Communication method for beam failure recovery

Abstract

The present invention is related to a communication method for beam failure recovery. The method the following steps: receiving, by a terminal device from a network device, a prediction configuration; receiving, by the terminal device, system-specific and / or beam-specific reference signals; configuring, by the terminal device, based on the prediction configuration an artificial intelligence / machine learning (AI / ML) based prediction model; predicting, by the terminal device, a beam failure event by applying the AI / ML based prediction model on the system-specific and / or beam-specific reference signals; and reporting, by the terminal device to the network device, the predicted beam failure event by sending a predicted beam failure report.

Description

[0001] 202404736

[0002] 1

[0003] Description

[0004] Communication Method for Beam Failure Recovery

[0005] The present disclosure relates to a system and a method for beam failure recovery in wireless communication networks. Specifically, the present disclosure pertains to the use of artificial intelligence and machine learning based prediction models for detecting and reporting beam failure events between terminal devices and network devices.

[0006] BACKGROUND

[0007] In the field of wireless communications, particularly in advanced cellular networks such as 5G and beyond, reliable and efficient communication between terminal devices and network infrastructure is of paramount importance. One of the key challenges in these networks is the maintenance of robust connections through directional beams, which are susceptible to failures due to factors such as user mobility, environmental changes, interference, and dynamic network conditions. In current systems, beam failure recovery mechanisms are typically reactive in nature, relying on the detection of a beam failure event by the terminal device before initiating recovery procedures. These procedures often involve the transmission of beam failure recovery requests, random access attempts, and the allocation of additional resources to re-establish the communication link. Known approaches generally utilize predefined thresholds and measurement reports based on signal quality indicators such as Reference Signal Received Power (RSRP) or Reference Signal Received Quality (RSRQ) to trigger recovery actions.

[0008] Despite the substantial advances in the field of beam management and failure recovery, existing solutions exhibit several limitations. Reactive recovery mechanisms can introduce significant latency, as the system only responds after a failure has occurred, potentially resulting in service interruptions, degraded user experience, and increased signaling overhead. Furthermore, the static nature of threshold-based detection may not adequately capture the complex and dynamic 202404028

[0009] 2 radio environments encountered in real-world deployments, leading to suboptimal resource utilization and unnecessary recovery attempts. While some systems attempt to enhance reliability by increasing the frequency of measurement and reporting, this approach can further burden network resources and terminal power consumption. Additionally, the allocation of resources for random access and uplink transmissions in response to beam failures is often not optimized for the specific context or predicted needs of the terminal device, which may result in inefficient use of network capacity.

[0010] It is therefore a technical problem underlying the present invention to provide improved methods and systems for beam failure recovery in wireless communication networks that at least partially overcome the disadvantages of known systems.

[0011] SUMMARY

[0012] It is an object of this invention to provide methods that improve the reliability and efficiency of beam failure recovery in wireless communication networks, that overcomes one or more of the disadvantages of known methods.

[0013] A first aspect of the invention provides a communication method for beam failure recovery, comprising: receiving, by a terminal device from a network device, a prediction configuration; receiving, by the terminal device, system-specific and / or beam-specific reference signals; configuring, by the terminal device, based on the prediction configuration an artificial intelligence I machine learning, AI / ML, based prediction model; predicting, by the terminal device, a beam failure event by applying the AI / ML based prediction model on the system-specific and / or beam-specific reference signals; and reporting, by the terminal device to the network device, the predicted beam failure event by sending a predicted beam failure report.

[0014] The prediction configuration may alternatively or optionally comprise measurement configuration. Thus, the terminal device may receive measurement and / or prediction configuration. 202404028

[0015] 3

[0016] The subject matter describes a method for improving the recovery process when a communication beam between a user device and a network base station is predicted to fail. In this context, a terminal device may refer to a user equipment (UE), such as a mobile phone or other wireless device, and a network device may refer to a base station, such as a gNB in 5G networks. A beam in this context refers to a directional radio signal used for communication between the terminal and the network device, and beam failure refers to a situation where the quality of this signal degrades below a usable threshold.

[0017] The terminal device may be a portable or stationary communications device, such as a mobile phone, car, smart home, laptop, smart wearables or the like. The terminal device may also be referred to as user equipment (UE).

[0018] The method begins with the terminal device receiving a prediction configuration from the network device. The prediction configuration may refer to a set of parameters or criteria, such as thresholds for the probability of beam failure, time horizons for prediction, thresholds for confidence level of model outputs, or other settings that guide how the terminal device should predict beam failures using artificial intelligence or machine learning (AI / ML) techniques.

[0019] The ML model may use a feedforward deep neural network (DNN) or long short-term memory (LSTM).

[0020] The trained model may be deployed on the terminal device, on the network device or on an edge cloud device.

[0021] Model inputs may be beam-specific reference signals and / or system-specific information (reference signals). The beam-specific reference signals may comprise any of one or more Synchronization Signal Blocks (SSBs) and / or Channel State Information Reference Signals (CSI-RS). The system-specific information may comprise any of one or more of phase tracking reference signal (PTRS), and / or positioning reference signal (PRS).

[0022] The AI / ML algorithm may further use at least one (terminal device) context information as model input. Context information refers to information that may be determined by the terminal device without the use of the communication network connection. Context information may comprise at least one of terminal device location, velocity, heading, device type, power status, mobility and / or environment information. 202404028

[0023] 4

[0024] Environment information may comprise information on Line-of-Sight (LOS) or No-Line-of-Sight (NLOS), map data, and / or network identifiers (e.g., PLMN ID, cell ID). Further, the UE may utilize recorded measurement data and / or prediction history.

[0025] Environment context information such as Line-of-Sight (LOS) or No-Line-of-Sight (NLOS) may be derived based on reference signals.

[0026] The AI / ML model may have a prediction target of beam index, K-top beams, and / or codebook entry for the next slot.

[0027] For each prediction of the model, the network device may collect current input features from the terminal device and / or the network.

[0028] The ML model may output a k-top beams or a best predicted beam for the upcoming slot and / or segment of the predicted trajectory.

[0029] The terminal device then receives system-specific and / or beam-specific reference signals. System-specific reference signals may refer to signals such as Synchronization Signal Blocks (SSBs) that are common to the entire cell, while beam-specific reference signals may refer to Channel State Information Reference Signals (CSI-RS) or other signals that are specific to a particular beam. These reference signals provide the terminal device with information about the quality and characteristics of the communication link.

[0030] Based on the received prediction configuration, the terminal device configures an AI / ML-based prediction model. This means that the terminal device sets up or adapts a machine learning model, such as a neural network, to process the reference signals and make predictions about future beam failures. The prediction model may use input features such as channel state information, device location, velocity, device context, and environmental factors to predict the likelihood of a beam failure event.

[0031] The terminal device then applies the AI / ML-based prediction model to the received reference signals to predict whether a beam failure event is likely to occur. If the model determines that a beam failure is probable within a certain time frame, as specified by the prediction configuration, the terminal device generates a predicted beam failure report.

[0032] Finally, the terminal device reports the predicted beam failure event to the network device by sending the predicted beam failure report. This proactive reporting allows 202404028

[0033] 5 the network device to take preparatory actions, such as reserving resources for rapid beam recovery or handover, before the actual beam failure occurs.

[0034] The advantages of this subject matter include a shift from reactive to proactive handling of beam failures. By predicting beam failures in advance and notifying the network device, the method reduces service interruption time and mitigates the negative impact on quality of service, especially for latency-critical applications. The use of AI / ML models enables more accurate and timely predictions based on a wide range of input data, leading to more efficient resource allocation and improved user experience. This approach also reduces the signaling overhead and measurement burden on the terminal device, as the prediction process can be tailored to use only the most relevant information, and the network can optimize its response based on the predicted events.

[0035] In a further embodiment, the method may further comprise that predicting a beam failure event further comprises: determining, by the terminal device, whether at least one failure probability of the prediction of the AI / ML based prediction model is above a predefined threshold of the prediction configuration.

[0036] This embodiment enables a more nuanced determination under which conditions the predicted beam failure report shall be sent. Therefore, the step of determining if an output the prediction model is above a predefined threshold (defined in the prediction configuration) may also be referred to as reporting criteria or reporting condition. Hence, only if the reporting criteria is met (e.g., when a predefined number of thresholds is met by the AI / ML model output), the predicted beam failure report may be sent.

[0037] The implementation introduces a refinement to the mechanism by which the terminal device predicts a beam failure event using the AI / ML based prediction model. Specifically, it adds a probabilistic assessment step, wherein the terminal device evaluates whether at least one failure probability output by the AI / ML prediction model exceeds a predefined threshold specified in the prediction configuration. This enables a more quantitative beam failure prediction, allowing the system to assess confidence levels in AI / ML model outputs rather than relying on binary decisions. Communication remains primarily between the terminal device and the network device, with the terminal device receiving the prediction 202404028

[0038] 6 configuration and reference signals, and reporting predicted beam failure events. The key feature is the threshold-based probability assessment, enabling the terminal device to trigger reports based on a predefined confidence threshold, which can be dynamically adjusted by the network device to suit operational requirements or network conditions. This mechanism improves adaptability and robustness by allowing real-time tuning of prediction criteria. The probabilistic evaluation also enables the terminal device to filter prediction results before reporting, optimizing signaling efficiency and reducing unnecessary reports. Overall, the integration of a threshold-based probability check within the AI / ML prediction workflow enhances decision-making and enables more precise, context-aware beam failure recovery in wireless communication systems.

[0039] In a further embodiment, the method may further comprise that the prediction configuration comprises at least one of a probability threshold of a predicted beam failure, and / or a time threshold of an observation time period and / or prediction horizon for the AI / ML based prediction model.

[0040] The observation time period may refer to the time period during which data is collected and analyzed by an AI / ML model to understand patterns, trends, and behaviors. This window may provide the historical data necessary for the model to learn and make informed predictions. The observation window may serve as the input data set for training the model, allowing it to capture relevant features and dependencies that will influence future predictions.

[0041] The prediction horizon may be the future time period for which the AI / ML model aims to forecast or predict outcomes based on the learned patterns from the observation time period. The prediction horizon may define the scope of the model's predictive capabilities. The prediction horizon may determine how far into the future the model's predictions extend, guiding decision-making processes and strategic planning based on anticipated conditions.

[0042] A prediction horizon time may differ depending on the terminal device’s mobility. Typical values may be a time of up to 400ms or even up to 800ms.

[0043] The implementation introduces specific enhancements to the communication method for beam failure recovery by detailing the content and structure of the prediction configuration exchanged between the network device and the terminal 202404028

[0044] 7 device. In the context of the method, the communication between the network device and the terminal device is characterized by the transmission of a prediction configuration from the network device to the terminal device, which serves as a foundational mechanism for enabling the terminal device to set up and operate an artificial intelligence or machine learning based prediction model. The new feature brought by this implementation is the explicit definition of the parameters that may be included within the prediction configuration, namely a probability threshold for a predicted beam failure, a time threshold defining the observation time period, and a prediction horizon for the Al or ML based prediction model. The probability threshold parameter allows the network device to specify the minimum likelihood at which a potential beam failure event, as predicted by the Al or ML model, should be considered significant or reliable enough, e.g., compared to a configured confidence level threshold, to trigger further action, such as reporting the event to the network device. This enables a more nuanced and adaptive approach to beam failure detection, as the sensitivity of the prediction model can be dynamically (adaptively) adjusted based on network conditions or service requirements. The time threshold of the observation time period provides a temporal boundary within which the reference signals are to be analyzed by the prediction model, ensuring that the model’s predictions are based on the most relevant and recent data, thereby improving the accuracy and timeliness of beam failure predictions. The prediction horizon parameter defines the future time interval over which the model is expected to forecast potential beam failures, allowing the terminal device to anticipate and report beam failures before they occur, thus facilitating proactive network management and reducing service interruptions. By specifying these parameters in the prediction configuration, the network device gains granular control over the operation of the terminal device’s prediction model, enabling tailored optimization for different deployment scenarios, user requirements, or network policies. The communication mechanism is thus enhanced by the inclusion of these configurable thresholds and time parameters, which are conveyed from the network device to the terminal device and directly influence the behavior of the Al or ML based prediction model. The new features introduced by this implementation provide significant technical advantages by allowing the prediction process to be both context-aware 202404028

[0045] 8 and adaptable, thereby improving the reliability and efficiency of beam failure recovery in advanced wireless communication systems.

[0046] In a further embodiment, the method may further comprise: receiving a resource configuration comprising a resource reserved for random access, RA, and / or uplink (UL) data transmissions of the terminal device; and performing RA and / or UL data transmissions using the reserved resource.

[0047] The receiving a resource configuration may happen after sending predicted beam failure report and before performing RA and / or UL data transmissions.

[0048] R, and / or UL data transmissions is performed using the resources reserved for RA and / or UL data transmissions.

[0049] The resource configuration reserved for the terminal device may comprise information on a new beam, RA configuration (e.g., for contention free RA), and / or a configured grant for UL data transmission assigned to the terminal device.

[0050] The implementation introduces additional mechanisms and features that enhance the communication method for beam failure recovery by specifying how the terminal device interacts with the network device regarding resource allocation and subsequent communication actions. Specifically, the implementation provides that the terminal device receives a resource configuration from the network device, where this configuration includes resources that are reserved for random access and / or uplink data transmissions. This mechanism establishes a direct communication pathway whereby the network device proactively informs the terminal device about specific resources that are set aside for its use, thereby facilitating efficient and prioritized access to the communication medium. The resource configuration may encompass parameters such as time, frequency, or code resources that are dedicated to the terminal device, ensuring that when a beam failure event is predicted or occurs, the terminal device can promptly initiate random access procedures or transmit uplink data without contention or delay. The new feature brought by this implementation is the integration of reserved resource allocation into the beam failure recovery process, which enables the terminal device to perform random access or uplink data transmissions using the reserved resource. This reserved allocation reduces the risk of collision with other devices and minimizes latency in re-establishing communication or reporting critical events, such 202404028

[0051] 9 as predicted beam failures. By ensuring that the terminal device has pre-assigned resources for these purposes, the method improves the reliability and responsiveness of the system, particularly in scenarios where rapid recovery from beam failure is essential for maintaining service continuity. The communication between the network device and the terminal device regarding resource configuration enables dynamic adaptation of resource usage based on the terminal device's needs and network status. Immediate random access or uplink transmissions using reserved resource allow the terminal device to respond quickly to predicted or actual beam failures, improving the robustness and efficiency of beam failure recovery. This is particularly valuable in dense networks or where low latency and high reliability are required, as it reduces delays and uncertainties of contention-based access. The implementation thus adds resource management and prioritized communication that strengthens the terminal device's ability to recover from beam failures and maintain connectivity.

[0052] In a further embodiment, the method may further comprise that the AI / ML based prediction model outputs at least one preferable beam for random access, RA, and / or UL data transmission; and the report information further comprises, information on at least one preferable beam for random access, RA, and / or UL data transmission predicted by the terminal device.

[0053] Preferable beam in this context refers to a result of a sophisticated AI / ML model that may be trained to accurately predict the at least one next „ best beam" by taking beam-specific reference signals and / or system-specific reference signals into consideration. Best beam may refer to a beam with most preferable properties, such as stability, given the current condition of the terminal device (e.g., high or low mobility, low energy, high throughput).

[0054] This embodiment may enable the terminal device to continue UL data transmissions in the preferable (next best) beam without having to performing RA procedure. The implementation specifies mechanisms and enhancements to the communication method for beam failure recovery by defining the nature and content of outputs generated by the AI / ML based prediction model and the information included in the report sent from the terminal device to the network device. In particular, the implementation provides that the AI / ML based prediction model 202404028

[0055] 10 predicts the occurrence of a beam failure event and identifies and outputs at least one preferable beam for random access procedures and / or uplink data transmission. In addition to detecting or forecasting a potential beam failure, the model determines alternative beams that are optimal or suitable for maintaining or restoring communication, thereby increasing the robustness and efficiency of the recovery process. The communication mechanism between the terminal device and the network device is correspondingly specified: the report transmitted by the terminal device includes information regarding the preferable beam or beams identified by the prediction model for use in random access or uplink data transmission. This information enables the network device to receive data indicating which beams the terminal device considers optimal for subsequent communication attempts, thereby allowing for informed and expedited coordination of recovery actions. The implementation provides integration of beam selection recommendations within the failure prediction and reporting process, leveraging the predictive capabilities of AI / ML to anticipate failures and to suggest remedial actions. This differs from aspects that focus only on detection and reporting of predicted beam failures without specifying inclusion of recommended beams for recovery. By including information about preferable beams for random access or uplink transmission within the report, the method enables efficient beam recovery, reduces latency, and improves reliability in scenarios where beam failure is imminent or has occurred. This enhancement enables a proactive and adaptive communication protocol between the terminal and network devices, utilizing AI / ML for prediction and for dynamic adaptation and optimization of beam management in real time.

[0056] In a further embodiment, the method may further comprise that the resource configuration further comprises a validity timer, and if the validity timer is not expired, performing RA and / or UL data transmission according to the reserved resource; or if the validity timer is expired, sending a new predicted beam failure report and releasing the initially reserved resource.

[0057] When the validity times is not expired, the terminal device may further stop or reset the validity timer after performing RA and / or UL data transmission according to the reserved resource. 202404028

[0058] 11

[0059] If the validity time expired, the terminal device may optionally check if a beam failure event is still predicted (e.g., if the reporting criteria is still met / valid), and if so, another predicted beam failure report may be sent. This predicted beam failure may still comprise a same preferable beam as indicated in the previous predicted beam failure report (plausibility check, helping to avoid unnecessarily performed RA; optionally, this may be transmitted to the network device using the UL data transmission). Alternatively, using a same preferable beam as indicated in the previous predicted beam failure report may be ruled out.

[0060] If the validity timer is still valid, resources reserved for random access, RA, and / or uplink, UL, data transmissions of the terminal device may be released.

[0061] The implementation introduces a mechanism in which the resource configuration, as provided to the terminal device, is augmented by the inclusion of a validity timer. This timer serves as a temporal control for the reserved resource allocated for random access or uplink data transmission following a predicted beam failure event. The communication between the terminal device and the network device is thus enhanced by the ability of the terminal device to autonomously manage the reserved resource based on the status of the validity timer. Specifically, as long as the validity timer remains unexpired, the terminal device is permitted to proceed with random access and / or uplink data transmission using the previously reserved resource, thereby ensuring efficient and timely use of network resources in response to a predicted beam failure. This mechanism allows for a seamless and responsive recovery process, minimizing latency and potential service disruption. In the event that the validity timer expires before the reserved resource are utilized, the terminal device is required to initiate a new communication with the network device by sending an updated predicted beam failure report. Concurrently, the terminal device releases the initially reserved resource, thereby preventing unnecessary occupation of network resources and enabling their reallocation for other purposes. This feature introduces a dynamic and adaptive approach to resource management in the context of beam failure recovery, as it ensures that reserved resource is only held for a defined period, and that the network is promptly informed of any changes in the prediction status. The validity timer thus acts as a safeguard against resource wastage and enhances the overall efficiency of the beam failure recovery process. The communication mechanisms involved include the initial reception of the 202404028

[0062] 12 resource configuration with the validity timer from the network device, the autonomous monitoring of the timer by the terminal device, the conditional execution of random access or uplink data transmission, and the subsequent reporting and release actions based on the timer's status. The new feature provided by this implementation is the integration of a time-based control for resource reservation, which enables more granular and efficient management of network resources in scenarios involving AI / ML-based beam failure prediction and recovery, thereby improving both network performance and user experience.

[0063] In a further embodiment, the method may further comprise that the system-specific and / or beam-specific reference signals comprise at least one of:

[0064] - beam-specific reference signals, comprising any of one or more Synchronization Signal Blocks (SSBs) and / or Channel State Information Reference Signals (CSI-RS); and / or

[0065] - system-specific reference signals, comprising phase tracking reference signal (PTRS), and / or positioning reference signal (PRS).

[0066] The newly introduced feature specifies that the system-specific and / or beam-specific reference signals, which are utilized by the terminal device, include at least one of the following: beam-specific reference signals, which may comprise one or more synchronization signal blocks or channel state information reference signals, and system-specific reference signals, which may comprise phase tracking reference signal (PTRS), and / or positioning reference signal (PRS). This enhancement directly impacts the mechanisms of communication between the terminal device and the network device by broadening the scope and granularity of the reference signals that are received and processed. In the context of the AI / ML-based prediction model, the inclusion of these specific types of reference signals enables the terminal device to leverage a richer and more diverse dataset for the purposes of predicting beam failure events. The beam-specific reference signals, such as synchronization signal blocks and channel state information reference signals, provide fine-grained, real-time information about the state and quality of specific beams, which is crucial for accurately assessing imminent beam failures. On the other hand, the system-specific reference signals, including parameters such as user equipment location, velocity, context information, and 202404028

[0067] 13 environmental factors, introduce a broader contextual awareness into the prediction process, allowing the AI / ML model to account for dynamic changes in the operating environment and user behavior that may influence beam stability. The communication mechanism is thus characterized by the terminal device receiving these enriched reference signals from the network device, integrating them into the AI / ML-based prediction model as configured by the prediction configuration, and subsequently utilizing this comprehensive input to enhance the accuracy and reliability of beam failure predictions. This, in turn, allows for more timely and contextually informed reporting of predicted beam failure events back to the network device. The new feature, therefore, not only refines the nature of the data exchanged between the terminal and network devices but also significantly augments the predictive capabilities of the AI / ML model by allowing it to process and learn from a multidimensional set of reference signals, ultimately improving the robustness and responsiveness of beam failure recovery mechanisms in the communication system.

[0068] A second aspect of the invention provides a communication method for beam failure recovery, comprising: sending, by a network device to a terminal device, a prediction configuration for configuring an artificial intelligence I machine learning, AI / ML, based prediction model; and receiving, by the network device, a predicted beam failure report from the terminal device, comprising a beam failure event predicted based on the prediction configuration.

[0069] The prediction configuration may comprise at least one of a probability threshold of a predicted beam failure, and / or a time threshold of an observation time period and / or prediction horizon and / or a confidence level threshold for the AI / ML based prediction model.

[0070] The prediction configuration may alternatively or optionally comprise measurement configuration. Thus, the network device may determine and send measurement and / or prediction configuration.

[0071] The network device may be a base station or the like. The network device may be also referred to as NodeB, gNB and so forth. The network device could also be a UE, acting as network device when communication with at least one other terminal device. 202404028

[0072] 14

[0073] The subject matter concerns a communication method designed to enhance the recovery process from beam failures in wireless networks, particularly in scenarios where maintaining continuous and high-quality service is critical. In this approach, a network device, which may refer to a base station or gNB, transmits a prediction configuration to a terminal device, which may refer to a user equipment (UE) such as a mobile phone or other wireless terminal. This prediction configuration is indicative of certain parameters: a probability threshold, an observation time period, and / or a prediction horizon. The probability threshold may refer to the minimum likelihood at which a beam failure event is considered significant enough to be reported. The observation time period may refer to the time period over which the terminal device monitors relevant metrics to assess the likelihood of beam failure, while the prediction horizon may refer to the future time interval for which the beam failure prediction is made. The confidence level threshold refers to the quality and / or reliability of the AI / ML model output.

[0074] Upon receiving this configuration, the terminal device utilizes artificial intelligence or machine learning (AI / ML) models to monitor and predict potential beam failures based on the configured parameters. When the terminal device predicts that a beam failure event is likely to occur, as determined by the probability threshold and within the specified observation time period or prediction horizon, it generates and transmits a predicted beam failure report to the network device. This report includes information about the anticipated beam failure event, enabling the network device to take proactive measures.

[0075] The advantages of this subject matter are significant. By enabling the terminal device to predict and report potential beam failures before they occur, the method shifts the recovery process from a reactive approach, where action is only taken after a failure has been detected, to a proactive approach. This proactive reporting allows the network device to reserve necessary resources and prepare for beam switching or recovery in advance, thereby reducing service interruption time. Such a reduction in interruption is particularly beneficial for latency-critical services, where even brief disruptions can degrade user experience. Furthermore, the use of AI / ML-based prediction leverages advanced data analysis and contextual awareness, potentially increasing the accuracy and timeliness of beam failure detection compared to traditional methods that rely solely on historical 202404028

[0076] 15 measurements or static thresholds. This approach also mitigates the impact on quality of service and supports more robust and reliable wireless communications in dynamic environments.

[0077] In a further embodiment, the method may further comprise that the method further comprises: upon receiving the predicted beam failure report, determining, by the network device, a resource configuration comprising a resource reserved for random access, RA, and / or uplink, UL, data transmission of the terminal device; and sending, by the network device, the resource configuration to the terminal device. Wherein the resource configuration may comprise information on a new beam assigned to the terminal device.

[0078] The implementation introduces additional communication mechanisms between the network device and the terminal device that build upon the Al-based beam failure prediction framework. Specifically, after the network device receives a predicted beam failure report from the terminal device, the network device determines a resource configuration for random access (RA) and / or uplink (UL) data transmission resources that are reserved for the terminal device. This determination process involves the network device assessing the predicted beam failure event and proactively identifying suitable RA and / or UL data transmission resources that will facilitate rapid recovery or continued communication in the event of actual beam failure. Once the appropriate resource configuration is determined, the network device communicates this resource configuration to the terminal device. The present communication mechanism is characterized by the network device transmitting specific resource allocation information, which may include parameters or identifiers related to RA and / or UL data transmission opportunities, directly to the terminal device. This ensures that the terminal device is pre-informed and prepared to utilize the reserved resource without delay if the predicted beam failure materializes. The new feature introduced by this implementation is the proactive reservation and explicit communication of RA and / or UL data transmission resources in response to a predicted beam failure, as opposed to merely reporting or reacting to beam failure events. This enhancement enables the network to not only anticipate potential disruptions based on Al-driven prediction but also to preemptively allocate and inform the terminal device of dedicated resources for 202404028

[0079] 16 rapid access or data transmission. As a result, the overall beam failure recovery process becomes more efficient and responsive, reducing latency and improving reliability in scenarios where beam failure is anticipated. The communication between the network device and the terminal device thus extends beyond the initial prediction configuration and reporting, encompassing a dynamic exchange where the network device takes an active role in resource management and recovery facilitation. This approach leverages the predictive capabilities of the terminal device and the resource allocation authority of the network device, resulting in a coordinated and intelligent beam failure recovery mechanism that enhances the robustness of the communication system.

[0080] In a further embodiment, the method may further comprise that the predicted beam event report further comprises information on at least one preferable beam for RA and / or UL data transmission predicted by the terminal device, and determining resource configuration further comprises: assigning at least one preferable beam preferred by the terminal device as the resource reserved for RA and / or UL data transmission.

[0081] The implementation introduces a specific mechanism in which the terminal device, upon predicting a beam failure, not only reports the predicted beam failure event to the network device but also includes in its report information regarding at least one preferable beam for random access (RA) and / or uplink (UL) data transmission. This preferable beam is identified by the terminal device using an AI / ML model and represents the beam or beams that are predicted to provide superior performance in terms of speed, stability, or noise characteristics. The communication between the terminal device and the network device is thus enhanced by the inclusion of this additional information in the predicted beam event report. Upon receiving this enriched report, the network device is able to determine and assign resources for RA and / or UL data transmission specifically in the preferable beam or beams indicated by the terminal device. This mechanism allows the network device to tailor its resource allocation in response to the terminal device’s AI / ML-based predictions, thereby improving the efficiency and reliability of beam failure recovery. The new feature brought by this implementation is the integration of AI / ML-driven beam selection into the beam failure recovery process, enabling the network device to 202404028

[0082] 17 make informed decisions about resource configuration based on the predicted optimal beams for communication. This not only enhances the adaptability of the system to dynamic channel conditions but also leverages the predictive capabilities of the terminal device to optimize communication performance, reducing the likelihood of further beam failures and improving overall system throughput and robustness. The preferable beam information, being derived from AI / ML analysis, reflects real-time assessments of which beams are likely to maintain superior link quality, thus enabling proactive and intelligent management of radio resources in response to anticipated beam failures.

[0083] In a further embodiment, the method may further comprise that the RA resource configuration further comprises that the RA resource configuration further comprises a validity timer, and when the validity timer expires, releasing the resource reserved for random RA and / or UL data transmission, and, when the terminal device performs RA and / or UL data transmission, the validity timer stops.

[0084] The implementation introduces a specific mechanism for managing the allocation and release of random access and uplink data transmission resources through the use of a validity timer within the RA resource configuration. The communication between the network device and the terminal device is enhanced by this timer-based control, which governs the lifecycle of the resource configuration. When the validity timer expires, the system is configured to automatically release the assigned random access and / or uplink data transmission resources, thereby ensuring that these resources are not held unnecessarily and can be efficiently reallocated for other operations or users. This mechanism prevents resource wastage and potential bottlenecks in scenarios where the terminal device no longer requires the allocated resources, either due to inactivity or completion of the intended transmission. Furthermore, the implementation specifies that the validity timer is paused or stopped when the terminal device is actively performing random access or uplink data transmission. This feature ensures that the resources remain available for the duration of active communication, preventing premature release that could disrupt ongoing transmissions. The introduction of the validity timer adds a dynamic and adaptive aspect to the resource management process, allowing the 202404028

[0085] 18 system to respond intelligently to the actual usage patterns of the terminal device. This not only optimizes resource utilization but also enhances the reliability and efficiency of the communication method, particularly in environments where beam failure recovery and rapid reallocation of resources are critical. The timer-based mechanism provides a clear and automated protocol for both the retention and release of communication resources, reducing the need for manual intervention or additional signaling between the network and terminal devices. By integrating this feature, the method achieves a balance between maintaining resource availability during active use and ensuring prompt release when resources are no longer needed, thus supporting both high system throughput and robust beam failure recovery. This improvement is particularly relevant in the context of Al-based beam failure prediction and reporting, as it complements the predictive allocation of resources with an intelligent release mechanism, further streamlining the overall communication process between the network device and the terminal device.

[0086] A third aspect of the invention provides a terminal device, configured to carry out the method according to any one of the implementations.

[0087] The aspect defines a terminal device that is specifically adapted to implement the method described in any of previous implementations. In this context, a terminal device may refer to a user equipment (UE) such as a mobile phone, tablet, or any wireless communication device capable of connecting to a radio access network. The method referenced involves an artificial intelligence or machine learning (AI / ML)-based approach to beam failure recovery in wireless communications, particularly in 5G or advanced networks.

[0088] The terminal device is configured to perform proactive beam failure recovery by utilizing AI / ML algorithms to predict the likelihood of an impending beam failure event. Beam failure may refer to a condition where the communication link between the terminal device and the base station (gNB) degrades or is lost due to changes in the radio environment, such as user movement or obstacles. Traditional methods for beam failure recovery are reactive, meaning they respond only after a failure has occurred, which can result in service interruption and degraded quality of service, especially for latency-sensitive applications. 202404028

[0089] 19

[0090] In contrast, the subject matter of this aspect enables the terminal device to monitor certain parameters, such as channel state information (CSI), device location, velocity, and environmental context, and apply an AI / ML-based prediction model. This model determines whether the probability of a beam failure event exceeds a configured threshold within a specified prediction horizon. If the prediction indicates a high likelihood of beam failure, the terminal device proactively reports this to the base station.

[0091] Upon receiving this report, the base station can reserve random access (RA) resources on a specific beam in advance, and configure a timer indicating how long these resources remain valid. The terminal device then uses this pre-reserved resource to quickly re-establish communication on a new beam if the predicted failure occurs, thereby minimizing service interruption.

[0092] The advantages of this approach include a proactive rather than reactive response to beam failure, significantly reduced interruption time during beam recovery, and improved quality of service for applications that are sensitive to latency. By leveraging AI / ML-based prediction, the terminal device can anticipate and mitigate potential connectivity issues before they impact user experience, rather than relying solely on historical measurements or reactive mechanisms. This results in a more robust and efficient communication system, particularly beneficial in dynamic environments where rapid mobility or changing conditions can frequently lead to beam failures.

[0093] A fourth aspect of the invention provides a network device configured to carry out the method according to any one of the implementations.

[0094] The subject matter concerns a network device that is specifically designed to implement a method for AI / ML-based beam failure recovery as described in earlier implementations. In this context, a network device may refer to a base station or gNB in a wireless communication system, such as those used in 5G networks. The device is configured, meaning it is equipped with the necessary hardware and software, to perform a series of operations that enable proactive management of beam failures using artificial intelligence and machine learning techniques.

[0095] The method carried out by this device involves receiving reports from user equipment (UE) that predict potential beam failures based on AI / ML models. These 202404028

[0096] 20 models may use input features such as channel state information, UE location, velocity, device context, and environmental factors to forecast the likelihood of a beam failure event within a specified prediction horizon. Upon receiving such a prediction report from the UE, the network device reserves random access (RA) resources on a particular beam, which allows the UE to quickly initiate a recovery procedure if the predicted beam failure occurs. The network device also configures a timer that defines the validity period of these reserved resource, ensuring efficient resource management and minimizing unnecessary resource occupation.

[0097] By implementing this approach, the network device enables a proactive rather than reactive response to beam failures. This results in significantly reduced service interruption times and helps maintain high quality of service, particularly for latency-sensitive applications. The use of AI / ML for event prediction and resource allocation optimizes network performance, reduces signaling overhead, and mitigates the negative impact of beam failures on user experience.

[0098] A fifth aspect of the invention provides a computer program, comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method according to any one of the implementations, or the method according to any one of the implementations.

[0099] The subject matter described is a computer program comprising instructions that, when executed by a computer, cause the computer to perform a method for proactive beam failure recovery in wireless communication systems, as defined in preceding implementations. The computer program may refer to software code or a set of executable instructions, which can be stored on a non-transitory medium and executed by a processor within a network node or user equipment.

[0100] The instructions in this program implement a method in which a user equipment (UE) utilizes artificial intelligence or machine learning (AI / ML) models to predict the likelihood of a beam failure event. A beam failure event may refer to a situation where the communication link between the UE and the base station (gNB) via a particular beam is likely to degrade or be lost. The prediction is based on configured criteria, which may include a probability threshold for beam failure and a look-ahead time period, and is enabled by configuration information provided by the gNB. The UE monitors relevant reference signals, such as synchronization signal blocks 202404028

[0101] 21

[0102] (SSB) or channel state information reference signals (CSI-RS), and applies the AI / ML model to determine whether the probability of an impending beam failure exceeds the configured threshold.

[0103] If the criteria are met, the UE proactively reports the predicted beam failure event to the gNB. Upon receiving this report, the gNB reserves random access (RA) resources in a specific beam, thereby preparing for a rapid recovery procedure. The UE may then trigger a random-access procedure to indicate a new beam to the gNB. The gNB also configures a timer value that defines the validity period for the reserved resource. The validity timer starts when the UE receives the RA resource configuration and stops when the UE performs the random-access procedure. If the timer expires before the procedure is performed, the UE releases the reserved resource.

[0104] The terms used in this context may refer to the following: UE may refer to user equipment, such as a mobile device; gNB may refer to a next-generation Node B, or base station; AI / ML event may refer to a trigger based on artificial intelligence or machine learning prediction; RA resources may refer to resources allocated for random access procedures; SSB and CSI-RS may refer to specific types of reference signals used for beam management; and validity timer may refer to a timer that determines how long the reserved resource remain available.

[0105] The advantages of this subject matter are significant. By enabling proactive, AI / ML-based prediction and reporting of beam failures, the procedure reduces service interruption time compared to conventional reactive methods. This improvement is particularly beneficial for latency-critical services, as it mitigates the impact of beam failures on quality of service. The approach also reduces unnecessary signaling and measurement overhead, as the prediction models can be tailored to minimize resource consumption. Overall, the computer program enables a more efficient, intelligent, and responsive beam failure recovery process in advanced wireless networks.

[0106] A sixth aspect of the invention provides a computer-readable storage medium comprising instructions which, when executed by a computer, cause the computer to carry out the method according to any one of the implementations, or the method according to any one of the implementations. 202404028

[0107] 22

[0108] The subject matter concerns a computer-readable storage medium that holds instructions which, when executed by a computer, enable the computer to perform a method as described in earlier implementations. The term computer-readable storage medium may refer to any physical device or medium capable of storing digital information, such as a hard drive, solid-state drive, optical disc, or memory card. The instructions stored on this medium are intended to be executed by a computer, which may refer to any programmable processing device, including servers, base stations, or user equipment (UE) in a wireless communication network.

[0109] The method performed by the computer, as referenced in previous implementations, relates to a proactive beam failure recovery procedure in a wireless communication system, particularly in the context of AI / ML-based mobility management. In this context, beam failure refers to the loss or degradation of a directional wireless link between a user equipment and a base station (gNB), which can lead to service interruption. Traditional beam failure recovery procedures are reactive, meaning they respond to failures after they occur, resulting in longer interruption times and degraded service quality, especially for latency-sensitive applications.

[0110] The subject matter enables a proactive approach, where the user equipment utilizes artificial intelligence or machine learning models to predict the likelihood of an impending beam failure based on various input parameters such as channel state information, device location, velocity, and environmental factors. Upon predicting a potential beam failure event, the user equipment reports this prediction to the base station, which then reserves random access (RA) resources on a specific beam in anticipation of the failure. The base station also configures a timer that determines the validity period for the reserved resource. This arrangement allows the user equipment to rapidly initiate a recovery process if the predicted failure occurs, thereby minimizing service interruption.

[0111] The advantage of this approach lies in its proactive nature, which significantly reduces the time required to recover from beam failures. By anticipating failures and preparing recovery resources in advance, the system can maintain higher service quality and reduce the impact of interruptions, which is particularly beneficial for delay-critical services. The use of AI / ML models for prediction further enhances the 202404028

[0112] 23 accuracy and efficiency of the recovery process, as it allows for more informed and timely decisions based on real-time and contextual data. The storage medium encapsulates these benefits by providing a means to deploy the described method in a practical and scalable manner within wireless communication networks.

[0113] Furthermore, a system, comprising of at least one terminal device according to the third aspect and a network device according to the fourth aspect, may be defined. Although, only described for one terminal device, the application is not limited thereto. The described method may be applied to networks with at least two terminal devices. The described method may also be applied to a network with more than one network device.

[0114] The provided solution enables a proactive approach to beam failure recovery by utilizing AI / ML-based prediction and event reporting, in contrast to the reactive procedures of the prior art. By predicting potential beam failures and reserving resources in advance, this solution significantly reduces service interruption time. This proactive strategy ensures improved quality of service, particularly for latency-critical applications. As a result, the disclosed methods offer a substantial improvement over traditional reactive recovery mechanisms by mitigating the negative impact of beam failures on user experience.

[0115] Figures

[0116] Fig. 1 provides a schematic flowchart illustrating a beam failure recovery method of a network device (gNB);

[0117] Fig. 2 provides a schematic flowchart illustrating a beam failure recovery method of a terminal device (UE);

[0118] Fig. 3 provides a schematic flowchart illustrating another beam failure recovery method of a terminal device (UE). 202404028

[0119] 24

[0120] Detailed description

[0121] Figure 1 illustrates a flowchart representing the iterative process performed by a gNB (network device) in the context of Al-based beam failure prediction and recovery according to the present disclosure and optional aspects. The process begins with the gNB, which initiates the procedure by optionally determining prediction configuration and report triggering conditions specifically for the reporting of Al-based beam failure predictions. This determination allows for setting criteria under which beam failure events will be predicted and subsequently reported by user equipment. Alternatively, a predefined (default) prediction configuration could be used.

[0122] Following this, the gNB provides the prediction configuration to the user equipment. This configuration includes the parameters and instructions necessary for the user equipment to configure an AI / ML-based prediction model to perform predictions related to beam failure events, as specified by the gNB.

[0123] The process then proceeds to a decision stage where the network device evaluates whether any user equipment report has been received. This decision node branches the process based on the presence or absence of such reports. If no user equipment reports are received, the process loops back, maintaining the current state and awaiting incoming reports. Alternatively, the gNB may also assess if a new prediction configuration shall be determined and sent.

[0124] If user equipment reports are received, the process advances to the next operation, wherein the gNB assigns random access (RA) resources and (optionally) configures a validity timer. The assignment of RA resources ensures that the user equipment is provided with the necessary resources for random access procedures, while the configuration of the validity timer establishes the temporal validity of these resource assignments. The assigned RA resources may also be a preferable beam for RA and / or UL data transmission predicted by the terminal device. Not illustrated herein is, that upon the validity timer expires, the assigned RA resource is released. The UE may start to perform RA and / or UL data transmission (and indicate this to the gNB) whilst the validity timer is not yet outdated.

[0125] The process concludes after the assignment of RA resources and the configuration of the validity timer, as indicated by the terminal node labeled "End". 202404028

[0126] 25

[0127] Figure 2 illustrates a flowchart depicting the iterative sequence of operations performed by a user equipment (UE) for beam failure prediction and reporting according to the present disclosure and optional aspects. The process commences with the UE in an initial state. The first decision block determines whether the UE has received a prediction configuration. If the configuration has not been received, the process loops back, maintaining the UE in a waiting state until the required configuration is obtained. Alternatively, the UE may also use a predetermined (default) prediction configuration, either from a previous communication with a gNB or an internally saved fallback configuration.

[0128] Upon receipt of the prediction configuration, the UE proceeds to execute a prediction of beam failure events. This prediction operation utilizes the received configuration parameters to assess the likelihood of beam failure, leveraging AI / ML-based models configured according to the prediction configuration, as indicated in the summary section.

[0129] Following the prediction step, the process advances to an optional second decision block where the UE evaluates whether the reporting criteria are fulfilled. The reporting criteria may be defined by thresholds or conditions specified within the prediction configuration, such as probability thresholds or observation time period parameters. Alternatively, predefined threshold criteria may be used.

[0130] If the reporting criteria are not fulfilled, the process returns to the prediction step, thereby enabling continuous or periodic assessment of beam failure events based on updated measurements or configurations. If the reporting criteria are fulfilled, the UE proceeds to report the predicted beam failure events. This reporting action involves transmitting a report containing the predicted beam failure information to the relevant network entity (gNB).

[0131] The process concludes after the reporting step, marking the end of the operational sequence for this particular prediction and reporting cycle. Throughout the flowchart, the logical flow ensures that the UE only reports predicted beam failures when both the necessary configuration is present and the defined reporting criteria are satisfied, thereby optimizing signaling efficiency and network resource utilization. 202404028

[0132] 26

[0133] Figure 3 illustrates an iterative flowchart depicting the procedure following the reporting of predicted beam failure events according to the present disclosure and optional aspects. The preceding steps up to “Report predicted beam failure events” are already described above for Figure 2. After the step of reporting predicted beam failure events, the process continues with determining as to whether a random access (RA) resource configuration has been received by the user equipment (UE). If the RA resource configuration has not been received, the process waits until such a configuration is available. Upon receipt of the RA resource configuration, the next (optional) operation is to evaluate whether the RA resource validity timer has expired.

[0134] If the validity timer has expired, the process proceeds to release the RA resources, thereby terminating the reserved allocation for random access. Furthermore, the UE may check if the report criteria (predicted beam failure threshold is met) is still met. If so, The UE sends a new predicted beam failure report to the gNB. If not, the process ends.

[0135] If the validity timer has not expired, the UE performs the RA and / or UL data transmissions procedure and simultaneously stops the validity timer. The process then transitions to the end state.

[0136] The flowchart thus details the conditional logic and operational sequence for handling RA resource allocation and release in response to predicted beam failure events, including the management of the validity timer associated with the RA resources. The interconnection between the decision nodes ensures that RA resources are only utilized within the validity period and are released promptly upon timer expiration, thereby optimizing resource usage in the context of beam failure recovery.

[0137] While the invention has been illustrated and described in detail in the drawings and foregoing description, such illustration and description are to be considered exemplary and not restrictive. The invention is not limited to the disclosed embodiments. In view of the foregoing description and drawings it will be evident to a person skilled in the art that various modifications may be made within the scope of the invention, as defined by the claims. 202404028

[0138] 27

[0139] Abbreviations:

[0140] Al artificial intelligence

[0141] ML machine learning

[0142] RA random access UL uplink

[0143] SSB synchronization signal block

[0144] CSI-RS channel state information resource reference signal

[0145] LOS Line-of-Sight

[0146] NLOS Non-Line-of-Sight

Claims

1. 20240402828Patent claims1. A communication method for beam failure recovery, comprising: receiving, by a terminal device from a network device, a prediction configuration; receiving, by the terminal device, system-specific and / or beam-specific reference signals; configuring, by the terminal device, based on the prediction configuration an artificial intelligence I machine learning, AI / ML, based prediction model; predicting, by the terminal device, a beam failure event by applying the AI / ML based prediction model on the system-specific and / or beam-specific reference signals; and reporting, by the terminal device to the network device, the predicted beam failure event by sending a predicted beam failure report.

2. The method according to claim ^characterized in that predicting a beam failure event further comprises: determining, by the terminal device, whether at least one failure probability of the prediction of the AI / ML based prediction model is above a predefined threshold of the prediction configuration.

3. The method according to any of the previous claims, characterized in that the prediction configuration comprises at least one of a probability threshold of a predicted beam failure, and / or a time threshold of an observation time period, and / or prediction horizon, and / or a confidence level threshold for the AI / ML based prediction model.

4. The method according to any of the previous claims, characterized in that the method further comprises: receiving a resource configuration comprising a resource reserved for random access, RA, and / or uplink, UL, data transmission of the terminal device; and performing RA and / or UL data transmissions using the reserved resource.202404028295. The method according to any of the previous claims, characterized in that the AI / ML based prediction model outputs at least one preferable beam for random access, RA, and / or UL data transmission; and the report information further comprises, information on at least one preferable beam for RA and / or UL data transmission predicted by the terminal device.

6. The method according to claim 5, characterized in that the resource configuration further comprises a validity timer, and if the validity timer is not expired, performing RA and / or UL data transmission according to the reserved resources; or if the validity timer is expired, sending a new predicted beam failure report and releasing the initially reserved resources.

7. The method according to any of the previous claims, characterized in that the system-specific and / or beam-specific reference signals comprise at least one of:- beam-specific reference signals, comprising any of one or more synchronization signal blocks, SSBs, and / or channel state information resource reference signal, CSI-RS; and- system-specific reference signals, comprising any of phase tracking reference signal, PTRS, and / or positioning reference signal, PRS.

8. A communication method for beam failure recovery, comprising: sending, by a network device to a terminal device, a prediction configuration for configuring an artificial intelligence I machine learning, AI / ML, based prediction model; and receiving, by the network device, a predicted beam failure report from the terminal device, comprising a beam failure event predicted based on the prediction configuration.

9. The method according to claim 8, characterized in that the method further comprises:20240402830 upon receiving the predicted beam failure report, determining, by the network device, a resource configuration comprising a resource reserved for random access, RA, and / or uplink, UL, data transmission of the terminal device; and sending, by the network device, the resource configuration to the terminal device.

10. The method according to claim 9, c h a r a c t e r i z e d i n that the predicted beam event report further comprises information on at least one preferable beam for RA and / or UL data transmission predicted by the terminal device, and determining a resource configuration further comprises assigning at least one preferable beam preferred by the terminal device as the resource reserved for RA and / or UL data transmission.11 . The method according to any of claims 9 or 10, c h a r a c t e r i z e d i n that the resource configuration further comprises a validity timer, when the validity timer expires, releasing the resource reserved for random RA and / or UL data transmission, and when the terminal device performs RA and / or UL data transmission, the validity timer stops.

12. A terminal device, configured to carry out the method according to any one of claims 1 to 7.

13. A network device configured to carry out the method according to any one of claims 8 to 11 .

14. A computer program, comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method according to any one of claims 1 to 7, or the method according to any one of claims 8 to 11 .

15. A computer-readable storage medium comprising instructions which, when executed by a computer, cause the computer to carry out the method according to any one of claims 1 to 7, or the method according to any one of claims 8 to 11 .

Images