Communication equipment and machine learning model monitoring method

Abstract

A method for configuring machine learning (ML) model monitoring, performed by at least one first node, includes providing, by a second node, auxiliary information for the at least one first node and / or the second node to monitor a plurality of ML models having a common portion.

Description

Technical Field

[0001] The present application relates to the field of wireless communication systems, and more specifically to communication devices and methods for machine learning (ML) model monitoring. For example, the present application relates to a study item description (SID) for Release 18 on artificial intelligence / machine learning (AI / ML) in the new radio (NR) air interface, which was determined at the 94e plenary meeting of the 3rd generation partnership project (3GPP) radio access network (RAN) in December 2022, with discussions led by RAN1 and started in May 2022. Specifically, the present application is related to enhanced channel state information (CSI) report feedback, beam management and / or positioning, where several ML models for CSI feedback can have a common CSI reconstruction part. Background Art

[0002] AI / ML technology has been applied in 3GPP RAN1, and several use cases have been identified for study, namely CSI feedback enhancement, beam management, and positioning. As stated in the 3GPP new SID, although specific AI / ML algorithms and models may be studied for evaluation purposes, AI / ML algorithms and models are implementation-specific and are not expected to be standardized. ML models first need to be trained and then deployed. After deployment, the ML model will enter the inference phase. During the inference phase, the ML model needs to be monitored to determine whether it is working properly. Recently, 3GPP RAN1 is discussing ML enhanced CSI feedback. The method compresses CSI on the user equipment (UE) side and restores CSI on the base station (gNB) side. The CSI feedback ML model with a common CSI reconstruction part was mentioned in the RAN 110 meeting, but there are some issues, such as how to determine whether the UE's ML model fails and how to deactivate the ML model with a common part with low signaling overhead. In addition, how to activate and deactivate multiple ML models with a common part also needs to be discussed.

[0003] Therefore, there is a need for communication devices and methods for machine learning (ML) model monitoring to solve the problems in the prior art, simplify the management of multiple ML models with common parts, provide methods for monitoring multiple ML models with common parts, reduce system signaling overhead, provide good communication performance and / or provide high reliability. Summary of the invention

[0004] The purpose of this application is to propose a communication device and method for machine learning (ML) model monitoring to solve the problems in the prior art, simplify the management of multiple ML models with a common part, provide a method for monitoring multiple ML models with a common part, reduce system signaling overhead, provide good communication performance and / or provide high reliability.

[0005] In a first aspect of the present application, a method for configuring machine learning (ML) model monitoring is performed by at least one first node, including providing auxiliary information through a second node, wherein the auxiliary information is used for the at least one first node and / or the second node to monitor multiple ML models having common parts.

[0006] In a second aspect of the present application, a first node includes a memory, a transceiver, and a processor coupled to the memory and the transceiver. The processor is configured to execute the above method.

[0007] In a third aspect of the present application, a method for configuring ML model monitoring is performed by a second node, comprising configuring auxiliary information to a first node and at least one third node, wherein the auxiliary information is used for the second node and / or the first node and the at least one third node to monitor multiple ML models having a common part.

[0008] In a fourth aspect of the present application, a second node includes a memory, a transceiver, and a processor coupled to the memory and the transceiver. The processor is configured to execute the above method.

[0009] In a fifth aspect of the present application, a non-transitory machine-readable storage medium stores instructions thereon, and when a computer executes the instructions, the instructions cause the computer to execute the above method.

[0010] In a sixth aspect of the present application, a chip includes a processor configured to call and run a computer program stored in a memory so that a device equipped with the chip executes the above method.

[0011] In a seventh aspect of the present application, a computer-readable storage medium stores a computer program, wherein the computer program enables a computer to execute the above method.

[0012] In an eighth aspect of the present application, a computer program product comprises a computer program, wherein the computer program enables a computer to execute the above method.

[0013] In a ninth aspect of the present application, a computer program is provided, wherein the computer program enables a computer to execute the above method. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present application or related technologies, the following briefly introduces the drawings to be described in the embodiments. Obviously, the drawings are only some embodiments of the present application, and ordinary technicians in this field can obtain other drawings based on these drawings without paying any price.

[0015] Figure 1 is a schematic diagram showing an example of a basic autoencoder model for enhancing CSI feedback according to an embodiment of the present application.

[0016] Figure 2 It is a block diagram showing one or more user equipment (UE) and a base station (such as a gNB) in a communication network system according to an embodiment of the present application.

[0017] Figure 3 It is a wireless communication flow chart showing the provision of auxiliary information by a second node or the execution of artificial intelligence (AI) / machine learning (ML) by a UE according to an embodiment of the present application.

[0018] Figure 4 It is a wireless communication flow chart showing configuration of UE auxiliary information for artificial intelligence (AI) / machine learning (ML) by a base station according to an embodiment of the present application.

[0019] Figure 5 is a schematic diagram showing an example of a basic autoencoder model for enhancing CSI feedback according to an embodiment of the present application.

[0020] Figure 6 It is a schematic diagram showing an example of a RAN intelligent function framework according to an embodiment of the present application.

[0021] Figure 7 It is a schematic diagram showing that encoders of two UEs share a common decoder at the gNB end according to an embodiment of the present application.

[0022] Figure 8 is a schematic diagram showing an example of dual-end model monitoring with a common part CSI reconstruction part according to an embodiment of the present application.

[0023] Fig. 9 is a flowchart showing an example of monitoring a dual-end model with a common part CSI reconstruction part according to an embodiment of the present application.

[0024] Fig.10 Detailed description is a block diagram showing a communication device (eg, UE) for wireless communication according to an embodiment of the present application.

[0025] Fig.11 is a block diagram showing a communication device (e.g., gNB) for wireless communication according to an embodiment of the present application.

[0026] Fig.12 is a block diagram showing a general ML / AI framework for NR air interface according to an embodiment of the present application.

[0027] Fig.13 is a block diagram of a system for wireless communication according to an embodiment of the present application. DETAILED DESCRIPTION

[0028] The technical content, structural features, objectives and effects of the present application are described in detail below in conjunction with the accompanying drawings. Specifically, the terms in the embodiments of the present application are only used to describe the purpose of specific embodiments, and are not intended to limit the present application.

[0029] AI / ML is introduced into the physical layer (PHY) and medium access control (MAC) layer to enhance system performance. Several use cases have been identified in 3GPP RAN1 for research, namely CSI feedback compression, beam management and positioning. ML learning models can be trained online or offline.

[0030] For machine learning models, training is first required. Training can be performed by one node, which can be a gNB, UE, or a third node. Training can be online or offline. After the model training is completed, it will be deployed. For enhanced CSI feedback, the ML model is a two-end model that will be deployed in the UE and gNB respectively. These CSI models need to be monitored to determine whether they are working properly.

[0031] Some embodiments of the present application discuss the case of CSI feedback enhancement. Figure 1 It is a schematic diagram of a basic autoencoder model for enhancing CSI feedback according to an embodiment of the present application. Figure 1 In some embodiments, the basic autoencoder model is shown in the figure. The encoder compresses the raw CSI-RS value (raw CSI for short) or the maximum eigenvector and reports the output to the gNB. The gNB will decompress it. The new CSI report is a CSI report that includes CSI feedback enhanced by the AI / ML model.

[0032] At the UE side, the input data is compressed and output to the channel. The input to the encoder can be a (maximum) eigenvector or a channel matrix. The compressed output is used as the input to the decoder and reconstructed at the gNB side. This application discusses various methods for training ML models, including training on the UE side and passing the model to the gNB, training on the gNB side and passing the model to the UE, joint training by the UE and gNB, and separate training by the UE and gNB. Some embodiments of this application mainly discuss joint training by the UE and gNB.

[0033] Figure 2 It is shown that according to some embodiments of the present application, at least one first node 10 (such as at least one user equipment UE), a second node 20 (such as a base station gNB) and at least one third node 30 (such as at least one user equipment UE) are used for a communication network system 40. The communication network system 40 includes at least one first node 10, a second node 20 and at least one third node 30. The first node 10 includes a memory 12, a transceiver 13 and a processor 11 coupled to the memory 12 and the transceiver 13. The second node 20 includes a memory 22, a transceiver 23 and a processor 21 coupled to the memory 22 and the transceiver 23. The third node 30 includes a memory 32, a transceiver 33 and a processor 31 coupled to the memory 32 and the transceiver 33. The processor 11, 21 or 31 is configured to implement the functions, processes and / or methods proposed in this description. The layers of the wireless interface protocol can be implemented in the processor 11, 21 or 31. The memory 12, 22 or 32 is operatively coupled to the processor 11, 21 or 31, and stores various information used to operate the processor 11, 21 or 31. The transceiver 13, 23 or 33 is operatively coupled to the processor 11, 21 or 31, and the transceiver 13, 23 or 33 is used to transmit and / or receive wireless signals.

[0034] The processor 11, 21 or 31 may include an application-specific integrated circuit (ASIC), other chipsets, logic circuits and / or data processing devices. The memory 12, 22 or 32 may include a read-only memory (ROM), a random access memory (RAM), a flash memory, a memory card, a storage medium and / or other storage devices. The transceiver 13, 23 or 33 may include a baseband circuit for processing radio frequency signals. When these embodiments are implemented in software form, the described techniques may be implemented by modules (such as procedures, functions, etc.). These modules may be stored in the memory 12, 22 or 32 and executed by the processor 11, 21 or 31. The memory 12, 22 or 32 may be implemented as a built-in or external processor 11, 21 or 31, in which case it may be communicatively coupled to the processor 11, 21 or 31 in various known ways.

[0035] In some embodiments, the processor 11 provides auxiliary information through the second node 20, wherein the auxiliary information is used for the processor 11 and / or the second node 20 to monitor multiple ML models having a common part. This can solve the problems in the prior art, simplify the management of multiple ML models having a common part, provide a method for monitoring multiple ML models having a common part, reduce system signaling overhead, provide good communication performance and / or provide high reliability.

[0036] In some embodiments, the processor 21 is used to configure auxiliary information to the first node 10 and the at least one third node 30, wherein the auxiliary information is used for the processor 21 and / or the first node 10 and the at least one third node 30 to monitor multiple ML models having a common portion. This can solve the problems in the prior art, reduce the management of multiple ML models having a common portion, provide a method for monitoring multiple ML models having a common portion, reduce system overhead, provide good communication performance and / or provide high reliability.

[0037] Figure 3 According to some embodiments of the present application, a method 300 for configuring machine learning (ML) model monitoring by at least one first node is shown. In some embodiments, the method 300 includes: step 302, providing auxiliary information through a second node, wherein the auxiliary information is used for at least one first node and / or a second node to monitor multiple ML models with a common part. This can solve the problems in the prior art, simplify the management of multiple ML models with a common part, provide a method for monitoring multiple ML models with a common part, reduce system signaling overhead, provide good communication performance and / or provide high reliability.

[0038] Figure 4 It is shown that, according to some embodiments of the present application, a method 400 for ML model monitoring configuration is performed by a second node. In some embodiments, the method 400 includes: step 402, configuring auxiliary information to the first node and at least one third node, wherein the auxiliary information is used for the second node and / or the first node and at least one third node to monitor multiple ML models with a common part. This can solve the problems in the prior art, simplify the management of multiple ML models with a common part, provide a method for monitoring multiple ML models with a common part, reduce system signaling overhead, provide good communication performance and / or provide high reliability.

[0039] Figure 6 It is a schematic diagram showing the RAN intelligent function framework according to an embodiment of the present application. Figure 6 It is shown that in some embodiments, the ML model needs to be monitored during the inference phase. The functional framework of RAN intelligence is provided by RAN3 and can be further modified to apply to RAN1. The deployed ML model will be monitored to check whether it is working properly. Usually, the performance of the ML model is compared with the standard. If the ML model does not work properly, the UE will switch to another ML model or fall back to a non-AI working mode. At the same time, the monitored ML model will be retrained.

[0040] Embodiment 1:

[0041] Figure 7 It is a schematic diagram showing a situation where two UEs share a common decoder on the gNB side according to an embodiment of the present application. Figure 7 As shown, as an example, there are multiple first nodes (which can be UEs) and one second node (which can be a gNB). The encoder of UE1 and the encoder of UE2 share a common decoder on the gNB side. The decoder corresponds to the CSI reconstruction part, and the encoder corresponds to the CSI generation part.

[0042] The gNB configures the encoders of multiple UEs (such as UE1 and UE2), and the configuration information should be transmitted via downlink in the form of a broadcast signal. This configuration information is training auxiliary information.

[0043] As an example, the auxiliary information may be included in DCI 2_0, DCI 2_x, or a new UE group common signal, hereinafter referred to as DCI 2_x. In the example, the auxiliary information includes at least one of the following: activation / enabling of a deployed ML model, deactivation / disabling of a deployed ML model, activation / enabling of ML model monitoring, deactivation / disabling of ML model monitoring, ML model label, and identification information.

[0044] In some examples, the UE supports multiple different types of ML models. In the auxiliary information, the ML model tag can indicate the type of ML model, such as {"CSI feedback", "beam management", "positioning"}. This is to distinguish multiple ML models in the UE so that the UE can determine which specific ML model the auxiliary information applies to. DCI 2_x needs to indicate which ML model should be configured. In some examples, this is achieved through a two-bit field in DCI 2_x.

[0045] 00: "CSI Feedback"

[0046] 01: "Beam Prediction Time"

[0047] 10: "Beam Prediction Spatial"

[0048] 11: Positioning

[0049] "CSIFeedback" indicates the ML model type used to enhance CSI feedback; "BeamPredictionTime" indicates the ML model type used for time-domain beam prediction; "BeamPredictionSpatial" indicates the ML model type used for spatial-domain beam prediction; "Positioning" indicates the ML model type used for positioning.

[0050] When this field is not configured in the assistance information or does not exist in the assistance information, the assistance information is applied to the CSI feedback ML model in the UE by default.

[0051] In some examples, the identification information includes identification data of the ML model part. For example, the identification data can be one of the following: an ID of the ML model common part, an index of the ML model common part, and a name of the ML model common part. In some examples, the ML model common part is used for CSI generation. In some examples, the ML model common part is used for CSI reconstruction. In some examples, DCI 2_x includes a one-bit field, where "1" indicates activation of multiple ML models of the relevant UE, and "0" indicates deactivation of the relevant multiple ML models. Activation means that after the ML model is deployed, it is activated to start working and enter the reasoning stage. Deactivation means that after the ML model is activated, it is deactivated and stops working or reasoning.

[0052] The auxiliary information is identified by the identification information of the second node. It includes at least one of the following: cell ID and RNTI. For example, the UE group common signaling is scrambled by the SFI-RNTI of DCI 2.0 or the new RNTI of DCI 2_x. SFI-RNTI means time slot format indication radio network temporary identifier, and RNTI means radio network temporary identifier.

[0053] Some examples relate to configuration of ML models without common parts. In some examples, auxiliary information is used to configure multiple UEs, each of which has an ML model or ML model part deployed. Some of these ML models or ML model parts are not paired with a common ML model part. Configuration (providing auxiliary information via broadcast-like signaling) includes at least one of the following: activation / enabling of deployed ML models, deactivation / disabling of deployed ML models, activation / enabling of ML model monitoring, deactivation / disabling of ML model monitoring, and identification information.

[0054] Embodiment 2:

[0055] In another example, the training auxiliary information is included in the SIB / MIB. For example, the training auxiliary information includes at least one of the following: activation / enabling of ML model training, deactivation / disabling of ML model training, and identification information of the second node.

[0056] For example, there may be a field in SIB(x) / MIB: ML-model-activation-monitoring-common{0 / 1}, where {0} indicates activation / enabling of ML model monitoring, such as for CSI feedback; and {1} indicates deactivation / disabling of ML model monitoring, such as for enhanced CSI feedback.

[0057] For example, there may be a field in SIB(x) / MIB: ML-model-activation-common{0 / 1}, where {0} indicates activation / enabling of the ML model, such as for enhanced CSI feedback; {1} indicates deactivation / disabling of the ML model, such as for CSI feedback. In this way, all UEs in the current cell can receive this information, and this information can be used to configure monitoring of the ML model.

[0058] Embodiment 3:

[0059] Figure 8 It is a schematic diagram illustrating an example of bilateral model monitoring with a common CSI reconstruction part according to an embodiment of the present application. Fig. 9 It is a flowchart illustrating an example of bilateral model monitoring with a common CSI reconstruction part according to an embodiment of the present application. Figure 8 and Fig. 9Some examples of model monitoring are shown below. Some examples for CSI feedback ML models with public parts are provided.

[0060] The common part may be a common CSI reconstruction part. A UE runs a ML model part for CSI generation. The ML model of the UE and the common part of the gNB are being monitored. If the ML model does not work properly or the UE needs to run a model with a different complexity, a model switch will be triggered. In some examples, if a participating UE runs a ML model part and it is determined through model monitoring that the ML model has failed, all ML models may be deactivated.

[0061] If a UE1 with an encoder paired with a gNB (with a common part decoder) is determined to be faulty through model monitoring, the encoder of UE1 will be deactivated and UE1 will switch to another ML model or fall back to non-AI working mode. At the same time, the common part of the gNB will also be deactivated, as well as other related UEs. The encoder or decoder (common model part) can be retrained after deactivation. Each model paired with the common part will be deactivated. The deactivation information is contained in broadcast-like signaling, which can be DCI 2_x.

[0062] If some UEs paired with a common ML model part in the gNB are determined to have ML model failures through model monitoring, these ML models will be deactivated. The deactivation information is included in broadcast-like signaling.

[0063] In some examples, the number of participating UEs can be a natural number or a percentage. For example, if within a time window, the deployed ML models of 20% of the participating UEs are indicated as failed by model monitoring, then the encoders of all UEs paired with the common ML model part will be deactivated. The deactivation information is included in the broadcast-like signaling, and the common model part of the gNB will also be deactivated.

[0064] In some examples, the time window is configured by the gNB to the participating UEs via RRC signaling / MAC-CE. In some examples, the time window is reported to the gNB by the participating UEs.

[0065] If all UE encoders paired with a common ML model part in the gNB are determined by model monitoring to have ML model failure within a time window, all ML models will be deactivated. The deactivation information is included in broadcast-like signaling. In some examples, the time window is configured by the gNB to the participating UEs via RRC signaling / MAC-CE. In some examples, the time window is reported to the gNB by the participating UEs.

[0066] Embodiment 4:

[0067] In some examples, for model selection, if there is a (backup) ML model, that model is selected; if there is no backup model, it falls back to a non-AI working mode; if there are multiple ML models, one ML model is randomly selected.

[0068] In some examples, for model selection, if there is a (backup) ML model, the model is selected; if there is no backup model, it falls back to a non-AI working mode; if there are multiple ML models, the ML model with a higher priority is selected.

[0069] The priority is determined by at least one of the following factors. For example, if the UE prefers a low-complexity model, the low-complexity model is prioritized. Complexity may include at least one of the following factors, such as FLOPs (floating-point operations per second), model size, number of model parameters, pre-processing overhead / complexity, and post-processing overhead / complexity. It can be a general model or a scenario-specific model (non-general model). In some examples, if the UE prefers a general model, the CSI feedback ML model with a common part is prioritized. The CSI feedback ML model with a common part can be regarded as a special general model. In some examples, if the UE prefers a scenario-specific model, the scenario-specific model is prioritized. Priority factors may also include power consumption, inference delay, whether the model can be post-processed / pre-processed, and whether the model can be fine-tuned.

[0070] UE Preference (UE Report):

[0071] In some examples, the priority of the ML model is determined by UE reporting. The UE reports its preference via UE capabilities at initial access.

[0072] gNB configuration:

[0073] The gNB can configure the priority of ML model selection through RRC signaling, MAC-CE or DCI field. If the UE does not report any preference or the gNB does not configure any model selection priority, the UE or gNB will randomly select an ML model when there are multiple alternative ML models to choose from.

[0074] Fig.10 1 is a block diagram of a communication device 1000 according to an embodiment of the present application. The communication device 1000 may be a first node, such as a UE. The second node may be a base station. The communication device 1000 includes a monitor 1001, which is used to provide auxiliary information by the second node, and the auxiliary information is used for at least one first node and / or second node to monitor multiple ML models having a common part.

[0075] Fig.111 is a block diagram of a communication device 1100 according to an embodiment of the present application. The communication device 1100 may be a second node, such as a base station. The first node and the third node may be UEs. The communication device 1100 includes a monitor 1101, which is used to configure auxiliary information to the first node and at least one third node, and the auxiliary information is used for the second node and / or the first node and at least one third node to monitor multiple ML models having a common part.

[0076] In some examples, at least one first node is a user equipment (UE), the second node is a base station, and at least one third node is another UE. The encoder of the UE and the encoder of at least one other UE share a common decoder on the base station. The encoder of the UE and the encoder of at least one other UE may be a channel state information (CSI) generation part and a CSI reconstruction part, respectively, and the common decoder of the base station corresponds to another part of the CSI generation part or the CSI reconstruction part. In some examples, the auxiliary information is included in UE group common signaling or broadcast signaling, which is included in downlink control information (DCI) 2_0 or DCI2_x, or the auxiliary information is included in a system information block (SIB) and / or a master information block (MIB).

[0077] In some examples, the auxiliary information includes at least one of: activation / enabling of ML model monitoring, deactivation / disabling of ML model monitoring, activation / enabling of deployed ML models, deactivation / disabling of deployed ML models, ML model tags, and identification information. In some examples, activation / enabling, deactivation / disabling of ML models, and activation / enabling and deactivation / disabling of ML model monitoring are fields in DCI 2_0 or DCI 2_x. In some examples, the auxiliary information has a field for indicating each ML model tag to identify different types of ML models, or when the field is not configured or does not exist in the auxiliary information, the auxiliary information defaults to the ML model of the CSI generation part of at least one first node and at least one third node.

[0078] In some examples, the identification information includes at least one of the following: a cell identifier (ID) or a radio network temporary identifier (RNTI), wherein the UE group common signaling is scrambled by a slot format indication radio network temporary identifier (SFI-RNTI) of DCI_2.0 or a new RNTI of DCI 2_x. In some examples, the identification information includes identification data of the ML model part, including at least one of the following: an ID of the ML model public part, an index of the ML model public part, and a name of the ML model public part. In some examples, the ML model public part is a public CSI generation part or a public CSI reconstruction part for the ML model.

[0079] In some examples, when the multiple ML models do not have a common part, the auxiliary information includes at least one of the following: activation / enabling of deployed ML models, deactivation / disabling of deployed ML models, activation / enabling of ML model monitoring, deactivation / disabling of ML model monitoring, ML model tags and identification information. In some examples, for an ML model with a common CSI reconstruction part, if at least one first node and at least one third node run the ML model part for CSI generation, the ML models of these first nodes and third nodes and the common part will be in a monitoring state. In some examples, for an ML model with a common CSI reconstruction part, if the ML model does not work properly, or at least one first node and at least one third node need to run models of different complexity, at least one first node, at least one third node and / or the second node will trigger a model switch.

[0080] In some examples, if at least one of the first nodes or the third node running a ML model portion paired with a common ML model portion is determined to be an ML model failure through model monitoring, all related ML models will be deactivated. In some examples, the ML model portions running on the first node and the third node are paired with the common portion, and when it is determined to be faulty through model monitoring, they will be deactivated within a time window, which can be configured to the relevant nodes by the second node through radio resource configuration (RRC) signaling or media access control control element (MAC-CE), or reported by the relevant node to the second node. In some examples, the ML model portions and / or the common portion in the first node and the third node will be retrained or re-monitored after deactivation. In some examples, for model selection, if there is a backup ML model, at least one of the first node, the second node, and the at least one third node will select the ML model; if there is no backup ML model, at least one of the first node, the second node, and the at least one third node will fall back to a non-artificial intelligence (AI) working mode; if there are multiple backup ML models, at least one of the first node, the second node, and the at least one third node will randomly select a backup ML model, or select a backup ML model with a higher priority.

[0081] In some examples, the priority of the ML model is determined by at least one of the following factors: the complexity of the ML model, including floating point operations per second (FLOPS), model size, number of model parameters, preprocessing overhead / complexity, post-processing overhead / complexity, general model, scenario-specific model, power consumption, inference delay, post-processing, pre-processing or model fine-tuning of the ML model. In some examples, the priority of the ML model is determined by a report of at least one of the first node, the second node and at least one third node. In some examples, at least one of the first node and at least one third node reports its preference through a capability report at the time of initial access, and the second node configures the priority of the ML model through RRC signaling, MAC-CE or DCI field. If none of the first node, the second node and the at least one third node determines the priority of the ML model, then when there are multiple ML models to choose from, at least one of the first node, the second node and the at least one third node will randomly select an ML model.

[0082] In some examples, a general framework including model monitoring is Fig.12 As shown, Figure 6Compared to , model monitoring is added. Model monitoring can trigger the retraining of ML models. In this example, model monitoring can trigger some actions such as model update, model activation, model deactivation, rollback, etc. These actions are completed by the executor. In another example, the executor and Fig.12 The model monitoring in is not connected, and the executors only perform the actions instructed by the model inference.

[0083] In summary, in certain embodiments of the present application, a method for monitoring multiple ML models having a common part is provided. The common part may be a CSI generation part or a CSI reconstruction part. When the ML model does not work properly, it should be switched to another model. For the CSI feedback ML model having a common part, a variety of model monitoring and model selection methods are provided. Certain embodiments of the present application have at least one of the following inventive effects: reducing the management overhead of multiple ML models having a common part. A method for monitoring multiple ML models having a common part is provided.

[0084] Fig.13 7 is a block diagram of an exemplary wireless communication system 700 according to an embodiment of the present application. The embodiments described herein may be implemented into a system using any appropriately configured hardware and / or software. Fig.13 A system 700 is shown, which includes a radio frequency (RF) circuit 710, a baseband circuit 720, an application circuit 730, a memory / storage device 740, a display 750, a camera 760, a sensor 770, and an input / output (I / O) interface 780, which are coupled to each other at least as shown. The application circuit 730 may include circuits, such as but not limited to one or more single-core or multi-core processors. The processor may include any combination of general-purpose processors and special-purpose processors, such as a graphics processor, an application processor. The processor may be coupled to a memory / storage device and configured to execute instructions stored in the memory / storage device to enable various applications and / or operating systems to run on the system.

[0085] The monitoring of the ML model is not limited to the UE or gNB, and the monitoring can be performed on a third node. The relevant signaling and data need to be reported to the third node. The third node can be a UE, a gNB, or a server. The method in the embodiment of the present application is also applicable. In this way, the signaling overhead between the gNB and the UE can be reduced.

[0086] While the present disclosure has been described in connection with what is considered to be the most practical and preferred embodiment, it is to be understood that the present disclosure is not limited to the disclosed embodiment, but is intended to cover various arrangements made without departing from the scope of the broadest interpretation of the appended claims.

Claims

1. A method for configuring machine learning (ML) model monitoring, performed by at least one first node, characterized in that: include: Auxiliary information is provided by the second node, wherein the auxiliary information is used for the at least one first node and / or the second node to monitor a plurality of ML models having a common portion.

2. The method according to claim 1, characterized in that The at least one first node is a user equipment UE, the second node is a base station, the at least one third node is at least one other UE, the encoder of the UE and the encoder of the at least one other UE share a common decoder of the base station, the encoder of the UE and the encoder of the at least one other UE involve one of a channel state information CSI generation part or a CSI reconstruction part, and the common decoder of the base station involves the other of the CSI generation part or the CSI reconstruction part.

3. The method according to claim 1 or 2, characterized in that: The auxiliary information is included in UE group common signaling or broadcast signaling, the UE group common signaling or the broadcast signaling is included in downlink control information DCI 2_0 or DCI 2_x, or the auxiliary information is included in a system information block SIB and / or a master information block MIB.

4. The method according to any one of claims 1 to 3, characterized in that The auxiliary information includes at least one of the following: activation / enabling of ML model monitoring, deactivation / disabling of the ML model monitoring, activation / enabling of a deployed ML model, deactivation / disabling of the deployed ML model, ML model label, and identification information.

5. The method according to claim 4, characterized in that The activation / enabling of the ML model, the deactivation / disabling of the ML model, and / or the activation / enabling and deactivation / disabling of the ML model monitoring are DCI fields in DCI 2_0 or DCI 2_x.

6. The method according to any one of claims 2 to 5, characterized in that The auxiliary information has a field for indicating each ML model label to identify different types of ML models; or when the field is not configured in the auxiliary information or the field is absent in the auxiliary information, the auxiliary information is used by default for at least one ML model used for the CSI generation part in the at least one first node and the at least one third node.

7. The method according to claim 4, characterized in that The identification information includes at least one of the following: a cell identifier ID or a radio network temporary identifier RNTI, wherein the UE group common signaling indicates through a time slot format that the radio network temporary identifier SFI-RNTI is scrambled as a new RNTI of DCI_2.0 or DCI 2_x.

8. The method according to claim 4 or 7, characterized in that: The identification information includes identification data of the ML model part, including at least one of the following: an ID of the ML model public part, an index of the ML model public part, and a name of the ML model public part.

9. The method according to claim 8, characterized in that The ML model common part is used for the ML model common CSI generation part or the ML model common CSI reconstruction part.

10. The method according to claim 4, characterized in that When the multiple ML models have no common parts, the auxiliary information includes at least one of the following: activation / enabling of the deployed ML model, deactivation / disabling of the deployed ML model, activation / enabling of the ML model monitoring, deactivation / disabling of the ML model monitoring, the ML model label, and the identification information.

11. The method according to any one of claims 10 to 12, characterized in that For the ML model having the common CSI reconstruction part, if at least one of the at least one first node and the at least one third node is running the ML model part for the CSI generation, the ML models of the at least one first node and the at least one third node and the common part will be monitored.

12. The method according to claim 11, characterized in that For the ML model with the common CSI reconstruction part, if the ML model does not work properly or the at least one first node and the at least one third node need to run models with different complexities, the at least one first node and the at least one third node and / or the second node will trigger model switching.

13. The method according to claim 11 or 12, characterized in that: If the ML model portion running in at least one involved node among the at least one first node and the at least one third node, paired with the common ML model portion, is determined to be an ML model failure through model monitoring, all involved ML models or all ML models will be deactivated.

14. The method according to claim 13, characterized in that If the ML model part running in at least one of the at least one first node and the at least one third node is paired with the common ML model part and is determined to be the ML model failure within a time window through the model monitoring, the time window is configured by the second node to at least one of the at least one first node and the at least one third node through radio resource configuration RRC signaling or media access control control element MAC-CE, or is reported to the second node by at least one of the at least one first node and the at least one third node.

15. The method according to claim 13 or 14, characterized in that After deactivation, the ML model portion and / or the common portion in the at least one first node and the at least one third node will be retrained or re-monitored.

16. The method according to any one of claims 2 to 10, characterized in that For model selection, if there is a backup machine learning ML model, one of the at least first node, the second node and the at least one third node selects the ML model; if there is no backup ML model, one of the at least first node, the second node and the at least one third node falls back to the non-artificial intelligence AI working mode; if there are multiple backup ML models, one of the at least first node, the second node and the at least one third node randomly selects a backup ML model, or selects a backup ML model with a higher priority.

17. The method according to claim 16, characterized in that The priority of the ML model is determined by at least one of the following factors: The complexity of the ML model, including floating-point operations per second (FLOPS), model size, number of model parameters, preprocessing overhead / complexity, postprocessing overhead / complexity; general-purpose model or scenario-specific model (non-general-purpose model); power consumption; inference latency; post-processing of the ML model; pre-processing of the ML model; or fine-tuning of the ML model.

18. The method according to claim 16 or 17, characterized in that The priority of the ML model is determined by reports of the at least one first node, the second node, and the at least one third node.

19. The method according to claim 18, characterized in that The at least one first node and the at least one third node report their preferences through capability reporting during initial access, and the second node configures the priority of the ML model through RRC signaling, MAC-CE or DCI field; If none of the first node, the second node, and the at least one third node determines the priority of the ML model, and there are multiple ML models to choose from, one of the at least first node, the second node, and the at least one third node randomly selects an ML model.

20. A method for configuring machine learning (ML) model monitoring, performed by a second node, characterized in that: include: Auxiliary information is configured to the first node and the at least one third node, wherein the auxiliary information is used for the second node and / or the first node and the at least one third node to monitor a plurality of ML models having a common portion.

21. A first node, characterized in that: include: Memory; Transceiver; as well as a processor coupled to the memory and the transceiver; The processor is configured to execute the method according to any one of claims 1 to 19.

22. A second node, characterized in that: include: Memory; Transceiver; as well as a processor coupled to the memory and the transceiver; The processor is configured to: Auxiliary information is configured to the first node and at least one third node, wherein the auxiliary information is used for the second node and / or the first node and the at least one third node to monitor a plurality of ML models having a common portion.

Images