AI / ML based bilateral model in multi-cell scenario

Abstract

Embodiments of the present disclosure relate to an artificial intelligence (AI) / machine learning (ML) based dual-sided model in a multi-cell scenario. In one aspect, a first network device receives, from a terminal device, an output of a first one or more machine learning models available at the terminal device. Further, the first network device receives, from a second network device, first information about a second one or more machine learning models available at the second network device. Further, the first network device obtains, based on the first information, second information about a performance of the second one or more machine learning models. The performance is determined based on the output.

Description

Technical Field

[0001] Various example embodiments relate to the field of communications, and particularly to devices, methods, apparatuses, and computer-readable storage media for two-sided models based on artificial intelligence (AI) / machine learning (ML) in multi-cell scenarios. Background Technology

[0002] A communication network can be viewed as a facility that enables communication between two or more communication devices or provides communication devices with access to a data network. Mobile or wireless communication networks are an example of communication networks.

[0003] Such communication networks operate according to standards, such as those issued by the 3rd Generation Partnership Project (3GPP) or the European Telecommunications Standards Institute (ETSI). Examples of such standards include the so-called fifth-generation (5G) standard or other standards issued by 3GPP. Summary of the Invention

[0004] Overall, the exemplary embodiments of this disclosure provide a technical solution related to AI / ML-based two-sided models in multi-cell scenarios.

[0005] In a first aspect, a first network device is provided. The first network device includes at least one processor and at least one memory storing instructions, which, when executed by the at least one processor, cause the first network device to at least: receive from a terminal device the output of one or more first machine learning models available at the terminal device; receive from a second network device first information regarding one or more second machine learning models available at the second network device; and, based on the first information, obtain second information regarding the performance of the second or more machine learning models, wherein the performance is determined based on the output.

[0006] In a second aspect, a second network device is provided. The second network device includes at least one processor and at least one memory storing instructions, which, when executed by the at least one processor, cause the second network device to at least: send first information to a first network device regarding a second or more machine learning models available at the second network device; and obtain second information regarding the performance of the second or more machine learning models, wherein the performance is determined based on the output of the first or more machine learning models available at a terminal device.

[0007] In a third aspect, a method is provided. The method includes: at a first network device, receiving from a terminal device the output of one or more first machine learning models available at the terminal device; at the first network device, receiving from a second network device first information regarding one or more second machine learning models available at the second network device; and at the first network device, based on the first information, obtaining second information regarding the performance of the second or more machine learning models, wherein the performance is determined based on the output.

[0008] In a fourth aspect, a method is provided. The method includes: at a second network device, sending first information about a second or more machine learning models available at the second network device to a first network device; and at the second network device, acquiring second information about the performance of the second or more machine learning models, wherein the performance is determined based on the output of the first or more machine learning models available at a terminal device.

[0009] In a fifth aspect, an apparatus is provided. The apparatus includes: components for receiving, at a first network device, the output of one or more first machine learning models available at the terminal device from a terminal device; components for receiving, at the first network device, first information about the second or more second machine learning models available at the second network device from a second network device; and components for obtaining, at the first network device, second information about the performance of the second or more second machine learning models based on the first information, wherein the performance is determined based on the output.

[0010] In a sixth aspect, an apparatus is provided. The apparatus includes: means for transmitting first information about a second or more machine learning models available at the second network device to a first network device; and means for obtaining second information about the performance of the second or more machine learning models at the second network device, wherein the performance is determined based on the output of the first or more machine learning models available at a terminal device.

[0011] In a seventh aspect, a non-transitory computer-readable medium is provided that stores instructions, which, when executed by a device, cause the device to perform at least the method according to any one of the third and fourth aspects described above.

[0012] In an eighth aspect, a computer program product is provided that includes program instructions for performing at least the method according to any one of the third and fourth aspects described above.

[0013] In a ninth aspect, a computer program including instructions is provided that, when executed by a device, causes the device to perform at least the method according to any one of the third and fourth aspects described above.

[0014] In a tenth aspect, a first network device is provided. The first network device includes a receiving circuitry configured to receive from a terminal device the output of one or more first machine learning models available at the terminal device; a receiving circuitry configured to receive from a second network device first information regarding one or more second machine learning models available at the second network device; and an acquiring circuitry configured to acquire second information regarding the performance of the second or more machine learning models based on the first information, wherein the performance is determined based on the output.

[0015] In an eleventh aspect, a second network device is provided. The second network device includes a transmitting circuitry configured to transmit first information to a first network device regarding a second or more machine learning models available at the second network device; and an acquiring circuitry configured to acquire second information regarding the performance of the second or more machine learning models, wherein the performance is determined based on the output of the first or more machine learning models available at a terminal device.

[0016] It should be understood that the summary portion is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0017] Some exemplary embodiments will now be described with reference to the accompanying drawings, in which:

[0018] Figure 1A The illustration shows an example environment in which example embodiments of the present disclosure may be implemented;

[0019] Figure 1B The illustration shows an example autoencoder architecture associated with some example embodiments of this disclosure;

[0020] Figure 2 The diagram illustrates the signaling flow between a first network device and a second network device according to some example embodiments of the present disclosure;

[0021] Figure 3A The illustrations illustrate examples of permitted model / function sharing according to some example embodiments of this disclosure;

[0022] Figure 3B The illustration shows an example communication process that allows model / function sharing according to some example embodiments of this disclosure;

[0023] Figure 3C The illustrations depict example inference strategies according to some example embodiments of the present disclosure;

[0024] Figure 3D The illustration shows another example communication process that allows model / function sharing according to some example embodiments of this disclosure;

[0025] Figure 3E The illustrations illustrate examples of non-model / function sharing according to some exemplary embodiments of this disclosure;

[0026] Figure 3F The illustration shows an example communication process that does not allow model / function sharing according to some example embodiments of this disclosure;

[0027] Figure 4 and Figure 5 The illustrations depict example communication processes according to some example embodiments of the present disclosure;

[0028] Figure 6 The illustration shows a flowchart of a method implemented at a first network device according to some embodiments of the present disclosure;

[0029] Figure 7 The illustration shows a flowchart of a method implemented at a second network device according to some embodiments of the present disclosure;

[0030] Figure 8 The illustration shows a simplified block diagram of a device suitable for implementing some example embodiments of the present disclosure; and

[0031] Figure 9 A block diagram illustrating an example of a computer-readable medium according to some exemplary embodiments of the present disclosure is shown.

[0032] Throughout the accompanying drawings, the same or similar reference numerals denote the same or similar elements. Detailed Implementation

[0033] The principles of this disclosure will now be described with reference to some exemplary embodiments. It should be understood that these embodiments are described for illustrative purposes and to assist those skilled in the art in understanding and implementing this disclosure, and do not constitute any limitation on the scope of this disclosure. The disclosure described herein can be implemented in various ways other than those described below.

[0034] In the following description and claims, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0035] In this disclosure, references to "an embodiment," "embodiment," and "example embodiment," etc., indicate that the described embodiment may include a particular feature, structure, or characteristic, but not every embodiment must include that particular feature, structure, or characteristic. Furthermore, such phrases do not necessarily refer to the same embodiment. Moreover, when a particular feature, structure, or characteristic is described in connection with an embodiment, those skilled in the art will understand that, whether explicitly described or not, combining it with other embodiments to affect such a feature, structure, or characteristic is within the scope of their knowledge.

[0036] It should be understood that although the terms “first” and “second”, etc., may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used only to distinguish one element from another. For example, without departing from the scope of the exemplary embodiments, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element. As used herein, the term “and / or” includes any and all combinations of one or more of the listed terms.

[0037] The terminology used herein is for describing particular embodiments and is not intended to limit the exemplary embodiments. The singular forms “a,” “an,” and “the” used herein also include the plural forms unless the context clearly indicates otherwise. Further understanding is that the terms “comprising,” “including,” “having,” “containing,” and / or “comprise”, as used herein, specify the presence of the stated features, elements, and / or components, but do not exclude the presence or addition of one or more other features, elements, components, and / or combinations thereof. As used herein, “at least one of the following: ” and “<at least one of a list of two or more elements>” and similar wording (where a list of two or more elements is connected by “and” or “or”) means at least any one of these elements, or at least any two or more of these elements, or at least all of these elements.

[0038] As used in this application, the term "circuit system" may refer to one or more or all of the following: (a) Pure hardware circuit implementation (such as implementations only in analog and / or digital circuit systems), and (b) A combination of hardware circuitry and software, such as (if applicable): (i) A combination of (multiple) analog and / or digital hardware circuits and software / firmware, and (ii) Any part of a hardware processor (including (multiple) digital signal processors), software, and (multiple) memories, which work together to enable a device (such as a mobile phone or server) to perform various functions, and (c) (Multiple) hardware circuits and / or (multiple) processors, such as (multiple) microprocessors or portions of (multiple) microprocessors, which require software (e.g., firmware) to operate, but the software may be absent when operation is not required.

[0039] The definition of "circuit system" applies to all uses of the term in this application, including in any claim. As another example, as used in this application, the term "circuit system" also covers only hardware circuitry or a processor (or processors) or a portion of hardware circuitry or a processor and its accompanying software and / or firmware. For example, if applicable to a particular claim element, the term "circuit system" also covers baseband integrated circuits or processor integrated circuits for mobile devices, or similar integrated circuits in servers, cellular network devices, or other computing or network devices.

[0040] As used herein, the term "communication network" refers to a network that conforms to any suitable communication standard, such as New Radio (NR), Long Term Evolution (LTE), LTE-A Advanced (LTE-A), Wideband Code Division Multiple Access (WCDMA), High-Speed ​​Packet Access (HSPA), Narrowband Internet of Things (NB-IoT), etc. Furthermore, communication between terminal devices and network devices in a communication network can be performed according to any suitable generation of communication protocol, including but not limited to third-generation (3G), 4G, 4.5G, 5G, or 6G communication protocols, and / or any other currently known or future-developed protocols. Embodiments of this disclosure can be applied to various communication systems. Given the rapid development of communications, there will naturally be communication technologies and systems that can be used to embody future types of communication technologies and systems. This should not be construed as limiting the scope of this disclosure to the systems described above.

[0041] As used herein, the term "network device" refers to a node in a communication network through which terminal devices can access and receive services. Network devices can refer to base stations (BS) or access points (APs), such as Node B (NodeB or NB), Radio Access Network (RAN) nodes, Evolved Node B (eNodeB or eNB), NR NB (also known as gNB), Remote Radio Unit (RRU), Radio Header (RH), infrastructure equipment for V2X (Vehicle-to-Everything) communication, Transmitter Receiver Point (TRP), Receiver Point (RP), Remote Radio Header (RRH), relay, Integrated Access and Backhaul (IAB) nodes, low-power nodes (such as femtoBS, picoBS), etc., depending on the terminology and technology used.

[0042] The term "terminal device" refers to any terminal device capable of wireless communication. As an example and not a limitation, a terminal device may also be referred to as a communication device, user equipment (UE), subscriber station (SS), portable subscriber station, mobile station (MS), or access terminal (AT). Terminal devices can include, but are not limited to, mobile phones, cellular phones, smartphones, Voice over IP (VoIP) phones, wireless local loop phones, tablets, wearable terminal devices, personal digital assistants (PDAs), portable computers, desktop computers, image capture terminal devices (such as digital cameras), gaming terminal devices, music storage and playback devices, in-vehicle wireless terminal devices, wireless endpoints, mobile stations, laptop embedded devices (LEE), laptop mounted devices (LME), USB dongles, smart devices, wireless customer premises equipment (CPE), Internet of Things (IoT) devices, watches or other wearable devices, head-mounted displays (HMDs), vehicles, drones, medical devices and applications (e.g., remote surgery), industrial devices and applications (e.g., robots and / or other wireless devices operating in industrial and / or automated processing chain environments), consumer electronics devices, devices operating on commercial and / or industrial wireless networks, etc. In the following description, the terms "terminal equipment", "communication equipment", "terminal", "user equipment" and "UE" are used interchangeably.

[0043] The principles and implementation of embodiments of this disclosure will now be described in detail with reference to the accompanying drawings. First, refer to... Figure 1A The illustration shows an example environment 100 in which example embodiments of the present disclosure may be implemented.

[0044] Environment 100 (which may be part of a communication network) includes terminal device 110 and network device 120 (also referred to as first network device 120) communicating with each other. Environment 100 may include one or more network devices 130-1, 130-2, ..., 130-N (collectively or individually referred to as network device 130 or second network device 130), where N represents any suitable positive integer. Terminal device 110 and / or first network device 120 may communicate with second network device 130. For example, one or more of terminal device 110, first network device 120, and second network device 130 may communicate with... Figure 1A Communicates with one or more other devices not shown in the diagram.

[0045] To send data and / or control information, terminal device 110 may communicate with first network device 120 or second network device 130. The link from first network device 120 or second network device 130 to terminal device 110 is called a downlink (DL), while the link from terminal device 110 to first network device 120 or second network device 130 is called an uplink (UL).

[0046] In some example embodiments, the first network device 120 may be the primary network device (also known as the serving network device) of the terminal device 110, and the second network device 130 may be a neighboring network device (also known as a non-serving network device), for example, a target set of neighboring network devices. As an example implementation, the terminal device 110 may perform cell (re)selection, and the (re)selected cell may be provided by the second network device 130. As another example implementation, the first network device 120 may switch the terminal device 110 to the second network device 130.

[0047] Despite Figure 1A The communication environment 100 describes terminal device 110, first network device 120, and second network device 130, but embodiments of this disclosure can be applied to any other suitable communication devices communicating with each other. That is, embodiments of this disclosure are not limited to... Figure 1A An exemplary scenario. In this regard, it should be noted that, although in Figure 1A The terminal device is schematically depicted as a mobile phone, and the first network device 120 and the second network device 130 are schematically depicted as base stations. It should be understood that these depictions are exemplary in nature and do not represent any limitation. In other embodiments, the first terminal device 110, the first network device 120, and the second network device 130 may be any other communication device, such as any other wireless communication device.

[0048] It should be understood that, such as Figure 1A The specific numbers of various communication devices and communication links shown are for illustrative purposes only and do not represent any limitation. Communication environment 100 may include any suitable number of communication devices and any suitable number of communication links for implementing embodiments of this disclosure. Furthermore, it should be understood that various wireless and wired communications (if desired) may exist between all communication devices.

[0049] Communications in Environment 100 may follow any suitable communication standards or protocols that are already in place or will be developed in the future, such as Universal Mobile Telecommunications System (UMTS), Long Term Evolution (LTE), LTE-Advanced (LTE-A), 5G New Radio (NR), 6G, Wi-Fi, and Global Microwave Access Interoperability (WiMAX) standards, and employ any suitable communication technologies, including, for example, Multiple-Input Multiple-Output (MIMO), Orthogonal Frequency Division Multiplexing (OFDM), Time Division Multiplexing (TDM), Frequency Division Multiplexing (FDM), Code Division Multiplexing (CDM), Bluetooth, ZigBee and Machine-Type Communications (MTC), Enhanced Mobile Broadband (eMBB), Massive Machine-Type Communications (mMTC), Ultra-Reliable Low-Latency Communications (URLLC), Carrier Aggregation (CA), Dual Connectivity (DC), and New Radio Unlicensed (NR-U) technologies.

[0050] The 18th edition (Rel-18) study of AI / ML for the NR air interface (FS_NR_AIML_Air) was completed at Radio Access Network Working Group 4 (RAN4) #109 meeting, and the consensus reached and the results of the study on outstanding issues are documented in Technical Report (TR) 38.843 (Version 18.0.0).

[0051] In addition, RAN#102 approved a new version 19 (Rel-19) work item (WI) for AI / ML of the NR air interface (NR_AIML_Air) to begin the standardization work for a general AI / ML framework for the air interface and to enable the use cases recommended in previous studies.

[0052] Generalization of AI / ML models / features is identified as one of the major challenges enabling AI / ML use cases. Generalization refers to the ability of a model and / or feature to appropriately adapt to new or previously unseen data derived from the same distribution used to create the model and / or feature. In other words, generalization examines the model / feature's ability, after being trained on the training set, to digest new data (primarily corresponding to new environments and / or scenarios) and to make correct predictions (on unknown / new environments or scenarios). If a model is trained too well on the training data, it will fail to generalize and will eventually make incorrect predictions when given new data. This will render the model ineffective, even if it can make correct predictions on the training dataset. This is called overfitting. In practice, if a model and / or feature has been trained for a given radio condition or specific parameter settings, encountering different radio conditions or parameter settings can severely impact the model / feature's performance.

[0053] Furthermore, some objectives related to Channel State Information (CSI) compression use cases are discussed. In the current 3GPP CSI feedback framework, the Rank Indicator (RI), Precoding Matrix Indicator (PMI), and Channel Quality Indicator (CQI) can be jointly reported to the gNB based on (multiple) configurations provided by the gNB. In the case of subband reporting, CQI feedback may require more resources. For codebook-based solutions, the UE determines the CQI used for reporting based on the precoding matrix indicated by the PMI and its associated receiver.

[0054] With the introduction of multi-user multiple-input multiple-output (MU-MIMO) systems, the overhead required to transmit high-resolution CSI feedback at high rank on the uplink has increased many times over. Therefore, AI / ML-based solutions can help reduce this overhead by compressing CSI reports. Various methods exist to achieve this using AI / ML-based solutions employing two-sided models (e.g., autoencoders).

[0055] Autoencoders are based on unsupervised learning techniques, where bottlenecks are imposed on the network to force a compressed knowledge representation of the original input. The main challenge remains reconstructing the original input. Figure 1B The diagram illustrates the architecture of an autoencoder. By definition, an autoencoder is designed to address the CSI feedback compression problem.

[0056] Before training the autoencoder, four hyperparameters need to be set, among others: 1) codeword / bottleneck size, 2) number of layers, 3) number of nodes per layer, and 4) the loss function to be used, such as mean squared error (MSE) or cosine similarity. The number of nodes per layer typically decreases with each subsequent layer of the encoder and increases in the decoder. The decoder is symmetrical to the encoder in terms of layer structure.

[0057] like Figure 1B As shown, an autoencoder consists of three parts: 1) an encoder, 2) a bottleneck (here, the codeword), and 3) a decoder. The encoder is designed to process the input data (e.g., the channel matrix) into a decoder. H The bottleneck (or feature vector) is compressed into codewords with a dimension smaller than the original information. The bottleneck (e.g., the codeword) is the compressed representation of the original information. Following the bottleneck is the decoder, i.e., the module used to decompress the codewords and reconstruct the data, i.e., the recovered information. .Then, and H The comparison is performed. This is a lossy process, and the recovered matrix... Possibly with H different.

[0058] The aforementioned CSI compression use case for enabling AI / ML in cellular communications features a cascaded, two-sided model. One model operates on the UE side to compress CSI, while the other model operates on the gNB side to decompress the compressed CSI received from the UE. When enabling AI / ML features involving two-sided models are deployed, the models at the UE and / or gNB may be updated frequently and independently at different times. Whenever a model is updated or the relevant gNB changes, alignment between the UE and gNB-side models is required to ensure interoperability and performance.

[0059] Specifically, in UE mobility scenarios, multiple UE-side models may need to interoperate with multiple different gNB-side models that the UE must use after mobility activities. However, the UE may typically not have any prior information about the multiple network-side models. Therefore, since the UE and / or network-side models / functions can vary between different gNBs due to deployment and vendor implementation, interoperability and joint performance between the UE and network-side models and functions cannot be guaranteed.

[0060] Therefore, for UE mobility scenarios, the UE and gNB (i.e., the primary gNB and (multiple) neighboring gNBs) must understand / evaluate the interoperability and performance characteristics of their corresponding models / functions in advance to enable AI / ML features on both sides. However, to date, there is no efficient method to support network-side interaction between the primary gNB and (multiple) neighboring gNBs to achieve the aforementioned interoperability and performance characteristics of the UE and network-side models / functions. Given the above, how to support network-side interaction to determine the interoperability and joint performance of the UE and network-side models / functions, particularly how the primary gNB will determine the joint performance of the UE-side models / functions (e.g., encoders) with the models / functions (e.g., decoders) of each / target neighboring gNB, and how the gNB performs optimizations related to handover scenarios with the terminal device, is an important problem that needs to be solved.

[0061] According to embodiments of this disclosure, a technical solution is provided to address the aforementioned problems. In this solution, a first network device receives the output of one or more first machine learning models available at a terminal device. Furthermore, the first network device receives first information from a second network device regarding a second or more machine learning models available at the second network device. Additionally, based on the first information, the first network device obtains second information regarding the performance of the second or more machine learning models. The performance is determined based on the output.

[0062] By acquiring second information about the performance of the second or more machine learning models based on first information received about the second or more machine learning models available at the second network device, the first network device can understand the performance of the models / functions available at the second network device (e.g., a neighboring cell (gNB)). In this way, the first network device can make better mobility decisions to switch terminals to a neighboring gNB without impacting performance or requiring a re-evaluation of the terminal-network side model pair performance after a cell change. Therefore, communication performance and efficiency can be improved.

[0063] In this disclosure, the terms “model” and “function” are used interchangeably in some cases.

[0064] In this disclosure, the terms “machine learning (ML) model” and “artificial intelligence (AI) model” are used interchangeably in some cases.

[0065] In this disclosure, the terms “performance,” “inference,” and “(multiple) test results” are used interchangeably in some cases.

[0066] In this disclosure, the terms "interoperability check" and "performance verification" may be used interchangeably in some cases.

[0067] Figure 2 The illustration shows a signaling flow 200 between a first network device and a second network device according to some example embodiments of the present disclosure. For discussion purposes, reference will be made to... Figure 1A Describe signaling flow 200.

[0068] like Figure 2 As shown, terminal device 110 sends (205) the output of one or more machine learning models (also referred to as the first one or more machine learning models) available at terminal device 110 to first network device 120. First network device 120 may include a master network device serving terminal device 110 or latched thereon by terminal device 110.

[0069] In some example embodiments, the first one or more machine learning models can be used as encoders to encode CSI, and in this case, the output can correspond to CSI encoded by the first one or more machine learning models. In other words, the output can include one or more compressed CSIs corresponding to the first one or more machine learning models. For example, if the terminal device 110 is capable of storing / using multiple AI / ML models / functions for dual-side use cases, the output can include multiple compressed CSIs corresponding to multiple terminal-side models / functions. The first one or more machine learning models can also be used for any other purpose (e.g., as encoders to encode any other information), and the scope of this disclosure is not limited in this respect.

[0070] In some example embodiments, the transmission of the output of a first or more machine learning models can be used to trigger a first network device 120 to obtain the performance of one or more machine learning models (also referred to as second or more machine learning models) available at a second network device 130. In this case, the transmission of the output of the first or more machine learning models can be used to trigger the initiation of model / functionality verification tests associated with the second or more machine learning models. The performance of the second or more machine learning models can be obtained based on the output of the first or more machine learning models. The second network device 130 may include adjacent network devices of the terminal device 110. Alternatively or additionally, the transmission of the output of the first or more machine learning models can be used to trigger the initiation of model / functionality verification tests associated with the second or more machine learning models and one or more machine learning models available at the first network device 120. It should be understood that although embodiments of this disclosure are discussed in the case where the transmission of the first transmission is used to trigger the acquisition of the performance of the second or more machine learning models available at the second network device 130 relative to the first or more machine learning models, embodiments of this disclosure can also be applied to the transmission of the first transmission being used to trigger the acquisition of the performance of the second or more machine learning models available at the second network device 130 and one or more machine learning models available at the first network device 120 relative to the first or more machine learning models. The scope of this disclosure is not limited in this respect.

[0071] In some example embodiments, performance may include the combined (i.e., overall) performance / compatibility (e.g., interoperability features) of a second or more machine learning models of the second network device 130 and a first or more machine learning models at the terminal device 110. As an example implementation, performance may be relative to the current model performance of the first network device 120, for example, relative to the performance of the current cell(s) models(s).

[0072] In an example embodiment where a first or more machine learning models are used as an encoder for encoding CSI, a second or more machine learning models can be used as a decoder for decoding the CSI encoded by the first or more machine learning models. In this case, performance can include joint performance (inference) of the two-sided model / function (i.e., encoder-decoder pair) and the network-side model / function. For example, performance can be associated with performance indicators of decoding the encoded CSI, such as decoding accuracy, decoding CQI, etc.

[0073] The transmission of the output of the first or more machine learning models used to trigger model / functional validation tests associated with the second or more machine learning models can be triggered in a variety of ways. For example, the transmission of the output of the first or more machine learning models can be triggered from the terminal side. As another example, the transmission of the output of the first or more machine learning models can be triggered from the network side.

[0074] In some example embodiments where the transmission of the output of the first or more machine learning models is triggered from the terminal side, terminal device 110 may send the output of the first or more machine learning models based on machine learning model updates at terminal device 110. For example, when a model update occurs at terminal device 110, terminal device 110 may trigger first network device 120 to acquire the performance of a second or more machine learning models available at second network device 130; in other words, trigger model / function verification testing. In this case, when there is a model / function update at terminal device 110, terminal device 110 may need to use the updated model / function. Terminal device 110 may then send an instruction (e.g., a message carrying the output of the first or more machine learning models) to first network device 120 to trigger model / function verification testing of the updated model / function available at terminal device 110 using network-side models / functions(s) of second network device 130.

[0075] In some example embodiments where the transmission of the output of the first or more machine learning models is triggered from the terminal side, terminal device 110 may transmit the output of the first or more machine learning models based on changes in channel conditions associated with terminal device 110. Changes in radio conditions may be associated with the mobility of terminal device 110. In some implementations, terminal device 110 may perform cell / beam-level radio measurements (e.g., Layer 1 (L1) / Layer 3 (L3) Reference Signal Received Power (RSRP)) in the primary / serving cell. When terminal device 110 determines that the measured RSRP level has decreased (e.g., a decrease in the RSRP level associated with the primary cell) (e.g., a decrease compared to a previous RSRP measurement, falling below a threshold), it may send an indication (e.g., a message carrying the output of the first or more machine learning models) to first network device 120 to trigger model / functional verification testing. In some implementations, terminal device 110 may perform radio measurements (e.g., RSRP measurements) associated with second network device 130. When terminal device 110 determines an increase in the measured RSRP level corresponding to one or more neighboring cells of second network device 130, terminal device 110 may send an indication (e.g., a message carrying the output of one or more first machine learning models) to first network device 120 to trigger a model / functional verification test. Alternatively or additionally, this trigger may also be generated by determining that terminal device 110 is moving toward the cell edge.

[0076] In some example embodiments where the transmission of the output of the first or more machine learning models is triggered from the terminal side, the terminal device 110 may send the output of the first or more machine learning models based on moving to a new cell. For example, whenever the terminal device 110 moves to a new cell, the new cell may become the primary cell, and the old primary cell may become one of one or more neighboring cells. In this case, the old set of neighboring cells may change because some cells in the old set may no longer be valid neighbors, and some new neighboring cells are added to the set of neighboring cells. In this case, the terminal device 110 may send an indication (e.g., a message carrying the output of the first or more machine learning models) to the first network device 120 to trigger model / functional verification testing.

[0077] In some example embodiments where the transmission of the output of the first or more machine learning models is triggered from the network side, the first network device 120 may send an instruction to the terminal device instructing the terminal device 110 to send the output of the first or more machine learning models, and then the terminal device 110 may send the output of the first or more machine learning models accordingly, as will be discussed below.

[0078] In some implementations, when a machine learning model / feature update / change occurs at the first network device 120 or the second network device 130, the first network device 120 may instruct the terminal device 110 to send the output of one or more first machine learning models to trigger a model / feature verification test. In this case, when there is a model / feature update at the first network device 120 or the second network device 130, the corresponding network device may need to use (multiple) updated network-side models / features. The network device (e.g., the first network device 120 or the second network device 130) can then trigger (multiple) model / feature verification tests for the updated network-side models / features for (multiple) terminal-side models / features.

[0079] In some implementations, when the first network device 120 determines that the terminal device 110 is moving toward the cell edge, it can instruct the terminal device 110 to send the output of one or more machine learning models to trigger the initiation of model / functional verification tests.

[0080] Upon receiving the output of one or more machine learning models from terminal device 110, first network device 120 may be triggered to acquire the performance (i.e., inference / test results) of a second or more machine learning models. Figure 2 As shown, the second network device 130 sends (210) information (also referred to as first information) about a second or more machine learning models available at the second network device 120. For example, the first information may include identification information of the second or more machine learning models, such as one or more model identifiers (IDs) of the second or more machine learning models.

[0081] In some example embodiments, the first network device 120 may send a request for first information to the second network device 130. Upon receiving the request, the second network device 130 may send the first information back to the first network device 120.

[0082] In some example embodiments, a request for first information may be based on a MODEL_INFO_REQUEST(model_type) message. This message may be intended for use by a first network device 120 to query information about a second or more machine learning models in a second network device 130. For example, this message may be intended to be used to query information about various model / feature pools (e.g., neighboring gNBs) in the second network device 130. The “model_type” field may include the type of model / feature(s)(s) for which this information is required, such as a DECODER.

[0083] In some example embodiments, the first information may be included in a MODEL_INFO_RESPONSE(status, model_info[model1, model2, …]) message. This message may be a response to a MODEL_INFO_REQUEST message. The “status” field may include the status of the response, for example, SUCCESS if the information was retrieved successfully, FAILURE if the request failed for some reason, or ACCESS_DENIED if the second network device 130 does not allow information sharing. The “model_info” field may include model-related information, such as identification information for one or more second machine learning models, such as a model ID.

[0084] like Figure 2 As shown, the first network device 120 obtains (215) information (also referred to as second information) about the performance of the second or more machine learning models based on the first information. Therefore, the second network device 130 obtains (220) the second information about the performance of the second or more machine learning models.

[0085] Depending on whether sharing of AI / ML model / function related information (e.g., model / function identifier (ID) used to identify the model / function) / files (e.g., one or more model / function parameters or complete model / function files) is allowed / supported between the first network device 120 and the second network device 130 (e.g., sharing of AI / ML model / function related information / files between gNBs), the methods used by the first network device 120 or the second network device 130 to obtain the performance of a second or more machine learning models can differ.

[0086] In some example embodiments, AI / ML model / function-related information / file sharing may be permitted between the first network device 120 and the second network device 130. In this case, the first network device 120 may be aware of a second or more machine learning models running on the second network device 130. For example, AI / ML model / function-related information / file sharing may be supported if the first network device 120 and the second network device 130 are from the same vendor. As another example, if the first network device 120 and the second network device 130 are from different vendors, AI / ML model / function-related information / file sharing may be supported if a consensus is reached between the different vendors, for example, within 3GPP. As yet another example, AI / ML model / function-related information / file sharing may be supported if the AI / ML model / function is standardized in 3GPP to assist the terminal device 110 during mobility scenarios.

[0087] In an example embodiment that allows sharing of AI / ML model / feature-related information / files, second information about the performance of a second or more machine learning models can be determined by the first network device 120 itself, as will be discussed in detail below.

[0088] In some implementations, a second or more machine learning models / functions in the pool of the second network device 130 may be identical to models / functions available at the first network device 120, and therefore one or more models / functions required by the second network device 130 may already be available at the first network device 120. In this case, the first network device 120 may determine, based on first information, that the second or more machine learning models are identical to one or more machine learning models available at the first network device 120, and therefore may determine one or more parameters (e.g., complete model / function files) of the second or more machine learning models based on one or more parameters (e.g., complete model / function files) of the one or more machine learning models available at the first network device 120. Then, the first network device 120 may determine second information regarding the performance of the second or more machine learning models based on the output of the first or more machine learning models and the available one or more parameters of the second or more machine learning models. In this case, the first network device 120 may perform corresponding interoperability checks (i.e., performance verification) locally. For example, the first network device 120 may evaluate / verify the joint performance of the first or more machine learning models / functions (e.g., encoders) with each of the second or more machine learning models / functions (e.g., decoders) in the model / function pool at the second network device 130. Then, the first network device 120 may send a performance report of the second or more machine learning models to the second network device 130. Alternatively or additionally, the first network device 120 may send the performance of the second or more machine learning models to the terminal device 110. For example, the performance of the second or more machine learning models shared with the terminal device 110 may include the joint performance of the ranking of the terminal-network side model pairs.

[0089] In some example implementations, the performance report of a second or more machine learning models can be based on a MODEL_PERFORMANCE_REPORT(report[]) message. This message can be sent from the first network device 120 to the second network device 130 to share the inference results after the inference of the second or more machine learning models / features relative to the first or more models has been performed at the first network device 120. The “report[]” field can include an aggregated report of the performance of various related models / features evaluated by the first network device 120.

[0090] In some implementations, at least one of the model / function pools at the second network device 130 and the first network device 120 may be different, and one or more parameters (e.g., complete model / function files of at least one of the second or more machine learning models at the second network device 130) may be shared by the second network device 130 and the first network device 120. In other words, one or more parameters (e.g., complete model / function files of at least one of the second or more machine learning models) may be obtained at the first network device 120. In this case, the first network device 120 may determine, based on first information, that at least one of the second or more machine learning models is different from one or more machine learning models available at the first network device 120. The first network device 120 may then send a request to the second network device 130 for one or more parameters (e.g., complete model / function files) of the at least one machine learning model, and then receive a response from the second network device 130 including one or more parameters (e.g., complete model files) of the at least one machine learning model.

[0091] In some example implementations, requests for one or more parameters of at least one machine learning model can be based on the MODEL_TRANSFER_REQUEST(model_id1, model_id2, …) message. This message can be intended to be used by a first network device 120 to request a second network device 130 to transfer models / features (i.e., one or more parameters, such as complete model / feature files) from its pool to the first network device 120 for two-sided model / feature performance verification. The “model_id” field can include a list of model / performance IDs of the models / features(s) to be transferred.

[0092] In some example implementations, a response to a request for one or more parameters of at least one machine learning model can be based on a MODEL_TRANSFER_RESPONSE(status, model) message. This message can be a response to a MODEL_TRANSFER_REQUEST. The "status" field can include the status of the response, for example, SUCCESS if the model / feature was retrieved successfully, and FAILURE if the request failed for some reason.

[0093] As described above, after querying and / or collecting information and / or files of a second or more machine learning models / functions existing in the pool of the second network device 130, the first network device 120 can then determine second information regarding the performance of the second or more machine learning models based on the output of the first or more machine learning models and one or more parameters of the second or more machine learning models (i.e., complete model / function files). In this case, the first network device 120 can perform a corresponding interoperability check (i.e., performance verification) at its endpoint. For example, the first network device 120 can evaluate / verify the joint performance of the first or more machine learning models / functions (e.g., encoders) with each of the second or more machine learning models / functions (e.g., decoders) in the model / function pool at the second network device 130. In other words, the first network device 120 can run inference on the second or more machine learning models / functions to evaluate the interoperability performance of(multiple) endpoint-side models / functions with(multiple) network-side models / functions available at the second network device 130. Alternatively or additionally, the first network device 120 can evaluate the interoperability performance of(multiple) endpoint-side models / functions with(multiple) network-side models / functions available at the first network device 120. Then, the first network device 120 may send a performance report of a second or more machine learning models to the second network device 130. For example, the performance report of the second or more machine learning models may be based on the MODEL_PERFORMANCE_REPORT(report[]) message as described above. Alternatively or additionally, the first network device 120 may send the performance of the second or more machine learning models to the terminal device 110. For example, the performance of the second or more machine learning models shared with the terminal device 110 may include the joint performance of the ranking of the terminal-network side model pair.

[0094] Figure 3A The diagram illustrates an example of how model / function sharing is allowed. For example... Figure 3A The UE shown can be an example implementation of terminal device 110, such as... Figure 3A The gNB1 shown may be an example implementation of the first network device 120, such as Figure 3A The gNB2 and gNB3 shown can be example implementations of network devices 130-1 and 130-2, respectively. In this case, gNB1 can be the UE's primary gNB, while gNB2 and gNB3 are neighboring gNBs.

[0095] like Figure 3AAs shown, the UE uses the activated model / function E1 (e.g., encoder E1) and can trigger the main gNB (i.e., gNB1) with AI / ML models / functions (e.g., decoders) D1, D2, D3 in its pool to run / initiate (multiple) joint performance (inference) tests of the dual-side models / functions (e.g., encoder-decoder pairs) of the UE-side models / functions and the network-side models / functions, wherein the network-side model / function pool contains the models / functions of the main gNB1 and / or the models / functions existing in the adjacent gNB pools (i.e., gNB2 and gNB3), wherein the model / function pools of gNB2 and gNB3 contain (D4, D5, D6) and (D7, D8, D9) respectively.

[0096] Models / functions or related information (e.g., model / function identifiers (IDs) used to identify models / functions) can be sent from each neighboring gNB (i.e., gNB2 and gNB3) to the primary gNB (i.e., gNB1). For example, gNB2 can send the complete file of the model in its model pool or the ID of the model in its model pool (i.e., D5, D6, D4) to gNB1. For example, gNB3 can send the complete file of the model in its model pool or the ID of the model in its model pool (i.e., D7, D8, D9) to gNB1. gNB1 can then use each model / function of the primary gNB1 and the neighboring gNBs (i.e., gNB2, gNB3) local to the primary gNB1 to infer the output of the model / function on the UE side. The primary gNB1 can run inference locally, and (multiple) corresponding inference / performance reports can be sent back to the neighboring gNBs (i.e., gNB2, gNB3) and / or the UE.

[0097] Now for reference Figures 3B to 3D Let's discuss an example communication process that allows model / function sharing. For example... Figures 3B to 3D The UE shown can be an example implementation of terminal device 110, such as... Figures 3B to 3D The gNB1 shown can be an example implementation of the first network device 120, such as... Figures 3B to 3D The gNB2 and gNB3 shown can be example implementations of network devices 130-1 and 130-2, respectively. gNB1 can be the UE's primary gNB, while gNB2 and gNB3 are neighboring gNBs.

[0098] First refer to Figure 3BIn this scenario, the adjacent gNBs (i.e., gNB2 and gNB3) and the primary gNB (i.e., gNB1) share the same model / function (i.e., decoder) pool. That is, all gNB1, gNB2, and gNB3 have the same decoder model / function pool—D1, D2, and D3—while the UE latched to gNB1 has encoder models / functions E1 and E2. The dual-side models / functions on both the terminal and network sides are associated with CSI compression and CSI decompression that enable AI / ML.

[0099] like Figure 3B As shown, in step 1, to evaluate the interoperability performance of the UE-side model / function (i.e., encoders E1 and E2) with the network-side model / function (i.e., decoders D1, D2, and D3) of the primary gNB1 and / or neighboring gNB2 and gNB3, the UE triggers a model / function verification test. As part of the test, the UE uses its available encoders E1 and / or E2 to compress (multiple) CSI feedbacks and sends the compressed outputs of E1 and / or E2 to gNB1.

[0100] In step 2, gNB1 sets D1 as the active model / function and D2 and D3 as inactive models / functions in its pool for the UE. gNB1 then triggers a MODEL_INFO_REQUEST message to its neighboring gNBs (i.e., both gNB2 and gNB3) to obtain information about the models / functions in their pools.

[0101] In step 3, gNB2 and gNB3 respond with details of the model / feature in their pools using the MODEL_INFO_RESPONSE message given to gNB1.

[0102] In step 4, after gNB1 receives information about the model / function and determines that the model in its pool is the same as the model in gNB1, it performs model / function inference on all / related models / functions in the pool based on the same compressed CSI feedback received from the UE in step 1. This inference strategy... Figure 3C As shown in the image.

[0103] In step 5, when the inference performance results / reports are available, gNB1 uses the MODEL_PERFORMANCE_REPORT message to send the report to gNB2 and gNB3, as well as to the UE, for various purposes, such as AI / ML model switching during handover, fallback to non-ML model during handover, cell selection, etc.

[0104] Now for reference Figure 3DIn this scenario, the model / function (decoder) pools at adjacent gNB2 and gNB3, as well as at the primary gNB1, are different; however, model / function file sharing is allowed between the primary gNB1 and adjacent gNB2 and gNB3. For example, this scenario could focus on deployments with localized models / functions and deployments from gNBs from different vendors. The model / function pools for gNB1, gNB2, and gNB3 are (D1, D2, D3), (D5, D6, D4), and (D7, D8, D9), respectively. UEs locked to gNB1 have encoder models / functions E1 and E2. The dual-side models / functions on both the terminal and network sides are associated with CSI compression and CSI decompression enabling AI / ML.

[0105] Figure 3D Steps 1 to 3 shown are Figure 3B Steps 1 through 3 are the same as shown. For simplicity, details will be omitted.

[0106] In step 4, when gNB1 receives information about the model / function and determines that the model in its pool is different from the models in gNB2 and gNB3, gNB1 requests the model / function file from gNB2 and gNB3 via the MODEL_TRANSFER_REQUEST message.

[0107] In step 5, gNB2 and gNB3 use the MODEL_TRANSFER_RESPONSE message to transmit their model / function to gNB1.

[0108] In step 6, when gNB1 receives the model / function from gNB2 and gNB3, it runs inference on all / related models / functions using the compressed CSI feedback received from the UE in step 1. When the inference performance results / report are available, gNB1 sends the report to gNB2 and gNB3 using the MODEL_PERFORMANCE_REPORT message, and also sends the report to the UE for various purposes, such as AI / ML model switching during handover, fallback to a non-ML model during handover, cell selection, etc. This inference strategy can be similar to... Figure 3C The difference in the inference strategies shown is that the models available at gNB2 and gNB3 are different from the models available at gNB1.

[0109] In some other example embodiments, AI / ML model / feature-related information / file sharing may not be permitted between the first network device 120 and the second network device 130, and in this case, the first network device 120 may be unaware of the models / features of the second network device 130. In other words, in this scenario (i.e., one or more parameters of the second or more machine learning models may not be available at the first network device 120), for example, if the first network device 120 and the second network device 130 are from two or more vendors, AI / ML model / feature-related information / file sharing may not be supported, thus limiting inter-vendor model sharing, for example, due to proprietary and security reasons.

[0110] In an example embodiment where sharing of AI / ML model / feature-related information / files is not permitted, the first network device 120 may query / request the performance (i.e., performance verification / inference) of a second network device 130. For example, the first network device 120 may query / request the joint performance of a first or more machine learning models / features (e.g., encoders) with each of the second or more machine learning models / features (e.g., decoders) in the model / feature pool at the second network device 130. In this case, the first network device 120 may query / request the second network device 130 for a corresponding interoperability check (i.e., joint performance verification) between the first or more machine learning models available on the terminal side and the second or more machine learning models available on the network side.

[0111] In an example embodiment where sharing of AI / ML model / feature-related information / files is not permitted, a first network device 120 may forward compressed input received from a terminal device 110 to a second network device 130 to allow the second network device 130 to determine the interoperability performance of(multiple) terminal-side models / features with(multiple) corresponding network-side models / features. The second network device 130 may determine the performance of a second or more machine learning models based on the output of the first or more machine learning models, and then send inference / test results regarding the performance of the second or more machine learning models to the first network device 120. Therefore, the first network device 120 may collect the inference / test results regarding the performance of the second or more machine learning models and then share them with the terminal device 110. For example, the performance of the second or more machine learning models shared with the terminal device 110 may include the joint performance ranking of the terminal-side and network-side model pairs.

[0112] In some example embodiments, the first network device 120 may send a request to the second network device 130 to evaluate the performance of a second or more machine learning models. This request may include the output of the first or more machine learning models. The second network device 130 may determine the performance of the second or more machine learning models based on the output of the first or more machine learning models, and then send a response to the first network device 120 including a performance evaluation report of the second or more machine learning models.

[0113] In some example implementations, requests to evaluate the performance of a second or more machine learning models can be based on a MODEL_PERFORMANCE_REQUEST(input_payload) message. This message can be intended to be used by the first network device 120 to request the second network device 130 to perform joint inference between the endpoint model / feature and its corresponding AI / ML models / features in its pool. The "input_payload" field can include input received from the endpoint device 110 (e.g., compressed CSI), which will be forwarded to the second network device 130 for inference.

[0114] In some example implementations, the response to a request to evaluate the performance of a second or more machine learning models can be based on a MODEL_PERFORMANCE_RESPONSE (report[]) message. This message can be sent by the second network device 130 to the first network device 120 along with a corresponding inference report as a response to a MODEL_PERFORMANCE_REQUEST message. The “report[]” field can include aggregated reports of various models / features that the second network device 130 tests by running inference on its models / features using the input received along with the MODEL_PERFORMANCE_REQUEST message.

[0115] The messages related to the interaction between the first network device 120 and the second network device 130 may be related to the Xn Application Protocol (XnAP).

[0116] Now for reference Figure 3E Let's discuss example illustrations where model / function sharing is not allowed. For example... Figure 3EAs shown, the UE can be an example implementation of terminal device 110, and gNB1 can be an example implementation of the first network device 120; gNB2 and gNB3 can be example implementations of network devices 130-1 and 130-2, respectively. The UE uses the active model / function E1 (e.g., encoder E1) and can trigger the main gNB (i.e., gNB1) with AI / ML models / functions (e.g., decoders) D1, D2, D3 in its pool to run / initiate (multiple) joint performance (inference) tests of the dual-side models / functions (e.g., encoder-decoder pairs) of the UE-side models / functions and the network-side models / functions, wherein the network-side model / function pool contains the models / functions of the main gNB1 and / or the models / functions existing in the pools of adjacent gNBs (i.e., gNB2 and gNB3), wherein the model / function pools of gNB2 and gNB3 contain (D4, D5, D6) and (D7, D8, D9), respectively.

[0117] like Figure 3E As shown, the primary gNB (i.e., gNB1) can provide compressed output (e.g., compressed CSI) received from the UE to neighboring gNBs (gNB2 and gNB3). Joint inference can be performed locally at the corresponding gNB2 and gNB3 using models / functions from the corresponding model / function pools of the neighboring gNBs (i.e., D4, D5, D6 for gNB2, and D7, D8, D9 for gNB3). Inference / performance reports can be shared with gNB1. After the inference / performance reports are available at the primary gNB1, a merged report can be sent to the UE.

[0118] Figure 3F The diagram illustrates an example communication process where model / function sharing is not allowed. For example... Figure 3F The UE shown can be an example implementation of terminal device 110, such as... Figure 3F The gNB1 shown may be an example implementation of the first network device 120. Figure 3F The gNB2 and gNB3 shown can be example implementations of network devices 130-1 and 130-2, respectively. In this case, gNB1 can be the UE's primary gNB, while gNB2 and gNB3 are neighboring gNBs.

[0119] exist Figure 3FIn this scenario, model / function information sharing may or may not be permitted (i.e., one or more models available at gNB2 and gNB3 can be shared with the primary gNB1, in other words, model identification information), but actual model / function sharing (i.e., sharing of specific model parameters / files) between adjacent gNBs and the primary gNB1 is not permitted. In this case, the model / function pools at adjacent gNB2 and gNB3 are different from the model / function pool at gNB1. Each gNB uses a different decoder model / function pool—gNB1 (D1, D2, D3); gNB2 (D5, D6, D4); gNB3 (D7, D8, D9). UEs locked to gNB1 have encoder models / functions E1 and E2. The dual-side models / functions on both the terminal and network sides are associated with CSI compression and CSI decompression enabling AI / ML.

[0120] Figure 3F Steps 1 and 2 shown are related to Figure 3B Steps 1 and 2 shown are the same. For simplicity, details will be omitted.

[0121] In step 3, both gNB2 and gNB3 respond only with the model / function information in the MODEL_INFO_RESPONSE message, without responding with the actual model / function file, because sharing the model / function is not allowed.

[0122] In step 4, when gNB1 determines that the actual model / function is unavailable in its pool, but only the model / function information is available, it requests gNB2 and gNB3 to run inference on the compressed CSI(s) received in step 1 using the decoder model / function(s) in their respective gNB pools. Additionally, the primary gNB1 forwards the same compressed CSI(s) received from the UE to neighboring gNBs using a MODEL_PERFORMANCE_REQUEST message.

[0123] In step 5, gNB2 and gNB3 use the decoders in their model / function pools to run inference with the input(s), and send performance feedback back to the main gNB1 using the MODEL_PERFORMANCE_RESPONSE message.

[0124] In step 6, when the primary gNB1 receives performance reports from all neighboring gNBs, it prepares performance reports for all models / functions in the pool and sends the reports to the UE for various purposes, such as AI / ML model switching during handover, fallback to a non-ML model during handover, cell selection, etc.

[0125] In some example embodiments, terminal device 110 may perform cell selection or cell reselection based on the performance of a second or more machine learning models received from first network device 120. In this case, terminal device 110 may know the performance of a second or more machine learning models available at second network device 130 for a given channel condition, and then, based on this, terminal device may perform cell selection or cell reselection accordingly.

[0126] In some example embodiments, if terminal device 110 determines that the performance of at least one machine learning model available at at least one network device, as indicated by the second information, is acceptable, it may select or reselect a cell of at least one of the network devices 130-1, 130-2, ..., 130-N for connection based on the performance. Whether the performance of the at least one machine learning model available at at least one network device is acceptable can be determined based on an acceptable level or a threshold. For example, if the joint performance is higher than an acceptable level, the joint performance is acceptable.

[0127] In some example embodiments, the second network device 130 may perform a system information broadcast indicating one or more machine learning models available at the corresponding network device. In this case, a broadcast of a list of models / functions (model / function IDs) available for the cell may be included in the system information broadcast (e.g., System Information Block (SIB4), System Information Block (SIB5), etc.). Therefore, the terminal device 110 may receive the system information broadcast and then select or reselect a cell based at least on the received system information broadcast. For example, if the machine learning models available at the corresponding network devices 130-1, 130-2, ..., 130-N are acceptable and their joint performance is better than that of one or more other network devices relative to the terminal-side models(s), and if the signal strength of the corresponding network device and one or more other network devices is the same, then the terminal device 110 may select or reselect a cell of the corresponding network device.

[0128] In some example embodiments, based on second information obtained regarding the performance of a second or more machine learning models available at the second network device 130, the first network device 120 can determine from the second network device 130 (i.e., from one or more network devices 130-1, 130-2, ..., 130-N) the network device to which the terminal device 110 should be switched (e.g., network device 130-1). The first network device 120 can then perform a handover process from the terminal device 110 to the second network device 130 (i.e., the determined network device, e.g., network device 130-1). Therefore, the terminal device 110 can perform a handover from the first network device 120 to the second network device 130 (i.e., the determined network device, e.g., network device 130-1).

[0129] Figure 4 The illustration shows an example communication process 400. It should be understood that process flow 400 can be viewed as follows: Figure 2 A more specific example of signaling flow 200 is shown. For example... Figure 4 The UE shown can be an example implementation of terminal device 110, such as... Figure 4 The gNB1 shown can be an example implementation of the first network device 120. Figure 4 The gNB2 and gNB3 shown can be example implementations of network devices 130-1 and 130-2, respectively. gNB1 and gNB2 have the same decoder model / function pool—D1, D2, and D3, while gNB3 has a different decoder model / function pool—D7, D8, and D9.

[0130] like Figure 4 As shown, in step 1, there is a trigger caused by any change in the bilateral model, such as a model update at the UE, a model update at (multiple) gNBs, or a RAN change. In step 2, the UE sends a trigger message with (multiple) compressed CSIs to gNB1.

[0131] In step 3, gNB1 sends a MODEL_INFO_REQUEST (model_type = decoder) message to gNB2. In step 4, gNB2 sends a MODEL_INFO_RESPONSE (status = success, model_info[D1, D2, D3]) message to gNB1. In step 5, gNB1 determines that gNB2's model is the same as gNB1's model and has no performance information regarding the current channel conditions. In step 6, gNB1 runs inference on decoder models D1, D2, and D3 and generates a performance report.

[0132] In step 7, gNB1 sends a MODEL_INFO_REQUEST (model_type = decoder) message to gNB3. In step 8, gNB3 sends a MODEL_INFO_RESPONSE (status = success, model_info[D7, D8, D9]) message to gNB1. In step 9, gNB1 determines that gNB3's model is new and has no performance information regarding the current channel conditions. In step 10, gNB1 sends a MODEL_TRANSFER_REQUEST (D7, D8, D9) message to gNB3. In step 11, gNB3 determines which model to transmit to gNB1. In step 12, gNB3 sends a MODEL_TRANSFER_RESPONSE (STATUS, MODELS[D7, D8, D9]) message to gNB1. In step 13, gNB1 runs inference on decoder models D7, D8, and D9 and generates a performance report.

[0133] In step 14, gNB1 sends a MODEL_PERFORMANCE_REPORT (report E1: [D1, D2, D3]) message to gNB2, and in step 15, gNB1 sends a MODEL_PERFORMANCE_REPORT (report E1: [D7, D8, D9]) message to gNB3. In step 16, gNB1 merges the reports from gNB2 and gNB3 and ranks the models based on KPIs. The ranking list is D7, D8, D9, D1, D2, D3.

[0134] In step 17, gNB1 sends a performance report to the UE for various purposes, such as ML model switching during handover, fallback to a non-ML model during handover, and cell selection. In step 18, the UE learns about the models available in neighboring gNBs and their performance for a given channel condition.

[0135] In step 19, during periodic updates or cell reselection processes, the UE reads SIB 4 to obtain neighboring cell information. In step 20, gNB2 performs an SIB broadcast (Physical Cell Identifier (PCI) = 2, Model = D1, D2). In step 21, gNB3 performs an SIB broadcast (PCI = 4, Model = D7, D8).

[0136] Steps 22-23 and 24-27 are optional. In some implementations related to cell (re)selection, steps 22-23 can be performed. In some implementations related to handover, steps 24-27 can be performed.

[0137] In step 22, after reading the SIB, the UE has two cells with similar signal strength, namely, cells PCI:2 and PCI:4. In step 23, since D7 and D8 performed better in step 16, the UE selects cell PCI:4 instead of cell PCI:2.

[0138] In step 24, the handover process is enabled between the UE and gNB1. In step 25, gNB1 must decide which gNB to hand over the UE to. In step 26, since models D7 and D8 performed better in step 16, gNB1 hands the UE over to gNB3 instead of gNB2. In step 27, the handover of the UE to gNB3 is performed.

[0139] It should be understood that the order of the above steps is for illustrative purposes only and does not imply any limitation on the scope of this disclosure, and the above steps can be performed in any order. For example, steps 3 and 7 can be performed in any order or in parallel. For example, steps 14 and 15 can be performed in any order or in parallel.

[0140] As referenced above Figures 2 to 3D The same operations and features are applicable to process 400 and have similar effects. For simplicity, details will be omitted.

[0141] Figure 5 The illustration depicts an example communication process 500. It should be understood that process flow 500 can be viewed as follows: Figure 2 A more specific example of signaling flow 200 is shown. For example... Figure 5 The UE shown can be an example implementation of terminal device 110, such as... Figure 5 The gNB1 shown can be an example implementation of the first network device 120. Figure 5 The gNB2 and gNB3 shown can be example implementations of network devices 130-1 and 130-2, respectively. gNB1 and gNB2 have the same decoder model / function pool—D1, D2, and D3, while gNB3 has a different decoder model / function pool—D7, D8, and D9.

[0142] like Figure 5 As shown, in step 1, there is a trigger caused by any change in the bilateral model, such as a model update at the UE, a model update at (multiple) gNBs, or a RAN change. In step 2, the UE sends a trigger message with (multiple) compressed CSIs to gNB1.

[0143] In step 3, gNB1 runs inference on decoder models D1, D2, and D3 and generates a performance report. In step 4, gNB1 sends a MODEL_INFO_REQUEST (model_type = decoder) message to gNB2. In step 5, gNB2 sends a MODEL_INFO_RESPONSE (state = INFO_ONLY, [D4, D5, D6]) message to gNB1. In step 6, gNB1 determines that gNB2 responds only with model information and infers that model sharing is not allowed.

[0144] In step 7, gNB1 sends a MODEL_PERFORMANCE_REQUEST message ((multiple) decoded CSI payloads, [D4, D5, D6]) to gNB2. In step 8, gNB2 runs inference on decoder models D4, D5, and D6 and generates a performance report. In step 9, gNB2 sends a MODEL_PERFORMANCE_RESPONSE message (report [D4, D5, D6]) to gNB1.

[0145] In step 10, gNB1 sends a MODEL_INFO_REQUEST (model_type = decoder) message to gNB3. In step 11, gNB3 sends a MODEL_INFO_RESPONSE (status = INFO_ONLY, [D7, D8, D9]) message to gNB1. In step 12, gNB1 determines that gNB3 responds only with model information and infers that model sharing is not allowed. In step 13, gNB1 sends a MODEL_PERFORMANCE_REQUEST ((multiple) decoded CSI payloads, [D7, D8, D9]) message to gNB3. In step 14, gNB3 runs inference on decoder models D7, D8, and D9 and generates a performance report. In step 15, gNB3 sends a MODEL_PERFORMANCE_RESPONSE (report [D7, D8, D9]) message to gNB1.

[0146] In step 16, gNB1 merges the reports from gNB1, gNB2, and gNB3, and ranks the models based on KPIs. The ranking list is D7, D8, D9, D4, D5, D6, D3, D2, and D1.

[0147] In step 17, gNB1 sends a performance report to the UE for various purposes, such as ML model switching during handover, fallback to a non-ML model during handover, and cell selection. In step 18, the UE learns about the models available in neighboring gNBs and their performance for a given channel condition.

[0148] In step 19, during periodic updates or cell reselection processes, the UE reads SIB 4 to obtain neighboring cell information. In step 20, gNB2 performs an SIB broadcast (Physical Cell Identifier (PCI) = 2, Model = D4, D5). In step 21, gNB3 performs an SIB broadcast (PCI = 4, Model = D7, D8).

[0149] Steps 22-23 and 24-27 are optional. In some implementations related to cell (re)selection, steps 22-23 can be performed. In some implementations related to handover, steps 24-27 can be performed.

[0150] In step 22, after reading the SIB, the UE has two cells with similar signal strength, namely, cells PCI:2 and PCI:4. In step 23, since D7 and D8 performed better in step 16, the UE selects cell PCI:4 instead of cell PCI:2.

[0151] In step 24, the handover process is enabled between the UE and gNB1. In step 25, gNB1 must decide which gNB to hand over the UE to. In step 26, since models D7 and D8 performed better in step 16, gNB1 hands the UE over to gNB3 instead of gNB2. In step 27, the handover of the UE to gNB3 is performed.

[0152] It should be understood that the order of the above steps is for illustrative purposes only and does not imply any limitation on the scope of this disclosure, and the above steps can be performed in any order. For example, steps 4 and 10 can be performed in any order or in parallel. (See above reference...) Figure 2 , Figure 3E and Figure 3F The same operations and features are applicable to process 500 and have similar effects. For simplicity, details will be omitted.

[0153] It should be understood that although the embodiments of this disclosure are discussed with regard to the CSI compression use case of enabling AI / ML, the embodiments of this disclosure can also be applied to other use cases.

[0154] According to the reference Figures 2 to 5Some embodiments allow for better mobility decisions by the primary network device to switch a terminal device to a neighboring network device without impacting performance or requiring a re-evaluation of the terminal network encoder-decoder pair performance after a cell change. For example, if the primary network device needs to select one of two target network devices to switch a terminal device, it can choose the network device with better performance for a given AI / ML model / feature at the terminal device. Furthermore, if the terminal device's movement / trajectory is directed towards a specific network device, the primary network device can determine the optimal network-side model / feature (decoder) from a pool of target network devices (e.g., a decoder set) to ensure the terminal device is paired with the best decoder after the switch. Additionally, by feeding model / feature performance information back to the terminal device, better mobility decisions can be made in idle-mode cell reselection / handover scenarios. For example, if the terminal device knows the performance of the models / features available for neighboring cells (network devices), it can choose the cell with the better-performing model during idle-mode cell reselection or connected-mode handover when other cell reselection parameters for different neighboring cells are nearly identical. In this way, communication performance and efficiency can be improved.

[0155] Figure 6 A flowchart 600 illustrating a method implemented at a first network device according to some embodiments of the present disclosure is shown. For discussion purposes, reference will be made to... Figure 1A Method 600 is described from the perspective of the first network device 120.

[0156] At box 610, the first network device 120 receives the output of one or more machine learning models available at the terminal device 110 from the terminal device 110.

[0157] At box 620, the first network device 120 receives first information from the second network device 130 about a second or more machine learning models available at the second network device 130.

[0158] At box 630, the first network device 120 obtains second information about the performance of one or more second machine learning models based on the first information. The performance is determined based on the output.

[0159] In some example embodiments, the first network device may further: determine, based on first information, that a second or more machine learning models are the same as one or more machine learning models available at the first network device 120; and determine one or more parameters of the second or more machine learning models based on one or more parameters of one or more machine learning models available at the first network device 120.

[0160] In some example embodiments, the first network device 120 may further: determine, based on first information, that at least one of the second or more machine learning models is different from one or more machine learning models available at the first network device 120; send a request to the second network device 130 for one or more parameters of the at least one machine learning model; and receive a response from the second network device 130 including one or more parameters of the at least one machine learning model.

[0161] In some example embodiments, obtaining second information about the performance of a second or more machine learning models may include: the first network device 120 determining the second information about the performance of the second or more machine learning models based on the output of the first or more machine learning models and one or more parameters of the second or more machine learning models.

[0162] In some example embodiments, the first network device 120 may further: determine, based on the first information, that one or more parameters of the second or more machine learning models are not available at the first network device 120. In some example embodiments, obtaining second information regarding the performance of the second or more machine learning models may include: sending a request to the second network device 130 to evaluate the performance of the second or more machine learning models, wherein the request includes the output of the first or more machine learning models; and receiving a response from the second network device 130 including a performance evaluation report of the second or more machine learning models.

[0163] In some example embodiments, the first network device 120 may also send a performance report of a second or more machine learning models to the second network device 130.

[0164] In some example embodiments, the first network device 120 may also determine the handover from the terminal device 110 to the second network device 130 based on the second information.

[0165] In some example embodiments, the first network device 120 may also send an instruction to the terminal device 110 to send first information based on at least one of the following: a machine learning model update at the first network device 120 or the second network device 130; or determining that the terminal device 110 is moving toward the cell edge.

[0166] In some example embodiments, the first network device 120 may also receive from the terminal device 110 an indication of the reason for sending the first indication, wherein the reason includes at least one of the following: an update of the machine learning model at the terminal device 110; a change in channel conditions associated with the terminal device 110; or the terminal device 110 moving to a new cell.

[0167] In some example embodiments, the first network device 120 may also send a request for the first information to the second network device 130.

[0168] In some example embodiments, the output may correspond to channel state information encoded by a first or more machine learning models.

[0169] In some example embodiments, a second or more machine learning models can be used as decoders for decoding encoded channel state information.

[0170] In some example embodiments, performance may be associated with a performance indicator of the decoding of encoded channel state information.

[0171] Those skilled in the art will understand that the above references Figures 2 to 5 All the operations and features described herein also apply to method 600 and have similar effects.

[0172] Figure 7 A flowchart 700 illustrating a method implemented at a second network device according to some embodiments of the present disclosure is shown. For discussion purposes, reference will be made to... Figure 1A Method 700 is described from the perspective of the second network device 130.

[0173] At box 710, the second network device 130 sends first information to the first network device 120 about a second or more machine learning models available at the second network device 130.

[0174] At box 720, the second network device 130 obtains second information about the performance of a second or more machine learning models, wherein the performance is determined based on the output of a first or more machine learning models available at terminal device 110.

[0175] In some example embodiments, the second network device 130 may also: receive a request from the first network device 120 for one or more parameters of a second or more machine learning models; and send a response to the first network device 120 including one or more parameters of the second or more machine learning models.

[0176] In some example embodiments, obtaining second information about the performance of a second or more machine learning models includes receiving a report about the performance of the second or more machine learning models from a first network device 120.

[0177] In some example embodiments, the second network device 130 may further: receive from the first network device 120 a request to evaluate the performance of a second or more machine learning models, wherein the request includes the output of the first or more machine learning models; and send a response to the first network device 120 including a performance evaluation report of the second or more machine learning models. In some example embodiments, obtaining second information about the performance of the second or more machine learning models may include: determining the performance of the second or more machine learning models by the second network device 130 based on the output of the first or more machine learning models.

[0178] In some example embodiments, the second network device 130 may also receive a request for the first information from the first network device 120.

[0179] In some example embodiments, the output may correspond to channel state information encoded by a first or more machine learning models.

[0180] In some example embodiments, a second or more machine learning models can be used as decoders for decoding encoded channel state information.

[0181] In some example embodiments, performance may be associated with a performance indicator of the decoding of encoded channel state information.

[0182] Those skilled in the art will understand that the above references Figures 2 to 5 All the operations and features described herein also apply to method 700 and have similar effects.

[0183] In some example embodiments, an apparatus capable of performing method 700 (e.g., first network device 120) may include components for performing corresponding steps of method 600. These components may be implemented in any suitable form. For example, the components may be implemented in a circuit system or a software module.

[0184] In some example embodiments, the apparatus includes: components for receiving from a terminal device the output of one or more first machine learning models available at the terminal device; components for receiving from a second network device first information about one or more second machine learning models available at the second network device; and components for obtaining second information about the performance of the second or more machine learning models based on the first information, wherein the performance is determined based on the output.

[0185] In some example embodiments, the apparatus further includes: components for determining, based on first information, that a second or more machine learning model is the same as one or more machine learning models available at the first network device; and components for determining one or more parameters of the second or more machine learning models based on one or more parameters of one or more machine learning models available at the first network device.

[0186] In some example embodiments, the apparatus further includes: components for determining, based on first information, that at least one of the second or more machine learning models is different from one or more machine learning models available at the first network device; components for sending a request to the second network device for one or more parameters of the at least one machine learning model; and components for receiving from the second network device a response including one or more parameters of the at least one machine learning model.

[0187] In some example embodiments, the component for obtaining second information about the performance of a second or more machine learning models may include: a component for determining, by a first network device, the second information about the performance of a second or more machine learning models based on the output of a first or more machine learning models and one or more parameters of the second or more machine learning models.

[0188] In some example embodiments, the apparatus further includes a component for determining one or more parameters of a second or more machine learning models that are not available at the first network device, based on the first information.

[0189] In some example embodiments, the components for obtaining second information about the performance of a second or more machine learning models may include: components for sending a request to a second network device for evaluating the performance of the second or more machine learning models, wherein the request includes the output of a first or more machine learning models; and components for receiving a response from the second network device including a performance evaluation report of the second or more machine learning models.

[0190] In some example embodiments, the apparatus further includes a component for sending a report on the performance of a second or more machine learning models to a second network device.

[0191] In some example embodiments, the apparatus further includes a component for determining a handover from a terminal device to a second network device based on second information.

[0192] In some example embodiments, the apparatus further includes: a component for sending an instruction to a terminal device to send first information based on at least one of the following: a machine learning model update at a first network device or a second network device; or determining that the terminal device is moving toward the cell edge.

[0193] In some example embodiments, the apparatus further includes: a component for receiving an indication of a reason for sending a first indication from a terminal device, wherein the reason includes at least one of the following: an update of a machine learning model at the terminal device; a change in channel conditions associated with the terminal device; or the terminal device moving to a new cell.

[0194] In some example embodiments, the apparatus further includes a component for sending a request for the first information to a second network device.

[0195] In some example embodiments, the output corresponds to channel state information encoded by a first or more machine learning models.

[0196] In some example embodiments, a second or more machine learning models are used as decoders for decoding encoded channel state information.

[0197] In some example implementations, performance is associated with a performance indicator for decoding encoded channel state information.

[0198] In some embodiments, the apparatus further includes components for performing additional steps of some embodiments of method 600. In some embodiments, the components include at least one processor and at least one memory including computer program code, the at least one memory and the computer program code being configured, together with the at least one processor, to achieve the performance of the apparatus.

[0199] In some example embodiments, an apparatus capable of performing method 700 (e.g., second network device 130) may include components for performing the corresponding steps of method 700. These components may be implemented in any suitable form. For example, the components may be implemented in a circuit system or a software module.

[0200] In some example embodiments, the apparatus includes: components for sending first information to a first network device about a second or more machine learning models available at a second network device; and components for obtaining second information about the performance of the second or more machine learning models, wherein the performance is determined based on the output of the first or more machine learning models available at a terminal device.

[0201] In some example embodiments, the apparatus further includes: a component for receiving a request from a first network device for one or more parameters of a second or more machine learning models; and a component for sending a response to the first network device including one or more parameters of the second or more machine learning models.

[0202] In some example embodiments, the component for obtaining second information about the performance of a second or more machine learning models includes: a component for receiving a report from a first network device about the performance of the second or more machine learning models.

[0203] In some example embodiments, the apparatus further includes: components for receiving from a first network device a request to evaluate the performance of a second or more machine learning models, wherein the request includes the output of the first or more machine learning models; and components for sending a response to the first network device including a performance evaluation report of the second or more machine learning models. In some example embodiments, the components for obtaining second information about the performance of the second or more machine learning models include components for determining the performance of the second or more machine learning models by the second network device based on the output of the first or more machine learning models.

[0204] In some example embodiments, the apparatus further includes components for receiving a request for first information from a first network device.

[0205] In some example embodiments, the output corresponds to channel state information encoded by a first or more machine learning models.

[0206] In some example embodiments, a second or more machine learning models are used as decoders for decoding encoded channel state information.

[0207] In some example implementations, performance is associated with a performance indicator for decoding encoded channel state information.

[0208] In some embodiments, the apparatus further includes components for performing additional steps of some embodiments of method 700. In some embodiments, the components include at least one processor and at least one memory including computer program code, the at least one memory and the computer program code being configured, together with the at least one processor, to achieve the performance of the apparatus.

[0209] Figure 8 A simplified block diagram of a device 800 suitable for implementing some example embodiments of the present disclosure is illustrated. The device 800 can be provided to implement a communication device, for example, such as... Figure 1AThe terminal device 110, the first network device 120, or the second network device 130 are shown. As shown, device 800 includes one or more processors 810, one or more memories 820 coupled to processor 810, and one or more communication modules 840 coupled to processor 810.

[0210] Communication module 840 is used for bidirectional communication. Communication module 840 has at least one antenna to facilitate communication. The communication interface can represent any interface required for communication with other network elements.

[0211] Processor 810 can be of any type suitable for a local technology network, and by way of non-limiting example, can include one or more of the following: general-purpose computer, special-purpose computer, microprocessor, digital signal processor (DSP), and processor based on a multi-core processor architecture. Device 800 can have multiple processors, such as application-specific integrated circuit chips that are time-dependent on a clock synchronized with the main processor.

[0212] Memory 820 may include one or more non-volatile memories and one or more volatile memories. Examples of non-volatile memories include, but are not limited to, read-only memory (ROM) 824, electrically programmable read-only memory (EPROM), flash memory, hard disk, compact disc (CD), digital video disk (DVD), and other magnetic and / or optical storage. Examples of volatile memories include, but are not limited to, random access memory (RAM) 822 and other volatile memories that do not persist during power outages.

[0213] Computer program 830 includes computer-executable instructions that are executed by the associated processor 810. Program 830 may be stored in ROM 824. Processor 810 may perform any suitable actions and processes by loading program 830 into RAM 822.

[0214] The embodiments of this disclosure can be implemented via program 830, enabling device 800 to execute reference... Figures 2 to 5 Any process discussed in this disclosure. Embodiments of this disclosure may also be implemented by hardware or by a combination of software and hardware.

[0215] In some example embodiments, program 830 may be tangibly contained in a computer-readable medium, which may be included in device 800 (such as memory 820) or other storage device accessible to device 800. Device 800 may load program 830 from the computer-readable medium into RAM 822 for execution. The computer-readable medium may include any type of tangible non-volatile memory, such as ROM, EPROM, flash memory, hard disk, CD, DVD, etc.

[0216] Figure 9 A block diagram illustrating an example of a computer-readable medium 900 according to some exemplary embodiments of the present disclosure is shown. A program 830 is stored on the computer-readable medium 900. It should be noted that although the computer-readable medium 900... Figure 9 The program is depicted in the form of a CD or DVD, but the computer-readable medium 900 may be any other form suitable for carrying or storing the program 830.

[0217] Generally, the various embodiments of this disclosure can be implemented using hardware or dedicated circuitry, software, logic, or any combination thereof. Some aspects can be implemented using hardware, while others can be implemented using firmware or software that can be executed by a controller, microprocessor, or other computing device. Although various aspects of the embodiments of this disclosure are illustrated and described as block diagrams, flowcharts, or using some other graphical representation, it should be understood that, as non-limiting examples, the blocks, apparatuses, systems, techniques, or methods described herein can be implemented using hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0218] This disclosure also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as instructions included in a program module, which execute in a device on a target real or virtual processor to perform the above-mentioned... Figure 6 and Figure 7 Any of the methods described herein. Typically, a program module includes routines, programs, libraries, objects, classes, components, data structures, etc., that perform a specific task or implement a specific abstract data type. In various embodiments, the functionality of a program module can be combined or split among program modules as needed. The machine-executable instructions of a program module can execute on a local or distributed device. In a distributed device, a program module can reside on both local and remote storage media.

[0219] Program code used to perform the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a stand-alone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0220] In the context of this disclosure, computer program code or related data may be carried by any suitable carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, etc.

[0221] Computer-readable media can be computer-readable signal media or computer-readable storage media. Computer-readable media can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any suitable combination of the foregoing. More specific examples of computer-readable storage media will include electrical connections having one or more wires, portable computer floppy disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable optical disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing. The term "non-transient" as used herein is a limitation on the medium itself (i.e., tangible, not signaling), not a limitation on the persistence of data storage (e.g., RAM and ROM).

[0222] Furthermore, although operations are described in a specific order, this should not be construed as requiring the operations to be performed in the specific order shown or sequentially, or to perform all of the shown operations to obtain the desired result. In some cases, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the foregoing discussion, these should not be construed as limiting the scope of this disclosure, but rather as descriptions of features that may be specific to particular embodiments. Certain features described in the context of a single embodiment may also be implemented in combination in a single embodiment. Conversely, the various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0223] Although this disclosure is described in language specific to structural features and / or methodological actions, it should be understood that the disclosure as defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are disclosed as exemplary forms of implementing the claims.

Claims

1. A first network device, comprising: At least one processor; as well as At least one memory storing instructions, which, when executed by the at least one processor, cause the first network device to at least: Receive the output of a first or more machine learning models available at the terminal device; Receive first information about a second or more machine learning models available at the second network device; as well as Based on the first information, second information is obtained regarding the performance of the second or more machine learning models, wherein the performance is determined based on the output.

2. The first network device of claim 1, wherein the first network device is further configured to perform at least one of the following: Based on the first information, it is determined that the second or more machine learning models are the same as one or more machine learning models available at the first network device; as well as Based on one or more parameters of the one or more machine learning models available at the first network device, determine one or more parameters of the second or more machine learning models; Based on the first information, it is determined that at least one of the second or more machine learning models is different from one or more machine learning models available at the first network device; Send a request to the second network device for one or more parameters of the at least one machine learning model; as well as Receive a response from the second network device including one or more parameters of the at least one machine learning model; Send a report on the performance of the second or more machine learning models to the second network device; Based on the first information, it is determined that one or more parameters of the second or more machine learning models are not available at the first network device; The second information regarding the performance of the second or more machine learning models includes: Send a request to the second network device for evaluating the performance of the second or more machine learning models, wherein the request includes the output of the first or more machine learning models; as well as Receive a response from the second network device that includes a performance evaluation report of the second or more machine learning models.

3. The first network device according to claim 2, wherein obtaining the second information regarding the performance of the second or more machine learning models comprises: The first network device determines the second information regarding the performance of the second or more machine learning models based on the output of the first or more machine learning models and the one or more parameters of the second or more machine learning models.

4. The first network device according to any one of claims 1 to 3, wherein the first network device is further configured to perform at least one of the following: Based on the second information, the handover from the terminal device to the second network device is determined; Based on at least one of the following, an instruction is sent to the terminal device instructing the terminal device to send the first information: a machine learning model update at the first network device or the second network device; or it is determined that the terminal device is moving toward the cell edge; The terminal device receives an indication of the reason for sending the first indication, wherein the reason includes at least one of the following: an update of the machine learning model at the terminal device; a change in channel conditions associated with the terminal device; or the terminal device moves to a new cell; or Send a request for the first information to the second network device.

5. The first network device according to any one of claims 1 to 3, wherein at least one of the following: The output corresponds to channel state information encoded by the first or more machine learning models; The second or more machine learning models are used as decoders for decoding encoded channel state information; or The performance is associated with a performance indicator for decoding the encoded channel state information.

6. A second network device, comprising: At least one processor; as well as At least one memory storing instructions, which, when executed by the at least one processor, cause the second network device to at least: Send first information to the first network device about a second or more machine learning models available at the second network device; as well as Obtain second information regarding the performance of the second or more machine learning models, wherein the performance is determined based on the output of the first or more machine learning models available at the terminal device.

7. The second network device of claim 6, wherein the second network device is further configured to perform at least one of the following: Receive a request from the first network device for one or more parameters of the second or more machine learning models; and Send a response to the first network device including the one or more parameters of the second or more machine learning models; Receive a request from the first network device for evaluating the performance of the second or more machine learning models, wherein the request includes the output of the first or more machine learning models; as well as Send a response to the first network device, including a performance evaluation report of the second or more machine learning models; The second information regarding the performance of the second or more machine learning models includes: The performance of the second or more machine learning models is determined by the second network device based on the output of the first or more machine learning models; or Receive a request for the first information from the first network device.

8. The second network device according to claim 6 or 7, wherein obtaining the second information regarding the performance of the second or more machine learning models comprises: Receive a report on the performance of the second or more machine learning models from the first network device.

9. A method for communication, comprising: At the first network device, the output of a first or more machine learning models available at the terminal device is received from the terminal device; At the first network device, first information about a second or more machine learning models available at the second network device is received from the second network device; as well as At the first network device, based on the first information, second information is obtained regarding the performance of the second or more machine learning models, wherein the performance is determined based on the output.

10. A method for communication, comprising: At the second network device, first information about a second or more machine learning models available at the second network device is sent to the first network device; as well as At the second network device, second information regarding the performance of the second or more machine learning models is obtained, wherein the performance is determined based on the output of the first or more machine learning models available at the terminal device.

Images