Communication device and method for AI / ML model replication of air interface

Abstract

The present application discloses a method of replicating an artificial intelligence (AI) / machine learning (ML) model in an air interface, performed by a first node, comprising: detecting signaling configured by a second node, where the signaling is associated with replicating at least one AI / ML model; copying an AI / ML model based on the signaling; and reporting the copied / duplicated AI / ML model based on the signaling.

Description

Technical Field

[0001] The present application relates to the field of wireless communication systems, and more specifically, to communication devices and methods for artificial intelligence (AI) / machine learning (ML) model replication for air interfaces. For example, the present application relates to the study item description (SID) of Release 18 on new radio (NR) air interface AI / ML established in the 3rd generation partnership project (3GPP) radio access network (RAN) plenary meeting 94e (December 2022). The discussion was led by RAN1 and started in May 2022. In particular, the present application covers one-sided models and two-sided models. Background Art

[0002] At the RAN1 94 meeting, ML was introduced to enhance the performance of the physical layer (PHY layer, or the first layer (L1 layer)), and it was agreed to study three use cases, namely enhanced channel state information (CSI) feedback, beam prediction, and positioning. Generally, all these use cases require a unified framework.

[0003] The following agreement was reached at the RAN1 110b-e meeting: For the UE-side model and the bilateral model, the following mechanisms are studied to support model selection, activation, deactivation, switching and fallback: 1. Determined by the network, including network-initiated and UE-initiated and network-requested. 2. Determined by the UE, including: event-triggered, configured by the network; UE decides and reports to the network; UE decides autonomously and reports to the network; UE decides autonomously but does not report to the network. In addition, the network-side model and other mechanisms are still under further study (FFS).

[0004] As can be seen from the above, multiple operations / behaviors have been defined. Therefore, during these operations, model replication is required to facilitate the execution of these operations. In addition, model replication also applies to operations not listed here, such as model deployment. A model can be trained on a node and then replicated and transferred to the UE for deployment.

[0005] In addition, the following agreement on ML model ID was reached at the RAN1 110b-e meeting. Model ID may be introduced. Agreement reached: Based on the AI / ML model having a model ID and its associated information and / or model functions, study the lifecycle management (LCM) procedure, at least applicable to some AI / ML operations; FFS: discuss the model ID and associated information and / or model functions in detail; FFS: study the use of LCM procedures based on model ID and associated information and / or model functions; FFS: study whether to support model ID; FFS: study specific operations applicable to AI / ML.

[0006] Some operations of ML models may conflict. One of the issues is "AI / ML model reasoning is interrupted due to model update". If the ML model is updated, it cannot perform reasoning at the same time. Therefore, the reasoning process is interrupted by the model update because the model update affects all ML models, some ML models, or the backbone of the ML model. Currently, there are few related technologies on this issue in the 3GPP existing technology.

[0007] Therefore, communication devices and methods for machine learning (ML) model monitoring are needed to solve the problems in the prior art, provide model distribution and transmission, reduce overhead, improve communication performance, and improve reliability. Summary of the invention

[0008] The purpose of this application is to propose a communication device and method for replicating artificial intelligence (AI) / machine learning (ML) models in the air interface to solve the problems in the prior art, provide model distribution and transmission, reduce overhead, improve communication performance, and improve reliability.

[0009] In a first aspect of the present application, a method for replicating an artificial intelligence (AI) / machine learning (ML) model in an air interface is provided, which is performed by a first node and includes: detecting signaling configured by a second node, wherein the signaling is related to replicating at least one AI / ML model; replicating the AI / ML model based on the signaling; and reporting the replicated / replicated AI / ML model based on the signaling.

[0010] In a second aspect of the present application, a first node is provided, comprising a memory; a transceiver; and a processor coupled to the memory and the transceiver, wherein the processor is configured to execute the above method.

[0011] In a third aspect of the present application, a method for replicating an artificial intelligence (AI) / machine learning (ML) model in an air interface is provided, which is executed by a second node, and includes configuring signaling to a first node, wherein the signaling is related to replicating at least one AI / ML model; controlling the first node to replicate the AI / ML model based on the signaling; and controlling the first node to report the replicated / repeated AI / ML model based on the signaling.

[0012] According to a fourth aspect of the present application, a second node is provided, comprising a memory; a transceiver; and a processor coupled to the memory and the transceiver, wherein the processor is configured to execute the above method.

[0013] In a fifth aspect of the present application, a non-temporary machine-readable storage medium is provided, in which instructions are stored, and when the instructions are executed by a computer, the computer executes the above method.

[0014] In a sixth aspect of the present application, a chip is provided, the chip comprising a processor, the processor being configured to call and run a computer program stored in a memory so that a device on which the chip is installed executes the above method.

[0015] In a seventh aspect of the present application, a computer-readable storage medium is provided, in which a computer program is stored, and the program can enable a computer to execute the above method.

[0016] In an eighth aspect of the present application, a computer program product is provided, which includes a computer program, and the program can enable a computer to execute the above method.

[0017] In a ninth aspect of the present application, a computer program is provided, which can enable a computer to execute the above method. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] 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.

[0019] Figure 1 The schematic diagram of the basic autoencoder model for enhancing CSI feedback according to an embodiment of the present application is exemplarily illustrated.

[0020] Figure 2 It is a block diagram of a communication node in a communication network system according to an embodiment of the present application.

[0021] Figure 3Exemplarily illustrated is a flowchart of a method for a first node to execute a method of replicating an artificial intelligence (AI) / machine learning (ML) model in the air interface according to an embodiment of the present application.

[0022] Figure 4 Exemplarily illustrated is a flowchart of a method for a second node to execute a method of replicating an AI / ML model in the air interface according to an embodiment of the present application.

[0023] Figure 5 Exemplarily illustrated is a flowchart of a basic autoencoder model for enhancing CSI feedback according to an embodiment of the present application.

[0024] Figure 6 Exemplarily illustrated is a schematic diagram of a functional framework of RAN intelligence according to an embodiment of the present application.

[0025] Figure 7 Exemplarily illustrated is a block diagram of a general framework of ML / AI for the NR air interface according to an embodiment of the present application.

[0026] Figure 8 Exemplarily illustrated is a flowchart of a reporting process of a replicated ML model according to an embodiment of the present application.

[0027] Fig. 9 Exemplarily illustrated is a flowchart of a process in which the CSI generation part is replicated and a paired model is transmitted to a third node according to an embodiment of the present application.

[0028] Fig.10 Exemplarily illustrated is a flowchart of a process in which the CSI generation part is replicated and a paired model is transmitted by a UE to a third node according to an embodiment of the present application.

[0029] Fig.11 Exemplarily illustrated is a block diagram of a communication device (such as a UE) according to an embodiment of the present application.

[0030] Fig.12 Exemplarily illustrated is a block diagram of a communication device (such as a gNB) according to an embodiment of the present application.

[0031] Fig.13 Exemplarily illustrated is a block diagram of a system for wireless communication according to an embodiment of the present application. Detailed implementation manners

[0032] The technical content, structural features, achieved objectives and effects of the present application are described in detail below with reference to the accompanying drawings. Specifically, the terms in the embodiments of the present application are only used for the purpose of describing specific embodiments and are not intended to limit the present application.

[0033] AI / ML is introduced into the physical layer (PHY layer) 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; positioning. ML models can be trained online or offline.

[0034] For the machine learning model, training is required first. The training can be performed by a node, which can be a gNB (next generation base station), UE (user equipment), or a third node. The training method can be online training or offline training. After the ML model training is completed, it will be deployed. For enhanced CSI feedback, the ML model is a bilateral model, so the model is deployed on the UE and gNB respectively. The operation of the CSI model will be monitored to ensure that it works properly.

[0035] Some embodiments of the present application discuss mechanisms for replicating AI / ML models over the air. Figure 1 The schematic diagram of the basic autoencoder model for enhancing CSI feedback according to an embodiment of the present application is exemplarily illustrated. Figure 1 In some embodiments, the basic model of the autoencoder is as follows: the encoder compresses the original CSI-RS value (referred to as original CSI) / maximum eigenvector and outputs it to the gNB; the gNB receives the report and decompresses it; the new CSI report is a CSI report containing CSI feedback enhanced by the AI / ML model.

[0036] At the UE side, the input signal is compressed and output to the channel. The input of the encoder can be the maximum eigenvector or the channel matrix; the compressed output is used as the input of the decoder and reconstructed at the gNB side. This application discusses a variety of ML model training methods, including training the ML model at the UE side and transmitting the model to the gNB; training the ML model at the gNB side and transmitting the model to the UE; joint training of the UE and gNB; independent training of the UE and gNB. Some embodiments of this application mainly discuss the joint training of the UE and gNB.

[0037] Figure 2An embodiment of the present application is schematically illustrated, in which in a communication network system 40, there is at least one first node 10, such as at least one user equipment (UE); a second node 20, such as a base station (e.g., gNB); and at least one third node 30, such as at least one user equipment (UE) or other communication devices. The communication network system 40 includes: at least one first node 10 (such as at least one UE); a second node 20 (such as a base station (gNB)); and at least one third node 30 (such as at least one UE or other communication devices). The first node 10 may include: a memory 12; a transceiver 13; and a processor 11, which is coupled to the memory 12 and the transceiver 13. The second node 20 may include: a memory 22; a transceiver 23; and a processor 21, which is coupled to the memory 22 and the transceiver 23. The third node 30 may include: a memory 32; a transceiver 33; and a processor 31, which is coupled to the memory 32 and the transceiver 33. The processor 11, 21 or 31 may be configured to execute the functions, processes and / or methods described in the present application. Each layer of the radio interface protocol may be implemented on 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 to support the operation of the processor 11, 21 or 31. The transceiver 13, 23 or 33 is operatively coupled to the processor 11, 21 or 31 and is used for transmitting and / or receiving wireless signals.

[0038] The processor 11, 21 or 31 may include an application-specific integrated circuit (ASIC); other chip sets; logic circuits; 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 the embodiment is implemented in software form, the technology may be implemented by modules (such as processes, functions, etc.) that execute corresponding functions. 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 integrated inside the processor 11, 21 or 31; or an external memory, which is communicatively connected to the processor 11, 21 or 31 through various known methods.

[0039] In some embodiments, the processor 11 is configured to detect signaling configured by the second node 20, wherein the signaling is related to duplicating at least one AI / ML model. The processor 11 is further configured to duplicate the AI / ML model based on the signaling, and the transceiver 13 is configured to report the duplicated / replicated AI / ML model based on the signaling. The solution can solve the problems in the prior art, provide model allocation and transmission, reduce overhead, improve communication performance, and improve reliability.

[0040] In some embodiments, the processor 21 is configured to configure signaling to the first node 10, wherein the signaling is related to copying at least one AI / ML model. The processor 21 is further configured to control the first node 10 to copy the AI / ML model based on the signaling, and the processor 21 is also configured to control the first node 10 to report the copied / duplicate AI / ML model based on the signaling. This solution can solve the problems in the prior art, provide model allocation and transmission, reduce overhead, improve communication performance, and improve reliability.

[0041] Figure 3 A method 300 based on an embodiment of the present application is schematically illustrated, and the method is used to perform replication of an AI / ML model by at least one first node (such as a UE) in an air interface. In some embodiments, the method 300 includes: step 302: detecting signaling configured by a second node, wherein the signaling is related to replicating at least one AI / ML model; step 304: replicating the AI / ML model based on the signaling; step 306: reporting the replicated / duplicate AI / ML model based on the signaling. The solution can solve the problems in the prior art, provide model allocation and transmission, reduce overhead, improve communication performance, and improve reliability.

[0042] Figure 4 A method 400 based on an embodiment of the present application is schematically illustrated, and the method is used for performing replication of an AI / ML model by a second node (such as a gNB) in an air interface. In some embodiments, the method 400 includes: step 402: configuring signaling to a first node, wherein the signaling is related to replicating at least one AI / ML model; step 404: controlling the first node to replicate the AI / ML model based on the signaling; step 406: controlling the first node to report the replicated / repeated AI / ML model based on the signaling. The solution can solve the problems in the prior art, provide model allocation and transmission, reduce overhead, improve communication performance, and improve reliability.

[0043] Figure 5 The flowchart of the basic autoencoder model for enhancing CSI feedback according to an embodiment of the present application is exemplarily illustrated. Figure 5In some embodiments, the basic model of the autoencoder is as follows: the encoder compresses the original CSI-RS value (referred to as original CSI) / maximum eigenvector and outputs it to the gNB; the gNB receives the report and decompresses it; the new CSI report is a CSI report containing CSI feedback enhanced by the AI / ML model.

[0044] Figure 6 The schematic diagram of the functional framework of RAN intelligence according to the embodiment of the present application is exemplarily illustrated. Figure 6 In some embodiments, the ML model needs to be monitored during the model inference process. RAN3 provides a functional framework of RAN intelligence, which can be further modified to apply to RAN1. The ML model needs to be monitored after deployment to check whether it is working properly. Typically, the performance of the ML model is compared with the standard. If the ML model does not work properly, the UE will do one of the following: switch to another ML model; fall back to a non-AI working mode. The monitored ML model will be retrained to restore normal function.

[0045] In some examples, general frameworks that include model monitoring include Figure 7 As shown. Figure 6 compared to, Figure 7 Added model monitoring. 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; model rollback, etc. These actions are performed by the executor. In another example, there is no direct connection between the executor and model monitoring (such as Figure 7 The executive only performs the operations indicated by the model reasoning.

[0046] In some embodiments of the present application, a method for replicating an ML model within an entity is provided. The benefits of model replication include: facilitating fine-tuning; facilitating model management and training. In order to achieve model replication, relevant signaling is designed, and the following contents are provided: allocation of model IDs; replication mechanism for multiple models; mechanism for gNB / UE to transmit replicated models. In addition, replication information (such as time and / or fine-tuning version) will be added to the model ID. It is necessary to copy the ML model and transmit it to the target node, which is mainly used for: model training; model generalization; model deployment, etc. The replicated model can be transmitted to another nearby UE, which will reduce the fine-tuning overhead of the UE; directly deploy and activate the replicated model. The technical effects achieved by some embodiments of the present application include at least one of the following: solving the model allocation problem, especially the allocation problem of model IDs. The model transmission problem is solved by model replication. To reduce overhead, in some embodiments, only the ID of the replicated model is transmitted to reduce the overhead of model transmission.

[0047] Example 1: Signaling design:

[0048] For an ML model (or part thereof), the ML model may be first deployed on the UE. The ML model (or part thereof) may be processed, such as fine-tuning; pre-processing; post-processing. The ML model (or part thereof) may be changed, such as model parameters change; model structure change; some model parameters and some model structure change, such as changes in the backbone of the ML model. In this case, the model is different from the state when it was initially deployed.

[0049] Whether an ML model can be copied is one of the model attributes or model descriptions, which can be configured by the gNB. For example, {1} indicates that the ML model can be copied; {0} indicates that the ML model cannot be copied. By default, the following contents of the ML model cannot be copied or are not allowed to be copied: the complete model; a partial model; a model structure; a model parameter. As an optional solution, by default, the following rule applies: only one of the complete model, partial model, model structure, or model parameters is allowed to be copied, and the others are not allowed to be copied.

[0050] In some examples, the signaling used to configure UE model replication includes: radio resource control (RRC) signaling; media access control-control element (MAC-CE); downlink control information (DCI). The signaling uses a single-bit field (One-BitField). By default, the following content cannot be copied or is not allowed to be copied: complete model; partial model; model structure; model parameters. As an optional solution, by default, the following rules apply: only one of the complete model, partial model, model structure, or model parameters is allowed to be copied, and other content is not allowed to be copied.

[0051] In some examples, the signaling used to configure UE model replication includes: Radio Resource Control (RRC) signaling; Media Access Control-Control Element (MAC-CE); Downlink Control Information (DCI). The signaling uses a two-bit field (Two-BitField), and its meaning is as follows: {00} means to copy the entire model (or part of the model); {01} means to copy only the model parameters; {10} means to copy only the model structure; {11} means to copy only the backbone of the ML model. By default, if the bit field is not configured, the complete ML model (or part of the model) is copied.

[0052] Figure 8A flowchart of the reporting process of the copied ML model according to an embodiment of the present application is exemplarily illustrated. The "Copy ML Model" signaling contains the copied model ID and indicates how to copy the ML model. In some examples, the UE reports the copied / duplicate ML model according to the configuration signaling, and the UE can report the complete model; only parameters; only model structure; only model backbone. In some examples, the gNB can configure a third node (such as a UE, an OTT server, or another gNB) to copy the ML model (or part thereof). The copying node does not necessarily have to be a UE.

[0053] In some examples, the Model ID is configured by the gNB, for example, in model copy signaling, the gNB sends a new Model ID to the UE, which is used for the newly copied model.

[0054] In some examples, the model ID is generated by the UE, and the ID of the copied ML model may be prefixed or suffixed based on the original ML model. For example, the original model ID is "xxxxxx"; the copied ML model ID may be "xxxxxx_copy001" or "copy001_xxxxxx".

[0055] In some examples, the model ID is configured by the gNB and generated by the UE. For example, the gNB configuration base ID is "xxxxxx" and the UE adds a prefix or suffix on this basis. For example, the gNB configuration ID is "yyyyyy", then the copied model ID can be "yyyyyy_copy001" or "copy001_yyyyyy".

[0056] Another way is to set the copied model ID within a specific range, for example, when the original ID "xxxxxx" is greater than "100000", the model ID is reserved for the copied model.

[0057] In some examples, the "copy" flag in the model ID indicates whether the model is a copy model or an original model. Other possible flags include: "C" (for copy) and "Dup" (for duplication).

[0058] Model Format:

[0059] In some examples, the copied model follows the format of the original model being copied. In some examples, this is the default setting, and if the copied model format is not configured, the format of the original model is used. In some examples, the copy signaling may indicate the format of the copied model, for example, ONNX; h5; Runtime Image, etc.

[0060] Example 2: Multiple Models:

[0061] The UE supports multiple ML models, and the gNB can configure the UE to copy multiple ML models through a single signaling. The signaling for configuring model copying includes the model IDs of the ML models to be copied, for example: {Model ID1 of Model 1, Model ID2 of Model 2, Model ID3 of Model 3, ……}. Similar to Embodiment 1, this signaling can be RRC signaling; MAC-CE; DCI.

[0062] In some examples, the signaling for configuring model copying includes the model IDs of the ML models to be copied and the model IDs of the copied models, for example {Model ID1 of Model 1, Model ID2 of Model 2, Model ID3 of Model 3, ……, Model ID11 of the copied model, Model ID21 of the copied model, Model ID31 of the copied model}.

[0063] In some examples, the signaling for configuring model copying includes the model IDs of the ML models to be copied and the model IDs of the copied models, for example {Model ID1 of Model 1, Model ID1 of the copied model, Model ID2 of the copied model, Model ID3 of the copied model}.

[0064] In some examples, there is only one original model ID, but it can be copied multiple times to generate multiple copied model IDs, for example {Model ID of Model 1, Model ID1 of the copied model, Model ID2 of the copied model, Model ID3 of the copied model}. Model ID1, Model ID2, and Model ID3 of the copied model are just three model IDs, and they are all linked to the same copied ML model.

[0065] For example, {Model ID of Model 1, Model ID1 of the copied model, Model ID2 of the copied model, Model ID3 of the copied model}. Model ID1, Model ID2, and Model ID3 of the copied model are three different model IDs, but they respectively correspond to three copies of the original ML model. This ML model can be copied at three different time points.

[0066] For example, this signaling is a broadcast type of signaling and is applicable to multiple UEs. Each UE will copy its ML model. The broadcast type of signaling can be DCI; RS; MAC-CE; RRC signaling.

[0067] Embodiment 3: Two-sided model for model transmission through gNB:

[0068] For the two-sided model, the above method can be used to replicate part of the model. Since the two-sided model works in pairs, it can be replicated in the following ways: copy part of the model; copy the entire model pair. The third node can be an OTT server (Over-the-Top Server); another gNB; another UE.

[0069] Fig. 9 The flowchart exemplarily illustrates the process in which the CSI generation part is copied and the pairing model is transmitted to the third node according to the embodiment of the present application. Fig. 9 In some embodiments, the following steps are performed: 1. The ML model on the UE is performing inference. 2. The UE is configured to copy the ML model; the gNB configures a new model ID to the UE for the copied ML model. 3. The UE performs model copying. 4. The UE reports the completion of the model copying to the gNB, including the PUCCH or PUSCH. 5. The gNB configures fine-tuning of the new copied model, and the configuration includes the ML model ID for specifying a specific ML model. 6. The UE performs the model fine-tuning procedure.

[0070] In some examples, ML models are paired on the gNB side. For ease of management, the paired ML models will obtain a new model ID, which is used to indicate the pairing relationship. For example, the ID format of the paired ML model is as follows: {ID of the CSI generation part, ID of the CSI reconstruction part}.

[0071] ID of the paired ML model: {ID of the CSI generation part, ID of the CSI reconstruction part}, which is transmitted to the third node together with the paired ML model.

[0072] Only the CSI generation part is copied and transmitted to the third node together with the paired ML model ID in the following format: {ID of CSI generation part, ID of CSI reconstruction part}. For example, in this copying method, one CSI reconstruction part can be paired with multiple CSI generation parts. The CSI generation part is transmitted to the third node together with the pairing information.

[0073] Only the CSI reconstruction part is copied and transmitted to the third node together with the paired ML model ID in the following format: {ID of CSI generation part, ID of CSI reconstruction part}. For example, in this copying method, one CSI generation part can be paired with multiple CSI reconstruction parts. The CSI reconstruction part is transmitted to the third node together with the pairing information.

[0074] Embodiment 4: Double-sided model in which the UE performs model transmission:

[0075] Fig.10The flowchart exemplarily illustrates the process in which the CSI generation part is copied and the pairing model is transmitted from the UE to the third node according to the embodiment of the present application. Fig.10 In some embodiments, the following steps are performed: 1. The ML model on the UE side is performing inference. 2. The UE is configured to copy the ML model; the gNB configures a new model ID to the UE for the copied ML model. 3. The UE performs model copying. 4. The UE reports the completion of the model copying to the gNB and transmits the copied model to the gNB via PUCCH or PUSCH. 5. The gNB fine-tunes the newly copied model. 6. The gNB transmits the copied model back to the UE. 7. The gNB triggers model switching. The trigger signal may be RRC signaling; MAC-CE; DCI. The model switching signaling includes the model ID of the ML model to be replaced; and the ID of the ML model after switching.

[0076] In some examples, ML models are paired on the gNB side. For ease of management, the paired ML models will obtain a new model ID to indicate the pairing relationship.

[0077] The ID format of the paired ML model is as follows: {ID of the CSI generation part, ID of the CSI reconstruction part}.

[0078] ID of the paired ML model: {ID of the CSI generation part, ID of the CSI reconstruction part}, which is transmitted to the third node together with the paired ML model.

[0079] Only the CSI generation part is copied and transmitted to the third node together with the paired ML model ID in the following format: {ID of CSI generation part, ID of CSI reconstruction part}. For example, in this copying method, one CSI reconstruction part can be paired with multiple CSI generation parts. The CSI generation part is transmitted to the third node together with the pairing information.

[0080] Only the CSI reconstruction part is copied and transmitted to the third node together with the paired ML model ID in the following format: {ID of CSI generation part, ID of CSI reconstruction part}. For example, in this copying method, one CSI generation part can be paired with multiple CSI reconstruction parts. The CSI reconstruction part is transmitted to the third node together with the pairing information.

[0081] Example 5: Additional information:

[0082] In some examples, time information may be included in the model copy, model ID, or copied model. For example, when the model ID is generated by the UE, the copied ML model ID may add a prefix or suffix to the original ML model ID, such as the original model ID: xxxxxx, the copied ML model ID: "xxxxxx_copy001_Oct26144000" or "copy001_xxxxxx_Oct26144000". Among them, Oct26144000 indicates that the model was copied at 14:40:00 on October 26. In some examples, the ML model ID may be sent to the gNB / UE via a single signaling. In other examples, the ML model ID may be sent via multiple signalings and merged into one ID. In some examples, time information is used to link the model with the system status, system configuration, and data collection time, so as to facilitate the tracing of the model generation process. In some examples, time information may be used for system rollback or system backup. In some examples, Oct26144000 indicates that the model was copied at 14:40:00 on October 26. In some examples, the time information is not included in the model ID but is stored in the model description information. The model description information records the time when the model is copied, which is used to indicate when the model is copied.

[0083] In some examples, the fine-tuning information may be included in the model ID or the model description information. For example, when the model ID is generated by the UE, the copied ML model ID may add a prefix or suffix based on the original ML model ID, such as the original model ID: xxxxxx, the copied ML model ID: "xxxxxx_copy001_TuningV2" or "copy001_xxxxxx_TuningV2". Among them, "TuningV2" indicates that the ML model has been fine-tuned twice. In some examples, "Tuning" is not necessarily spelled strictly according to the letters of the word, and its essence is to indicate that the model has been fine-tuned through a label. In some examples, the training information may also be included in the model ID or the model description information. For example, when the model ID is generated by the UE, the copied ML model ID may add a prefix or suffix based on the original ML model ID, such as the original model ID: xxxxxx, the copied ML model ID: "xxxxxx_copy001_trainingV2" or "copy001_xxxxxx_trainingV2". Among them, "trainingV2" indicates that the ML model has been trained twice. In some examples, "Training" is not necessarily spelled strictly according to the letters of the word. Its essence is to indicate through the label that the model has been trained. In some examples, at least one of the following information is included: fine-tuning information, timestamp, training information. This information can be stored in the model ID or model description information.

[0084] Fig.11 It is a block diagram of a communication device 1100 based on an embodiment of the present application. The communication device 1100 can be used as a first node, such as a UE. The second node can be a base station. The communication device 1100 includes the following components: a detector 1101, used to detect signaling configured by the second node, wherein the signaling is related to copying at least one AI / ML model. A copying unit 1102, used to copy the AI / ML model based on the signaling. A reporter 1103, used to report the copied / duplicate AI / ML model based on the signaling. This solution can solve the problems in the prior art, provide model allocation and transmission, reduce overhead, improve communication performance, and improve reliability.

[0085] Fig.12 It is a block diagram of a communication device 1200 based on an embodiment of the present application. The communication device 1200 can be used as a second node, such as a base station. The first node and the third node can be UEs. The communication device 1200 includes the following components: a configurator 1201, used to configure signaling to the first node, wherein the signaling is related to copying at least one AI / ML model. A controller 1202, used to control the first node to copy the AI / ML model based on the signaling; used to control the first node to report the copied / repeated AI / ML model based on the signaling. This solution can solve the problems in the prior art, provide model allocation and transmission, reduce overhead, improve communication performance, and improve reliability.

[0086] In some examples, the signaling may include radio resource control (RRC) signaling; media access control-control element (MAC-CE); downlink control information (DCI); reference signal signaling. In some examples, the signaling is broadcast-type signaling. In some examples, by default, the signaling is used to indicate copying a complete model portion. In some examples, the signaling may be a single-bit field or multiple bits, and the multiple bits are used to indicate copying a complete model portion; copying only parameters; copying a model structure; or copying a backbone portion of an AI / ML model. In some examples, when reporting a copied / duplicate AI / ML model based on the signaling, the report content may include reporting a complete model portion; reporting only parameters; reporting a model structure; or reporting a backbone portion of an AI / ML model.

[0087] In some examples, at least one of the following conditions is met: the signaling includes a model identifier (ID) for copying / forging the AI / ML model; wherein the model ID is generated by the first node, and the copied / forged AI / ML model is generated using the model ID and a prefix or suffix is ​​added to the original AI / ML model; wherein the model ID is configured by the second node and generated by the first node; and, the copied / forged model ID is within a specific range and is reserved for the copied model.

[0088] In some examples, the signaling is a single signaling for configuring the first node to replicate multiple AI / ML models. In some examples, the signaling includes a model ID of the AI / ML model to be replicated and a model ID of the replicated / forked model. In some examples, if the original model has only one ID, the unique ID can be replicated multiple times to be used for each replicated / forked model ID.

[0089] In some examples, the AI / ML model is paired at a first node or a second node. In some examples, the paired AI / ML model has a new ID to indicate the pairing relationship. In some examples, one or more channel state information (CSI) generation parts are copied at a first node, and one or more CSI reconstruction parts are paired at a first node or a second node. In some examples, a CSI reconstruction part is paired with multiple CSI generation parts. In some examples, the copied / replicated AI / ML model and / or the paired AI / ML model is transmitted to a third node. In some examples, the signaling, model ID and / or model description information of the copied / replicated AI / ML model includes fine-tuning information, timestamps and / or training information. In some examples, time information is used to associate the AI / ML model with the system state, configuration, and time of model replication, and / or time information is used for system rollback or backup.

[0090] In summary, in some embodiments of the present application, a method for replicating an ML model in an entity is provided. Model replication facilitates fine-tuning, model management, and training, and related signaling is designed. A model ID allocation scheme is provided, and a replication method for multiple models is given. A method for delivering a replicated model through a gNB / UE is also provided. At least replication information (such as time and / or fine-tuning version) is added to the model ID. When necessary, the replicated ML model can be transmitted to the target node for model training, model generalization, deployment, etc. The replicated model can be transmitted to another nearby UE, thereby reducing the fine-tuning overhead of the UE, or directly deploying and activating the replicated model. Some embodiments of the present application may have at least one of the following inventive effects: 1. The problem of model allocation is solved, especially the problem of model ID allocation. 2. The problem of model transmission is solved by replicating the model. To reduce overhead, in some embodiments, only the ID of the replicated model is transmitted.

[0091] 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.13A 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.

[0092] Monitoring of the ML model is not limited to the UE or gNB. Monitoring can be performed on a third node, and 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 invention is applicable. In this way, the signaling overhead between the gNB and the UE can be reduced.

[0093] 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 replicating an artificial intelligence AI / machine learning ML model in an air interface, executed by a first node, It is characterized in that include: detecting signaling configured by a second node, wherein the signaling is related to replicating at least one AI / ML model; Replicate the AI / ML model based on the signaling; and AI / ML models that replicate / duplicate based on the signaling reports.

2. The method according to claim 1, It is characterized in that The signaling includes radio resource control RRC signaling, medium access control-control element MAC-CE, downlink control information DCI or reference signal signaling.

3. The method according to claim 1 or 2, It is characterized in that The signaling is broadcast-like signaling.

4. The method according to any one of claims 1 to 3, It is characterized in that By default, the signaling is used to indicate that a complete model part is copied.

5. The method according to any one of claims 1 to 4, It is characterized in that The signaling is a bit field or multiple bits, and the multiple bits are used to indicate copying the complete model part, copying only parameters, copying the model structure, or copying the backbone of the AI / ML model.

6. The method according to claim 5, It is characterized in that Reporting the replicated / duplicate AI / ML model based on the signaling includes reporting the complete model portion, reporting only the parameters, reporting the model structure, or reporting the backbone of the AI / ML model.

7. The method according to any one of claims 1 to 6, It is characterized in that At least one of the following is met: The signaling includes a model identifier ID for the copied / duplicate AI / ML model; The model ID is generated by the first node, and the copied / duplicate AI / ML model uses the model ID and adds a prefix or suffix to the original AI / ML model; The model ID is configured by the second node and generated by the first node; as well as The copied / duplicate model ID is within a specific range, and the model ID is reserved for the copied model.

8. The method according to any one of claims 1 to 7, It is characterized in that The signaling is a single signaling for configuring the first node to replicate multiple AI / ML models.

9. The method according to claim 8, It is characterized in that The signaling includes the model ID of the AI / ML model to be copied and the copied / duplicate model ID.

10. The method according to claim 8 or 9, It is characterized in that If the original model has a unique ID, then the unique ID is copied multiple times and used for each copied / duplicate model ID.

11. The method according to any one of claims 1 to 10, It is characterized in that The AI / ML model is paired at the first node or the second node.

12. The method according to claim 11, It is characterized in that The paired AI / ML models have a new ID indicating the paired relationship.

13. The method according to claim 11 or 12, It is characterized in that One or more channel state information (CSI) generation parts are replicated at the first node, and one or more CSI reconstruction parts are paired at the first node or the second node.

14. The method according to claim 13, It is characterized in that One CSI reconstruction part is paired with the multiple CSI generation parts.

15. The method according to any one of claims 11 to 14, It is characterized in that The copied / duplicate AI / ML model and / or paired AI / ML model is transmitted to a third node.

16. The method according to any one of claims 7 to 15, It is characterized in that The signaling, the model ID and / or the model description information of the copied / duplicate AI / ML model includes fine-tuning information, timestamp and / or training information.

17. The method according to claim 16, It is characterized in that The time information associates the AI / ML model with system status, configuration, and the time when the model was copied, and / or the time information is used for system rollback or backup.

18. A method for replicating an artificial intelligence AI / machine learning ML model in an air interface, executed by a second node, It is characterized in that include: configuring signaling to the first node, wherein the signaling is related to replicating at least one AI / ML model; controlling the first node to replicate the AI / ML model based on the signaling; and Control the first node to replicate / duplicate the AI / ML model based on the signaling report.

19. A first node, It is 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 17.

20. A second node, It is characterized in that include: Memory; Transceiver; as well as a processor coupled to the memory and the transceiver; Wherein, the processor is configured to: configuring signaling to the first node, wherein the signaling is related to replication of the AI / ML model; controlling the first node to replicate the AI / ML model based on the signaling; and Control the first node to replicate / duplicate the AI / ML model based on the signaling report.

Images