] In some examples, the method further comprises: if, within the recovery time period, the performance of the prediction functionality is determined to be equal to or higher than the second performance threshold, enabling an active mode for the prediction functionality; or if, within the recovery time period, the performance of the prediction functionality is determined to be lower than the second performance threshold, deactivating the prediction functionality.

[0165] In some examples, the method further comprises determining the performance of the predicted functionality by performing recovery verification using the one or more reference signal resource for performing measurements in the recovery mode.

[0166] In some examples, the one or more reference signal resources for performing measurements in the recovery mode comprises at least one of: reference signals for performing inference measurements; or measurements, suitable for recovery verification, provided by one or more active functionalities associated with the first apparatus.

[0167] In some examples, the method further comprises, during the recovery time period, determining performance of the prediction functionality, by performing: performing beam predictions or beam inference using the prediction functionality in an inference stage, wherein the beam predictions comprise predicted at least one reference signal measurements; obtaining first at least one reference signal measurements; and comparing the beam predictions or beam inference with the first at least one reference signal measurements; wherein the performance of the prediction functionality is determined based on said comparison.

[0168] In some examples, the first configuration is received via a radio resource control message from the second apparatus and the radio resource control message relates to an initial configuration for the first apparatus.

[0169] In some examples, the indication further indicates that the prediction functionality is entered into an active mode if failure recovery is successful, wherein in the active mode, the prediction functionality is applicable.

[0170] In some examples, the indication is carried in a UE assistance information or a RRC reconfiguration complete message.

[0171] According to a nineteenth aspect, there is provided a computer program product comprising a set of instructions which, when executed on an apparatus, is configured to cause the apparatus to carry out the method of any preceding method definition according to the eighteenth aspect.

[0172] According to a twentieth aspect, there is provided a non-transitory computer readable medium comprising program instructions stored thereon for performing a method, comprising: determining a recovery time period for a prediction functionality, wherein the prediction functionality is implemented by one or more artificial intelligence or machine learning (AI / ML) models, wherein the recovery time period comprises a recovery evaluation time period during which the first apparatus performs performance evaluation of the prediction functionality; determining whether or not a recovery condition is satisfied during the recovery evaluation time period; and transmitting, to a second apparatus, an indication indicating whether or not the prediction functionality is applicable within the recovery time period.

[0173] The program instructions of the twentieth aspect may also perform operations according to any preceding method definition of the eighteenth aspect.

[0174] According to a twenty-first aspect, there is provided an apparatus comprising: means for determining a recovery time period for a prediction functionality, wherein the prediction functionality is implemented by one or more artificial intelligence or machine learning (AI / ML) models, wherein the recovery time period comprises a recovery evaluation time period during which the first apparatus performs performance evaluation of the prediction functionality; means for determining whether or not a recovery condition is satisfied during the recovery evaluation time period; and means for transmitting, to a second apparatus, an indication indicating whether or not the prediction functionality is applicable within the recovery time period.

[0175] According to a twenty-second aspect, this specification describes an apparatus configured to perform any method as described with reference to the eighteenth aspect.

[0176] According to a twenty-third aspect, this specification describes an apparatus comprising: at least one processor; and at least one memory including computer program code which, when executed by the atleast one processor, causes the apparatus to perform operations according to any preceding method definition of the eighteenth aspect.

[0177] According to an twenty-fourth aspect, this specification describes an apparatus comprising: a first module configured to determine a recovery time period for a prediction functionality, wherein the prediction functionality is implemented by one or more artificial intelligence or machine learning (AI / ML) models, wherein the recovery time period comprises a recovery evaluation time period during which the first apparatus performs performance evaluation of the prediction functionality; a second module configured to determine whether or not a recovery condition is satisfied during the recovery evaluation time period; and a third module configured to transmit, to a second apparatus, an indication indicating whether or not the prediction functionality is applicable within the recovery time period.

[0178] According to a twenty-fifth aspect, this specification describes a second apparatus comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the second apparatus to: send, to a first apparatus, a first configuration, wherein the first configuration relates to a recovery mode for prediction functionality, wherein the prediction functionality is implemented by one or more artificial intelligence or machine learning (AI / ML) models; and receive, from the first apparatus, an indication indicating whether or not the prediction functionality is applicable within the recovery time period.

[0179] In some examples, the first configuration comprises at least one of: a failure detection time period; a recovery time period; a recovery evaluation time period; a first performance threshold of failure recovery for the prediction functionality; a second performance threshold of failure recovery for the prediction functionality; or one or more reference signal resources for performing measurements in the recovery mode.

[0180] In some examples, the one or more reference signal resources for performing measurements in the recovery mode comprises at least one of: reference signals for performing inference measurements; or measurements, suitable for recovery verification, provided by one or more active functionalities associated with the first apparatus.

[0181] In some examples, the indication is carried in a UE assistance information or a RRC reconfiguration complete message.

[0182] According to a twenty-sixth aspect, this specification describes a method, performed at a second apparatus, comprising: sending, to a first apparatus, a first configuration, wherein the first configuration relates to a recovery mode for prediction functionality, wherein the prediction functionality is implemented by one or more artificial intelligence or machine learning (AI / ML) models; and receiving, from the firstapparatus, an indication indicating whether or not the prediction functionality is applicable within the recovery time period.

[0183] In some examples, the first configuration comprises at least one of: a failure detection time period; a recovery time period; a recovery evaluation time period; a first performance threshold of failure recovery for the prediction functionality; a second performance threshold of failure recovery for the prediction functionality; or one or more reference signal resources for performing measurements in the recovery mode.

[0184] In some examples, the one or more reference signal resources for performing measurements in the recovery mode comprises at least one of: reference signals for performing inference measurements; or measurements, suitable for recovery verification, provided by one or more active functionalities associated with the first apparatus.

[0185] In some examples, the indication is carried in a UE assistance information or a RRC reconfiguration complete message.

[0186] According to a twenty-seventh aspect, there is provided a computer program product comprising a set of instructions which, when executed on an apparatus, is configured to cause the apparatus to carry out the method of any preceding method definition according to the twenty-sixth aspect.

[0187] According to a twenty-eighth aspect, there is provided a non-transitory computer readable medium comprising program instructions stored thereon for performing a method, comprising: sending, to a first apparatus, a first configuration, wherein the first configuration relates to a recovery mode for prediction functionality, wherein the prediction functionality is implemented by one or more artificial intelligence or machine learning (AI / ML) models; and receiving, from the first apparatus, an indication indicating whether or not the prediction functionality is applicable within the recovery time period.

[0188] The program instructions of the twenty-eighth aspect may also perform operations according to any preceding method definition of the twenty-sixth aspect.

[0189] According to a twenty-ninth aspect, there is provided an apparatus comprising: means for sending, to a first apparatus, a first configuration, wherein the first configuration relates to a recovery mode for prediction functionality, wherein the prediction functionality is implemented by one or more artificial intelligence or machine learning (AI / ML) models; and means for receiving, from the first apparatus, an indication indicating whether or not the prediction functionality is applicable within the recovery time period.

[0190] According to a thirtieth aspect, this specification describes an apparatus configured to perform any method as described with reference to the twenty-sixth aspect.

[0191] According to a thirty-first aspect, this specification describes an apparatus comprising: at least one processor; and at least one memory including computer program code which, when executed by the atleast one processor, causes the apparatus to perform operations according to any preceding method definition of the twenty-sixth aspect.

[0192] According to a thirty-second aspect, this specification describes an apparatus comprising: a first module configured to send, to a first apparatus, a first configuration, wherein the first configuration relates to a recovery mode for prediction functionality, wherein the prediction functionality is implemented by one or more artificial intelligence or machine learning (AI / ML) models; and a second module configured to receive, from the first apparatus, an indication indicating whether or not the prediction functionality is applicable within the recovery time period.

Claims

Claims1. A first apparatus comprising:at least one processor; andat least one memory storing instructions that, when executed by the at least one processor, cause the first apparatus to:determine a recovery time period for a prediction functionality, wherein the prediction functionality is implemented by one or more artificial intelligence or machine learning (AI / ML) models, wherein the recovery time period comprises a recovery evaluation time period during which the first apparatus performs performance evaluation of the prediction functionality;determine whether or not a recovery condition is satisfied during the recovery evaluation time period; andtransmit, to a second apparatus, an indication indicating whether or not the prediction functionality is applicable within the recovery time period.

2. A first apparatus as claimed in claim 1 , wherein the recovery time period is determined based on the recovery evaluation time period and a first processing time.

3. A first apparatus as claimed in claim 2, wherein the recovery time period is based on an equation:T = Kx Ceil((T1+D) / K), where T1 is the recovery evaluation time period, K is a scaling factor to T1, and D is a first processing time.

4. A first apparatus as claimed in any one of the preceding claims, wherein the instructions further cause the first apparatus to receive a first configuration from the second apparatus, wherein the first configuration relates to a recovery mode for the prediction functionality.

5. A first apparatus as claimed in claim 4, wherein the first configuration comprises at least one of:a failure detection time period;a recovery time period;a recovery evaluation time period;a first performance threshold of failure recovery for the prediction functionality;a second performance threshold of failure recovery for the prediction functionality; orone or more reference signal resources for performing measurements in the recovery mode.

6. A first apparatus as claimed in claim 5, wherein the instructions further cause the first apparatus to perform failure recovery during the recovery time period based on the first configuration, wherein performing failure recovery comprises:detecting a failure has occurred in the first model during the failure detection time period; performing failure recovery of the first model during the recovery time period.

7. A first apparatus as claimed in any one of the preceding claims, wherein determining whether the recovery condition is satisfied comprises:during a failure detection time period, performing an evaluation of the prediction functionality based on a comparison of a performance of the prediction functionality with a first performance threshold.

8. A first apparatus as claimed in claim 6 or 7, wherein performing failure recovery based on the first configuration comprises:during the recovery evaluation time period, performing an evaluation of the prediction functionality based on a comparison of a performance of the prediction functionality with the second performance threshold.

9. A first apparatus as claimed in claim 8, wherein the instructions further cause the first apparatus to:if, within the recovery time period, the performance of the prediction functionality is determined to be equal to or higher than the second performance threshold, enable an active mode for the prediction functionality; orif, within the recovery time period, the performance of the prediction functionality is determined to be lower than the second performance threshold, deactivate the prediction functionality.

10. A first apparatus as claimed in any one of the preceding claims, wherein the instructions further cause the first apparatus to determine the performance of the predicted functionality by performing recovery verification using the one or more reference signal resource for performing measurements in the recovery mode.

11. A first apparatus as claimed in claim 10, wherein the one or more reference signal resources for performing measurements in the recovery mode comprises at least one of:reference signals for performing inference measurements; ormeasurements, suitable for recovery verification, provided by one or more active functionalities associated with the first apparatus.

12. A first apparatus as claimed in any one of the preceding claims, wherein the instructions further cause the first apparatus to, during the recovery time period, determine performance of the prediction functionality, by performing:performing beam predictions or beam inference using the prediction functionality in an inference stage, wherein the beam predictions comprise predicted at least one reference signal measurements; obtaining first at least one reference signal measurements; andcomparing the beam predictions or beam inference with the first at least one reference signal measurements; wherein the performance of the prediction functionality is determined based on said comparison.

13. A first apparatus as claimed in any one of claims 4 to 12, wherein the first configuration is received via a radio resource control message from the second apparatus and the radio resource control message relates to an initial configuration for the first apparatus.

14. A first apparatus as claimed in any one of the preceding claims, wherein the indication further indicates that the prediction functionality is entered into an active mode if failure recovery is successful, wherein in the active mode, the prediction functionality is applicable.

15. A first apparatus as claimed in any one of the preceding claims, wherein the indication is carried in a user equipment (UE) assistance information (UAI) or a radio resource control (RRC) reconfiguration complete message.

16. A second apparatus comprising:at least one processor; andat least one memory storing instructions that, when executed by the at least one processor, cause the second apparatus to:send, to a first apparatus, a first configuration, wherein the first configuration relates to a recovery mode for prediction functionality, wherein the prediction functionality is implemented by one or more artificial intelligence or machine learning (AI / ML) models; andreceive, from the first apparatus, an indication indicating whether or not the prediction functionality is applicable within the recovery time period.

17. A second apparatus as claimed in claim 16, wherein the first configuration comprises at least one of:a failure detection time period;a recovery time period;a recovery evaluation time period;a first performance threshold of failure recovery for the prediction functionality;a second performance threshold of failure recovery for the prediction functionality; orone or more reference signal resources for performing measurements in the recovery mode.

18. A second apparatus as claimed in claim 17, the one or more reference signal resources for performing measurements in the recovery mode comprises at least one of:reference signals for performing inference measurements; ormeasurements, suitable for recovery verification, provided by one or more active functionalities associated with the first apparatus.

19. A second apparatus as claimed in any one of the claims 16 to 18, wherein the indication is carried in a user equipment (UE) assistance information (UAI) or a radio resource control (RRC) reconfiguration complete message.

20. A method, performed at a first apparatus, comprising:determining a recovery time period for a prediction functionality, wherein the prediction functionality is implemented by one or more artificial intelligence or machine learning (AI / ML) models, wherein the recovery time period comprises a recovery evaluation time period during which the first apparatus performs performance evaluation of the prediction functionality;determining whether a recovery condition is satisfied during the recovery evaluation time period; andtransmitting, to a second apparatus, an indication indicating whether the prediction functionality is applicable within the recovery time period.

21. A method, performed at a second apparatus, comprising:sending, to a first apparatus, a first configuration, wherein the first configuration relates to a recovery mode for prediction functionality, wherein the prediction functionality is implemented by one or more artificial intelligence or machine learning (AI / ML) models; andreceiving, from the first apparatus, an indication indicating whether the prediction functionality is applicable within the recovery time period.