Wireless communication system and method for determining ai / ML model in ue movement process

Abstract

The present application discloses a method of determining an artificial intelligence (AI) / machine learning (ML) model in a wireless communication system based on information, comprising receiving, by a third node, information from a first node and / or a second node, where the information is used for the third node to determine the AI / ML model in the wireless communication system. Further, a method of processing an AI / ML model, performed by a first node, includes the first node receiving a message from a third node and / or a second node, and the first node altering the AI / ML model or not altering the AI / ML model.

Description

Technical Field

[0001] The present application relates to the field of wireless communication systems, and more specifically, to a wireless communication system and a method for determining an artificial intelligence (AI) / machine learning (ML) model during the movement of a user equipment (UE). More specifically, the goal is to ensure that the AI / ML model can work properly after the handover is completed. The present application can reduce latency and ensure service continuity of UE mobility in an AI / ML-based wireless communication system. Background Art

[0002] Our society is undergoing a digital revolution. In the past few years, artificial intelligence (AI) and machine learning (ML) methods have been widely used in various industries to drive innovation and improve process efficiency. However, in terms of network design, with the sharp increase in data volume and the increase in network complexity, traditional methods cannot provide fast solutions in many cases. Therefore, AI / ML will become an indispensable technology for improving the performance of future wireless communication networks.

[0003] To support the integration of AI / ML in cellular network design, Rel.15 introduced the network data analytics function (NWDAF), which was enhanced in Rel.16 and Rel.17 organized by SA2. It is mainly used in the 5G core network (5G core, 5GC) to improve its functions. At the same time, it also standardizes the optimization of data collection and storage, training and ML model retrieval. In November 2019, 3GPP SA1 discussed the application scenarios and potential performance requirements of the 5G system (5GS) for the distribution and transmission of AI / ML models in the TR22.874 research report. From the perspective of the radio access network (RAN), RAN3 approved a research project on AI / ML empowering RAN at the Rel.17 meeting. The research TR37.817 aims to explore the collection of wireless big data and its application in network automation and intelligence, including use case definition, processes and information interaction required for different use cases. Subsequently, in order to explore the potential optimization effect of AI / ML algorithms on air interface, Rel.18 supported a new research project led by RAN1 at the 3GPP RAN plenary meeting. The study involved three use cases, including CSI feedback enhancement, beam management, and positioning accuracy improvement, and divided the discussion into two parts: general topics and specific use case topics. After two meetings, RAN2 related research was launched in October 2022, focusing on processes and signaling based on RAN1 research and agreements.

[0004] Many companies mentioned that handover is a factor that affects model selection. At the RAN1#110bis meeting, a preliminary agreement was reached on the model selection mechanism for the UE-side model and the dual-end model. More specifically, the decision on model selection can be made by the network or the UE.

[0005] In fact, in an air interface wireless communication system based on AI / ML, the channel may deteriorate due to the mobility of the UE, which may cause handover to occur. When the UE enters a new serving gNB, the system can adopt the traditional method or the AI / ML method in certain use cases. If the AI / ML method is adopted, triggering model selection is a candidate solution to ensure service continuity. However, the details of model selection related to UE mobility have not been discussed yet. In addition, since the UE enters a new serving gNB, the AI / ML model and its characteristics used on the old serving gNB may be unfamiliar to the new serving gNB, so model transmission may be involved during the UE movement process. In addition, considering that the large size of the AI / ML model will generate signaling overhead, model reconfiguration is also a feasible solution.

[0006] Therefore, based on the above discussion, during the UE mobility process, the AI / ML model determination method for the new serving gNB should be studied. Summary of the invention

[0007] The purpose of the present application is to propose a wireless communication system and a method for determining an artificial intelligence (AI) / machine learning (ML) model during the movement of a user equipment (UE), so as to provide a method for determining an AI / ML model during the movement of the UE, which has not been discussed in current research, and to ensure service continuity during the movement of the UE and improve switching performance.

[0008] In a first aspect of the present application, a method for determining an artificial intelligence (AI) / machine learning (ML) model in a wireless communication system based on information includes receiving information from a first node and / or a second node by a third node, wherein the information is used by the third node to determine the AI / ML model in the wireless communication system.

[0009] In a second aspect of the present application, a method for processing an AI / ML model is performed by a first node, including the first node receiving a message from a third node and / or a second node; and the first node changing the AI / ML model or not changing the AI / ML model.

[0010] In a third aspect of the present application, a wireless communication system includes a memory, a transceiver, and a processor coupled to the memory and the transceiver. The processor is configured to execute the above method.

[0011] In a fourth aspect of the present application, a non-transitory machine-readable storage medium stores instructions, which, when executed by a computer, cause the computer to perform the above method.

[0012] In a fifth 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 having the chip installed therein executes the above method.

[0013] In a sixth aspect of the present application, a computer readable storage medium stores a computer program, which enables a computer to execute the above method.

[0014] In a seventh aspect of the present application, a computer program product comprises a computer program, which enables a computer to execute the above method.

[0015] In an eighth aspect of the present application, a computer program enables a computer to execute the above method. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] 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.

[0017] Figure 1 This is a block diagram of one or more user equipments communicating with a network / gNB in ​​a communication network system according to an embodiment of the present application.

[0018] Figure 2A It is a flowchart of a method for determining an artificial intelligence (AI) / machine learning (ML) model in a wireless communication system based on information according to an embodiment of the present application.

[0019] Figure 2B It is a flowchart of a method for processing an AI / ML model by a first node according to an embodiment of the present application.

[0020] Figure 3A It is a schematic diagram of an example of the functional location of an AI / ML model applicable to a single-end model according to an embodiment of the present application.

[0021] Figure 3B It is a schematic diagram of an example of the functional location of an AI / ML model applicable to a dual-end model according to an embodiment of the present application.

[0022] Figure 4 It is a schematic diagram of an example of the functional location of an AI / ML model applicable to a dual-end model according to an embodiment of the present application.

[0023] Figure 5 It is a schematic diagram of an example of the functional location of an AI / ML model applicable to a single-end model according to an embodiment of the present application.

[0024] Figure 6 It is a schematic diagram of an example of the functional location of an AI / ML model applicable to a single-end model according to an embodiment of the present application.

[0025] Figure 7 is a system block diagram for wireless communication according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] 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.

[0027] Figure 1 It is described that in some embodiments of the present application, a communication network system 40 includes one or more user equipment (UE) 10 and a network / gNB 20 for communication. The communication network system 40 includes one or more UEs 10 and a network / gNB 20. One or more UEs 10 may include a memory 12, a transceiver 13, and a processor 11 coupled to the memory 12 and the transceiver 13. The network / gNB 20 may include a memory 22, a transceiver 23, and a processor 21 coupled to the memory 22 and the transceiver 23. The processor 11 or 21 may be configured to perform the proposed functions, processes and / or methods described in this description. Each layer of the wireless interface protocol may be implemented by the processor 11 or 21. The memory 12 or 22 is operatively coupled to the processor 11 or 21 and stores various information to support the operation of the processor 11 or 21. The transceiver 13 or 23 is operatively coupled to the processor 11 or 21 and is responsible for sending and / or receiving wireless signals.

[0028] The processor 11 or 21 may include an application-specific integrated circuit (ASIC), other chipsets, logic circuits and / or data processing devices. The memory 12 or 22 may include a read-only memory (ROM), a random access memory (RAM), flash memory, a memory card, a storage medium and / or other storage devices. The transceiver 13 or 23 may include a baseband circuit to process radio frequency signals. When these embodiments are implemented in software, the techniques described herein may be implemented by modules (e.g., procedures, functions, etc.) that perform the functions described. These modules may be stored in the memory 12 or 22 and executed by the processor 11 or 21. The memory 12 or 22 may be integrated inside the processor 11 or 21, or may be located outside the processor 11 or 21, in which case they may be communicatively connected to the processor 11 or 21 by various known methods.

[0029] Figure 2AA method 200 is described for determining an artificial intelligence (AI) / machine learning (ML) model based on information in a wireless communication system according to some embodiments of the present application. In some embodiments, the method 200 includes: step 202, receiving information from a first node and / or a second node by a third node, wherein the information is used by the third node to determine the AI / ML model in the wireless communication system.

[0030] In some embodiments, the information is sent from the first node and / or the second node to the third node and includes a message containing the AI / ML model and / or AI / ML model-related auxiliary information. In some embodiments, the AI / ML model is transmitted from the second node to the third node; and / or from the first node to the third node. In some embodiments, the AI / ML model-related auxiliary information includes one of the following: model ID; model group ID; or parameters and / or structure of the AI / ML model. In some embodiments, if the third node receives the parameters and / or structure of the AI / ML model, and if the third node has model relevance, the third node determines an AI / ML model. In some embodiments, if the third node receives the parameters and / or structure of the AI / ML model, the third node determines an AI / ML model.

[0031] In some embodiments, the third node determines the AI / ML model by model selection and / or model reconfiguration and / or model update. In some embodiments, the model relevance represents the range of some parameters and / or structures of the AI / ML model. In some embodiments, the model relevance is configured by the network, or fixed, or indicated by the second node or the first node. In some embodiments, the model reconfiguration and / or update includes the following options: determined by the third node; or based on model relevance.

[0032] Figure 2B 300 is a flowchart of a method for processing an AI / ML model by a first node according to an embodiment of the present application. In some embodiments, method 300 includes: step 302, the first node receives a message from a third node and / or a second node; and the first node changes the AI / ML model or does not change the AI / ML model.

[0033] In some embodiments, the message includes at least one of the following: a first indicator indicating whether the AI / ML model has been changed, and the user equipment UE knows whether the target gNB uses a new AI / ML model based on the first indicator; an AI / ML model ID that uniquely identifies the AI / ML model; a second indicator indicating that the UE uses an old AI / ML model or uses the new AI / ML model. In some embodiments, the first node changes the AI / ML model or does not change the AI / ML model, including: if the target gNB sends the second indicator, the UE changes or does not change the new AI / ML model according to the requirements of the second indicator; or if the target gNB does not send the second indicator, but the first node only knows that the AI / ML model in the target gNB has changed based on the first indicator, then the first node decides whether to change the new AI / ML model.

[0034] In addition, the wireless communication system 40 is configured to execute the above method 200 or the above method 300. The wireless communication system 400 is also configured to execute the methods in some of the following embodiments.

[0035] In some embodiments, the location of the AI / ML model function is applicable to a single-end model or a dual-end model, and the wireless communication system includes a base station and a UE. In some embodiments, the AI / ML model function includes an AI / ML model training function and an AI / ML model reasoning function. In some embodiments, for the location of the AI / ML model function of the single-end model, the AI / ML model training function is deployed in the base station, and the AI / ML model reasoning function is deployed in the UE; or the AI / ML model training function and the AI / ML model reasoning function are both deployed in the base station or the UE. In some embodiments, for the location of the AI / ML model function of the dual-end model, the AI / ML model training function is deployed in the base station or the UE and the base station, and the AI / ML model reasoning function is deployed in the UE and the base station.

[0036] In some embodiments, the source base station or UE is capable of sending a message containing auxiliary information related to the AI / ML model and / or transferring the old used AI / ML model to the target base station. In some embodiments, when the source gNB decides to initiate a handover procedure for the UE based on a related AI / ML model report message, or the UE is notified that it has moved to another base station, the source gNB or UE sends a message containing auxiliary information related to the AI / ML model to the target base station, and then the target gNB selects an AI / ML model or reconfigures the AI / ML model based on the received message, so that the UE can continue to work based on AI / ML operations.

[0037] In some embodiments, the received message includes a model ID for model selection, a model group ID for model selection, a model relevance for model selection, and / or information for model reconfiguration. In some embodiments, when the source base station decides to initiate a handover process for the UE based on a related AI / ML model report message, or the UE is notified that it has moved to another base station, the source base station or the UE sends the AI / ML model to the target gNB, and then the target base station uses the old AI / ML model to enable the UE to continue to work based on AI / ML operations. In some embodiments, the AI / ML model is transmitted using a new message, and / or an enhanced traditional message is used to transmit the AI / ML model. In some embodiments, the source base station or the UE sends only part of the AI / ML model to the target base station.

[0038] In some embodiments, the target gNB sends information about the status of the AI / ML model to the source base station, the source base station forwards the information to the UE, and the UE makes a decision based on the information. In some embodiments, the target gNB sends information about the status of the AI / ML model directly to the UE, and the UE makes a decision based on the information. In some embodiments, the status of the AI / ML model includes: the target base station selects a new AI / ML model, the target base station reconfigures the AI / ML model, or the target base station receives an old AI / ML model.

[0039] In some embodiments, the information includes at least one of the following: an indicator for indicating whether the AI / ML model has been changed; an AI / ML model ID that uniquely identifies the AI / ML model; and an indicator for indicating whether the UE uses an old AI / ML model or a new AI / ML model. In some embodiments, the auxiliary information related to the AI / ML model includes at least one of the following: a UE ID that uniquely identifies the UE; an AI / ML model ID that uniquely identifies the AI / ML model; an AI / ML model group ID for identifying an AI / ML model group; the type of the AI / ML model; the complexity of the AI / ML model; the generalization capability of the AI / ML model; and the key performance indicators (KPIs) of the AI / ML model.

[0040] The present invention relates to a wireless communication system in which an AI / ML algorithm runs on the RAN. The relevant new study project (SID) of AI / ML on the NR air interface has been approved at the 2022 123GPP RAN Plenary Meeting 94e, and RAN1 has taken the lead in relevant discussions. From the perspective of RAN2, the study was launched in October 2022, focusing mainly on the procedures and signaling based on the RAN1 study. The present invention provides a method for determining the AI / ML model during UE mobility. More specifically, the goal is to ensure that the AI / ML model can work properly after the handover is completed. The method provided can reduce latency and ensure service continuity of UE mobility in an AI / ML-based wireless communication system.

[0041] Since UE mobility may cause handovers, for AI / ML-based wireless communication systems, methods for determining the AI / ML model for the serving gNB must be supported to ensure service continuity. This may include the following aspects for different scenarios: model transmission, model selection, and model (re)configuration. In the present invention, model transmission, model selection, and model reconfiguration are involved. In the following sections, "source gNB" is used to represent the old serving gNB, and "target gNB" is used to represent the new serving gNB.

[0042] In order to ensure that the AI / ML-based method can work properly after the UE moves and the handover is completed, an important issue is to determine the AI / ML model for the wireless system. The present application provides a method for determining the AI / ML model, including: a method for selecting, transmitting and / or reconfiguring the AI / ML model; and the transmission and configuration of auxiliary information related to the AI / ML model. The effect of the present invention is to provide a method for determining the AI / ML model during the movement of the UE, which has not been discussed in the current research, and ensure service continuity during the movement of the UE and improve the switching performance.

[0043] Embodiment 1:

[0044] Consider the following scenario: the AI / ML model training function is deployed in the gNB, while the AI / ML model inference function is deployed on the UE and gNB side. More specifically, the scenario can be a single-end model and / or a dual-end model, and the location of the AI / ML model function is shown in Figure 3. Figure 3A and Figure 3B The following table describes the location of AI / ML model functions in different scenarios. Figure 3A The location of AI / ML model functions in the . Figure 3BIn the , the location of the AI / ML model function is applicable to the dual-end model. When the system initiates a handover procedure due to UE mobility, in order to maintain service continuity of AI / ML operations in wireless communication, the source gNB or UE can send a message containing auxiliary information related to the AI / ML model and / or transfer the old AI / ML model to the target gNB. The following methods can be considered.

[0045] Method 1: The source gNB or UE sends auxiliary information to the target gNB:

[0046] In method 1, when the source gNB decides to initiate a handover procedure for the UE based on a report message related to the AI / ML model, or the UE is notified that it has moved to another gNB, the source gNB or UE can send a message containing auxiliary information related to the AI / ML model to the target gNB, and then the target gNB can select an AI / ML model or reconfigure the AI / ML model based on the received message to enable the UE to continue to work based on AI / ML operations. To achieve this goal, the following alternatives can be considered.

[0047] Alternative 1: Model selection based on model ID:

[0048] In some examples, the target gNB receives a message containing an AI / ML model ID, which is a unified identifier for AI / ML models worldwide. The target gNB then searches for the AI / ML model in its own system based on the AI / ML model ID. If there is an AI / ML model with the same model ID as the old AI / ML model on the target gNB side, it means that the target gNB side uses the same AI / ML model as the source gNB side, and the model is also used on the UE side. Therefore, the target gNB selects the AI / ML model, enabling the UE to continue working based on AI / ML operations.

[0049] The message received by the target gNB is provided by the source gNB and / or the UE. In addition, the AI / ML model can be trained in other entities, such as third-party servers, other gNBs, OAM (operation and maintenance), OTT (Internet service provider), etc. Since the training of AI / ML models consumes a lot of resources, in order to avoid repeated training, the trained AI / ML models can be transferred between different entities in some cases. Therefore, the AI / ML model may exist in different entities. The model ID can be a relative ID or an absolute ID, but in either case, the target gNB can ensure the selection of a unique AI / ML model based on the model ID.

[0050] Alternative 2: Model selection based on model group ID:

[0051] In some examples, the target gNB receives a message including an AI / ML model group ID, which indicates that a group of AI / ML models with the same specific functions and / or the same characteristics share the same AI / ML model group ID. The target gNB then selects an AI / ML model from its own system based on the AI / ML model group ID. If the AI / ML model group ID on the target gNB side is the same as the old AI / ML model used, it means that the target gNB side has an AI / ML model with the same specific functions and / or the same characteristics as the old AI / ML model used on the source gNB side, and these models can also be used on the UE side. Once the AI / ML model in the target gNB has the same model group ID as the old AI / ML model used, the model will be regarded as a candidate AI / ML model. Subsequently, the target gNB selects an AI / ML model from the candidate AI / ML models for the UE to continue working based on AI / ML operations.

[0052] Since there may be multiple candidate AI / ML models, the model selection rules are as follows: If there is only one candidate AI / ML model, the candidate AI / ML model can be directly regarded as the selected AI / ML model. If there are multiple candidate AI / ML models, the target gNB can randomly select an AI / ML model from the candidate AI / ML models. The message received by the target gNB is provided by the source gNB and / or the UE.

[0053] Alternative 3: Model selection based on model correlation:

[0054] In some examples, the target gNB performs model selection during the handover process based on model relevance, which indicates the range of some parameters and / or structures of the AI / ML model. If the parameters and / or structure of a certain AI / ML model satisfy the model relevance, it will be considered as a candidate AI / ML model. More specifically, the source gNB or UE provides the target gNB with relevant parameters or structures of the old AI / ML model, such as the complexity, generalization capability, key performance indicators (KPI) of the AI / ML model, etc., which are considered as the benchmark / reference point of model relevance. For example, if the number of layers of the old AI / ML model is 5, the benchmark / reference point of the number of layers of the model is 5.

[0055] Based on the benchmark / reference floating range, it is assumed that the floating range of model relevance is plus or minus 1. Therefore, the number of model layers is in the interval [4, 6], that is, AI / ML models with 4, 5, and 6 layers can be regarded as candidate AI / ML models by the target gNB.

[0056] Due to the different nature of AI / ML related parameters and structures, model correlation can be calculated in various ways, such as addition, subtraction, multiplication, division, etc., and can be a single method or a combination of multiple methods.

[0057] In addition, model relevance can be configured by the network, fixed, or indicated by the source gNB or UE. Since there may be multiple candidate AI / ML models, the model selection rules are as follows: If there is only one candidate AI / ML model, the candidate AI / ML model can be directly regarded as the selected AI / ML model. If there are multiple candidate AI / ML models, the target gNB can randomly select an AI / ML model from the candidate AI / ML models. The message received by the target gNB is provided by the source gNB and / or UE.

[0058] Alternative 4: Model Reconfiguration:

[0059] In some examples, the source gNB or UE sends the target gNB parameters and / or structural information related to the AI / ML model, such as the complexity, generalization capability, key performance indicators (KPI), etc. of the AI / ML model, so that the target gNB can reconfigure the AI / ML model according to the provided information, or change some AI / ML models in its own system. For example, if the number of layers of the provided model is 5, and the number of layers of the AI / ML model on the target gNB side does not meet the requirement, but it has an AI / ML model with 6 layers, the target gNB adjusts the number of layers of the AI / ML model from 6 to 5 through reconfiguration.

[0060] Since the target gNB may store multiple AI / ML models, it is necessary to determine which AI / ML model should be reconfigured. The options are as follows: Autonomous decision by the target gNB. The target gNB determines the AI / ML model to be reconfigured based on its own implementation. Model relevance selection. The model relevance is set for model selection. The target gNB selects one among the candidate AI / ML models to be reconfigured or changed. For details on the model range, see Alternative 3. The message received by the target gNB is provided by the source gNB and / or the UE.

[0061] One or a combination of the above alternatives can ensure that the UE operates normally when based on AI / ML and maintain service continuity during UE mobility.

[0062] As described above, in method 1, there are multiple alternatives that can be used to achieve the goal during UE mobility. It should be noted that all alternatives use auxiliary information contained in the message sent by the source gNB or UE to the target gNB. Specifically, the message may include at least one of the following auxiliary information: UE ID, which uniquely identifies the UE; AI / ML model ID, which uniquely identifies the AI / ML model or relatively identifies the AI / ML model; AI / ML model group ID, which identifies a group of AI / ML models that have similar characteristics in the same scenario; AI / ML model type, which indicates the type of AI / ML model; AI / ML model complexity, such as computational effort (FLOPs), parameter scale, etc.; AI / ML model generalization capability, which refers to the ability of AI / ML models to be applicable to different use cases; AI / ML model key performance indicators (KPIs): such as rate, accuracy, power consumption, reliability, etc.; AI / ML model format, which refers to the transmission format of the AI / ML model, which can be a private format, an open format (such as ONNX), or a new format defined in 3GPP.

[0063] Method 2: AI / ML model is sent from source gNB to target gNB:

[0064] In method 2, when the source gNB decides to initiate a handover procedure for the UE based on the relevant AI / ML model report message, or the UE is notified that it has moved to another gNB, the source gNB or UE can directly send the AI / ML model to the target gNB. The target gNB then uses the old AI / ML model to enable the UE to continue to work based on AI / ML operations. Regarding the transmission of the AI / ML model, the following alternatives can be considered.

[0065] Alternative 1: New messages for transmitting AI / ML models:

[0066] Case 1: For AI / ML model transmission between source gNB and target gNB, a new Xn message should be introduced for transmission between source gNB and target gNB. The Xn message may contain at least one of the following auxiliary information:

[0067] UE ID, uniquely identifies UE; AI / ML model ID, uniquely identifies AI / ML model or relatively identifies AI / ML model; AI / ML model group ID, identifies a group of AI / ML models that have similar characteristics in the same scenario; AI / ML model type, indicates the type of AI / ML model; the complexity of the AI / ML model, such as the amount of computation (FLOPs), parameter scale, etc.; the generalization capability of the AI / ML model, refers to the ability of the AI / ML model to be applicable to different use cases; the key performance indicators (KPIs) of the AI / ML model, such as rate, accuracy, power consumption, reliability, etc.; the format of the AI / ML model, refers to the transmission format of the AI / ML model, which can be a private format, an open format (such as ONNX), or a new format defined in 3GPP.

[0068] Case 2: For AI / ML model transmission between UE and target gNB:

[0069] In order to transmit the AI / ML model between the UE and the target gNB, a new RRC message should be introduced for transmission between the UE and the target gNB. At the same time, the message can be transmitted after the UE accesses the target gNB. The RRC message may contain at least one of the following auxiliary information: UE ID, which uniquely identifies the UE; AI / ML model ID, which uniquely identifies the AI / ML model or relatively identifies the AI / ML model; AI / ML model group ID, which identifies a group of AI / ML models that have similar characteristics in the same scenario; AI / ML model type, which indicates the type of AI / ML model; the complexity of the AI / ML model, such as the amount of computation (FLOPs), parameter scale, etc.; the generalization ability of the AI / ML model, which refers to the ability of the AI / ML model to be applicable to different use cases; the key performance indicators (KPIs) of the AI / ML model, such as rate, accuracy, power consumption, reliability, etc.; the format of the AI / ML model, which refers to the transmission format of the AI / ML model, which can be a private format, an open format (such as ONNX), or a new format defined in 3GPP.

[0070] Alternative 2: Enhance traditional messages to support AI / ML models:

[0071] Case 1: For AI / ML model transmission between source gNB and target gNB, the existing Xn message should be reused for transmitting AI / ML model between source gNB and target gNB. When the source gNB decides to perform handover, the handover request is the first Xn message sent by the source gNB to the target gNB. Therefore, some AI / ML related information should be introduced in the Xn message.

[0072] The switching request message may include at least one of the following auxiliary information: AI / ML model ID, which uniquely identifies the AI / ML model; AI / ML model group ID, which identifies a group of AI / ML models that have similar characteristics in the same scenario; AI / ML model type, which indicates the type of AI / ML model; the complexity of the AI / ML model, such as the amount of computation (FLOP), parameter scale, etc.; the generalization capability of the AI / ML model, which refers to the ability of the AI / ML model to be applicable to different use cases; the key performance indicators (KPIs) of the AI / ML model, such as rate, accuracy, power consumption, reliability, etc.; the format of the AI / ML model, which refers to the transmission format of the AI / ML model, which can be a private format, an open format (such as ONNX), or a new format defined in 3GPP.

[0073] Case 2: For AI / ML model transmission between UE and target gNB, the existing RRC message should be reused for transmitting AI / ML model between UE and target gNB. Since the RRC Complete message is the first message sent by UE to target gNB, some AI / ML related information should be introduced in the RRC message.

[0074] The RRC completion message may include at least one of the following auxiliary information: UE ID, which uniquely identifies the UE; AI / ML model ID, which uniquely identifies the AI / ML model; AI / ML model group ID, which identifies a group of AI / ML models that have similar characteristics in the same scenario; AI / ML model type, which indicates the type of AI / ML model; the complexity of the AI / ML model, such as the amount of computation (FLOP), parameter scale, etc.; the generalization capability of the AI / ML model, which refers to the ability of the AI / ML model to be applicable to different use cases; the key performance indicators (KPIs) of the AI / ML model, such as rate, accuracy, power consumption, reliability, etc.; the format of the AI / ML model, which refers to the transmission format of the AI / ML model, which can be a private format, an open format (such as ONNX), or a new format defined in 3GPP.

[0075] Method 3: Partial AI / ML model is sent from source gNB or UE to target gNB:

[0076] In some examples, the source gNB or UE sends to the target gNB a portion of the old AI / ML model, such as a backbone network, a portion of the main AI / ML model structure and parameters (e.g., number of layers, activation function, optimizer), the complexity of the AI / ML model, the generalization capability of the AI / ML model, the KPI of the AI / ML model, the type of the AI / ML model, etc. The target gNB reconfigures or changes the AI / ML model based on the received main AI / ML model portion to adapt to the continued operation of the UE.

[0077] Embodiment 2:

[0078] When handover occurs, according to the above discussion, the target gNB may have the following three AI / ML model situations based on different information: (1) The target gNB selects a new AI / ML model. (2) The target gNB reconfigures the AI / ML model. (3) The target gNB receives the old AI / ML model. In cases (1) and (2), the AI / ML model used in the system is partially or completely changed, and to some extent, the AI / ML model can be regarded as a new AI / ML model. In case (3), the AI / ML model used in the system has not changed, and to some extent, the AI / ML model is still the old AI / ML model.

[0079] Subsequently, the target gNB will notify the UE of the relevant AI / ML model situation to let the UE know whether the AI / ML model on the target gNB side has changed. In addition, if the target gNB adopts a new AI / ML model, it may take one of the following approaches: ask the UE whether to change the currently used AI / ML model; directly notify the UE to change the currently used AI / ML model; let the UE decide whether to continue using the old AI / ML model or switch to the new AI / ML model. In order to achieve this goal, the following alternatives need to be considered. In fact, there are two ways for the target gNB to notify the UE: the target gNB notifies the UE through the source gNB; the target gNB directly notifies the UE.

[0080] Case 1: The source gNB notifies the UE:

[0081] The target gNB sends the relevant information of the AI / ML model to the source gNB, which then forwards this information to the UE, which makes a decision based on this information. This information includes at least one of the following:

[0082] An indicator to indicate whether the AI / ML model has changed. Based on this indicator, the UE can determine whether the target gNB uses a new AI / ML model. For example, if the indicator is set to 0, it means that the target gNB still uses the old AI / ML model; if the indicator is set to 1, it means that the target gNB uses a new AI / ML model.

[0083] AI / ML model ID is used to uniquely identify the AI / ML model. Based on this information, the UE can also determine whether the AI / ML model has changed and can obtain the ID of the new AI / ML model.

[0084] An indicator used to indicate whether the UE continues to use the old AI / ML model or switches to the new AI / ML model. For example, if the indicator is set to 0, it means that the UE can continue to use the old AI / ML model; if the indicator is set to 1, it means that the UE should use the new AI / ML model. Another way is that if the indicator is configured, the UE continues to use the old AI / ML model, otherwise it switches to the new AI / ML model.

[0085] In addition, the above information between the source gNB and the target gNB can be included in the handover command, which may include the following forms: as part of the Xn message handover request confirmation; as a new dedicated RRC message and included in the Xn message handover request confirmation; as a new dedicated RRC message and included in the new Xn message; the above information between the source gNB and the UE can be included in the RRC reconfiguration message, or sent as a new dedicated RRC message.

[0086] Case 2: The target gNB notifies the UE directly:

[0087] The target gNB sends the relevant information of the AI / ML model directly to the UE. More specifically, this process occurs after the UE accesses the target gNB, and the UE makes a decision based on this information. The information includes at least one of the following:

[0088] An indicator to indicate whether the AI / ML model has changed. Based on this indicator, the UE can determine whether the target gNB uses a new AI / ML model. For example, if the indicator is set to 0, it means that the target gNB still uses the old AI / ML model; if the indicator is set to 1, it means that the target gNB uses a new AI / ML model.

[0089] AI / ML model ID is used to uniquely identify the AI / ML model. Based on this information, the UE can also determine whether the AI / ML model has changed and can obtain the ID of the new AI / ML model.

[0090] An indicator used to indicate whether the UE continues to use the old AI / ML model or switches to the new AI / ML model. For example, if the indicator is set to 0, it means that the UE can continue to use the old AI / ML model; if the indicator is set to 1, it means that the UE should use the new AI / ML model. Another way is that if the indicator is configured, the UE continues to use the old AI / ML model, otherwise it switches to the new AI / ML model.

[0091] After receiving the information sent by the target gNB, the UE makes a decision: if the target gNB sends a clear indicator, the UE will decide whether to request a new AI / ML model based on the indicator; if the target gNB does not send a clear indicator, but the UE only knows that the AI / ML model has changed in the target gNB, then the UE can decide whether to request a new AI / ML model based on its own needs. In addition, the above information between the target gNB and the UE can be included in the RRC reconfiguration message or sent as a new dedicated RRC message.

[0092] Embodiment 3:

[0093] Consider the following scenario: AI / ML model training functions are deployed on both the gNB and UE sides, and AI / ML model inference functions are also deployed on both the gNB and UE sides. More specifically, this scenario is applicable to the two-end model, which can be used for joint training and independent training. The location of the AI / ML model functions in the two-end model is as follows: Figure 4 shown.

[0094] Sub-embodiment 1:

[0095] When the system initiates a handover process due to UE mobility, in order to maintain service continuity of AI / ML operations in wireless communication in Example 3, the source gNB may send a message containing auxiliary information related to the AI / ML model and / or transfer the old AI / ML model stored on the source gNB side to the target gNB. In addition, since the two parts of the training model are deployed on different sides (gNB and UE), no model transmission occurs between the UE and the gNB. The following methods can be considered:

[0096] Method 1: The source gNB sends auxiliary information to the target gNB:

[0097] In method 1, when the source gNB decides to initiate a handover procedure for the UE based on a report message related to the AI / ML model, the source gNB may send a message containing auxiliary information related to the AI / ML model to the target gNB. Subsequently, the target gNB may select an AI / ML model or reconfigure the AI / ML model based on the received message so that the UE continues to operate based on the AI / ML operation.

[0098] The alternative scheme in this method is the same as Example 1-Method 1, but it should be noted that the auxiliary information is only provided by the source gNB and does not involve the UE.

[0099] Method 2: AI / ML model is sent from source gNB to target gNB:

[0100] In method 2, when the source gNB decides to start a handover process for the UE based on a report message related to the AI / ML model, the source gNB can directly send the AI / ML model to the target gNB. The target gNB then uses the old AI / ML model to enable the UE to continue to operate based on AI / ML operations. The following alternatives can be considered for the transmission of the AI / ML model: The alternatives of this method are the same as those of embodiment 1-method 2, but it should be noted that: the auxiliary information is only provided by the source gNB and does not involve the UE; the AI / ML model is only transmitted through the Xn message.

[0101] Sub-embodiment 2:

[0102] In addition, when the handover occurs, according to the above discussion, the target gNB may have the following three AI / ML model situations based on different information: The target gNB selects a new AI / ML model. The target gNB reconfigures the AI / ML model. The target gNB receives the old AI / ML model. In cases (1) and (2), the AI / ML model used in the system is partially or completely changed, and to some extent, the AI / ML model can be regarded as a new AI / ML model. In case (3), the AI / ML model used in the system has not changed, and to some extent, the AI / ML model is still the old AI / ML model.

[0103] Subsequently, the target gNB will notify the UE of the AI / ML model, so that the UE knows whether the AI / ML model on the target gNB side has changed. To achieve this goal, the following alternatives need to be considered. In fact, there are two ways for the target gNB to notify the UE: the target gNB notifies the UE through the source gNB; the target gNB directly notifies the UE.

[0104] Case 1: The source gNB notifies the UE:

[0105] The target gNB sends the relevant information of the AI / ML model to the source gNB, which then forwards this information to the UE. The information includes at least one of the following:

[0106] An indicator to indicate whether the AI / ML model has changed. Based on this indicator, the UE can determine whether the target gNB uses a new AI / ML model. For example, if the indicator is set to 0, it means that the target gNB still uses the old AI / ML model; if the indicator is set to 1, it means that the target gNB uses a new AI / ML model.

[0107] AI / ML model ID is used to uniquely identify the AI / ML model. Based on this information, the UE can also determine whether the AI / ML model has changed and can obtain the ID of the new AI / ML model.

[0108] After receiving the information forwarded by the source gNB, the UE can learn whether the AI / ML model on the target gNB has changed. Then, the UE can notify the target gNB that it has learned about the AI / ML model.

[0109] In addition, the above information between the source gNB and the target gNB can be included in the following messages: as part of the handover command and included in the Xn message handover request confirmation; as a new dedicated RRC message and included in the Xn message handover request confirmation; as a new dedicated RRC message and included in a new Xn message; the above information between the UE and the target gNB can be included in the RRC reconfiguration message or the RRC completion message, or sent as a new dedicated RRC message.

[0110] Case 2: The target gNB notifies the UE directly:

[0111] The target gNB sends the relevant information of the AI / ML model directly to the UE. More specifically, this process occurs after the UE accesses the target gNB, and the UE makes a decision based on this information. The information includes at least one of the following:

[0112] An indicator to indicate whether the AI / ML model has changed. Based on this indicator, the UE can determine whether the target gNB uses a new AI / ML model. For example, if the indicator is set to 0, it means that the target gNB still uses the old AI / ML model; if the indicator is set to 1, it means that the target gNB uses a new AI / ML model.

[0113] AI / ML model ID is used to uniquely identify the AI / ML model. Based on this information, the UE can also determine whether the AI / ML model has changed and can obtain the ID of the new AI / ML model.

[0114] After receiving the information sent by the target gNB, the UE can determine whether the AI / ML model on the target gNB side has changed. Then, the UE can notify the target gNB that it has learned about the AI / ML model.

[0115] In addition, the above information between the UE and the target gNB can be included in the following messages: RRC reconfiguration message; new dedicated RRC message.

[0116] Embodiment 4:

[0117] Consider the following scenario: AI / ML model training functions are deployed on the gNB side, and AI / ML model inference functions are also deployed on the gNB side. More specifically, this scenario is applicable to a single-end model, where the location of the AI / ML model functions is as follows: Figure 5The method for determining the AI / ML model and the UE notification method are the same as those in Example 3.

[0118] Embodiment 5:

[0119] Consider the following scenario: AI / ML model training functions are deployed on the gNB side, and AI / ML model inference functions are also deployed on the gNB side. More specifically, this scenario applies to a single-end model, where the location of the AI / ML model functions is as follows: Figure 6 shown.

[0120] In this embodiment, when the UE moves to the target gNB, since both the model training function and the inference function are located in the gNB, if the current AI / ML model can operate normally, there is no need to transfer the model. However, for other AI / ML related functions (such as data collection), the target gNB may still need to collect data. In addition, the AI / ML model may need to be optimized or reconfigured, and the target gNB, as one of the entities, may need to obtain relevant AI / ML model auxiliary information.

[0121] After accessing the target gNB, the UE will send a message containing auxiliary information related to the AI / ML model to the target gNB. The message may contain at least one of the following auxiliary information: UE ID, which uniquely identifies the UE; AI / ML model ID, which uniquely identifies the AI / ML model; AI / ML model group ID, which identifies a group of AI / ML models that have similar characteristics in the same scenario; AI / ML model type, which indicates the type of AI / ML model; the complexity of the AI / ML model, such as the amount of computation (FLOP), parameter scale, etc.; the generalization capability of the AI / ML model, which refers to the ability of the AI / ML model to be applicable to different use cases; the key performance indicators (KPIs) of the AI / ML model, such as rate, accuracy, power consumption, reliability, etc.; the format of the AI / ML model, which refers to the transmission format of the AI / ML model, which can be a private format, an open format (such as ONNX), or a new format defined in 3GPP.

[0122] In addition, the above information between the UE and the target gNB can be included in the following messages: RRC completion message; new dedicated RRC message.

[0123] In summary, in order to ensure the normal operation of the AI / ML method after the UE moves, the key is to determine the AI / ML model used by the wireless system. The present application provides a method for selecting, transmitting and / or reconfiguring an AI / ML model, as well as a method for transmitting and configuring auxiliary information related to the AI / ML model. The present invention can provide a method for determining an AI / ML model during UE movement, which has not been involved in current research. It can ensure the continuity of services during UE movement and improve switching performance.

[0124] Figure 7 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. Figure 7 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.

[0125] 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 determining an artificial intelligence AI / machine learning ML model in a wireless communication system based on information, characterized in that: include: Information from the first node and / or the second node is received by a third node, wherein the information is used by the third node to determine an AI / ML model in the wireless communication system.

2. The method for determining an AI / ML model in a wireless communication system based on information according to claim 1, characterized in that: The information is sent from the first node and / or the second node to the third node, and includes a message containing the AI / ML model and / or auxiliary information related to the AI / ML model.

3. The method for determining an AI / ML model in a wireless communication system based on information according to claim 2, characterized in that: The AI / ML model is transmitted from the second node to the third node; and / or from the first node to the third node.

4. The method for determining an AI / ML model in a wireless communication system based on information according to claim 2, characterized in that: The AI / ML model-related auxiliary information includes one of the following: Model ID; Model Group ID; or Parameters and / or structure of the AI / ML model.

5. The method for determining an AI / ML model in a wireless communication system based on information according to claim 4, characterized in that: If the third node receives the parameters and / or structure of the AI / ML model, and if the third node has model relevance, the third node determines an AI / ML model.

6. The method for determining an AI / ML model in a wireless communication system based on information according to claim 4, characterized in that: If the third node receives the parameters and / or structure of the AI / ML model, the third node determines an AI / ML model.

7. The method for determining an AI / ML model in a wireless communication system based on information according to claim 4, characterized in that: The third node determines the AI / ML model through model selection and / or model reconfiguration and / or model update.

8. The method for determining an AI / ML model in a wireless communication system based on information according to claim 5, characterized in that: The model dependencies represent the range of some parameters and / or structures of the AI / ML model.

9. The method for determining an AI / ML model in a wireless communication system based on information according to claim 5, characterized in that: The model dependency is configured by the network, or is fixed, or is indicated by the second node or the first node.

10. The method for determining an AI / ML model in a wireless communication system based on information according to claim 7, characterized in that: The model reconfiguration and / or updating includes the following options: At the sole discretion of the third node; or Based on model correlation.

11. A method for processing an AI / ML model, performed by a first node, characterized in that: include: The first node receives a message from the third node and / or the second node; as well as The first node modifies the AI / ML model or does not modify the AI / ML model.

12. The method according to claim 11, characterized in that The message includes at least one of the following: A first indicator, indicating whether the AI / ML model is changed, and the user equipment UE knows whether the target gNB uses a new AI / ML model based on the first indicator; AI / ML model ID, which uniquely identifies the AI / ML model; The second indicator indicates that the UE uses the old AI / ML model or the new AI / ML model.

13. The method according to claim 12, characterized in that The first node changes the AI / ML model or does not change the AI / ML model, including: If the target gNB sends the second indicator, the UE changes or does not change the new AI / ML model according to the requirement of the second indicator; or If the target gNB does not send the second indicator, but the first node knows that the AI / ML model in the target gNB has changed based only on the first indicator, the first node decides whether to change the new AI / ML model.

14. A wireless communication system, characterized in that: include: Memory; Transceiver; as well as a processor connected to the memory and the transceiver; The processor is configured to execute the method according to any one of claims 1 to 13.

15. A non-transitory machine-readable storage medium, characterized in that: Instructions are stored which, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 13.

16. A chip, characterized in that: include: The processor is configured to call and run the computer program stored in the memory so that the device on which the chip is installed executes the method according to any one of claims 1 to 13.

17. A computer-readable storage medium, characterized in that: A computer program is stored, the computer program causing a computer to execute the method according to any one of claims 1 to 13.

18. A computer program product, characterized in that It comprises a computer program causing a computer to execute the method according to any one of claims 1 to 13.

19. A computer program, characterized in that The computer program causes a computer to execute the method according to any one of claims 1 to 13.

Images