The communication system is also able to communicate with other networks, such as a public switched telephone network or the Internet 312, or utilize services provided by them. The communication network may also be able to support the usage of cloud services, for example at least part of core network operations may be carried out as a cloud service (this is depicted in FIG. 3 by “cloud” 314). The communication system may also comprise a central control entity, or a like, providing facilities for networks of different operators to cooperate for example in spectrum sharing.

Edge cloud may be brought into radio access network (RAN) by utilizing network function virtualization (NFV) and software defined networking (SDN). Using edge cloud may mean access node operations to be carried out, at least partly, in a server, host or node operationally coupled to a remote radio head or base station comprising radio parts. It is also possible that node operations will be distributed among a plurality of servers, nodes or hosts. Application of cloud RAN architecture enables RAN real time functions being carried out at the RAN side (in a distributed unit, DU) and non-real time functions being carried out in a centralized manner (in a centralized unit, CU 308).

It should also be understood that the distribution of labor between core network operations and base station operations may differ from that of the LTE or even be non-existent. Some other technology advancements probably to be used are Big Data and all-IP, which may change the way networks are being constructed and managed. 5G (or new radio, NR) networks are being designed to support multiple hierarchies, where MEC servers can be placed between the core and the base station or nodeB (gNB). It should be appreciated that MEC can be applied in 4G networks as well. The gNB is a next generation Node B (or, new Node B) supporting the 5G network (i.e., the NR).

5G may also utilize non-terrestrial nodes 306, e.g. access nodes, to enhance or complement the coverage of 5G service, for example by providing backhauling, wireless access to wireless devices, service continuity for machine-to-machine (M2M) communication, service continuity for Internet of Things (IoT) devices, service continuity for passengers on board of vehicles, ensuring service availability for critical communications and/or ensuring service availability for future railway/maritime/aeronautical communications. The non-terrestrial nodes may have fixed positions with respect to the Earth surface or the non-terrestrial nodes may be mobile non-terrestrial nodes that may move with respect to the Earth surface. The non-terrestrial nodes may comprise satellites and/or HAPS s. Satellite communication may utilize geostationary earth orbit (GEO) satellite systems, but also low earth orbit (LEO) satellite systems, in particular mega-constellations (systems in which hundreds of (nano)satellites are deployed). Each satellite in the mega-constellation may cover several satellite-enabled network entities that create on-ground cells. The on-ground cells may be created through an on-ground relay node 304 or by a gNB located on-ground or in a satellite.

A person skilled in the art appreciates that the depicted system is only an example of a part of a radio access system and in practice, the system may comprise a plurality of (e/g)NodeBs, the user device may have access to a plurality of radio cells and the system may comprise also other apparatuses, such as physical layer relay nodes or other network elements, etc. At least one of the (e/g)NodeBs or may be a Home(e/g)nodeB. Additionally, in a geographical area of a radio communication system a plurality of different kinds of radio cells as well as a plurality of radio cells may be provided. Radio cells may be macro cells (or umbrella cells) which are large cells, usually having a diameter of up to tens of kilometers, or smaller cells such as micro-, femto- or picocells. The (e/g)NodeBs of FIG. 1 may provide any kind of these cells. A cellular radio system may be implemented as a multilayer network including several kinds of cells. Typically, in multilayer networks, one access node provides one kind of a cell or cells, and thus a plurality of (e/g)NodeB s are required to provide such a network structure.

For fulfilling the need for improving the deployment and performance of communication systems, the concept of “plug-and-play” (e/g)NodeBs has been introduced. Typically, a network which is able to use “plug-and-play” (e/g)Node Bs, includes, in addition to Home (e/g)NodeBs (H(e/g)nodeBs), a home node B gateway, or HNB-GW (not shown in FIG. 1). A HNB Gateway (HNB-GW), which is typically installed within an operator's network may aggregate traffic from a large number of HNBs back to a core network.

The Radio Resource Control (RRC) protocol is used in various wireless communication systems for defining the air interface between the UE and a base station, such as eNB/gNB. This protocol is specified by 3GPP in in TS 36.331 for LTE and in TS 38.331 for 5G. In terms of the RRC, the UE may operate in LTE and in 5G in an idle mode or in a connected mode, wherein the radio resources available for the UE are dependent on the mode where the UE at present resides. In 5G, the UE may also operate in inactive mode. In the RRC idle mode, the UE has no connection for communication, but the UE is able to listen to page messages. The UE may communicate with the eNB/gNB via various logical channels like Broadcast Control Channel (BCCH), Paging Control Channel (PCCH), Common Control Channel (CCCH), Dedicated Control Channel (DCCH), Dedicated Traffic Channel (DTCH).

The transitions between the states are controlled by a state machine of the RRC. When the UE is powered up, it is in a disconnected mode/idle mode. The UE may transit to RRC connected mode with an initial attach or with a connection establishment. If there is no activity from the UE for a short time, eNB/gNB may suspend its session by moving to RRC Inactive and can resume its session by moving to RRC connected mode. The UE can move to the RRC idle mode from the RRC connected mode or from the RRC inactive mode.

The actual user and control data from network to the UEs is transmitted via downlink physical channels, which in 5G include Physical downlink control channel (PDCCH) which carries the necessary downlink control information (DCI), Physical Downlink Shared Channel (PDSCH), which carries the user data and system information for user, and Physical broadcast channel (PBCH), which carries the necessary system information to enable a UE to access the 5G network.

The user and control data from UE to the network is transmitted via uplink physical channels, which in 5G include Physical Uplink Control Channel (PUCCH), which is used for uplink control information including HARQ feedback acknowledgments, scheduling request, and downlink channel-state information for link adaptation, Physical Uplink Shared Channel (PUSCH), which is used for uplink data transmission, and Physical Random Access Channel (PRACH), which is used by the UE to request connection setup referred to as random access.

Future devices, i.e. both UEs and network elements, will host Artificial Intelligence/Machine Learning (AI/ML) algorithms, which require a significant amount of resources in both training and inference phase. The AI/ML algorithms may be implemented, for example, by neural networks.

A neural network (NN) is a computation graph consisting of several layers of computation. Each layer consists of one or more units, where each unit performs an elementary computation. A unit is connected to one or more other units, and the connection may have associated with a weight. The weight may be used for scaling the signal passing through the associated connection. Weights are learnable parameters, i.e., values which can be learned from training data. There may be other learnable parameters, such as those of batch-normalization layers.

Two of the most widely used architectures for neural networks are feed-forward and recurrent architectures. Feed-forward neural networks are such that there is no feedback loop: each layer takes input from one or more of the layers before and provides its output as the input for one or more of the subsequent layers. Also, units inside a certain layer take input from units in one or more of preceding layers and provide output to one or more of following layers.

Initial layers (those close to the input data) extract semantically low-level features, and intermediate and final layers extract more high-level features. After the feature extraction layers there may be one or more layers performing a certain task, such as classification, semantic segmentation, object detection, denoising, style transfer, super-resolution, etc. In recurrent neural nets, there is a feedback loop, so that the network becomes stateful, i.e., it is able to memorize information or a state.

Neural networks are being utilized in an ever-increasing number of applications for many different types of device, such as mobile phones. Examples include image and video analysis and processing, social media data analysis, device usage data analysis, etc.

One of the important properties of neural networks (and other machine learning tools) is that they are able to learn properties from input data, either in supervised way or in unsupervised way. Such learning is a result of a training algorithm, or of a meta-level neural network providing the training signal.

In general, the training algorithm consists of changing some properties of the neural network so that its output is as close as possible to a desired output. For example, in the case of classification of objects in images, the output of the neural network can be used to derive a class or category index which indicates the class or category that the object in the input image belongs to. Training usually happens by minimizing or decreasing the output's error, also referred to as the loss. Examples of losses are mean squared error, cross-entropy, etc. In recent deep learning techniques, training is an iterative process, where at each iteration the algorithm modifies the weights of the neural net to make a gradual improvement of the network's output, i.e., to gradually decrease the loss.

Herein, term “model” may refer to “neural network”, and the weights of neural networks are sometimes referred to as learnable parameters or simply as parameters.

Training a neural network is an optimization process. The goal of the optimization or training process is to make the model learn the properties of the data distribution from a limited training dataset. In other words, the goal is to learn to use a limited training dataset in order to learn to generalize to previously unseen data, i.e., data which was not used for training the model. This is usually referred to as generalization. In practice, data may be split into at least two sets, the training set and the validation set. The training set is used for training the network, i.e., to modify its learnable parameters in order to minimize the loss. The validation set is used for checking the performance of the network on data, which was not used to minimize the loss, as an indication of the final performance of the model. In particular, the errors on the training set and on the validation set are monitored during the training process to understand the following things:

- If the network is learning at all—in this case, the training set error should decrease, otherwise the model is in the regime of underfitting.
- If the network is learning to generalize—in this case, also the validation set error needs to decrease and to be not too much higher than the training set error. If the training set error is low, but the validation set error is much higher than the training set error, or it does not decrease, or it even increases, the model is in the regime of overfitting. This means that the model has just memorized the training set's properties and performs well only on that set but performs poorly on a set not used for tuning its parameters.

Highly complex networks, such as 5G NR, generate very large RRC configuration messages for defining the air interface, which may comprise a huge amount of data, comprising a plurality of information elements consecutive to, or nested within, each other. RRC Reconfiguration message is one of the most complicated messages used in 5G NR, extending up to 8K bytes and even more as Rel-16 devices allow DL/UL RRC message segmentation enabling segmentation of RRC messages up to 45 KB. For example, in LTE typically the sizes of RRC configuration messages used to be around 4 KB, and thus 5G NR may introduce a ten-fold increase in the size.

The resource consuming AI/ML algorithms of the future UEs and network elements, combined with the processing of very large RRC configuration messages, the probability of having issues like overheating and overloading of computing resources (hardware and software) is very likely to increase.

Moreover, it may happen that some of the RRC configurations are not suitable for the network and/or the UE as they, as such, may create some adversity; high resource consumption at UE/network, overheating situation, etc. In such a situation, such as internal overheating, the UE may respond by triggering a UE Assistance Information to the network. The network may, in response, trim the RRC configuration to allow the UE to avert the situation. However, this method is purely reactive, and an anomalous operating condition has already taken place in the UE, thereby complicating to perform any corrective measures.

Accordingly, there is a need for a method for proactively prevent such anomalous operating conditions to take place.

In the following, an enhanced method configuring a user equipment (UE) will be described in more detail, in accordance with various embodiments.

The method, which is disclosed in the flow chart of FIG. 4 as reflecting the operation of an apparatus, such as a user equipment (UE), wherein the method comprises establishing, by said apparatus, a connection to a network; predicting, by an algorithm-based model implemented in said apparatus, whether any of configurable features of the apparatus infers a predetermined likelihood to cause an anomalous operating condition in said apparatus, and signaling information about said configurable features to the network.

Thus, the UE may comprise a model for predicting anomalous operating condition caused by certain configurable features or their combination in advance. As there is no certain way to find out, if a UE will face an anomalous situation, such as an overload of processing resources and/or overheating, an algorithm-based model designed to predict such situations (anomaly detection) is implemented in the UE, aiming to handle the situation proactively. The results of the prediction are signalled to the network, which may utilize the information for the decision making of what kind of configuration to assign the UE (for example, during normal operation in a cell or during a handover) to ensure that any adverse situations are avoided.

The model may be implemented, for example, as a neural network.

Hence, the UE and/or a network element, such as a gNB, may deploy a model that is trained to detect anomalies, i.e., it receives triggers from the UE regarding overheating conditions, high resource consumption and other potential adverse situations, such as if particular UE model capabilities creating an anomalous operating condition. The triggers are labelled and used as input data for the model training phase.

FIGS. 5a and 5b, as well as FIGS. 5c and 5d, show the training and inference of a ML-based model as a simplified block for a UE and a network element, such as a gNB, respectively.

In the training phase (FIG. 5a), the ML-based model implemented in a UE may be trained using various data as input data in the ML algorithm training, which may include, for example, adjustable RRC Reconfiguration parameters, such as a list of possible capabilities and/or their combination, but also feedback from executing model inference, as described below. This may include, for example, band(s) or band combinations for a set of frequency carriers and/or carrier bandwidths and/or modulations. The input data for the training may also include UE Radio measurements & Performance measurements. As a result of the training, a trained ML model is produced, which may be executed by the UE. It is also possible that UE sends the trained ML model to the gNB to be executed there. In this case, a trained ML Model is deployed at the gNB, or at gNB-CU in case of split architecture

The trained ML model is used in the inference phase (FIG. 5b.), which may take place, for example, at the time of UE reconfiguration, upon receiving additional signaling, such as a proposed configuration, upon anomaly detection inference at UE, or upon final configuration of the UE. The input data for inference may, for example, trigger a need to reconfigure the UE. The input data may be, for example, a configuration proposed by the gNB, which may include a proposal for a new configuration for one or more of the following: band(s) or band combinations for a set of frequency carriers and/or carrier bandwidths and/or modulations.

The output of the trained ML model in the inference phase may comprise, for example, one or more of the following:

- a list of possible problematic capabilities or their combinations
  - Band(s) or band combinations for a set of frequency carriers and/or carrier bandwidths and/or modulations
- a probability that the proposed RRC Reconfiguration parameter causes anomalous operation
- an anomaly notification to the gNB to enable the gNB to perform an optimal reconfiguration, e.g. to switch to a lower order RRC Configuration.

The Model Inference output may be also used as feedback in the input of an ML Algorithm training to facilitate the training or retraining of an ML Model. For example, the output of FIG. 5b may be used in the input of FIG. 5a.

For predicting the probability for an anomalous situation, the trained ML model of the UE may monitor the number of times over a total number of anomalies that a certain anomaly is observed for a given set of band(s) or band combinations, for a set of frequency carriers, and/or carrier bandwidths, or modulations. The probability may be calculated for each capability separately (e.g., per band or per modulation or per frequency carrier), for each feature separately (e.g., per overheating or in-device co-existence detection), or each feature specific parameter separately (e.g. number of MIMO layers, number of carriers). The UE may then create a distribution of anomalies (e.g. per capability). The UE may then can characterize certain capability, i.e., a band, band combination, frequency carrier, carrier bandwidth and modulation that has the highest probability of anomaly, as a problematic capability. Herein, the definition of an anomaly may be provided by the network.

If the UE has sent the trained ML model to the gNB, the gNB may run/execute the UE-trained ML model to determine the best configuration for a given UE. Thus, the gNB can internally determine the best configuration for a given UE.

For the ML-based model implemented in a gNB, the training phase (FIG. 5c) input data may include, for example, an RRC Reconfiguration that the gNB provides to the UE, or alternative RRC Reconfiguration parameters that the UE may have indicated to prefer.

The trained ML model is used in the inference phase (FIG. 5d.), which may take place, for example, in a situation, where the gNB can predict the best configuration parameters for a given UE or class of UEs. The input data for inference may include, for example, an input provided by the UE to the gNB on the preferred or predicted preferred RRC configuration parameters, or a proposed RRC Reconfiguration that gNB has originally provide to the UE.

The output of the trained ML model in the inference phase may comprise, for example, one or more of the following:

- a preferred RRC Configuration per UE or per class of UEs
- an indication, determined based on the preferred UE configuration, whether to maintain the originally proposed configuration or to update the original configuration to another preferred by the UE.

Again, the Model Inference output may be also used as feedback in the input of an ML Algorithm training to facilitate the training or retraining of an ML Model. For example, the output of FIG. 5d may be used in the input of FIG. 5c.

For predicting the preferred RRC Configuration per UE or per a class of UEs, the gNB may store for a given UE or a group of UEs alternative preferred configuration parameters. The trained ML model of the gNB may create a probability with which certain alternative parameters are chosen by a UE or a group of UEs for a given state (e.g., power level, radio conditions, etc.) or for a state over a group of UEs (e.g., having common radio conditions). The trained ML model of the gNB may calculate a preference probability based on the number of times a certain configuration is selected by a UE or a group of UEs for a given state as opposed to the rest of the configurations. The trained ML model of the gNB may then predict that when a UE or a set of UEs is in a certain condition (radio condition, power level, etc.), it has a higher probability to select or at least prefer certain parameters. Based on the prediction, the gNB may either keep the original configuration or may propose an updated configuration.

The model may also be implemented, for example, as a rule-based model.

Hence, instead of training a neural network, a rule-based (i.e. non-learned) network, such as a Hidden Markov Model (HMM) or a Bayesian network, may be used for predicting the potential anomalous operating conditions. In a Bayesian network, an event that occurred may be modelled by a probabilistic graphical model that represents a set of variables and their conditional dependencies, wherein the likelihood that anyone(s) of several possible known causes was the contributing factor to the event may be predicted. Thus, a Bayesian network can be used to compute the probabilities of various anomalous operating condition emerging based on various configurable features of the UE.

According to an embodiment, said anomalous operating condition comprises one or more of the following:

- overheating of the apparatus;
- overload of processing resources.

According to an embodiment, said configurable features causing said anomalous operating condition comprise one or more of the following:

- frequency band(s) or band combinations for a set of frequency carriers and/or carrier bandwidths and/or modulations;
- Artificial Intelligence/Machine Learning (AI/ML) algorithms implemented in said apparatus;
- Machine Learning (ML) models processed in said apparatus;
- any combination of the above.

There are various scenarios where the model and its results can be utilised. According to a first scenario, the network requests potential feature sets and their combinations that are prone to high resource usage or may create anomalous situation during runtime from the UE.

According to an embodiment, the method comprises receiving a request, from the network, for informing one or more configurable features of the apparatus potentially causing an anomalous operating condition in said apparatus.

According to an embodiment, said request identifies one or more of said configurable features.

According to an embodiment, the method comprises running said model for the identified one or more features; and signaling information whether any of said identified one or more features of the apparatus infers a predetermined likelihood to cause an anomalous operating condition in said apparatus to the network.

The flow chart of FIG. 6 illustrates a method according to some of the above embodiments. At an initial stage, the UE has established a connection to the network, and the UE has been authenticated (600). This enables the network, such as an access point, e.g. a gNB, to send a request (602) to UE to inform one or more configurable features of the apparatus potentially causing an anomalous operating condition in the UE. The request may be referred to as, for example, UE Capability Enquiry. The request may include a filter setting for the UE to limit the configurable features to a specific subset of features. For example, the filter may request the UE to inform potential band(s) or band combinations for a set of frequency carriers and/or carrier bandwidths and/or modulations where it is likely that anomalous operation may occur. As another example, the network may request the UE to include information about a prediction relating any AI/ML assisted algorithm(s) in the UE possibly causing anomalous operating conditions when operating under potential band(s) or band combinations for a set of frequency carriers and/or carrier bandwidths and/or modulations. The filter setting requested by the network may be based, for example, on information predicted based on a UE model identifier.

Based on the request, the UE compiles (604) at least one, but possibly a list of several possible problematic capabilities and/or their combinations by executing the algorithm underlying the model that provides the prediction. The UE sends a response message (606) to the network. The response may be referred to as, for example, UE Capability Information. The response may include said list of predicted problematic capabilities. The list may be included in a separate capability container, which may be organized, for example, in a two-dimensional manner, where a first dimension represents the number of anomalous feature sets per band, which may potentially cause high resource utilization in the UE. Each feature set may comprise various features, such as carrier bandwidths, modulations, and/or MIMO layers, indicated for example as, 1400 MHz with 64 QAM. It is also possible that the feature set per band itself may not face any anomalous situation but only when it participates with another band in a band combination. For such cases, the second dimension is required to reflect the combination of different feature sets in given band. For example, 1000 MHz with 64 QAM in Band X and 400 MHz with 256 QAM in Band Y may be seen as potential combination for overheating in the UE.

Based on the content of the response message, the network, such as an access point, e.g. a gNB, stores (608) the list of the possible problematic capabilities and/or their combinations for said UE. Based on the information, the network and UE can proactively deal with anomalous situations (610), for example by the network adjusting RRC reconfigurations in accordance with the information.

According to a second scenario, the UE may dynamically flag features potentially causing an anomalous operating condition, for example, based on RRC reconfiguration parameters suggested by the network, thus providing feedback to the network about potentially problematic issues that may arise upon certain frequency carriers or their combinations during measurement reporting or upon RRC reconfiguration. Based on the received feedback, the network may allow the UE to switch to use a lower order configuration upon encountering an anomalous operating condition identified to be caused by certain feature(s).

According to an embodiment, the method comprises receiving RRC reconfiguration parameters from the network; running said model for the RRC reconfiguration parameters; and signaling information whether any of said RRC reconfiguration parameters infers a predetermined likelihood to cause an anomalous operating condition in said apparatus to the network.

According to an embodiment, the method comprises receiving an indication about RRC reconfiguration parameters selected by the network in a MAC Control Element; and acknowledging the RRC reconfiguration parameters in a RRC Reconfiguration complete message to the network.

According to an embodiment, the method comprises detecting a predetermined likelihood to an anomalous operating condition to take place in said apparatus; sending a request to the network for switching to a lower order RRC configuration; receiving an indication about the lower order RRC reconfiguration parameters selected by the network in a MAC Control Element; and acknowledging the lower order RRC reconfiguration parameters in a RRC Reconfiguration complete message to the network.

Some of the above embodiments relating to the UE flagging the features potentially causing an anomalous operating condition are illustrated in the flow chart of FIG. 7. At an initial stage, the UE and the network may be aware of the problematic capabilities and/or their combinations for said UE (700), and accordingly also the preferable radio reconfiguration parameters for said UE. This enables to network to test various adjustments in radio reconfiguration parameters and their effects to UE performance in terms of inferring possible anomalous operating conditions. Thus, the network sends a RRC Reconfiguration message to the UE (702), wherein the message comprises a dedicated configuration (e.g. adjustment of one or more parameters by a delta value). The RRC Reconfiguration message may also include a request to run the model according to the adjusted parameters and inform the results. The information field of the request may be referred to as, for example, ReportAnomalyInd.

Based on the adjusted parameters, the UE executes the algorithm underlying the model and compiles (704) a list of possible problematic capabilities and/or their combinations. However, the UE does not activate the RRC configuration according to adjusted parameters yet. The UE sends a RRC Reconfiguration Complete message (706) to the network. In the message, the UE specifically labels the alternative configurations using specific labels (such as AlternativeDedicatedConfiguration(s)) that allows the gNB to further activate one of the alternative configurations later on.

The network, such as the gNB, stores (708) the results of the alternative configurations along with the known preferable radio reconfiguration parameters and thus learns how said UE, which may belong to a different UE class or have different UE capabilities, would react, in terms of different performance criteria e.g., energy consumption/overheating/etc., if the reference configuration were to be used. For example, the gNB may learn the preference of different UEs by calculating a probability with which a given UE class or a given UE prefers a given RRC Configuration parameter or a combination of parameters.

Based on the radio reconfiguration parameters stored for said UE, the network may decide to continue with its original proposed configuration in the RRC Reconfiguration message (702), or the network may decide to already activate one of the known preferable radio reconfigurations for the UE. This may be performed by signaling a MAC Control element (CE) (710) to the UE, which may include the index of the selected configuration. The UE activates the configuration according to the index and sends back a RRC Reconfiguration Complete message to the network (712).

The UE then operates normally (714), until at some time instance, the UE predicts an anomalous situation (716) based on its local model and proactively indicates this to the network (718). The message used herein may be referred to as, for example, UE Anomaly Assistance Information, and it may include a request to activate a specific index out of the plurality of alternative dedicated configurations. The network accepts the request (720) and signals the activation of a new configuration by indicating the index of the selected configuration in a MAC CE (722) to the UE. The UE switches to the new configuration according to the index so as to avert the predicted anomaly and sends back a RRC Reconfiguration Complete message to the network (722).

It is noted that the method illustrated in the flow chart of FIG. 7 may also be simplified such that the UE sends only the RRC Reconfiguration Complete message including the labels of the alternative configurations (706) to the network and the network trains a model that allows predictive functionality for detecting possible anomalies before preparing the RRC reconfiguration for a UE. This implies that the network just collects feedback from the UE but does not continue further from the step 708 now including the training of the model.

According to a third scenario, the data gather by the UE about its performance in various operating conditions and identified anomalous operating conditions may be utilized in a handover situation to ensure a resource efficient handover configuration for the UE.

Some embodiments relating to the resource efficient handover configuration are illustrated in the flow chart of FIG. 8. The UE may be configured, typically under the network control, to tag the performance of all used radio configurations in terms of how they are performing compared to the performance expected based on the model. The data collected by this procedure may be referred to as, for example, UE Performance History Information.

The UE may collect this data according to a predetermined setting, or as a response to a signalling from the network. For example, the UE may collect the data across 16 or 32 radio reconfigurations, or the UE may be configured to log all information when adding component carriers of a given bandwidth etc.

In FIG. 8, the source gNB of the HO process may request the collected data from the UE (800), wherein a signalling, herein referred to as UE Performance History Information request, is sent to the UE. The request may either trigger the data collecting at the UE to be started or it may configure the UE to provide the previously collected data. Once the UE has collected the UE Performance History Information (802), the results are sent to the source gNB (804), for example in a signalling referred to as UE Performance History Information Update. In another embodiment, gNB can instruct the UE to report the UE Performance History information together with RRC measurement report.

The source gNB may have its own algorithm-based model, which has been trained and/or updated based on UE Performance History Information obtained from a plurality of UEs previously. Based on the now received UE Performance History Information, the source gNB may then further train and/or update its model (806).

As a part of the handover process, the source gNB shares the UE Performance History Information, possibly supplemented with a Resource Efficient Recommendation with potential target gNB(s), for example in the Handover Request message (808), allowing the target gNB to determine a resource efficient HO configuration. The target gNB acknowledges the determined HO configuration in the Handover Request Acknowledge message (810), which include a container possibly listing aspects of the configuration, which are meant to avoid one or more specific anomalous behavior situations.

It is noted that while FIG. 8 depicts the method using inter-gNB handover signaling, the embodiments are equally applicable in intra-gNB handover signaling.

An apparatus, such as a UE, according to an aspect comprises means for establishing a connection to a network; means for predicting, by an algorithm-based model implemented in said apparatus, whether any of configurable features of the apparatus infers a predetermined likelihood to cause an anomalous operating condition in said apparatus; and means for signaling information about said configurable features to the network.

According to an embodiment, said anomalous operating condition comprises one or more of the following:

- overheating of the apparatus;
- overload of processing resources.

According to an embodiment, said configurable features causing said anomalous operating condition comprise one or more of the following:

- frequency band(s) or band combinations for a set of frequency carriers and/or carrier bandwidths and/or modulations;
- Artificial Intelligence/Machine Learning (AI/ML) algorithms implemented in said apparatus;
- Machine Learning (ML) models processed in said apparatus;
- any combination of the above.

According to an embodiment, the apparatus comprises means for receiving a request, from the network, for informing one or more configurable features of the apparatus potentially causing an anomalous operating condition in said apparatus.