Event detection for channel state information compression

Abstract

This application relates to event detection for channel state information (CSI) compression. This application relates to apparatus and components, including means, systems, and methods for event detection associated with CSI compression.

Description

[0001] Cross-references to related applications

[0002] This application claims priority to Greek application No. 20250100066, filed on January 31, 2025, entitled “Event Detection for ChannelState Information Compression,” the entire contents of which are incorporated herein by reference for all purposes. Technical Field

[0003] This application relates generally to communication networks, and more specifically to event detection for channel state information compression. Background Technology

[0004] The 3rd Generation Partnership Project (3GPP) Technical Specifications (TS) define the standards for wireless networks. These TS descriptions cover various aspects of signaling services through systems that combine wireless networks. Attached Figure Description

[0005] Figure 1 Examples of network environments based on some implementation schemes are provided.

[0006] Figure 2 Example procedures for network-side monitoring of channel state information (CSI) compression performance according to some implementations of this paper are illustrated.

[0007] Figure 3 Example procedures for UE-side monitoring of CSI compression performance are illustrated according to some implementations of this paper.

[0008] Figure 4 Example systems for event detection are illustrated according to some implementation schemes.

[0009] Figure 5 Example procedures for supporting event detection associated with CSI compression, according to some implementation schemes, are illustrated.

[0010] Figure 6 The operational flow / algorithm structure according to some implementation schemes is illustrated.

[0011] Figure 7 Another operational flow / algorithm structure based on some implementation schemes is illustrated.

[0012] Figure 8 Another operational flow / algorithm structure based on some implementation schemes is illustrated.

[0013] Figure 9 Examples of user equipment based on some implementation schemes are shown.

[0014] Figure 10 Examples of network devices based on some implementation schemes are shown. Detailed Implementation

[0015] The following detailed description refers to the accompanying drawings. The same reference numerals may be used to identify the same or similar elements in different drawings. In the following description, specific details, such as particular structures, architectures, interfaces, and techniques, are set forth for illustrative and non-limiting purposes to provide a thorough understanding of various aspects of the various embodiments. However, it will be apparent to those skilled in the art that various aspects of the various embodiments may be practiced in other examples departing from these specific details. In some instances, descriptions of well-known devices, circuits, and methods have been omitted so as not to obscure the description of the various embodiments with unnecessary detail. For the purposes of this document, the phrases “A / B” and “A or B” refer to (A), (B), or (A and B); and the phrase “based on A” means “at least partially based on A,” for example, it can be “based only on A” or it can be “partially based on A.”

[0016] The following is a glossary of terms that may be used in this disclosure.

[0017] As used herein, the term "circuit" refers to a portion of or includes said hardware component configured to provide the described functionality. Hardware components may include electronic circuitry, logic circuitry, processors (shared, dedicated, or grouped) or memories (shared, dedicated, or grouped), application-specific integrated circuits (ASICs), field-programmable devices (FPDs) (e.g., field-programmable gate arrays (FPGAs), programmable logic devices (PLDs), complex PLDs (CPLDs), high-capacity PLDs (HCPLDs), structured ASICs, or programmable system-on-a-chip (SoCs)), or digital signal processors (DSPs). In some embodiments, the circuit may execute one or more software or firmware programs to provide at least some of the described functionality. The term "circuit" may also refer to a combination of one or more hardware elements (or combinations of circuits used in electrical or electronic systems) and program code for executing the functionality. In these embodiments, the combination of hardware elements and program code may be referred to as a particular type of circuit.

[0018] As used herein, the term "processor circuit" means, is part of, or includes the following: a circuit capable of sequentially and automatically performing a series of arithmetic or logical operations or recording, storing, or transmitting digital data. The term "processor circuit" can refer to an application processor, baseband processor, central processing unit (CPU), graphics processing unit, single-core processor, dual-core processor, triple-core processor, quad-core processor, or any other device capable of executing or otherwise operating computer-executable instructions (such as program code, software modules, and / or functional procedures).

[0019] As used herein, the term "interface circuit" refers to, is part of, or includes a circuit that enables the exchange of information between two or more components or devices. The term "interface circuit" can also refer to one or more hardware interfaces, such as buses, I / O interfaces, peripheral component interfaces, and network interface cards.

[0020] As used herein, the term "user equipment" or "UE" refers to equipment having radio communication capabilities that allow a user to access network resources within a communication network. The term "user equipment" or "UE" may be considered synonymous with and may be referred to as a client, mobile phone, mobile device, mobile terminal, user terminal, mobile unit, mobile station, mobile user, subscriber, user, remote station, access agent, user agent, receiver, radio equipment, reconfigurable radio equipment, or reconfigurable mobile device. Furthermore, the term "user equipment" or "UE" can include any type of wireless / wired equipment or any computing device that includes a wireless communication interface.

[0021] As used herein, the term "computer system" means any type of interconnected electronic device, computer device, or component thereof. Additionally, the term "computer system" or "system" may refer to various components of a computer that are communicatively coupled to each other. Furthermore, the term "computer system" or "system" may refer to multiple computer devices or multiple computing systems that are communicatively coupled to each other and configured to share computing resources or network resources.

[0022] As used herein, the term "resource" means a physical or virtual device, a physical component or asset or virtual component or asset within a computing or network environment, or a physical or virtual component or component within a device or component that is accessible to or usable by the device or component. Resources may include, but are not limited to, memory space / utilization, processor / CPU time, processor / CPU utilization, processor and accelerator load, hardware time or utilization, power, input / output operations, port or network sockets, channel / link allocation, throughput, or workload units. "Hardware resource" may refer to computing resources, storage resources, or networking resources provided by physical hardware components. "Virtualization resource" may refer to computing resources, storage resources, or networking resources provided by virtualization infrastructure to an application, device, or system. The term "communication resource" may refer to resources accessible to or usable by a computer device / system to transmit information through channels of a communication network. For example, communication resources may include, but are not limited to, time / frequency resources, code resources, modulation resources, etc. The term "system resource" may refer to any kind of shared entity used to provide services and may include computing resources or network resources. System resources can be viewed as a coherent set of functions, network data objects, or services that can be accessed through a server, where such system resources reside on a single host or multiple hosts and are clearly identifiable.

[0023] As used herein, the term "channel" refers to any tangible or intangible transmission medium used to transmit data or data streams. The term "channel" may be synonymous or equivalent with "communication channel," "data communication channel," "transmission channel," "data transmission channel," "access channel," "data access channel," "link," "data link," "carrier," "radio frequency carrier," or any other similar term indicating a means or medium through which data is transmitted. Additionally, as used herein, the term "link" refers to a connection between two devices used for transmitting and receiving information.

[0024] As used in this article, the terms "instantiate" and "instantiate" refer to the creation of an instance. "Instance" also refers to the concrete occurrence of an object, which may occur, for example, during the execution of program code.

[0025] The term "connection" can refer to an established signaling relationship between two or more elements at a common communication protocol layer through a communication channel, link, interface, or reference point.

[0026] As used herein, the term "network element" refers to physical or virtualized equipment or infrastructure used to provide wired or wireless communication network services. The term "network element" may be considered synonymous with or referred to as a networked computer, network hardware, network equipment, network node, or virtualized network function.

[0027] The term "information element" refers to a structured element that contains one or more fields. The term "field" refers to the individual content of an information element, or a data element that contains content. An information element may include one or more additional information elements.

[0028] Figure 1 A network environment 100 according to some implementation schemes is illustrated. Network environment 100 may include a user equipment (UE) 104 communicatively coupled to a base station 108 of a radio access network (RAN) 110. UE 104 and base station 108 may communicate via a 3GPPTS-compatible air interface, such as an interface defining a fifth-generation (5G) new radio (NR) system or a subsequent system (e.g., a sixth-generation (6G) system). Base station 108 may provide user plane (UP) and control plane (CP) protocol termination to UE 104.

[0029] Network environment 100 may also include a core network (CN) 112. For example, CN 112 may include a 5th generation core network (5GC), a 6th generation core network (6GC), or a later generation core network. CN 112 may be coupled to base station 108 via fiber optic or wireless backhaul. CN 112 may provide functionality to UE 104 via base station 108. These functions may include managing subscriber profile information, subscriber location, service authentication, or switching of voice and data sessions.

[0030] The network environment 100 may also include an external data network 120, which can be accessed by the UE 104 via RAN 110.

[0031] In some embodiments, network environment 100 may further include UE 106. UE 106 may be coupled to UE 104 via a sidelink interface. In some embodiments, UE 106 may act as a relay node to communicatively couple UE 104 to RAN 110. In other embodiments, UE 106 and UE 104 may represent terminating nodes of a communication link. For example, UE 104 and 106 may exchange data with each other.

[0032] UE 104 may generate Channel State Information (CSI) based on measurements of one or more reference signals transmitted by RAN 110 (e.g., by base station 108) and report the CSI to RAN 110. The one or more reference signals may include, for example, a CSI-Reference Signal (CSI-RS) and / or a Synchronization Signal Block (SSB, also known as a Synchronization Signal / Physical Broadcast Channel (PBCH) block). The CSI may include, for example, a Channel Quality Indicator (CQI), a Pre-decoding Matrix Indicator (PMI), a CSI-RS Resource Indicator (CRI), an SSB Resource Indicator (SSBRI), a Layer Indicator (LI), a Rank Indicator (RI), a Layer 1 (L1) Reference Signal Received Power (RSRP), an L1 Signal-to-Interference-plus-Noise Ratio (SINR), a Capability Index, and / or time-domain attributes.

[0033] In the implementation, UE 104 can use an artificial intelligence (AI) / machine learning (ML) model to compress CSI. UE 104 can report the compressed CSI to base station 108. Base station 108 can use a reconstruction model (e.g., an AI / ML model) to reconstruct CSI based on the compressed CSI (referred to as restored CSI or reconstructed CSI). Compressed CSI using an AI / ML model can achieve more accurate beamforming information and / or reduce signaling overhead compared to reporting uncompressed CSI.

[0034] UE 104 and / or base station 108 may monitor the performance of the CSI compression model, for example, based on Key Performance Indicators (KPIs). For example, a KPI may include the squared generalized cosine similarity (SGCS) between the recovered CSI and the uncompressed CSI (e.g., ground-based CSI measurements). SGCS can be a value between 0 and 1, where 1 indicates that the recovered CSI perfectly corresponds to the uncompressed CSI, and values ​​closer to 0 indicate a lower similarity between the recovered CSI and the uncompressed CSI.

[0035] Figure 2 An example procedure 200 for network-side monitoring of CSI compression performance is illustrated. Aspects of procedure 200 can be performed by a UE (e.g., UE 104) and / or a network (e.g., RAN 110, such as base station 108).

[0036] At step 204 of procedure 200, the UE can receive a reference signal from the network. The reference signal may include, for example, CSI-RS and / or SSB.

[0037] At step 208 of process 200, the UE may generate a CSI measurement (e.g., a ground-based CSI measurement). For example, the CSI measurement may include one or more pre-decoding vectors V. In this example, the pre-decoding vector V may correspond to an estimated channel H, for example, where H = U∑V. For example, the pre-decoding vector V may be an eigenvector of the channel H.

[0038] At 212 of process 200, the encoder can generate CSI feedback C based on the pre-decoded vector V. The encoder can use a trained AI / ML model (e.g., a CSI compression model) to generate CSI feedback C. CSI feedback C can be compressed, for example, by latent space encoding of the pre-decoded vector V. Although process 200 is illustrated as CSI feedback C being encoded from the pre-decoded vector V, in other embodiments, CSI feedback can be encoded from the channel H.

[0039] At position 216, CSI feedback C and CSI measurement V can be sent to the network. In some implementations, an eT2-type high-resolution codebook can be used to report the CSI measurement V, which can provide more accurate reporting than a conventional eT2 codebook.

[0040] At 220 of process 200, the decoder on the network side (e.g., in the base station) can decode the compressed CSI feedback C to generate the recovered CSI. (Also known as CSI reconstruction).

[0041] At point 224 of process 200, the network computes the recovery of CSI. The SGCS is the distance between the CSI measurement V and the CSI measurement. The SGCS can indicate the recovery of CSI. The closer the value is to the CSI measurement V, the more accurate the CSI compression model (e.g., including encoding and / or decoding) is. In some implementations, in addition to or instead of SGCS, another intermediate KPI, such as Normalized Mean Squared Error (NMSE), can be used.

[0042] The network can identify one or more problems (e.g., data drift, model degradation, hardware failure) based on SGCS and / or take one or more actions (e.g., retrain the CSI compressed model, reconfigure the CSI compressed model, switch the CSI compressed model to a different model, and / or deactivate the CSI compressed model).

[0043] In the example, SGCS can correspond to:

[0044]

[0045] Where W corresponds to the vector of CSI measurement V, and W' corresponds to the vector of recovered CSI. The vector and / or N sb This corresponds to the number of subbands associated with CSI.

[0046] Figure 3 An example procedure 300 for UE-side monitoring of CSI compression performance is illustrated. Various aspects of procedure 200 can be performed by a UE (e.g., UE 104) and / or a network (e.g., RAN 110, such as base station 108).

[0047] Operations 304, 308, and 312 of process 300 can be similar to corresponding operations 204, 208, and 212 of process 200. For example, the UE can receive a reference signal at 304 and generate a CSI measurement V (e.g., a ground-based CSI measurement) at 308. At 312, the UE can use an AI / ML-trained encoder (e.g., employing a CSI compression model) to generate compressed CSI feedback C based on the CSI measurement V. The compressed CSI feedback C can be a latent space representation of the CSI measurement V.

[0048] At step 316 of procedure 300, the UE can send compressed CSI feedback C to the network. Since the monitoring of CSI compression performance is performed on the UE side, the UE does not need to send CSI measurement V to the network.

[0049] At position 320, the network can decode the compressed CSI feedback C to generate the recovered CSI. At position 324, the network can send a recovery CSI to the UE. At 328, the UE can recover CSI. The SGCS (or another intermediate KPI) is calculated using CSI measurement V. At point 332, the UE can report the SGCS to the network.

[0050] Similar to process 200, the network can identify one or more problems (e.g., data drift, model degradation, hardware failure) based on SGCS and / or take one or more actions (e.g., retrain the CSI compressed model, reconfigure the CSI compressed model, switch the CSI compressed model to a different model, and / or deactivate the CSI compressed model).

[0051] However, additional mechanisms may be needed to assess the accuracy of the monitoring process (e.g., accurately identifying problems and / or appropriate actions). It has been proposed to average the per-sample difference between the Genie SGCS and the calculated SGCS based on ground-based reports and SGCS estimates (e.g., E[|SGCS]). Genie –SGCS 测量The SGCS accuracy used for model monitoring is calculated using the SGCS calculation method. However, while this metric reflects the accuracy of the SGCS calculation, it fails to reflect the performance of decision-making processes that are important for monitoring CSI compression performance (e.g., identifying problems with the model and / or appropriate actions).

[0052] For example, a CSI compression model can be trained for a specific environment (e.g., including a set of propagation conditions such as channel type, indoor / outdoor, line-of-sight (LoS) or non-LoS). CSI compression models typically perform well in the environment in which they are trained (e.g., SGCS values ​​are close to 1). However, a CSI compression model may or may not work well in other environments. Furthermore, SGCS values ​​(or other KPIs) may have different value distributions across different samples within the same environment, making it difficult to obtain reliable estimates of SGCS for event prediction.

[0053] The implementation described herein provides a mechanism for performing event detection to monitor the performance of a CSI compression model. Event detection can identify problems with the CSI compression model and / or trigger appropriate actions. Event detection can identify anomalies, model degradation, and / or environmental changes and / or react to them. In the implementation, event detection may include direct event detection based on compressed CSI(C) transmitted from the UE to the network and / or indirect event detection based on intermediate KPIs such as SGCS and / or another KPI.

[0054] Figure 4 An example of a system 400 for event detection according to some embodiments is illustrated. UE 404 (e.g., corresponding to UE 104) may include encoder 412 to generate compressed CSI(C), for example, based on pre-decoded vector V and / or channel H as discussed above. UE 404 may transmit the compressed CSI to network 408 (e.g., corresponding to base station 108). Decoder 416 of network 408 may generate a recovered CSI based on the received compressed CSI.

[0055] System 400 may also include event detector 420. Event detector 420 may be implemented in UE 404, network 408 and / or another device communicating with UE 404 and / or network 408. Event detector 420 may include direct event detector 424 and / or indirect event detector 428. It should be noted that some implementations may include direct event detector 424 without indirect event detector 428, indirect event detector 428 without direct event detector 424, or both direct event detector 424 and indirect event detector 428.

[0056] The direct event detector 424 can receive compressed CSICs and detect events directly based on the compressed CSICs. For example, the direct event detector 424 can use an event detection model (e.g., an AI / ML model) to predict the occurrence of events based on the compressed CSICs. Events can be, for example, data distribution drift, model degradation, hardware failure, environmental changes, and / or anomalous behavior. The direct event detector 424 can additionally or alternatively recommend actions to resolve events, such as retraining, reconfiguring, switching, and / or deactivating (e.g., reverting to a traditional CSI report) the compressed CSI model.

[0057] The indirect event detector 428 may include a KPI estimator 432 to estimate one or more KPIs based on compressed CSI C. The KPI estimator 432 may use a trained AI / ML model to estimate one or more KPIs. The indirect event detector 432 may also include a detection logic unit 436 to detect events based on one or more estimated KPIs. In some embodiments, the one or more estimated KPIs may include estimated SGCS.

[0058] The direct event detector 424 and / or the indirect event detector 428 (e.g., the KPI estimator 432 and / or the event detection logic unit 436) can be trained using training data. In some embodiments, the training data may include inputs and associated output labels. Each input may include one or more compressed CSIs, such as a set of compressed CSIs over a time window. For example, each input may include a vector corresponding to an observation window. The output labels may indicate whether an event should be detected for the associated input. For example, the output labels may be a binary classifier with a first value indicating that an event was detected (e.g., logic 1) or a second value indicating that an event was not detected (e.g., logic 0). The output may indicate whether the event occurred within a prediction window, which may overlap with or not overlap with the observation window (e.g., the prediction window may be after the observation window).

[0059] In some implementations, the output label may be based on one or more KPIs. For example, the one or more KPIs used for the output label may include throughput-related KPIs, which can represent system performance and indicate whether an event requiring action has occurred. These one or more throughput-related KPIs may include, for example, spectral efficiency, channel quality indicator (CQI), mutual information (e.g., an internal UE metric indicating communication throughput), etc.

[0060] Not all environmental changes lead to performance degradation. For example, a CSI compression model trained on a dataset associated with a first environment may still perform well in another environment (e.g., generalize to another environment) to achieve inference performance. In the implementation, it may be beneficial for the event detector 420 not to predict events in this case (e.g., to avoid the overhead involved in switching and / or retraining the model). Output labels of the training data based on one or more throughput-related KPIs can enable the event detector 420 to detect events based on associated (e.g., predicted) performance degradation, rather than just environmental changes.

[0061] In the example, if one or more (e.g., any or all) of the throughput-related KPIs decrease more than a threshold within a time interval and / or if the value of one or more of the throughput-related KPIs is less than a threshold, a first output label (e.g., indicating that an event was detected) can be assigned to the input of the training data. Additionally or alternatively, if one or more (e.g., any or all) of the throughput-related KPIs decrease less than a threshold and / or if the value of one or more of the throughput-related KPIs is greater than a threshold, a second output label (e.g., indicating that an event was not detected) can be assigned to the input of the training data.

[0062] In some implementations, the output labels of the training data can be associated with specific types of events, such as data drift, model degradation, environmental changes, hardware failures, and / or anomalies. This enables the event detection model to detect the type of event. In other implementations, the event detection model can detect the occurrence of an event without distinguishing between different types of events.

[0063] In some implementations, model performance metrics can be used to train and / or evaluate the event detection model of the event detector 420. For example, model performance metrics may include F1 scores, precision metrics, and / or recall metrics. In some implementations, corresponding counters may be maintained based on whether the event detection model predicts an event and whether the event actually occurs (e.g., a "Genie" result).

[0064] Table 1 illustrates examples of four counters, n0, n1, n2, and n3.

[0065] Predicted negative Predicted positive Genie negative n0 n1 Genie positive n2 n3

[0066] Table 1

[0067] Counter n0 refers to the situation where the event detection model does not predict the event and the event does not actually occur. It should be noted that in some implementations, counter n0 may not be used to calculate model performance metrics. Therefore, counter n0 may or may not be maintained during the training of the event detection model.

[0068] Counter n1 refers to the case where the event detection model predicts an event but the event does not actually occur (e.g., a false alarm event prediction). Counter n2 refers to the case where the event detection model does not predict an event but the event does actually occur (e.g., the model misses a real event). Counter n3 refers to the case where the event detection model predicts an event and the event does actually occur (e.g., the model correctly predicts the event).

[0069] Precision metrics can correspond to the number of true positives divided by the total number of predicted positives (e.g., n³ / (n¹+n³)). Precision metrics indicate the proportion of actual correct predictions made by the model. Therefore, precision metrics measure how accurate the model is when it identifies an event as positive.

[0070] Recall can be defined as the number of true positives divided by the total number of actual positives (e.g., n³ / (n²+n³)). Recall can also measure a model's ability to capture positive cases (also known as sensitivity or true positive rate), for example, the percentage of all actual positive cases correctly predicted by the model.

[0071] The F1 score can be a function of both precision and recall metrics. For example, the F1 score can be equal to 2 * precision * recall / (precision + recall). Therefore, the F1 score can be a single metric that combines recall and precision using the harmonic mean.

[0072] In some implementations, the event detection model can be trained based on the F1 score, for example, to optimize the F1 score. For instance, the event detection model could be an AI / ML classifier model with weights adjusted based on the F1 score.

[0073] In some implementations, the event detection model can be trained online during deployment (e.g., based on real-world data). For example, online training can be based on one or more throughput-related KPIs that can be monitored by the user experience (UE) on a continuous basis. For example, the one or more throughput-related KPIs used for online training may include instantaneous throughput, throughput trend analysis (e.g., short-term declines and / or increases), and / or time-averaged throughput over a time window (e.g., a sliding window).

[0074] Online training allows event detection models to adapt to real-time changes, for example, aligning the model with current network conditions to provide more accurate event predictions. Additionally, online training allows for continuous validation of the model's predictions against actual throughput measurements.

[0075] In some implementations, the UE can use a reinforcement learning (RL) framework, where events and / or actions determined by the model receive feedback based on real-time throughput changes. This feedback can be used to update the model for future decisions. For example, one or more reward functions can be implemented for the model to incentivize maintaining or increasing throughput.

[0076] In some implementations, the output of the event detector may include the probability of an event (e.g., rather than a binary value indicating whether the event is predicted to occur). For example, the probability may correspond to values ​​from 0 to 1, where a value closer to one indicates a higher probability of the event occurring, and a value closer to zero indicates a lower probability of the event occurring. In other implementations, the output of the event detector may be a binary value indicating whether the event is predicted to occur.

[0077] In some implementations, when the event detector is implemented at the UE, the UE can send an indication of the detected event to the network. For example, the indication of the detected event can be sent to a lifecycle management (LCM) entity of the network, which may be implemented in the base station and / or core network. In some implementations, the UE can indicate a recommended action based on the detected event (e.g., as recommended by an event detection model). In other implementations, the network can determine the action to be taken based on the indication of the detected event received from the UE.

[0078] Figure 5 An example process 500 for supporting event detection associated with CSI compression, according to some implementation schemes, is illustrated. Aspects of process 500 may be performed by a UE (e.g., UE 104) and / or a network (e.g., RAN 110 and / or base station 108) or components thereof. For example, Figure 5 The signaling can be with Figure 4 This is used in conjunction with system 400. It should be noted that in some implementations, one or more operations of process 500 may be omitted or modified. Additionally, some implementations may include one or more additional operations as described herein.

[0079] In the example, at step 504, the UE can send an indication to the network of a detected event. This indication may include a probability of the detected event and / or a binary value indicating whether the event was detected. In some implementations, the indication may include an identifier for the event.

[0080] In some implementations, at 508, the network can request event-related information from the UE, such as one or more KPIs. The network can allocate resources in which the UE will provide the requested information. At 512, the UE can (e.g., within the allocated resources) send the requested information to the network.

[0081] At point 516, the network can determine the action to be taken based on the detected event and / or additional information (e.g., KPIs) received from the UE at point 512. In some implementations, the action determination can be performed by the network's LCM entity. For example, the action may include reconfiguring the CSI compression model, retraining the CSI compression model, switching to a different CSI compression model, and / or deactivating the CSI compression model (e.g., reverting to traditional CSI reporting).

[0082] At point 520, the network can send instructions for actions to the UE. In some implementations, when the action includes retraining the CSI compression model, the network can provide retraining information, such as information related to the retraining process, and / or resource allocation for one or more reference signals (e.g., CSI-RS) used for retraining. When the action includes switching to a different model, the network can indicate the model ID of the new model to be used for CSI compression. When the action includes deactivating the current model (e.g., as part of switching to a new model or reverting to legacy operation), the network can indicate the model ID of the current model to be deactivated.

[0083] Figure 6 An operational flow / algorithm structure 600 according to some embodiments is illustrated. In some embodiments, the operational flow / algorithm structure 600 may be implemented by a UE (such as, for example, UE 104, UE 900) or a component thereof (e.g., processor 904A). In other embodiments, the operational flow / algorithm structure 600 may be executed by another device (or a component thereof), such as another device of a base station (e.g., base station 108), a RAN (e.g., RAN 110), or a device of a core network (e.g., core network 112).

[0084] The operation flow / algorithm structure 600 may include: at 604, obtaining the compressed CSI transmitted from the UE to the network. For example, the compressed CSI can be encoded using a CSI compression model as discussed herein.

[0085] The operation flow / algorithm structure 600 may further include, at 608, using an AI / ML event detection model based on compressed CSI to detect events indicating performance degradation of the CSI compression model used to generate the compressed CSI. The AI / ML event detection model can detect events directly and / or indirectly based on compressed CSI. In an example of indirect detection, the AI / ML event detection model can estimate one or more intermediate KPIs (e.g., SGCS) and detect events based on one or more intermediate KPIs. In some implementations, the AI / ML event detection model can be trained using training data that classifies events based on one or more throughput-related KPIs (e.g., spectral efficiency, CQI, etc.) using associated output labels.

[0086] The operation flow / algorithm structure 600 may further include, at 612, generating an indication of the detected event for transmission to the network. In some embodiments, the indication may be a binary value indicating the probability of the event or whether the event was detected. In some embodiments, the indication may be an indication of the type of event (e.g., data drift, model degradation, hardware failure, environmental change, and / or anomaly).

[0087] Figure 7 Another operational flow / algorithm structure 700 according to some embodiments is illustrated. In some embodiments, the operational flow / algorithm structure 700 may be implemented by a UE (such as, for example, UE 104, UE 900) or a component thereof (e.g., processor 904A). In other embodiments, the operational flow / algorithm structure 700 may be executed by another device (or a component thereof), such as another device of a base station (e.g., base station 108), a RAN (e.g., RAN 110), or a device of a core network (e.g., core network 112).

[0088] The operation flow / algorithm structure 700 may include: at 704, obtaining training data, which includes inputs corresponding to the compressed CSI on the observation window and associated output labels, the associated output labels indicating whether an event has occurred based on one or more throughput-related key performance indicators (KPIs). The one or more throughput-related KPIs may include, for example, spectral efficiency or channel quality indicators. In some implementations, the event detection model may be further trained based on precision metrics, recall metrics, and / or F1 scores (which may be based on precision and recall metrics).

[0089] The operation process / algorithm structure 700 may also include: at 708, using training data to train an event detection model, wherein the event detection model is used to detect performance degradation of the CSI compression model. Performance degradation may include, for example, data drift, model degradation, hardware failure, environmental changes, and / or anomalies.

[0090] In some implementations, the operation flow / algorithm structure 700 may further include: obtaining one or more additional throughput-related KPIs when deploying the event detection model, and performing online training of the event detection model based on the one or more additional throughput-related KPIs.

[0091] Figure 8 Another operational flow / algorithm structure 800 according to some implementation schemes is illustrated. The operational flow / algorithm structure 800 may be implemented by network devices such as, for example, devices of RAN 110 (e.g., base station 108), network device 1000 or its components (e.g., processor 1004A).

[0092] The operation flow / algorithm structure 800 may include, at 804, receiving from the UE the probability that an event associated with performance degradation of the CSI compression model used by the UE to generate the compressed CSI has occurred. In some implementations, the network device may also receive from the UE an indication of the event type (e.g., data drift, model degradation, hardware failure, environmental change, and / or anomaly).

[0093] The operation flow / algorithm structure 800 may further include: at 808, receiving one or more KPIs associated with an event from the UE. In some implementations, the network device may (e.g., based on the probability that the received event has occurred) generate a request for one or more KPIs to send to the UE.

[0094] The operation process / algorithm structure 800 may also include, at 812, determining the action to be taken based on the probability and one or more KPIs. For example, the action may include retraining the CSI compression model, reconfiguring the CSI compression model, switching to a different CSI compression model, or deactivating the CSI compression model.

[0095] The operation flow / algorithm structure 800 may also include: at 816, generating an instruction for the action to be sent to the UE.

[0096] In the example, when the action includes retraining the CSI compression model, the network device can allocate reference signal resources to the UE for one or more reference signals to be used for retraining the CSI compression model.

[0097] Figure 9UE 900 is illustrated according to some implementation schemes. UE 900 may be similar to UE 104 and / or UE 106 and is substantially interchangeable with them.

[0098] UE 900 can be any mobile or non-mobile computing device, such as, for example, a mobile phone, computer, tablet, industrial wireless sensor (e.g., microphone, carbon dioxide sensor, pressure sensor, humidity sensor, thermometer, motion sensor, accelerometer, laser scanner, fluid level sensor, stock sensor, voltmeter / ammeter, or actuator), video surveillance / monitoring device (e.g., camera or camcorder), wearable device (e.g., smartwatch), or Internet of Things device.

[0099] UE 900 may include a processor 904, RF interface circuitry 908, memory / storage device 912, user interface 916, sensor 920, drive circuitry 922, power management integrated circuit (PMIC) 924, antenna 926, and battery 928. The components of UE 900 may be implemented as integrated circuits (ICs), portions of integrated circuits, discrete electronic devices or other modules, logic components, hardware, software, firmware, or combinations thereof. Figure 9 The block diagram is intended to show a high-level view of some of the components of the UE 900. However, some of the components shown may be omitted, additional components may be present, and different arrangements of the components shown may occur in other specific implementations.

[0100] The components of UE 900 can be coupled to a variety of other components via one or more interconnects 932, which can represent any type of interface, input / output, bus (local, system, or extension), transmit line, trace, or optical connector, allowing various circuit components (on common or different chips or chipsets) to interact with each other.

[0101] Processor 904 may include processor circuitry, such as, for example, baseband processor circuitry (BB) 904A, central processing unit circuitry (CPU) 904B, and graphics processing unit circuitry (GPU) 904C. Processor 904 may include any type of circuitry or processor circuitry that executes or otherwise operates computer-executable instructions (such as program code, software modules, or functional processes from memory / storage device 912) to cause UE 900 to perform operations as described herein (e.g., operations associated with event detection for CSI compression). Processor 904 may also include interface circuitry 904D for enabling communication by, for example, communicatively coupling the processor circuitry to one or more other components of UE 900.

[0102] In some implementations, the baseband processor 904A can access the communication protocol stack 936 in the memory / storage device 912 to communicate over a 3GPP-compliant network. Generally, the baseband processor 904A can access the communication protocol stack 936 to perform the following operations: user plane functions at the PHY, MAC, RLC, PDCP, SDAP, and PDU layers; and control plane functions at the PHY, MAC, RLC, PDCP, RRC, and NAS layers. In some implementations, PHY layer operations may additionally / optionally be performed by components of the RF interface circuitry 908.

[0103] The baseband processor 904A can generate or process baseband signals or waveforms that carry information in 3GPP-compliant networks. In some implementations, the waveforms used for NR can be based on cyclic prefix OFDM (CP-OFDM) in the uplink or downlink, and Discrete Fourier Transform Extended OFDM (DFT-S-OFDM) in the uplink.

[0104] The memory / storage device 912 may include one or more non-transitory computer-readable media, which include instructions (e.g., a communication protocol stack 936) that can be executed by one or more processors in processor 904 to cause the UE 900 to perform operations as described herein (e.g., operations associated with event detection for CSI compression).

[0105] The memory / storage device 912 includes any type of volatile or non-volatile memory that can be distributed throughout the UE 900. In some embodiments, some of the memory / storage devices 912 may be located on the processor 904 itself (e.g., the memory / storage device 912 may be part of a chipset corresponding to the baseband processor 904A), while other memory / storage devices 912 are located external to the processor 904 but are accessible via a memory interface. The memory / storage device 912 may include any suitable volatile or non-volatile memory, such as, but not limited to, dynamic random access memory (DRAM), static random access memory (SRAM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), flash memory, solid-state memory, or any other type of memory device technology.

[0106] RF interface circuitry 908 may include transceiver circuitry and a radio frequency front-end module (RFEM) that allows UE 900 to communicate with other devices via a radio access network. RF interface circuitry 908 may include various components arranged in the transmit or receive path. These components may include, for example, switches, mixers, amplifiers, filters, synthesizer circuitry, and control circuitry.

[0107] In the receiving path, the RFEM can receive the radiated signal from the air interface via antenna 926 and continue to filter and amplify the signal (using a low-noise amplifier). This signal can be provided to the receiver of the transceiver, which down-converts the RF signal into a baseband signal that is provided to the baseband processor of processor 904.

[0108] In the transmission path, the transceiver's transmitter up-converts the baseband signal received from the baseband processor and provides the RF signal to the RFEM. The RFEM can then amplify the RF signal using a power amplifier before it is radiated across the air interface via antenna 926.

[0109] In various implementations, the RF interface circuit 908 can be configured to transmit / receive signals in a manner compatible with NR access technology.

[0110] Antenna 926 may include antenna elements to convert electrical signals into radio waves to travel through the air and to convert received radio waves back into electrical signals. These antenna elements may be arranged in one or more antenna panels. Antenna 926 may have omnidirectional, directional, or combinations thereof antenna panels to enable beamforming and multiple-input multiple-output (MIMO) communication. Antenna 926 may include a microstrip antenna, a printed antenna fabricated on the surface of one or more printed circuit boards, a patch antenna, or a phased array antenna. Antenna 926 may have one or more panels designed for a specific frequency band (including bands in FR1 or FR2).

[0111] User interface 916 includes various input / output (I / O) devices designed to enable users to interact with UE 900. User interface 916 includes input device circuitry and output device circuitry. Input device circuitry includes any physical or virtual components for accepting input, particularly including one or more physical or virtual buttons (e.g., a reset button), a physical keyboard, keypad, mouse, touchpad, touchscreen, microphone, scanner, or headset, etc. Output device circuitry includes any physical or virtual components for displaying or otherwise conveying information (such as sensor readings, actuator positions, or other similar information). Output device circuitry may include any number or combination of audio or visual displays, particularly including one or more simple visual outputs / indicators (e.g., binary status indicators such as light-emitting diodes (LEDs) and multi-character visual outputs), or more complex outputs such as display devices or touchscreens (e.g., liquid crystal displays (LCDs), LED displays, quantum dot displays, and projectors), wherein the output of characters, graphics, multimedia objects, etc., is generated or produced through the operation of UE 900.

[0112] Sensor 920 may include devices, modules, or subsystems designed to detect events or changes in their environment and transmit information about the detected events (sensor data) to other devices, modules, or subsystems. Examples of such sensors include: inertial measurement units including accelerometers, gyroscopes, or magnetometers; microelectromechanical systems (MEMS) or nanoelectromechanical systems (NEM) including 3-axis accelerometers, 3-axis gyroscopes, or magnetometers; level sensors; flow sensors; temperature sensors (e.g., thermistors); pressure sensors; barometric pressure sensors; gravimeters; altimeters; image capture devices (e.g., cameras or lensless aperture sensors); light detection and ranging sensors; proximity sensors (e.g., infrared radiation detectors, etc.); depth sensors; ambient light sensors; ultrasonic transceivers; and microphones or other similar audio capture devices.

[0113] The driving circuitry 922 may include software and hardware elements for controlling specific devices embedded in, attached to, or otherwise communicatively coupled to the UE 900. The driving circuitry 922 may include various drivers that allow other components to interact with or control various input / output (I / O) devices that may exist within or be connected to the UE 900. For example, the driving circuitry 922 may include: a display driver for controlling and allowing access to a display device; a touchscreen driver for controlling and allowing access to a touchscreen interface; a sensor driver for obtaining sensor readings of a sensor 920 and controlling and allowing access to the sensor 920; a driver for obtaining actuator positioning of an electromechanical component or controlling and allowing access to an electromechanical component; a camera driver for controlling and allowing access to an embedded image capture device; and an audio driver for controlling and allowing access to one or more audio devices.

[0114] The PMIC 924 manages the power supplied to various components of the UE 900. Specifically, relative to the processor 904, the PMIC 924 controls power source selection, voltage scaling, battery charging, or DC-DC conversion.

[0115] Battery 928 can power UE 900, but in some examples, UE 900 may be installed and deployed in a fixed location and may have a power source coupled to the power grid. Battery 928 may be a lithium-ion battery, a metal-air battery such as a zinc-air battery, an aluminum-air battery, a lithium-air battery, etc. In some specific implementations, such as in vehicle-based applications, battery 928 may be a typical lead-acid automotive battery.

[0116] Figure 10 A network device 1000 according to some implementation schemes is illustrated. The network device 1000 may be similar to the components of the base station 108 and / or the CN 112, and is substantially interchangeable with the components of the base station and / or the CN.

[0117] Network device 1000 may include processor 1004, RF interface circuit 1008 (if implemented as a base station), core network (CN) interface circuit 1014, memory / storage device circuit 1012 and antenna structure 1026.

[0118] The components of network device 1000 can be coupled to various other components through one or more interconnectors 1028.

[0119] The processor 1004, RF interface circuit 1008, memory / storage device circuit 1012 (including communication protocol stack 1010), antenna structure 1026, and interconnect 1028 can be similar to those relative to... Figure 9 Elements with similar names as shown and described.

[0120] Processor 1004 may include processor circuitry, such as, for example, baseband processor circuitry (BB) 1004A, central processing unit circuitry (CPU) 1004B, and graphics processing unit circuitry (GPU) 1004C. Processor 1004 may include any type of circuitry or processor circuitry that executes or otherwise operates computer-executable instructions (such as program code, software modules, or functional processes from memory / storage device circuitry 1012) to cause network device 1000 to perform operations as described herein (e.g., operations associated with event detection for CSI compression). Processor 1004 may also include interface circuitry 1004D for communicatively coupling the processor circuitry to one or more other components of network device 1000.

[0121] The CN interface circuit 1014 can provide connectivity to a core network (e.g., a 5GC using a 5G core network (5GC) compatible network interface protocol (such as Carrier Ethernet) or some other suitable protocol). Network connectivity can be provided to / from network device 1000 via fiber optic or wireless backhaul. The CN interface circuit 1014 may include one or more dedicated processors or FPGAs for communicating using one or more of the aforementioned protocols. In some implementations, the CN interface circuit 1014 may include multiple controllers for providing connectivity to other networks using the same or different protocols.

[0122] As is widely recognized, the use of personally identifiable information should comply with privacy policies and practices that are generally accepted to meet or exceed industry or governmental requirements for protecting user privacy. Specifically, personally identifiable information data should be managed and processed to minimize the risk of unintentional or unauthorized access or use, and the nature of authorized use should be clearly explained to users.

[0123] For one or more embodiments, at least one of the components shown in one or more of the foregoing figures may be configured to perform one or more operations, techniques, processes, or methods described in the Embodiments section below. For example, the baseband circuitry described above in conjunction with one or more of the foregoing figures may be configured to operate according to one or more examples below. As another example, circuitry associated with the UE, base station, or network element described above in conjunction with one or more of the foregoing figures may be configured to operate according to one or more embodiments described in the Embodiments section below.

[0124] As is widely recognized, the use of personally identifiable information should comply with privacy policies and practices that are generally accepted to meet or exceed industry or governmental requirements for protecting user privacy. Specifically, personally identifiable information data should be managed and processed to minimize the risk of unintentional or unauthorized access or use, and the nature of authorized use should be clearly explained to users.

[0125] For one or more embodiments, at least one of the components shown in one or more of the foregoing figures may be configured to perform one or more operations, techniques, processes, or methods described in the Embodiments section below. For example, the baseband circuitry described above in conjunction with one or more of the foregoing figures may be configured to operate according to one or more examples below. As another example, circuitry associated with the UE, base station, or network element described above in conjunction with one or more of the foregoing figures may be configured to operate according to one or more embodiments described in the Embodiments section below.

[0126] Example

[0127] The following sections provide further exemplary implementations.

[0128] Example 1 includes a method comprising: obtaining compressed channel state information (CSI) transmitted from a user equipment (UE) to a network; using an artificial intelligence (AI) / machine learning (ML) event detection model based on the compressed CSI to detect events indicating performance degradation of a CSI compression model used to generate the compressed CSI; and generating an indication of the detected events for transmission to the network.

[0129] Example 2 includes the method according to Example 1 or some other example herein, wherein the AI / ML event detection model is used to detect the event directly based on the compressed CSI.

[0130] Example 3 includes the method according to Example 1 or some other embodiment herein, wherein, in order to detect the event, the AI / ML event detection model is used to: estimate a key performance indicator (KPI) based on the compressed CSI; and detect the event based on the estimated KPI.

[0131] Example 4 includes the method according to Example 3 or some other embodiment herein, wherein the KPI includes the squared generalized cosine similarity (SGCS) between the uncompressed CSI corresponding to the compressed CSI and the recovered CSI decoded from the compressed CSI.

[0132] Example 5 includes the method according to Example 1 or some other embodiment herein, wherein the AI / ML event detection model is trained using training data, which is classified based on one or more throughput-related key performance indicators (KPIs) using associated output labels.

[0133] Example 6 includes the method according to Example 1 or some other embodiment herein, wherein the AI / ML event detection model is trained based on an F1 score, a precision metric, or a recall metric.

[0134] Example 7 includes the method according to Example 1 or some other example herein, the method further comprising: monitoring one or more throughput-related KPIs when the AI / ML event detection model is activated; and performing online training of the AI / ML event detection model based on the one or more throughput-related KPIs.

[0135] Example 8 includes the method according to Example 1 or some other embodiment of this document, wherein the AI / ML event detection model is implemented by the UE.

[0136] Example 9 includes the method according to Example 1 or some other embodiment of this document, the method further comprising: receiving from the network a request for information associated with the event, wherein the information includes one or more key performance indicators (KPIs); and generating a message including the one or more KPIs for transmission to the network.

[0137] Example 10 includes the method according to Example 1 or some other embodiment herein, wherein the indication of the event indicates the type of the event, wherein the type includes data drift, model degradation, or environmental change.

[0138] Example 11 includes the method according to Example 1 or some other embodiment herein, the method further including receiving from the network an indication of an action to be taken based on the event, wherein the action includes retraining the CSI compression model, reconfiguring the CSI compression model, switching to a different CSI compression model, or deactivating the CSI compression model.

[0139] Example 12 includes one or more computer-readable media having instructions that, when executed, cause processing circuitry to: obtain training data including inputs corresponding to compressed channel state information (CSI) on an observation window and associated output labels, the associated output labels indicating whether an event has occurred based on one or more throughput-related key performance indicators (KPIs); and use the training data to train an event detection model, wherein the event detection model is used to detect performance degradation of the CSI compression model.

[0140] Example 13 includes one or more computer-readable media according to Example 12 or some other embodiment herein, wherein the one or more throughput-related KPIs include spectral efficiency or channel quality indicators.

[0141] Example 14 includes one or more computer-readable media according to Example 12 or some other embodiment herein, wherein the instructions, when executed, further cause the processing circuitry to train the event detection model based on an F1 score, a precision metric, or a recall metric.

[0142] Example 15 includes one or more computer-readable media according to Example 12 or some other embodiment herein, wherein detecting the performance degradation includes detecting data drift, model degradation, or environmental changes.

[0143] Example 16 includes one or more computer-readable media according to Example 12 or some other embodiment herein, wherein the instructions, when executed, further cause the processing circuitry to: obtain one or more additional throughput-related KPIs when deploying the event detection model; and perform online training of the event detection model based on the one or more additional throughput-related KPIs.

[0144] Example 17 includes a method comprising: receiving from a user equipment (UE) a probability that an event has occurred associated with performance degradation of a CSI compression model used by the UE to generate compressed channel state information (CSI); receiving from the UE one or more key performance indicators (KPIs) associated with the event; determining an action to be taken based on the probability and the one or more KPIs, wherein the action includes retraining the CSI compression model, reconfiguring the CSI compression model, switching to a different CSI compression model, or deactivating the CSI compression model; and generating an instruction for the action to be sent to the UE.

[0145] Example 18 includes the method according to Example 17 or some other embodiment herein, the method further including generating a request for the one or more KPIs for sending to the UE.

[0146] Example 19 includes the method according to Example 17 or some other embodiment herein, wherein the action includes retraining the CSI compression model, and wherein the method further includes allocating reference signal resources to the UE for one or more reference signals to be used to retrain the CSI compression model.

[0147] Example 20 includes the method according to Example 17 or some other embodiment herein, the method further including receiving from the UE an indication of the type of the event, wherein the type includes data drift, model degradation, or environmental change.

[0148] Another embodiment may include an apparatus comprising one or more elements for performing the method described or associated with any one of embodiments 1 to 20 or any other method or process described herein.

[0149] Another embodiment may include one or more non-transitory computer-readable media, the one or more non-transitory computer-readable media including instructions that, when executed by one or more processors of an electronic device, cause the electronic device to perform one or more elements of the method or any other method or process described herein according to any one of embodiments 1 to 20.

[0150] Another embodiment may include an apparatus comprising one or more elements for performing the methods described or associated with any one of embodiments 1 to 20 or any other methods or processes described herein.

[0151] Another embodiment may include the methods, techniques or processes described or associated with any one of embodiments 1 to 20 or any part or component thereof.

[0152] Another embodiment may include an apparatus comprising: one or more processors and one or more computer-readable media, the one or more computer-readable media including instructions that, when executed by the one or more processors, cause the one or more processors to perform a method, technique, or process described or associated with any one or more of embodiments 1 to 20.

[0153] Another embodiment may include signals described or associated with any one of embodiments 1 to 20 or any part or component thereof.

[0154] Another embodiment may include datagrams, information elements, packets, frames, segments, PDUs, or messages described or associated with any one of embodiments 1 to 20 or any part or component thereof, or otherwise described in this disclosure.

[0155] Another embodiment may include a signal encoded with data as described or associated with any one of embodiments 1 to 20 or a portion or component thereof, or otherwise described in this disclosure.

[0156] Another embodiment may include signals encoded as datagrams, IEs, packets, frames, segments, PDUs, or messages as described or associated with any one of embodiments 1 to 20 or any part or component thereof, or otherwise described in this disclosure.

[0157] Another embodiment may include an electromagnetic signal carrying computer-readable instructions, wherein execution of the computer-readable instructions by one or more processors will cause the one or more processors to perform a method, technique or process described or associated with any one or a portion of Embodiments 1 to 20.

[0158] Another embodiment may include a computer program comprising instructions, wherein execution of the program by a processing element will cause the processing element to perform a method, technique, or process described or associated with any one or a portion thereof according to Embodiments 1 to 20.

[0159] Another embodiment may include signals in a wireless network as shown and described herein.

[0160] Another embodiment may include a method for communicating in a wireless network as shown and described herein.

[0161] Another embodiment may include a system for providing wireless communication as shown and described herein.

[0162] Another embodiment may include a device for providing wireless communication as shown and described herein.

[0163] Unless otherwise expressly stated, any of the embodiments described above may be combined with any other embodiment (or combination of embodiments). The foregoing description of one or more specific embodiments provides illustration and description, but is not intended to be exhaustive or to limit the scope of the embodiments to the precise forms disclosed. In view of the teachings above, modifications and variations are possible, or modifications and variations may be obtained from the practice of various embodiments.

[0164] Although the above embodiments have been described in considerable detail, many variations and modifications will become apparent to those skilled in the art once the above disclosure is fully understood. It is intended that the following claims be construed as encompassing all such variations and modifications.

Claims

1. A method, the method comprising: Obtain Compressed Channel State Information (CSI) transmitted from User Equipment (UE) to the network; An artificial intelligence (AI) / machine learning (ML) event detection model based on the compressed CSI is used to detect events that indicate performance degradation of the CSI compression model used to generate the compressed CSI; as well as Generate indications for the detected events to be sent to the network.

2. The method according to claim 1, wherein the AI / ML event detection model is used to detect the event directly based on the compressed CSI.

3. The method of claim 1, wherein, in order to detect the event, the AI / ML event detection model is used for: Based on the compressed CSI, estimate Key Performance Indicators (KPIs); and The event is detected based on the estimated KPI.

4. The method of claim 3, wherein the KPI includes the squared generalized cosine similarity (SGCS) between the uncompressed CSI corresponding to the compressed CSI and the recovered CSI decoded from the compressed CSI.

5. The method of claim 1, wherein the AI / ML event detection model is trained using training data, the training data being classified based on one or more throughput-related key performance indicators (KPIs) using associated output labels.

6. The method of claim 1, wherein the AI / ML event detection model is trained based on an F1 score, a precision metric, or a recall metric.

7. The method according to claim 1, further comprising: When the AI / ML event detection model is activated, one or more throughput-related KPIs are monitored; as well as Based on one or more throughput-related KPIs, perform online training of the AI / ML event detection model.

8. The method according to claim 1, wherein the AI / ML event detection model is implemented by the UE.

9. The method according to claim 1, further comprising: Receive a request from the network for information associated with the event, wherein the information includes one or more key performance indicators (KPIs); as well as Generate a message including one or more of the KPIs for sending to the network.

10. The method of claim 1, wherein the indication of the event indicates the type of the event, wherein the type includes data drift, model degradation, or environmental change.

11. The method of claim 1, further comprising receiving from the network an indication of an action to be taken based on the event, wherein the action includes retraining the CSI compression model, reconfiguring the CSI compression model, switching to a different CSI compression model, or deactivating the CSI compression model.

12. One or more computer-readable media having instructions that, when executed, cause processing circuitry to perform the following operations: Obtain training data, which includes inputs corresponding to Compressed Channel State Information (CSI) on the observation window and associated output labels, the associated output labels indicating whether an event has occurred based on one or more throughput-related Key Performance Indicators (KPIs); and The training data is used to train an event detection model, which is used to detect performance degradation of the CSI compression model.

13. The one or more computer-readable media of claim 12, wherein the one or more throughput-related KPIs include spectral efficiency or channel quality indicators.

14. The one or more computer-readable media of claim 12, wherein the instructions, when executed, further cause the processing circuitry to train the event detection model based on an F1 score, a precision metric, or a recall metric.

15. One or more computer-readable media according to claim 12, wherein detecting the performance degradation includes detecting data drift, model degradation, or environmental changes.

16. The one or more computer-readable media of claim 12, wherein the instructions, when executed, further cause the processing circuitry to: When deploying the event detection model, obtain one or more additional throughput-related KPIs; and The event detection model is trained online based on one or more additional throughput-related KPIs.

17. A method, the method comprising: The probability that an event has occurred that is associated with the performance degradation of the CSI compression model used by the User Equipment (UE) to generate Compressed Channel State Information (CSI); Receive one or more key performance indicators (KPIs) associated with the event from the UE; The action to be taken is determined based on the probability and one or more KPIs, wherein the action includes retraining the CSI compression model, reconfiguring the CSI compression model, switching to a different CSI compression model, or deactivating the CSI compression model; as well as Generate an instruction for the action to be sent to the UE.

18. The method of claim 17, further comprising generating a request for the one or more KPIs for sending to the UE.

19. The method of claim 17, wherein the action includes retraining the CSI compression model, and wherein the method further includes allocating reference signal resources to the UE for one or more reference signals to be used for retraining the CSI compression model.

20. The method of claim 17, further comprising receiving from the UE an indication of the type of the event, wherein the type includes data drift, model degradation, or environmental change.

Images